aboutsummaryrefslogtreecommitdiffstats
diff options
context:
space:
mode:
Diffstat
-rw-r--r--arch/x86/Kconfig.cpu54
-rw-r--r--arch/x86/boot/cpu.c17
-rw-r--r--arch/x86/boot/mkcpustr.c40
-rw-r--r--arch/x86/es7000/Makefile5
-rw-r--r--arch/x86/es7000/es7000.h114
-rw-r--r--arch/x86/ia32/ia32_signal.c51
-rw-r--r--arch/x86/kernel/Makefile3
-rw-r--r--arch/x86/kernel/cpu/Makefile34
-rw-r--r--arch/x86/kernel/cpu/addon_cpuid_features.c88
-rw-r--r--arch/x86/kernel/cpu/amd.c548
-rw-r--r--arch/x86/kernel/cpu/amd_64.c224
-rw-r--r--arch/x86/kernel/cpu/centaur.c4
-rw-r--r--arch/x86/kernel/cpu/centaur_64.c6
-rw-r--r--arch/x86/kernel/cpu/cmpxchg.c72
-rw-r--r--arch/x86/kernel/cpu/common.c973
-rw-r--r--arch/x86/kernel/cpu/common_64.c765
-rw-r--r--arch/x86/kernel/cpu/cpu.h19
-rw-r--r--arch/x86/kernel/cpu/cyrix.c23
-rw-r--r--arch/x86/kernel/cpu/feature_names.c84
-rw-r--r--arch/x86/kernel/cpu/intel.c364
-rw-r--r--arch/x86/kernel/cpu/intel_64.c95
-rw-r--r--arch/x86/kernel/cpu/intel_cacheinfo.c169
-rw-r--r--arch/x86/kernel/cpu/mcheck/mce_64.c2
-rw-r--r--arch/x86/kernel/cpu/mkcapflags.pl32
-rw-r--r--arch/x86/kernel/cpu/powerflags.c20
-rw-r--r--arch/x86/kernel/cpu/transmeta.c32
-rw-r--r--arch/x86/kernel/cpu/umc.c3
-rw-r--r--arch/x86/kernel/e820.c28
-rw-r--r--arch/x86/kernel/es7000_32.c (renamed from arch/x86/es7000/es7000plat.c)85
-rw-r--r--arch/x86/kernel/genx2apic_cluster.c7
-rw-r--r--arch/x86/kernel/genx2apic_phys.c7
-rw-r--r--arch/x86/kernel/i387.c154
-rw-r--r--arch/x86/kernel/process.c1
-rw-r--r--arch/x86/kernel/sigframe.h14
-rw-r--r--arch/x86/kernel/signal_32.c45
-rw-r--r--arch/x86/kernel/signal_64.c95
-rw-r--r--arch/x86/kernel/traps_32.c1
-rw-r--r--arch/x86/kernel/traps_64.c6
-rw-r--r--arch/x86/kernel/vmlinux_32.lds.S9
-rw-r--r--arch/x86/kernel/vmlinux_64.lds.S9
-rw-r--r--arch/x86/kernel/xsave.c316
-rw-r--r--arch/x86/kvm/vmx.h15
-rw-r--r--arch/x86/lib/Makefile3
-rw-r--r--arch/x86/lib/usercopy_32.c7
-rw-r--r--arch/x86/mach-generic/Makefile1
-rw-r--r--arch/x86/mm/init_64.c56
-rw-r--r--arch/x86/pci/i386.c3
-rw-r--r--arch/x86/pci/mmconfig-shared.c12
-rw-r--r--arch/x86/power/cpu_32.c7
-rw-r--r--arch/x86/power/cpu_64.c7
-rw-r--r--include/asm-generic/vmlinux.lds.h1
-rw-r--r--include/asm-x86/bugs.h5
-rw-r--r--include/asm-x86/cpufeature.h118
-rw-r--r--include/asm-x86/e820.h2
-rw-r--r--include/asm-x86/i387.h84
-rw-r--r--include/asm-x86/msr-index.h16
-rw-r--r--include/asm-x86/processor-cyrix.h8
-rw-r--r--include/asm-x86/processor-flags.h1
-rw-r--r--include/asm-x86/processor.h27
-rw-r--r--include/asm-x86/sigcontext.h87
-rw-r--r--include/asm-x86/sigcontext32.h6
-rw-r--r--include/asm-x86/thread_info.h1
-rw-r--r--include/asm-x86/ucontext.h6
-rw-r--r--include/asm-x86/xcr.h49
-rw-r--r--include/asm-x86/xsave.h118
-rw-r--r--include/linux/ioport.h3
-rw-r--r--include/linux/percpu.h7
-rw-r--r--kernel/resource.c68
68 files changed, 3019 insertions, 2317 deletions
diff --git a/arch/x86/Kconfig.cpu b/arch/x86/Kconfig.cpu
index 60a85768cfcb..f8843c3ae77d 100644
--- a/arch/x86/Kconfig.cpu
+++ b/arch/x86/Kconfig.cpu
@@ -419,6 +419,60 @@ config X86_DEBUGCTLMSR
419 def_bool y 419 def_bool y
420 depends on !(MK6 || MWINCHIPC6 || MWINCHIP2 || MWINCHIP3D || MCYRIXIII || M586MMX || M586TSC || M586 || M486 || M386) 420 depends on !(MK6 || MWINCHIPC6 || MWINCHIP2 || MWINCHIP3D || MCYRIXIII || M586MMX || M586TSC || M586 || M486 || M386)
421 421
422menuconfig PROCESSOR_SELECT
423 default y
424 bool "Supported processor vendors" if EMBEDDED
425 help
426 This lets you choose what x86 vendor support code your kernel
427 will include.
428
429config CPU_SUP_INTEL
430 default y
431 bool "Support Intel processors" if PROCESSOR_SELECT
432 help
433 This enables extended support for Intel processors
434
435config CPU_SUP_CYRIX_32
436 default y
437 bool "Support Cyrix processors" if PROCESSOR_SELECT
438 depends on !64BIT
439 help
440 This enables extended support for Cyrix processors
441
442config CPU_SUP_AMD
443 default y
444 bool "Support AMD processors" if PROCESSOR_SELECT
445 help
446 This enables extended support for AMD processors
447
448config CPU_SUP_CENTAUR_32
449 default y
450 bool "Support Centaur processors" if PROCESSOR_SELECT
451 depends on !64BIT
452 help
453 This enables extended support for Centaur processors
454
455config CPU_SUP_CENTAUR_64
456 default y
457 bool "Support Centaur processors" if PROCESSOR_SELECT
458 depends on 64BIT
459 help
460 This enables extended support for Centaur processors
461
462config CPU_SUP_TRANSMETA_32
463 default y
464 bool "Support Transmeta processors" if PROCESSOR_SELECT
465 depends on !64BIT
466 help
467 This enables extended support for Transmeta processors
468
469config CPU_SUP_UMC_32
470 default y
471 bool "Support UMC processors" if PROCESSOR_SELECT
472 depends on !64BIT
473 help
474 This enables extended support for UMC processors
475
422config X86_DS 476config X86_DS
423 bool "Debug Store support" 477 bool "Debug Store support"
424 default y 478 default y
diff --git a/arch/x86/boot/cpu.c b/arch/x86/boot/cpu.c
index 75298fe2edca..6ec6bb6e9957 100644
--- a/arch/x86/boot/cpu.c
+++ b/arch/x86/boot/cpu.c
@@ -59,17 +59,18 @@ int validate_cpu(void)
59 u32 e = err_flags[i]; 59 u32 e = err_flags[i];
60 60
61 for (j = 0; j < 32; j++) { 61 for (j = 0; j < 32; j++) {
62 int n = (i << 5)+j; 62 if (msg_strs[0] < i ||
63 if (*msg_strs < n) { 63 (msg_strs[0] == i && msg_strs[1] < j)) {
64 /* Skip to the next string */ 64 /* Skip to the next string */
65 do { 65 msg_strs += 2;
66 msg_strs++; 66 while (*msg_strs++)
67 } while (*msg_strs); 67 ;
68 msg_strs++;
69 } 68 }
70 if (e & 1) { 69 if (e & 1) {
71 if (*msg_strs == n && msg_strs[1]) 70 if (msg_strs[0] == i &&
72 printf("%s ", msg_strs+1); 71 msg_strs[1] == j &&
72 msg_strs[2])
73 printf("%s ", msg_strs+2);
73 else 74 else
74 printf("%d:%d ", i, j); 75 printf("%d:%d ", i, j);
75 } 76 }
diff --git a/arch/x86/boot/mkcpustr.c b/arch/x86/boot/mkcpustr.c
index bbe76953bae9..8ef60f20b371 100644
--- a/arch/x86/boot/mkcpustr.c
+++ b/arch/x86/boot/mkcpustr.c
@@ -15,33 +15,33 @@
15 15
16#include <stdio.h> 16#include <stdio.h>
17 17
18#include "../kernel/cpu/feature_names.c" 18#include "../kernel/cpu/capflags.c"
19
20#if NCAPFLAGS > 8
21# error "Need to adjust the boot code handling of CPUID strings"
22#endif
23 19
24int main(void) 20int main(void)
25{ 21{
26 int i; 22 int i, j;
27 const char *str; 23 const char *str;
28 24
29 printf("static const char x86_cap_strs[] = \n"); 25 printf("static const char x86_cap_strs[] = \n");
30 26
31 for (i = 0; i < NCAPINTS*32; i++) { 27 for (i = 0; i < NCAPINTS; i++) {
32 str = x86_cap_flags[i]; 28 for (j = 0; j < 32; j++) {
33 29 str = x86_cap_flags[i*32+j];
34 if (i == NCAPINTS*32-1) { 30
35 /* The last entry must be unconditional; this 31 if (i == NCAPINTS-1 && j == 31) {
36 also consumes the compiler-added null character */ 32 /* The last entry must be unconditional; this
37 if (!str) 33 also consumes the compiler-added null
38 str = ""; 34 character */
39 printf("\t\"\\x%02x\"\"%s\"\n", i, str); 35 if (!str)
40 } else if (str) { 36 str = "";
41 printf("#if REQUIRED_MASK%d & (1 << %d)\n" 37 printf("\t\"\\x%02x\\x%02x\"\"%s\"\n",
42 "\t\"\\x%02x\"\"%s\\0\"\n" 38 i, j, str);
43 "#endif\n", 39 } else if (str) {
44 i >> 5, i & 31, i, str); 40 printf("#if REQUIRED_MASK%d & (1 << %d)\n"
41 "\t\"\\x%02x\\x%02x\"\"%s\\0\"\n"
42 "#endif\n",
43 i, j, i, j, str);
44 }
45 } 45 }
46 } 46 }
47 printf("\t;\n"); 47 printf("\t;\n");
diff --git a/arch/x86/es7000/Makefile b/arch/x86/es7000/Makefile
deleted file mode 100644
index 3ef8b43b62fc..000000000000
--- a/arch/x86/es7000/Makefile
+++ /dev/null
@@ -1,5 +0,0 @@
1#
2# Makefile for the linux kernel.
3#
4
5obj-$(CONFIG_X86_ES7000) := es7000plat.o
diff --git a/arch/x86/es7000/es7000.h b/arch/x86/es7000/es7000.h
deleted file mode 100644
index 4e62f6fa95b8..000000000000
--- a/arch/x86/es7000/es7000.h
+++ /dev/null
@@ -1,114 +0,0 @@
1/*
2 * Written by: Garry Forsgren, Unisys Corporation
3 * Natalie Protasevich, Unisys Corporation
4 * This file contains the code to configure and interface
5 * with Unisys ES7000 series hardware system manager.
6 *
7 * Copyright (c) 2003 Unisys Corporation. All Rights Reserved.
8 *
9 * This program is free software; you can redistribute it and/or modify it
10 * under the terms of version 2 of the GNU General Public License as
11 * published by the Free Software Foundation.
12 *
13 * This program is distributed in the hope that it would be useful, but
14 * WITHOUT ANY WARRANTY; without even the implied warranty of
15 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.
16 *
17 * You should have received a copy of the GNU General Public License along
18 * with this program; if not, write the Free Software Foundation, Inc., 59
19 * Temple Place - Suite 330, Boston MA 02111-1307, USA.
20 *
21 * Contact information: Unisys Corporation, Township Line & Union Meeting
22 * Roads-A, Unisys Way, Blue Bell, Pennsylvania, 19424, or:
23 *
24 * http://www.unisys.com
25 */
26
27/*
28 * ES7000 chipsets
29 */
30
31#define NON_UNISYS 0
32#define ES7000_CLASSIC 1
33#define ES7000_ZORRO 2
34
35
36#define MIP_REG 1
37#define MIP_PSAI_REG 4
38
39#define MIP_BUSY 1
40#define MIP_SPIN 0xf0000
41#define MIP_VALID 0x0100000000000000ULL
42#define MIP_PORT(VALUE) ((VALUE >> 32) & 0xffff)
43
44#define MIP_RD_LO(VALUE) (VALUE & 0xffffffff)
45
46struct mip_reg_info {
47 unsigned long long mip_info;
48 unsigned long long delivery_info;
49 unsigned long long host_reg;
50 unsigned long long mip_reg;
51};
52
53struct part_info {
54 unsigned char type;
55 unsigned char length;
56 unsigned char part_id;
57 unsigned char apic_mode;
58 unsigned long snum;
59 char ptype[16];
60 char sname[64];
61 char pname[64];
62};
63
64struct psai {
65 unsigned long long entry_type;
66 unsigned long long addr;
67 unsigned long long bep_addr;
68};
69
70struct es7000_mem_info {
71 unsigned char type;
72 unsigned char length;
73 unsigned char resv[6];
74 unsigned long long start;
75 unsigned long long size;
76};
77
78struct es7000_oem_table {
79 unsigned long long hdr;
80 struct mip_reg_info mip;
81 struct part_info pif;
82 struct es7000_mem_info shm;
83 struct psai psai;
84};
85
86#ifdef CONFIG_ACPI
87
88struct oem_table {
89 struct acpi_table_header Header;
90 u32 OEMTableAddr;
91 u32 OEMTableSize;
92};
93
94extern int find_unisys_acpi_oem_table(unsigned long *oem_addr);
95#endif
96
97struct mip_reg {
98 unsigned long long off_0;
99 unsigned long long off_8;
100 unsigned long long off_10;
101 unsigned long long off_18;
102 unsigned long long off_20;
103 unsigned long long off_28;
104 unsigned long long off_30;
105 unsigned long long off_38;
106};
107
108#define MIP_SW_APIC 0x1020b
109#define MIP_FUNC(VALUE) (VALUE & 0xff)
110
111extern int parse_unisys_oem (char *oemptr);
112extern void setup_unisys(void);
113extern int es7000_start_cpu(int cpu, unsigned long eip);
114extern void es7000_sw_apic(void);
diff --git a/arch/x86/ia32/ia32_signal.c b/arch/x86/ia32/ia32_signal.c
index f1a2ac777faf..8d64c1bc8474 100644
--- a/arch/x86/ia32/ia32_signal.c
+++ b/arch/x86/ia32/ia32_signal.c
@@ -179,9 +179,10 @@ struct sigframe
179 u32 pretcode; 179 u32 pretcode;
180 int sig; 180 int sig;
181 struct sigcontext_ia32 sc; 181 struct sigcontext_ia32 sc;
182 struct _fpstate_ia32 fpstate; 182 struct _fpstate_ia32 fpstate_unused; /* look at kernel/sigframe.h */
183 unsigned int extramask[_COMPAT_NSIG_WORDS-1]; 183 unsigned int extramask[_COMPAT_NSIG_WORDS-1];
184 char retcode[8]; 184 char retcode[8];
185 /* fp state follows here */
185}; 186};
186 187
187struct rt_sigframe 188struct rt_sigframe
@@ -192,8 +193,8 @@ struct rt_sigframe
192 u32 puc; 193 u32 puc;
193 compat_siginfo_t info; 194 compat_siginfo_t info;
194 struct ucontext_ia32 uc; 195 struct ucontext_ia32 uc;
195 struct _fpstate_ia32 fpstate;
196 char retcode[8]; 196 char retcode[8];
197 /* fp state follows here */
197}; 198};
198 199
199#define COPY(x) { \ 200#define COPY(x) { \
@@ -215,7 +216,7 @@ static int ia32_restore_sigcontext(struct pt_regs *regs,
215 unsigned int *peax) 216 unsigned int *peax)
216{ 217{
217 unsigned int tmpflags, gs, oldgs, err = 0; 218 unsigned int tmpflags, gs, oldgs, err = 0;
218 struct _fpstate_ia32 __user *buf; 219 void __user *buf;
219 u32 tmp; 220 u32 tmp;
220 221
221 /* Always make any pending restarted system calls return -EINTR */ 222 /* Always make any pending restarted system calls return -EINTR */
@@ -259,26 +260,12 @@ static int ia32_restore_sigcontext(struct pt_regs *regs,
259 260
260 err |= __get_user(tmp, &sc->fpstate); 261 err |= __get_user(tmp, &sc->fpstate);
261 buf = compat_ptr(tmp); 262 buf = compat_ptr(tmp);
262 if (buf) { 263 err |= restore_i387_xstate_ia32(buf);
263 if (!access_ok(VERIFY_READ, buf, sizeof(*buf)))
264 goto badframe;
265 err |= restore_i387_ia32(buf);
266 } else {
267 struct task_struct *me = current;
268
269 if (used_math()) {
270 clear_fpu(me);
271 clear_used_math();
272 }
273 }
274 264
275 err |= __get_user(tmp, &sc->ax); 265 err |= __get_user(tmp, &sc->ax);
276 *peax = tmp; 266 *peax = tmp;
277 267
278 return err; 268 return err;
279
280badframe:
281 return 1;
282} 269}
283 270
284asmlinkage long sys32_sigreturn(struct pt_regs *regs) 271asmlinkage long sys32_sigreturn(struct pt_regs *regs)
@@ -350,7 +337,7 @@ badframe:
350 */ 337 */
351 338
352static int ia32_setup_sigcontext(struct sigcontext_ia32 __user *sc, 339static int ia32_setup_sigcontext(struct sigcontext_ia32 __user *sc,
353 struct _fpstate_ia32 __user *fpstate, 340 void __user *fpstate,
354 struct pt_regs *regs, unsigned int mask) 341 struct pt_regs *regs, unsigned int mask)
355{ 342{
356 int tmp, err = 0; 343 int tmp, err = 0;
@@ -380,7 +367,7 @@ static int ia32_setup_sigcontext(struct sigcontext_ia32 __user *sc,
380 err |= __put_user((u32)regs->flags, &sc->flags); 367 err |= __put_user((u32)regs->flags, &sc->flags);
381 err |= __put_user((u32)regs->sp, &sc->sp_at_signal); 368 err |= __put_user((u32)regs->sp, &sc->sp_at_signal);
382 369
383 tmp = save_i387_ia32(fpstate); 370 tmp = save_i387_xstate_ia32(fpstate);
384 if (tmp < 0) 371 if (tmp < 0)
385 err = -EFAULT; 372 err = -EFAULT;
386 else { 373 else {
@@ -401,7 +388,8 @@ static int ia32_setup_sigcontext(struct sigcontext_ia32 __user *sc,
401 * Determine which stack to use.. 388 * Determine which stack to use..
402 */ 389 */
403static void __user *get_sigframe(struct k_sigaction *ka, struct pt_regs *regs, 390static void __user *get_sigframe(struct k_sigaction *ka, struct pt_regs *regs,
404 size_t frame_size) 391 size_t frame_size,
392 void **fpstate)
405{ 393{
406 unsigned long sp; 394 unsigned long sp;
407 395
@@ -420,6 +408,11 @@ static void __user *get_sigframe(struct k_sigaction *ka, struct pt_regs *regs,
420 ka->sa.sa_restorer) 408 ka->sa.sa_restorer)
421 sp = (unsigned long) ka->sa.sa_restorer; 409 sp = (unsigned long) ka->sa.sa_restorer;
422 410
411 if (used_math()) {
412 sp = sp - sig_xstate_ia32_size;
413 *fpstate = (struct _fpstate_ia32 *) sp;
414 }
415
423 sp -= frame_size; 416 sp -= frame_size;
424 /* Align the stack pointer according to the i386 ABI, 417 /* Align the stack pointer according to the i386 ABI,
425 * i.e. so that on function entry ((sp + 4) & 15) == 0. */ 418 * i.e. so that on function entry ((sp + 4) & 15) == 0. */
@@ -433,6 +426,7 @@ int ia32_setup_frame(int sig, struct k_sigaction *ka,
433 struct sigframe __user *frame; 426 struct sigframe __user *frame;
434 void __user *restorer; 427 void __user *restorer;
435 int err = 0; 428 int err = 0;
429 void __user *fpstate = NULL;
436 430
437 /* copy_to_user optimizes that into a single 8 byte store */ 431 /* copy_to_user optimizes that into a single 8 byte store */
438 static const struct { 432 static const struct {
@@ -447,7 +441,7 @@ int ia32_setup_frame(int sig, struct k_sigaction *ka,
447 0, 441 0,
448 }; 442 };
449 443
450 frame = get_sigframe(ka, regs, sizeof(*frame)); 444 frame = get_sigframe(ka, regs, sizeof(*frame), &fpstate);
451 445
452 if (!access_ok(VERIFY_WRITE, frame, sizeof(*frame))) 446 if (!access_ok(VERIFY_WRITE, frame, sizeof(*frame)))
453 goto give_sigsegv; 447 goto give_sigsegv;
@@ -456,8 +450,7 @@ int ia32_setup_frame(int sig, struct k_sigaction *ka,
456 if (err) 450 if (err)
457 goto give_sigsegv; 451 goto give_sigsegv;
458 452
459 err |= ia32_setup_sigcontext(&frame->sc, &frame->fpstate, regs, 453 err |= ia32_setup_sigcontext(&frame->sc, fpstate, regs, set->sig[0]);
460 set->sig[0]);
461 if (err) 454 if (err)
462 goto give_sigsegv; 455 goto give_sigsegv;
463 456
@@ -521,6 +514,7 @@ int ia32_setup_rt_frame(int sig, struct k_sigaction *ka, siginfo_t *info,
521 struct rt_sigframe __user *frame; 514 struct rt_sigframe __user *frame;
522 void __user *restorer; 515 void __user *restorer;
523 int err = 0; 516 int err = 0;
517 void __user *fpstate = NULL;
524 518
525 /* __copy_to_user optimizes that into a single 8 byte store */ 519 /* __copy_to_user optimizes that into a single 8 byte store */
526 static const struct { 520 static const struct {
@@ -536,7 +530,7 @@ int ia32_setup_rt_frame(int sig, struct k_sigaction *ka, siginfo_t *info,
536 0, 530 0,
537 }; 531 };
538 532
539 frame = get_sigframe(ka, regs, sizeof(*frame)); 533 frame = get_sigframe(ka, regs, sizeof(*frame), &fpstate);
540 534
541 if (!access_ok(VERIFY_WRITE, frame, sizeof(*frame))) 535 if (!access_ok(VERIFY_WRITE, frame, sizeof(*frame)))
542 goto give_sigsegv; 536 goto give_sigsegv;
@@ -549,13 +543,16 @@ int ia32_setup_rt_frame(int sig, struct k_sigaction *ka, siginfo_t *info,
549 goto give_sigsegv; 543 goto give_sigsegv;
550 544
551 /* Create the ucontext. */ 545 /* Create the ucontext. */
552 err |= __put_user(0, &frame->uc.uc_flags); 546 if (cpu_has_xsave)
547 err |= __put_user(UC_FP_XSTATE, &frame->uc.uc_flags);
548 else
549 err |= __put_user(0, &frame->uc.uc_flags);
553 err |= __put_user(0, &frame->uc.uc_link); 550 err |= __put_user(0, &frame->uc.uc_link);
554 err |= __put_user(current->sas_ss_sp, &frame->uc.uc_stack.ss_sp); 551 err |= __put_user(current->sas_ss_sp, &frame->uc.uc_stack.ss_sp);
555 err |= __put_user(sas_ss_flags(regs->sp), 552 err |= __put_user(sas_ss_flags(regs->sp),
556 &frame->uc.uc_stack.ss_flags); 553 &frame->uc.uc_stack.ss_flags);
557 err |= __put_user(current->sas_ss_size, &frame->uc.uc_stack.ss_size); 554 err |= __put_user(current->sas_ss_size, &frame->uc.uc_stack.ss_size);
558 err |= ia32_setup_sigcontext(&frame->uc.uc_mcontext, &frame->fpstate, 555 err |= ia32_setup_sigcontext(&frame->uc.uc_mcontext, fpstate,
559 regs, set->sig[0]); 556 regs, set->sig[0]);
560 err |= __copy_to_user(&frame->uc.uc_sigmask, set, sizeof(*set)); 557 err |= __copy_to_user(&frame->uc.uc_sigmask, set, sizeof(*set));
561 if (err) 558 if (err)
diff --git a/arch/x86/kernel/Makefile b/arch/x86/kernel/Makefile
index a07ec14f3312..c9be69fedb70 100644
--- a/arch/x86/kernel/Makefile
+++ b/arch/x86/kernel/Makefile
@@ -38,7 +38,7 @@ obj-y += tsc.o io_delay.o rtc.o
38 38
39obj-$(CONFIG_X86_TRAMPOLINE) += trampoline.o 39obj-$(CONFIG_X86_TRAMPOLINE) += trampoline.o
40obj-y += process.o 40obj-y += process.o
41obj-y += i387.o 41obj-y += i387.o xsave.o
42obj-y += ptrace.o 42obj-y += ptrace.o
43obj-y += ds.o 43obj-y += ds.o
44obj-$(CONFIG_X86_32) += tls.o 44obj-$(CONFIG_X86_32) += tls.o
@@ -69,6 +69,7 @@ obj-$(CONFIG_KEXEC) += machine_kexec_$(BITS).o
69obj-$(CONFIG_KEXEC) += relocate_kernel_$(BITS).o crash.o 69obj-$(CONFIG_KEXEC) += relocate_kernel_$(BITS).o crash.o
70obj-$(CONFIG_CRASH_DUMP) += crash_dump_$(BITS).o 70obj-$(CONFIG_CRASH_DUMP) += crash_dump_$(BITS).o
71obj-$(CONFIG_X86_NUMAQ) += numaq_32.o 71obj-$(CONFIG_X86_NUMAQ) += numaq_32.o
72obj-$(CONFIG_X86_ES7000) += es7000_32.o
72obj-$(CONFIG_X86_SUMMIT_NUMA) += summit_32.o 73obj-$(CONFIG_X86_SUMMIT_NUMA) += summit_32.o
73obj-y += vsmp_64.o 74obj-y += vsmp_64.o
74obj-$(CONFIG_KPROBES) += kprobes.o 75obj-$(CONFIG_KPROBES) += kprobes.o
diff --git a/arch/x86/kernel/cpu/Makefile b/arch/x86/kernel/cpu/Makefile
index ee76eaad3001..7f0b45a5d788 100644
--- a/arch/x86/kernel/cpu/Makefile
+++ b/arch/x86/kernel/cpu/Makefile
@@ -3,22 +3,30 @@
3# 3#
4 4
5obj-y := intel_cacheinfo.o addon_cpuid_features.o 5obj-y := intel_cacheinfo.o addon_cpuid_features.o
6obj-y += proc.o feature_names.o 6obj-y += proc.o capflags.o powerflags.o common.o
7 7
8obj-$(CONFIG_X86_32) += common.o bugs.o 8obj-$(CONFIG_X86_32) += bugs.o cmpxchg.o
9obj-$(CONFIG_X86_64) += common_64.o bugs_64.o 9obj-$(CONFIG_X86_64) += bugs_64.o
10obj-$(CONFIG_X86_32) += amd.o 10
11obj-$(CONFIG_X86_64) += amd_64.o 11obj-$(CONFIG_CPU_SUP_INTEL) += intel.o
12obj-$(CONFIG_X86_32) += cyrix.o 12obj-$(CONFIG_CPU_SUP_AMD) += amd.o
13obj-$(CONFIG_X86_32) += centaur.o 13obj-$(CONFIG_CPU_SUP_CYRIX_32) += cyrix.o
14obj-$(CONFIG_X86_64) += centaur_64.o 14obj-$(CONFIG_CPU_SUP_CENTAUR_32) += centaur.o
15obj-$(CONFIG_X86_32) += transmeta.o 15obj-$(CONFIG_CPU_SUP_CENTAUR_64) += centaur_64.o
16obj-$(CONFIG_X86_32) += intel.o 16obj-$(CONFIG_CPU_SUP_TRANSMETA_32) += transmeta.o
17obj-$(CONFIG_X86_64) += intel_64.o 17obj-$(CONFIG_CPU_SUP_UMC_32) += umc.o
18obj-$(CONFIG_X86_32) += umc.o
19 18
20obj-$(CONFIG_X86_MCE) += mcheck/ 19obj-$(CONFIG_X86_MCE) += mcheck/
21obj-$(CONFIG_MTRR) += mtrr/ 20obj-$(CONFIG_MTRR) += mtrr/
22obj-$(CONFIG_CPU_FREQ) += cpufreq/ 21obj-$(CONFIG_CPU_FREQ) += cpufreq/
23 22
24obj-$(CONFIG_X86_LOCAL_APIC) += perfctr-watchdog.o 23obj-$(CONFIG_X86_LOCAL_APIC) += perfctr-watchdog.o
24
25quiet_cmd_mkcapflags = MKCAP $@
26 cmd_mkcapflags = $(PERL) $(srctree)/$(src)/mkcapflags.pl $< $@
27
28cpufeature = $(src)/../../../../include/asm-x86/cpufeature.h
29
30targets += capflags.c
31$(obj)/capflags.c: $(cpufeature) $(src)/mkcapflags.pl FORCE
32 $(call if_changed,mkcapflags)
diff --git a/arch/x86/kernel/cpu/addon_cpuid_features.c b/arch/x86/kernel/cpu/addon_cpuid_features.c
index a6ef672adbba..0d9c993aa93e 100644
--- a/arch/x86/kernel/cpu/addon_cpuid_features.c
+++ b/arch/x86/kernel/cpu/addon_cpuid_features.c
@@ -7,6 +7,8 @@
7#include <asm/pat.h> 7#include <asm/pat.h>
8#include <asm/processor.h> 8#include <asm/processor.h>
9 9
10#include <mach_apic.h>
11
10struct cpuid_bit { 12struct cpuid_bit {
11 u16 feature; 13 u16 feature;
12 u8 reg; 14 u8 reg;
@@ -48,6 +50,92 @@ void __cpuinit init_scattered_cpuid_features(struct cpuinfo_x86 *c)
48 } 50 }
49} 51}
50 52
53/* leaf 0xb SMT level */
54#define SMT_LEVEL 0
55
56/* leaf 0xb sub-leaf types */
57#define INVALID_TYPE 0
58#define SMT_TYPE 1
59#define CORE_TYPE 2
60
61#define LEAFB_SUBTYPE(ecx) (((ecx) >> 8) & 0xff)
62#define BITS_SHIFT_NEXT_LEVEL(eax) ((eax) & 0x1f)
63#define LEVEL_MAX_SIBLINGS(ebx) ((ebx) & 0xffff)
64
65/*
66 * Check for extended topology enumeration cpuid leaf 0xb and if it
67 * exists, use it for populating initial_apicid and cpu topology
68 * detection.
69 */
70void __cpuinit detect_extended_topology(struct cpuinfo_x86 *c)
71{
72#ifdef CONFIG_SMP
73 unsigned int eax, ebx, ecx, edx, sub_index;
74 unsigned int ht_mask_width, core_plus_mask_width;
75 unsigned int core_select_mask, core_level_siblings;
76
77 if (c->cpuid_level < 0xb)
78 return;
79
80 cpuid_count(0xb, SMT_LEVEL, &eax, &ebx, &ecx, &edx);
81
82 /*
83 * check if the cpuid leaf 0xb is actually implemented.
84 */
85 if (ebx == 0 || (LEAFB_SUBTYPE(ecx) != SMT_TYPE))
86 return;
87
88 set_cpu_cap(c, X86_FEATURE_XTOPOLOGY);
89
90 /*
91 * initial apic id, which also represents 32-bit extended x2apic id.
92 */
93 c->initial_apicid = edx;
94
95 /*
96 * Populate HT related information from sub-leaf level 0.
97 */
98 core_level_siblings = smp_num_siblings = LEVEL_MAX_SIBLINGS(ebx);
99 core_plus_mask_width = ht_mask_width = BITS_SHIFT_NEXT_LEVEL(eax);
100
101 sub_index = 1;
102 do {
103 cpuid_count(0xb, sub_index, &eax, &ebx, &ecx, &edx);
104
105 /*
106 * Check for the Core type in the implemented sub leaves.
107 */
108 if (LEAFB_SUBTYPE(ecx) == CORE_TYPE) {
109 core_level_siblings = LEVEL_MAX_SIBLINGS(ebx);
110 core_plus_mask_width = BITS_SHIFT_NEXT_LEVEL(eax);
111 break;
112 }
113
114 sub_index++;
115 } while (LEAFB_SUBTYPE(ecx) != INVALID_TYPE);
116
117 core_select_mask = (~(-1 << core_plus_mask_width)) >> ht_mask_width;
118
119#ifdef CONFIG_X86_32
120 c->cpu_core_id = phys_pkg_id(c->initial_apicid, ht_mask_width)
121 & core_select_mask;
122 c->phys_proc_id = phys_pkg_id(c->initial_apicid, core_plus_mask_width);
123#else
124 c->cpu_core_id = phys_pkg_id(ht_mask_width) & core_select_mask;
125 c->phys_proc_id = phys_pkg_id(core_plus_mask_width);
126#endif
127 c->x86_max_cores = (core_level_siblings / smp_num_siblings);
128
129
130 printk(KERN_INFO "CPU: Physical Processor ID: %d\n",
131 c->phys_proc_id);
132 if (c->x86_max_cores > 1)
133 printk(KERN_INFO "CPU: Processor Core ID: %d\n",
134 c->cpu_core_id);
135 return;
136#endif
137}
138
51#ifdef CONFIG_X86_PAT 139#ifdef CONFIG_X86_PAT
52void __cpuinit validate_pat_support(struct cpuinfo_x86 *c) 140void __cpuinit validate_pat_support(struct cpuinfo_x86 *c)
53{ 141{
diff --git a/arch/x86/kernel/cpu/amd.c b/arch/x86/kernel/cpu/amd.c
index 18514ed26104..32e73520adf7 100644
--- a/arch/x86/kernel/cpu/amd.c
+++ b/arch/x86/kernel/cpu/amd.c
@@ -1,13 +1,22 @@
1#include <linux/init.h> 1#include <linux/init.h>
2#include <linux/bitops.h> 2#include <linux/bitops.h>
3#include <linux/mm.h> 3#include <linux/mm.h>
4
4#include <asm/io.h> 5#include <asm/io.h>
5#include <asm/processor.h> 6#include <asm/processor.h>
6#include <asm/apic.h> 7#include <asm/apic.h>
7 8
9#ifdef CONFIG_X86_64
10# include <asm/numa_64.h>
11# include <asm/mmconfig.h>
12# include <asm/cacheflush.h>
13#endif
14
8#include <mach_apic.h> 15#include <mach_apic.h>
16
9#include "cpu.h" 17#include "cpu.h"
10 18
19#ifdef CONFIG_X86_32
11/* 20/*
12 * B step AMD K6 before B 9730xxxx have hardware bugs that can cause 21 * B step AMD K6 before B 9730xxxx have hardware bugs that can cause
13 * misexecution of code under Linux. Owners of such processors should 22 * misexecution of code under Linux. Owners of such processors should
@@ -24,26 +33,273 @@
24extern void vide(void); 33extern void vide(void);
25__asm__(".align 4\nvide: ret"); 34__asm__(".align 4\nvide: ret");
26 35
27static void __cpuinit early_init_amd(struct cpuinfo_x86 *c) 36static void __cpuinit init_amd_k5(struct cpuinfo_x86 *c)
28{ 37{
29 if (cpuid_eax(0x80000000) >= 0x80000007) { 38/*
30 c->x86_power = cpuid_edx(0x80000007); 39 * General Systems BIOSen alias the cpu frequency registers
31 if (c->x86_power & (1<<8)) 40 * of the Elan at 0x000df000. Unfortuantly, one of the Linux
32 set_cpu_cap(c, X86_FEATURE_CONSTANT_TSC); 41 * drivers subsequently pokes it, and changes the CPU speed.
42 * Workaround : Remove the unneeded alias.
43 */
44#define CBAR (0xfffc) /* Configuration Base Address (32-bit) */
45#define CBAR_ENB (0x80000000)
46#define CBAR_KEY (0X000000CB)
47 if (c->x86_model == 9 || c->x86_model == 10) {
48 if (inl (CBAR) & CBAR_ENB)
49 outl (0 | CBAR_KEY, CBAR);
33 } 50 }
34
35 /* Set MTRR capability flag if appropriate */
36 if (c->x86_model == 13 || c->x86_model == 9 ||
37 (c->x86_model == 8 && c->x86_mask >= 8))
38 set_cpu_cap(c, X86_FEATURE_K6_MTRR);
39} 51}
40 52
41static void __cpuinit init_amd(struct cpuinfo_x86 *c) 53
54static void __cpuinit init_amd_k6(struct cpuinfo_x86 *c)
42{ 55{
43 u32 l, h; 56 u32 l, h;
44 int mbytes = num_physpages >> (20-PAGE_SHIFT); 57 int mbytes = num_physpages >> (20-PAGE_SHIFT);
45 int r;
46 58
59 if (c->x86_model < 6) {
60 /* Based on AMD doc 20734R - June 2000 */
61 if (c->x86_model == 0) {
62 clear_cpu_cap(c, X86_FEATURE_APIC);
63 set_cpu_cap(c, X86_FEATURE_PGE);
64 }
65 return;
66 }
67
68 if (c->x86_model == 6 && c->x86_mask == 1) {
69 const int K6_BUG_LOOP = 1000000;
70 int n;
71 void (*f_vide)(void);
72 unsigned long d, d2;
73
74 printk(KERN_INFO "AMD K6 stepping B detected - ");
75
76 /*
77 * It looks like AMD fixed the 2.6.2 bug and improved indirect
78 * calls at the same time.
79 */
80
81 n = K6_BUG_LOOP;
82 f_vide = vide;
83 rdtscl(d);
84 while (n--)
85 f_vide();
86 rdtscl(d2);
87 d = d2-d;
88
89 if (d > 20*K6_BUG_LOOP)
90 printk("system stability may be impaired when more than 32 MB are used.\n");
91 else
92 printk("probably OK (after B9730xxxx).\n");
93 printk(KERN_INFO "Please see http://membres.lycos.fr/poulot/k6bug.html\n");
94 }
95
96 /* K6 with old style WHCR */
97 if (c->x86_model < 8 ||
98 (c->x86_model == 8 && c->x86_mask < 8)) {
99 /* We can only write allocate on the low 508Mb */
100 if (mbytes > 508)
101 mbytes = 508;
102
103 rdmsr(MSR_K6_WHCR, l, h);
104 if ((l&0x0000FFFF) == 0) {
105 unsigned long flags;
106 l = (1<<0)|((mbytes/4)<<1);
107 local_irq_save(flags);
108 wbinvd();
109 wrmsr(MSR_K6_WHCR, l, h);
110 local_irq_restore(flags);
111 printk(KERN_INFO "Enabling old style K6 write allocation for %d Mb\n",
112 mbytes);
113 }
114 return;
115 }
116
117 if ((c->x86_model == 8 && c->x86_mask > 7) ||
118 c->x86_model == 9 || c->x86_model == 13) {
119 /* The more serious chips .. */
120
121 if (mbytes > 4092)
122 mbytes = 4092;
123
124 rdmsr(MSR_K6_WHCR, l, h);
125 if ((l&0xFFFF0000) == 0) {
126 unsigned long flags;
127 l = ((mbytes>>2)<<22)|(1<<16);
128 local_irq_save(flags);
129 wbinvd();
130 wrmsr(MSR_K6_WHCR, l, h);
131 local_irq_restore(flags);
132 printk(KERN_INFO "Enabling new style K6 write allocation for %d Mb\n",
133 mbytes);
134 }
135
136 return;
137 }
138
139 if (c->x86_model == 10) {
140 /* AMD Geode LX is model 10 */
141 /* placeholder for any needed mods */
142 return;
143 }
144}
145
146static void __cpuinit init_amd_k7(struct cpuinfo_x86 *c)
147{
148 u32 l, h;
149
150 /*
151 * Bit 15 of Athlon specific MSR 15, needs to be 0
152 * to enable SSE on Palomino/Morgan/Barton CPU's.
153 * If the BIOS didn't enable it already, enable it here.
154 */
155 if (c->x86_model >= 6 && c->x86_model <= 10) {
156 if (!cpu_has(c, X86_FEATURE_XMM)) {
157 printk(KERN_INFO "Enabling disabled K7/SSE Support.\n");
158 rdmsr(MSR_K7_HWCR, l, h);
159 l &= ~0x00008000;
160 wrmsr(MSR_K7_HWCR, l, h);
161 set_cpu_cap(c, X86_FEATURE_XMM);
162 }
163 }
164
165 /*
166 * It's been determined by AMD that Athlons since model 8 stepping 1
167 * are more robust with CLK_CTL set to 200xxxxx instead of 600xxxxx
168 * As per AMD technical note 27212 0.2
169 */
170 if ((c->x86_model == 8 && c->x86_mask >= 1) || (c->x86_model > 8)) {
171 rdmsr(MSR_K7_CLK_CTL, l, h);
172 if ((l & 0xfff00000) != 0x20000000) {
173 printk ("CPU: CLK_CTL MSR was %x. Reprogramming to %x\n", l,
174 ((l & 0x000fffff)|0x20000000));
175 wrmsr(MSR_K7_CLK_CTL, (l & 0x000fffff)|0x20000000, h);
176 }
177 }
178
179 set_cpu_cap(c, X86_FEATURE_K7);
180}
181#endif
182
183#if defined(CONFIG_NUMA) && defined(CONFIG_X86_64)
184static int __cpuinit nearby_node(int apicid)
185{
186 int i, node;
187
188 for (i = apicid - 1; i >= 0; i--) {
189 node = apicid_to_node[i];
190 if (node != NUMA_NO_NODE && node_online(node))
191 return node;
192 }
193 for (i = apicid + 1; i < MAX_LOCAL_APIC; i++) {
194 node = apicid_to_node[i];
195 if (node != NUMA_NO_NODE && node_online(node))
196 return node;
197 }
198 return first_node(node_online_map); /* Shouldn't happen */
199}
200#endif
201
202/*
203 * On a AMD dual core setup the lower bits of the APIC id distingush the cores.
204 * Assumes number of cores is a power of two.
205 */
206static void __cpuinit amd_detect_cmp(struct cpuinfo_x86 *c)
207{
208#ifdef CONFIG_X86_HT
209 unsigned bits;
210
211 bits = c->x86_coreid_bits;
212
213 /* Low order bits define the core id (index of core in socket) */
214 c->cpu_core_id = c->initial_apicid & ((1 << bits)-1);
215 /* Convert the initial APIC ID into the socket ID */
216 c->phys_proc_id = c->initial_apicid >> bits;
217#endif
218}
219
220static void __cpuinit srat_detect_node(struct cpuinfo_x86 *c)
221{
222#if defined(CONFIG_NUMA) && defined(CONFIG_X86_64)
223 int cpu = smp_processor_id();
224 int node;
225 unsigned apicid = hard_smp_processor_id();
226
227 node = c->phys_proc_id;
228 if (apicid_to_node[apicid] != NUMA_NO_NODE)
229 node = apicid_to_node[apicid];
230 if (!node_online(node)) {
231 /* Two possibilities here:
232 - The CPU is missing memory and no node was created.
233 In that case try picking one from a nearby CPU
234 - The APIC IDs differ from the HyperTransport node IDs
235 which the K8 northbridge parsing fills in.
236 Assume they are all increased by a constant offset,
237 but in the same order as the HT nodeids.
238 If that doesn't result in a usable node fall back to the
239 path for the previous case. */
240
241 int ht_nodeid = c->initial_apicid;
242
243 if (ht_nodeid >= 0 &&
244 apicid_to_node[ht_nodeid] != NUMA_NO_NODE)
245 node = apicid_to_node[ht_nodeid];
246 /* Pick a nearby node */
247 if (!node_online(node))
248 node = nearby_node(apicid);
249 }
250 numa_set_node(cpu, node);
251
252 printk(KERN_INFO "CPU %d/%x -> Node %d\n", cpu, apicid, node);
253#endif
254}
255
256static void __cpuinit early_init_amd_mc(struct cpuinfo_x86 *c)
257{
258#ifdef CONFIG_X86_HT
259 unsigned bits, ecx;
260
261 /* Multi core CPU? */
262 if (c->extended_cpuid_level < 0x80000008)
263 return;
264
265 ecx = cpuid_ecx(0x80000008);
266
267 c->x86_max_cores = (ecx & 0xff) + 1;
268
269 /* CPU telling us the core id bits shift? */
270 bits = (ecx >> 12) & 0xF;
271
272 /* Otherwise recompute */
273 if (bits == 0) {
274 while ((1 << bits) < c->x86_max_cores)
275 bits++;
276 }
277
278 c->x86_coreid_bits = bits;
279#endif
280}
281
282static void __cpuinit early_init_amd(struct cpuinfo_x86 *c)
283{
284 early_init_amd_mc(c);
285
286 /* c->x86_power is 8000_0007 edx. Bit 8 is constant TSC */
287 if (c->x86_power & (1<<8))
288 set_cpu_cap(c, X86_FEATURE_CONSTANT_TSC);
289
290#ifdef CONFIG_X86_64
291 set_cpu_cap(c, X86_FEATURE_SYSCALL32);
292#else
293 /* Set MTRR capability flag if appropriate */
294 if (c->x86 == 5)
295 if (c->x86_model == 13 || c->x86_model == 9 ||
296 (c->x86_model == 8 && c->x86_mask >= 8))
297 set_cpu_cap(c, X86_FEATURE_K6_MTRR);
298#endif
299}
300
301static void __cpuinit init_amd(struct cpuinfo_x86 *c)
302{
47#ifdef CONFIG_SMP 303#ifdef CONFIG_SMP
48 unsigned long long value; 304 unsigned long long value;
49 305
@@ -54,7 +310,7 @@ static void __cpuinit init_amd(struct cpuinfo_x86 *c)
54 * Errata 63 for SH-B3 steppings 310 * Errata 63 for SH-B3 steppings
55 * Errata 122 for all steppings (F+ have it disabled by default) 311 * Errata 122 for all steppings (F+ have it disabled by default)
56 */ 312 */
57 if (c->x86 == 15) { 313 if (c->x86 == 0xf) {
58 rdmsrl(MSR_K7_HWCR, value); 314 rdmsrl(MSR_K7_HWCR, value);
59 value |= 1 << 6; 315 value |= 1 << 6;
60 wrmsrl(MSR_K7_HWCR, value); 316 wrmsrl(MSR_K7_HWCR, value);
@@ -64,209 +320,119 @@ static void __cpuinit init_amd(struct cpuinfo_x86 *c)
64 early_init_amd(c); 320 early_init_amd(c);
65 321
66 /* 322 /*
67 * FIXME: We should handle the K5 here. Set up the write
68 * range and also turn on MSR 83 bits 4 and 31 (write alloc,
69 * no bus pipeline)
70 */
71
72 /*
73 * Bit 31 in normal CPUID used for nonstandard 3DNow ID; 323 * Bit 31 in normal CPUID used for nonstandard 3DNow ID;
74 * 3DNow is IDd by bit 31 in extended CPUID (1*32+31) anyway 324 * 3DNow is IDd by bit 31 in extended CPUID (1*32+31) anyway
75 */ 325 */
76 clear_cpu_cap(c, 0*32+31); 326 clear_cpu_cap(c, 0*32+31);
77 327
78 r = get_model_name(c); 328#ifdef CONFIG_X86_64
329 /* On C+ stepping K8 rep microcode works well for copy/memset */
330 if (c->x86 == 0xf) {
331 u32 level;
79 332
80 switch (c->x86) { 333 level = cpuid_eax(1);
81 case 4: 334 if((level >= 0x0f48 && level < 0x0f50) || level >= 0x0f58)
82 /* 335 set_cpu_cap(c, X86_FEATURE_REP_GOOD);
83 * General Systems BIOSen alias the cpu frequency registers
84 * of the Elan at 0x000df000. Unfortuantly, one of the Linux
85 * drivers subsequently pokes it, and changes the CPU speed.
86 * Workaround : Remove the unneeded alias.
87 */
88#define CBAR (0xfffc) /* Configuration Base Address (32-bit) */
89#define CBAR_ENB (0x80000000)
90#define CBAR_KEY (0X000000CB)
91 if (c->x86_model == 9 || c->x86_model == 10) {
92 if (inl (CBAR) & CBAR_ENB)
93 outl (0 | CBAR_KEY, CBAR);
94 }
95 break;
96 case 5:
97 if (c->x86_model < 6) {
98 /* Based on AMD doc 20734R - June 2000 */
99 if (c->x86_model == 0) {
100 clear_cpu_cap(c, X86_FEATURE_APIC);
101 set_cpu_cap(c, X86_FEATURE_PGE);
102 }
103 break;
104 }
105
106 if (c->x86_model == 6 && c->x86_mask == 1) {
107 const int K6_BUG_LOOP = 1000000;
108 int n;
109 void (*f_vide)(void);
110 unsigned long d, d2;
111
112 printk(KERN_INFO "AMD K6 stepping B detected - ");
113
114 /*
115 * It looks like AMD fixed the 2.6.2 bug and improved indirect
116 * calls at the same time.
117 */
118
119 n = K6_BUG_LOOP;
120 f_vide = vide;
121 rdtscl(d);
122 while (n--)
123 f_vide();
124 rdtscl(d2);
125 d = d2-d;
126
127 if (d > 20*K6_BUG_LOOP)
128 printk("system stability may be impaired when more than 32 MB are used.\n");
129 else
130 printk("probably OK (after B9730xxxx).\n");
131 printk(KERN_INFO "Please see http://membres.lycos.fr/poulot/k6bug.html\n");
132 }
133
134 /* K6 with old style WHCR */
135 if (c->x86_model < 8 ||
136 (c->x86_model == 8 && c->x86_mask < 8)) {
137 /* We can only write allocate on the low 508Mb */
138 if (mbytes > 508)
139 mbytes = 508;
140
141 rdmsr(MSR_K6_WHCR, l, h);
142 if ((l&0x0000FFFF) == 0) {
143 unsigned long flags;
144 l = (1<<0)|((mbytes/4)<<1);
145 local_irq_save(flags);
146 wbinvd();
147 wrmsr(MSR_K6_WHCR, l, h);
148 local_irq_restore(flags);
149 printk(KERN_INFO "Enabling old style K6 write allocation for %d Mb\n",
150 mbytes);
151 }
152 break;
153 }
154
155 if ((c->x86_model == 8 && c->x86_mask > 7) ||
156 c->x86_model == 9 || c->x86_model == 13) {
157 /* The more serious chips .. */
158
159 if (mbytes > 4092)
160 mbytes = 4092;
161
162 rdmsr(MSR_K6_WHCR, l, h);
163 if ((l&0xFFFF0000) == 0) {
164 unsigned long flags;
165 l = ((mbytes>>2)<<22)|(1<<16);
166 local_irq_save(flags);
167 wbinvd();
168 wrmsr(MSR_K6_WHCR, l, h);
169 local_irq_restore(flags);
170 printk(KERN_INFO "Enabling new style K6 write allocation for %d Mb\n",
171 mbytes);
172 }
173
174 break;
175 }
176
177 if (c->x86_model == 10) {
178 /* AMD Geode LX is model 10 */
179 /* placeholder for any needed mods */
180 break;
181 }
182 break;
183 case 6: /* An Athlon/Duron */
184
185 /*
186 * Bit 15 of Athlon specific MSR 15, needs to be 0
187 * to enable SSE on Palomino/Morgan/Barton CPU's.
188 * If the BIOS didn't enable it already, enable it here.
189 */
190 if (c->x86_model >= 6 && c->x86_model <= 10) {
191 if (!cpu_has(c, X86_FEATURE_XMM)) {
192 printk(KERN_INFO "Enabling disabled K7/SSE Support.\n");
193 rdmsr(MSR_K7_HWCR, l, h);
194 l &= ~0x00008000;
195 wrmsr(MSR_K7_HWCR, l, h);
196 set_cpu_cap(c, X86_FEATURE_XMM);
197 }
198 }
199
200 /*
201 * It's been determined by AMD that Athlons since model 8 stepping 1
202 * are more robust with CLK_CTL set to 200xxxxx instead of 600xxxxx
203 * As per AMD technical note 27212 0.2
204 */
205 if ((c->x86_model == 8 && c->x86_mask >= 1) || (c->x86_model > 8)) {
206 rdmsr(MSR_K7_CLK_CTL, l, h);
207 if ((l & 0xfff00000) != 0x20000000) {
208 printk ("CPU: CLK_CTL MSR was %x. Reprogramming to %x\n", l,
209 ((l & 0x000fffff)|0x20000000));
210 wrmsr(MSR_K7_CLK_CTL, (l & 0x000fffff)|0x20000000, h);
211 }
212 }
213 break;
214 } 336 }
337 if (c->x86 == 0x10 || c->x86 == 0x11)
338 set_cpu_cap(c, X86_FEATURE_REP_GOOD);
339#else
340
341 /*
342 * FIXME: We should handle the K5 here. Set up the write
343 * range and also turn on MSR 83 bits 4 and 31 (write alloc,
344 * no bus pipeline)
345 */
215 346
216 switch (c->x86) { 347 switch (c->x86) {
217 case 15: 348 case 4:
218 /* Use K8 tuning for Fam10h and Fam11h */ 349 init_amd_k5(c);
219 case 0x10:
220 case 0x11:
221 set_cpu_cap(c, X86_FEATURE_K8);
222 break; 350 break;
223 case 6: 351 case 5:
224 set_cpu_cap(c, X86_FEATURE_K7); 352 init_amd_k6(c);
353 break;
354 case 6: /* An Athlon/Duron */
355 init_amd_k7(c);
225 break; 356 break;
226 } 357 }
358
359 /* K6s reports MCEs but don't actually have all the MSRs */
360 if (c->x86 < 6)
361 clear_cpu_cap(c, X86_FEATURE_MCE);
362#endif
363
364 /* Enable workaround for FXSAVE leak */
227 if (c->x86 >= 6) 365 if (c->x86 >= 6)
228 set_cpu_cap(c, X86_FEATURE_FXSAVE_LEAK); 366 set_cpu_cap(c, X86_FEATURE_FXSAVE_LEAK);
229 367
230 display_cacheinfo(c); 368 if (!c->x86_model_id[0]) {
231 369 switch (c->x86) {
232 if (cpuid_eax(0x80000000) >= 0x80000008) 370 case 0xf:
233 c->x86_max_cores = (cpuid_ecx(0x80000008) & 0xff) + 1; 371 /* Should distinguish Models here, but this is only
372 a fallback anyways. */
373 strcpy(c->x86_model_id, "Hammer");
374 break;
375 }
376 }
234 377
235#ifdef CONFIG_X86_HT 378 display_cacheinfo(c);
236 /*
237 * On a AMD multi core setup the lower bits of the APIC id
238 * distinguish the cores.
239 */
240 if (c->x86_max_cores > 1) {
241 int cpu = smp_processor_id();
242 unsigned bits = (cpuid_ecx(0x80000008) >> 12) & 0xf;
243 379
244 if (bits == 0) { 380 /* Multi core CPU? */
245 while ((1 << bits) < c->x86_max_cores) 381 if (c->extended_cpuid_level >= 0x80000008) {
246 bits++; 382 amd_detect_cmp(c);
247 } 383 srat_detect_node(c);
248 c->cpu_core_id = c->phys_proc_id & ((1<<bits)-1);
249 c->phys_proc_id >>= bits;
250 printk(KERN_INFO "CPU %d(%d) -> Core %d\n",
251 cpu, c->x86_max_cores, c->cpu_core_id);
252 } 384 }
385
386#ifdef CONFIG_X86_32
387 detect_ht(c);
253#endif 388#endif
254 389
255 if (cpuid_eax(0x80000000) >= 0x80000006) { 390 if (c->extended_cpuid_level >= 0x80000006) {
256 if ((c->x86 == 0x10) && (cpuid_edx(0x80000006) & 0xf000)) 391 if ((c->x86 >= 0x0f) && (cpuid_edx(0x80000006) & 0xf000))
257 num_cache_leaves = 4; 392 num_cache_leaves = 4;
258 else 393 else
259 num_cache_leaves = 3; 394 num_cache_leaves = 3;
260 } 395 }
261 396
262 /* K6s reports MCEs but don't actually have all the MSRs */ 397 if (c->x86 >= 0xf && c->x86 <= 0x11)
263 if (c->x86 < 6) 398 set_cpu_cap(c, X86_FEATURE_K8);
264 clear_cpu_cap(c, X86_FEATURE_MCE);
265 399
266 if (cpu_has_xmm2) 400 if (cpu_has_xmm2) {
401 /* MFENCE stops RDTSC speculation */
267 set_cpu_cap(c, X86_FEATURE_MFENCE_RDTSC); 402 set_cpu_cap(c, X86_FEATURE_MFENCE_RDTSC);
403 }
404
405#ifdef CONFIG_X86_64
406 if (c->x86 == 0x10) {
407 /* do this for boot cpu */
408 if (c == &boot_cpu_data)
409 check_enable_amd_mmconf_dmi();
410
411 fam10h_check_enable_mmcfg();
412 }
413
414 if (c == &boot_cpu_data && c->x86 >= 0xf && c->x86 <= 0x11) {
415 unsigned long long tseg;
416
417 /*
418 * Split up direct mapping around the TSEG SMM area.
419 * Don't do it for gbpages because there seems very little
420 * benefit in doing so.
421 */
422 if (!rdmsrl_safe(MSR_K8_TSEG_ADDR, &tseg)) {
423 printk(KERN_DEBUG "tseg: %010llx\n", tseg);
424 if ((tseg>>PMD_SHIFT) <
425 (max_low_pfn_mapped>>(PMD_SHIFT-PAGE_SHIFT)) ||
426 ((tseg>>PMD_SHIFT) <
427 (max_pfn_mapped>>(PMD_SHIFT-PAGE_SHIFT)) &&
428 (tseg>>PMD_SHIFT) >= (1ULL<<(32 - PMD_SHIFT))))
429 set_memory_4k((unsigned long)__va(tseg), 1);
430 }
431 }
432#endif
268} 433}
269 434
435#ifdef CONFIG_X86_32
270static unsigned int __cpuinit amd_size_cache(struct cpuinfo_x86 *c, unsigned int size) 436static unsigned int __cpuinit amd_size_cache(struct cpuinfo_x86 *c, unsigned int size)
271{ 437{
272 /* AMD errata T13 (order #21922) */ 438 /* AMD errata T13 (order #21922) */
@@ -279,10 +445,12 @@ static unsigned int __cpuinit amd_size_cache(struct cpuinfo_x86 *c, unsigned int
279 } 445 }
280 return size; 446 return size;
281} 447}
448#endif
282 449
283static struct cpu_dev amd_cpu_dev __cpuinitdata = { 450static struct cpu_dev amd_cpu_dev __cpuinitdata = {
284 .c_vendor = "AMD", 451 .c_vendor = "AMD",
285 .c_ident = { "AuthenticAMD" }, 452 .c_ident = { "AuthenticAMD" },
453#ifdef CONFIG_X86_32
286 .c_models = { 454 .c_models = {
287 { .vendor = X86_VENDOR_AMD, .family = 4, .model_names = 455 { .vendor = X86_VENDOR_AMD, .family = 4, .model_names =
288 { 456 {
@@ -295,9 +463,11 @@ static struct cpu_dev amd_cpu_dev __cpuinitdata = {
295 } 463 }
296 }, 464 },
297 }, 465 },
466 .c_size_cache = amd_size_cache,
467#endif
298 .c_early_init = early_init_amd, 468 .c_early_init = early_init_amd,
299 .c_init = init_amd, 469 .c_init = init_amd,
300 .c_size_cache = amd_size_cache, 470 .c_x86_vendor = X86_VENDOR_AMD,
301}; 471};
302 472
303cpu_vendor_dev_register(X86_VENDOR_AMD, &amd_cpu_dev); 473cpu_dev_register(amd_cpu_dev);
diff --git a/arch/x86/kernel/cpu/amd_64.c b/arch/x86/kernel/cpu/amd_64.c
deleted file mode 100644
index d1692b2a41ff..000000000000
--- a/arch/x86/kernel/cpu/amd_64.c
+++ /dev/null
@@ -1,224 +0,0 @@
1#include <linux/init.h>
2#include <linux/mm.h>
3
4#include <asm/numa_64.h>
5#include <asm/mmconfig.h>
6#include <asm/cacheflush.h>
7
8#include <mach_apic.h>
9
10#include "cpu.h"
11
12int force_mwait __cpuinitdata;
13
14#ifdef CONFIG_NUMA
15static int __cpuinit nearby_node(int apicid)
16{
17 int i, node;
18
19 for (i = apicid - 1; i >= 0; i--) {
20 node = apicid_to_node[i];
21 if (node != NUMA_NO_NODE && node_online(node))
22 return node;
23 }
24 for (i = apicid + 1; i < MAX_LOCAL_APIC; i++) {
25 node = apicid_to_node[i];
26 if (node != NUMA_NO_NODE && node_online(node))
27 return node;
28 }
29 return first_node(node_online_map); /* Shouldn't happen */
30}
31#endif
32
33/*
34 * On a AMD dual core setup the lower bits of the APIC id distingush the cores.
35 * Assumes number of cores is a power of two.
36 */
37static void __cpuinit amd_detect_cmp(struct cpuinfo_x86 *c)
38{
39#ifdef CONFIG_SMP
40 unsigned bits;
41#ifdef CONFIG_NUMA
42 int cpu = smp_processor_id();
43 int node = 0;
44 unsigned apicid = hard_smp_processor_id();
45#endif
46 bits = c->x86_coreid_bits;
47
48 /* Low order bits define the core id (index of core in socket) */
49 c->cpu_core_id = c->initial_apicid & ((1 << bits)-1);
50 /* Convert the initial APIC ID into the socket ID */
51 c->phys_proc_id = c->initial_apicid >> bits;
52
53#ifdef CONFIG_NUMA
54 node = c->phys_proc_id;
55 if (apicid_to_node[apicid] != NUMA_NO_NODE)
56 node = apicid_to_node[apicid];
57 if (!node_online(node)) {
58 /* Two possibilities here:
59 - The CPU is missing memory and no node was created.
60 In that case try picking one from a nearby CPU
61 - The APIC IDs differ from the HyperTransport node IDs
62 which the K8 northbridge parsing fills in.
63 Assume they are all increased by a constant offset,
64 but in the same order as the HT nodeids.
65 If that doesn't result in a usable node fall back to the
66 path for the previous case. */
67
68 int ht_nodeid = c->initial_apicid;
69
70 if (ht_nodeid >= 0 &&
71 apicid_to_node[ht_nodeid] != NUMA_NO_NODE)
72 node = apicid_to_node[ht_nodeid];
73 /* Pick a nearby node */
74 if (!node_online(node))
75 node = nearby_node(apicid);
76 }
77 numa_set_node(cpu, node);
78
79 printk(KERN_INFO "CPU %d/%x -> Node %d\n", cpu, apicid, node);
80#endif
81#endif
82}
83
84static void __cpuinit early_init_amd_mc(struct cpuinfo_x86 *c)
85{
86#ifdef CONFIG_SMP
87 unsigned bits, ecx;
88
89 /* Multi core CPU? */
90 if (c->extended_cpuid_level < 0x80000008)
91 return;
92
93 ecx = cpuid_ecx(0x80000008);
94
95 c->x86_max_cores = (ecx & 0xff) + 1;
96
97 /* CPU telling us the core id bits shift? */
98 bits = (ecx >> 12) & 0xF;
99
100 /* Otherwise recompute */
101 if (bits == 0) {
102 while ((1 << bits) < c->x86_max_cores)
103 bits++;
104 }
105
106 c->x86_coreid_bits = bits;
107
108#endif
109}
110
111static void __cpuinit early_init_amd(struct cpuinfo_x86 *c)
112{
113 early_init_amd_mc(c);
114
115 /* c->x86_power is 8000_0007 edx. Bit 8 is constant TSC */
116 if (c->x86_power & (1<<8))
117 set_cpu_cap(c, X86_FEATURE_CONSTANT_TSC);
118
119 set_cpu_cap(c, X86_FEATURE_SYSCALL32);
120}
121
122static void __cpuinit init_amd(struct cpuinfo_x86 *c)
123{
124 unsigned level;
125
126#ifdef CONFIG_SMP
127 unsigned long value;
128
129 /*
130 * Disable TLB flush filter by setting HWCR.FFDIS on K8
131 * bit 6 of msr C001_0015
132 *
133 * Errata 63 for SH-B3 steppings
134 * Errata 122 for all steppings (F+ have it disabled by default)
135 */
136 if (c->x86 == 0xf) {
137 rdmsrl(MSR_K8_HWCR, value);
138 value |= 1 << 6;
139 wrmsrl(MSR_K8_HWCR, value);
140 }
141#endif
142
143 /* Bit 31 in normal CPUID used for nonstandard 3DNow ID;
144 3DNow is IDd by bit 31 in extended CPUID (1*32+31) anyway */
145 clear_cpu_cap(c, 0*32+31);
146
147 /* On C+ stepping K8 rep microcode works well for copy/memset */
148 if (c->x86 == 0xf) {
149 level = cpuid_eax(1);
150 if((level >= 0x0f48 && level < 0x0f50) || level >= 0x0f58)
151 set_cpu_cap(c, X86_FEATURE_REP_GOOD);
152 }
153 if (c->x86 == 0x10 || c->x86 == 0x11)
154 set_cpu_cap(c, X86_FEATURE_REP_GOOD);
155
156 /* Enable workaround for FXSAVE leak */
157 if (c->x86 >= 6)
158 set_cpu_cap(c, X86_FEATURE_FXSAVE_LEAK);
159
160 level = get_model_name(c);
161 if (!level) {
162 switch (c->x86) {
163 case 0xf:
164 /* Should distinguish Models here, but this is only
165 a fallback anyways. */
166 strcpy(c->x86_model_id, "Hammer");
167 break;
168 }
169 }
170 display_cacheinfo(c);
171
172 /* Multi core CPU? */
173 if (c->extended_cpuid_level >= 0x80000008)
174 amd_detect_cmp(c);
175
176 if (c->extended_cpuid_level >= 0x80000006 &&
177 (cpuid_edx(0x80000006) & 0xf000))
178 num_cache_leaves = 4;
179 else
180 num_cache_leaves = 3;
181
182 if (c->x86 >= 0xf && c->x86 <= 0x11)
183 set_cpu_cap(c, X86_FEATURE_K8);
184
185 /* MFENCE stops RDTSC speculation */
186 set_cpu_cap(c, X86_FEATURE_MFENCE_RDTSC);
187
188 if (c->x86 == 0x10) {
189 /* do this for boot cpu */
190 if (c == &boot_cpu_data)
191 check_enable_amd_mmconf_dmi();
192
193 fam10h_check_enable_mmcfg();
194 }
195
196 if (c == &boot_cpu_data && c->x86 >= 0xf && c->x86 <= 0x11) {
197 unsigned long long tseg;
198
199 /*
200 * Split up direct mapping around the TSEG SMM area.
201 * Don't do it for gbpages because there seems very little
202 * benefit in doing so.
203 */
204 if (!rdmsrl_safe(MSR_K8_TSEG_ADDR, &tseg)) {
205 printk(KERN_DEBUG "tseg: %010llx\n", tseg);
206 if ((tseg>>PMD_SHIFT) <
207 (max_low_pfn_mapped>>(PMD_SHIFT-PAGE_SHIFT)) ||
208 ((tseg>>PMD_SHIFT) <
209 (max_pfn_mapped>>(PMD_SHIFT-PAGE_SHIFT)) &&
210 (tseg>>PMD_SHIFT) >= (1ULL<<(32 - PMD_SHIFT))))
211 set_memory_4k((unsigned long)__va(tseg), 1);
212 }
213 }
214}
215
216static struct cpu_dev amd_cpu_dev __cpuinitdata = {
217 .c_vendor = "AMD",
218 .c_ident = { "AuthenticAMD" },
219 .c_early_init = early_init_amd,
220 .c_init = init_amd,
221};
222
223cpu_vendor_dev_register(X86_VENDOR_AMD, &amd_cpu_dev);
224
diff --git a/arch/x86/kernel/cpu/centaur.c b/arch/x86/kernel/cpu/centaur.c
index a0534c04d38a..89bfdd9cacc6 100644
--- a/arch/x86/kernel/cpu/centaur.c
+++ b/arch/x86/kernel/cpu/centaur.c
@@ -289,7 +289,6 @@ static void __cpuinit init_c3(struct cpuinfo_x86 *c)
289 if (c->x86_model >= 6 && c->x86_model < 9) 289 if (c->x86_model >= 6 && c->x86_model < 9)
290 set_cpu_cap(c, X86_FEATURE_3DNOW); 290 set_cpu_cap(c, X86_FEATURE_3DNOW);
291 291
292 get_model_name(c);
293 display_cacheinfo(c); 292 display_cacheinfo(c);
294} 293}
295 294
@@ -475,6 +474,7 @@ static struct cpu_dev centaur_cpu_dev __cpuinitdata = {
475 .c_early_init = early_init_centaur, 474 .c_early_init = early_init_centaur,
476 .c_init = init_centaur, 475 .c_init = init_centaur,
477 .c_size_cache = centaur_size_cache, 476 .c_size_cache = centaur_size_cache,
477 .c_x86_vendor = X86_VENDOR_CENTAUR,
478}; 478};
479 479
480cpu_vendor_dev_register(X86_VENDOR_CENTAUR, &centaur_cpu_dev); 480cpu_dev_register(centaur_cpu_dev);
diff --git a/arch/x86/kernel/cpu/centaur_64.c b/arch/x86/kernel/cpu/centaur_64.c
index 1d181c40e2e1..a1625f5a1e78 100644
--- a/arch/x86/kernel/cpu/centaur_64.c
+++ b/arch/x86/kernel/cpu/centaur_64.c
@@ -16,9 +16,10 @@ static void __cpuinit early_init_centaur(struct cpuinfo_x86 *c)
16 16
17static void __cpuinit init_centaur(struct cpuinfo_x86 *c) 17static void __cpuinit init_centaur(struct cpuinfo_x86 *c)
18{ 18{
19 early_init_centaur(c);
20
19 if (c->x86 == 0x6 && c->x86_model >= 0xf) { 21 if (c->x86 == 0x6 && c->x86_model >= 0xf) {
20 c->x86_cache_alignment = c->x86_clflush_size * 2; 22 c->x86_cache_alignment = c->x86_clflush_size * 2;
21 set_cpu_cap(c, X86_FEATURE_CONSTANT_TSC);
22 set_cpu_cap(c, X86_FEATURE_REP_GOOD); 23 set_cpu_cap(c, X86_FEATURE_REP_GOOD);
23 } 24 }
24 set_cpu_cap(c, X86_FEATURE_LFENCE_RDTSC); 25 set_cpu_cap(c, X86_FEATURE_LFENCE_RDTSC);
@@ -29,7 +30,8 @@ static struct cpu_dev centaur_cpu_dev __cpuinitdata = {
29 .c_ident = { "CentaurHauls" }, 30 .c_ident = { "CentaurHauls" },
30 .c_early_init = early_init_centaur, 31 .c_early_init = early_init_centaur,
31 .c_init = init_centaur, 32 .c_init = init_centaur,
33 .c_x86_vendor = X86_VENDOR_CENTAUR,
32}; 34};
33 35
34cpu_vendor_dev_register(X86_VENDOR_CENTAUR, &centaur_cpu_dev); 36cpu_dev_register(centaur_cpu_dev);
35 37
diff --git a/arch/x86/kernel/cpu/cmpxchg.c b/arch/x86/kernel/cpu/cmpxchg.c
new file mode 100644
index 000000000000..2056ccf572cc
--- /dev/null
+++ b/arch/x86/kernel/cpu/cmpxchg.c
@@ -0,0 +1,72 @@
1/*
2 * cmpxchg*() fallbacks for CPU not supporting these instructions
3 */
4
5#include <linux/kernel.h>
6#include <linux/smp.h>
7#include <linux/module.h>
8
9#ifndef CONFIG_X86_CMPXCHG
10unsigned long cmpxchg_386_u8(volatile void *ptr, u8 old, u8 new)
11{
12 u8 prev;
13 unsigned long flags;
14
15 /* Poor man's cmpxchg for 386. Unsuitable for SMP */
16 local_irq_save(flags);
17 prev = *(u8 *)ptr;
18 if (prev == old)
19 *(u8 *)ptr = new;
20 local_irq_restore(flags);
21 return prev;
22}
23EXPORT_SYMBOL(cmpxchg_386_u8);
24
25unsigned long cmpxchg_386_u16(volatile void *ptr, u16 old, u16 new)
26{
27 u16 prev;
28 unsigned long flags;
29
30 /* Poor man's cmpxchg for 386. Unsuitable for SMP */
31 local_irq_save(flags);
32 prev = *(u16 *)ptr;
33 if (prev == old)
34 *(u16 *)ptr = new;
35 local_irq_restore(flags);
36 return prev;
37}
38EXPORT_SYMBOL(cmpxchg_386_u16);
39
40unsigned long cmpxchg_386_u32(volatile void *ptr, u32 old, u32 new)
41{
42 u32 prev;
43 unsigned long flags;
44
45 /* Poor man's cmpxchg for 386. Unsuitable for SMP */
46 local_irq_save(flags);
47 prev = *(u32 *)ptr;
48 if (prev == old)
49 *(u32 *)ptr = new;
50 local_irq_restore(flags);
51 return prev;
52}
53EXPORT_SYMBOL(cmpxchg_386_u32);
54#endif
55
56#ifndef CONFIG_X86_CMPXCHG64
57unsigned long long cmpxchg_486_u64(volatile void *ptr, u64 old, u64 new)
58{
59 u64 prev;
60 unsigned long flags;
61
62 /* Poor man's cmpxchg8b for 386 and 486. Unsuitable for SMP */
63 local_irq_save(flags);
64 prev = *(u64 *)ptr;
65 if (prev == old)
66 *(u64 *)ptr = new;
67 local_irq_restore(flags);
68 return prev;
69}
70EXPORT_SYMBOL(cmpxchg_486_u64);
71#endif
72
diff --git a/arch/x86/kernel/cpu/common.c b/arch/x86/kernel/cpu/common.c
index 4e456bd955bb..7581b62df184 100644
--- a/arch/x86/kernel/cpu/common.c
+++ b/arch/x86/kernel/cpu/common.c
@@ -1,28 +1,62 @@
1#include <linux/init.h> 1#include <linux/init.h>
2#include <linux/kernel.h>
3#include <linux/sched.h>
2#include <linux/string.h> 4#include <linux/string.h>
5#include <linux/bootmem.h>
6#include <linux/bitops.h>
7#include <linux/module.h>
8#include <linux/kgdb.h>
9#include <linux/topology.h>
3#include <linux/delay.h> 10#include <linux/delay.h>
4#include <linux/smp.h> 11#include <linux/smp.h>
5#include <linux/module.h>
6#include <linux/percpu.h> 12#include <linux/percpu.h>
7#include <linux/bootmem.h>
8#include <asm/processor.h>
9#include <asm/i387.h> 13#include <asm/i387.h>
10#include <asm/msr.h> 14#include <asm/msr.h>
11#include <asm/io.h> 15#include <asm/io.h>
16#include <asm/linkage.h>
12#include <asm/mmu_context.h> 17#include <asm/mmu_context.h>
13#include <asm/mtrr.h> 18#include <asm/mtrr.h>
14#include <asm/mce.h> 19#include <asm/mce.h>
15#include <asm/pat.h> 20#include <asm/pat.h>
16#include <asm/asm.h> 21#include <asm/asm.h>
22#include <asm/numa.h>
17#ifdef CONFIG_X86_LOCAL_APIC 23#ifdef CONFIG_X86_LOCAL_APIC
18#include <asm/mpspec.h> 24#include <asm/mpspec.h>
19#include <asm/apic.h> 25#include <asm/apic.h>
20#include <mach_apic.h> 26#include <mach_apic.h>
27#include <asm/genapic.h>
21#endif 28#endif
22 29
30#include <asm/pda.h>
31#include <asm/pgtable.h>
32#include <asm/processor.h>
33#include <asm/desc.h>
34#include <asm/atomic.h>
35#include <asm/proto.h>
36#include <asm/sections.h>
37#include <asm/setup.h>
38
23#include "cpu.h" 39#include "cpu.h"
24 40
41static struct cpu_dev *this_cpu __cpuinitdata;
42
43#ifdef CONFIG_X86_64
44/* We need valid kernel segments for data and code in long mode too
45 * IRET will check the segment types kkeil 2000/10/28
46 * Also sysret mandates a special GDT layout
47 */
48/* The TLS descriptors are currently at a different place compared to i386.
49 Hopefully nobody expects them at a fixed place (Wine?) */
25DEFINE_PER_CPU(struct gdt_page, gdt_page) = { .gdt = { 50DEFINE_PER_CPU(struct gdt_page, gdt_page) = { .gdt = {
51 [GDT_ENTRY_KERNEL32_CS] = { { { 0x0000ffff, 0x00cf9b00 } } },
52 [GDT_ENTRY_KERNEL_CS] = { { { 0x0000ffff, 0x00af9b00 } } },
53 [GDT_ENTRY_KERNEL_DS] = { { { 0x0000ffff, 0x00cf9300 } } },
54 [GDT_ENTRY_DEFAULT_USER32_CS] = { { { 0x0000ffff, 0x00cffb00 } } },
55 [GDT_ENTRY_DEFAULT_USER_DS] = { { { 0x0000ffff, 0x00cff300 } } },
56 [GDT_ENTRY_DEFAULT_USER_CS] = { { { 0x0000ffff, 0x00affb00 } } },
57} };
58#else
59DEFINE_PER_CPU_PAGE_ALIGNED(struct gdt_page, gdt_page) = { .gdt = {
26 [GDT_ENTRY_KERNEL_CS] = { { { 0x0000ffff, 0x00cf9a00 } } }, 60 [GDT_ENTRY_KERNEL_CS] = { { { 0x0000ffff, 0x00cf9a00 } } },
27 [GDT_ENTRY_KERNEL_DS] = { { { 0x0000ffff, 0x00cf9200 } } }, 61 [GDT_ENTRY_KERNEL_DS] = { { { 0x0000ffff, 0x00cf9200 } } },
28 [GDT_ENTRY_DEFAULT_USER_CS] = { { { 0x0000ffff, 0x00cffa00 } } }, 62 [GDT_ENTRY_DEFAULT_USER_CS] = { { { 0x0000ffff, 0x00cffa00 } } },
@@ -56,17 +90,150 @@ DEFINE_PER_CPU(struct gdt_page, gdt_page) = { .gdt = {
56 [GDT_ENTRY_ESPFIX_SS] = { { { 0x00000000, 0x00c09200 } } }, 90 [GDT_ENTRY_ESPFIX_SS] = { { { 0x00000000, 0x00c09200 } } },
57 [GDT_ENTRY_PERCPU] = { { { 0x00000000, 0x00000000 } } }, 91 [GDT_ENTRY_PERCPU] = { { { 0x00000000, 0x00000000 } } },
58} }; 92} };
93#endif
59EXPORT_PER_CPU_SYMBOL_GPL(gdt_page); 94EXPORT_PER_CPU_SYMBOL_GPL(gdt_page);
60 95
61__u32 cleared_cpu_caps[NCAPINTS] __cpuinitdata; 96#ifdef CONFIG_X86_32
62
63static int cachesize_override __cpuinitdata = -1; 97static int cachesize_override __cpuinitdata = -1;
64static int disable_x86_serial_nr __cpuinitdata = 1; 98static int disable_x86_serial_nr __cpuinitdata = 1;
65 99
66struct cpu_dev *cpu_devs[X86_VENDOR_NUM] = {}; 100static int __init cachesize_setup(char *str)
101{
102 get_option(&str, &cachesize_override);
103 return 1;
104}
105__setup("cachesize=", cachesize_setup);
106
107static int __init x86_fxsr_setup(char *s)
108{
109 setup_clear_cpu_cap(X86_FEATURE_FXSR);
110 setup_clear_cpu_cap(X86_FEATURE_XMM);
111 return 1;
112}
113__setup("nofxsr", x86_fxsr_setup);
114
115static int __init x86_sep_setup(char *s)
116{
117 setup_clear_cpu_cap(X86_FEATURE_SEP);
118 return 1;
119}
120__setup("nosep", x86_sep_setup);
121
122/* Standard macro to see if a specific flag is changeable */
123static inline int flag_is_changeable_p(u32 flag)
124{
125 u32 f1, f2;
126
127 asm("pushfl\n\t"
128 "pushfl\n\t"
129 "popl %0\n\t"
130 "movl %0,%1\n\t"
131 "xorl %2,%0\n\t"
132 "pushl %0\n\t"
133 "popfl\n\t"
134 "pushfl\n\t"
135 "popl %0\n\t"
136 "popfl\n\t"
137 : "=&r" (f1), "=&r" (f2)
138 : "ir" (flag));
139
140 return ((f1^f2) & flag) != 0;
141}
142
143/* Probe for the CPUID instruction */
144static int __cpuinit have_cpuid_p(void)
145{
146 return flag_is_changeable_p(X86_EFLAGS_ID);
147}
148
149static void __cpuinit squash_the_stupid_serial_number(struct cpuinfo_x86 *c)
150{
151 if (cpu_has(c, X86_FEATURE_PN) && disable_x86_serial_nr) {
152 /* Disable processor serial number */
153 unsigned long lo, hi;
154 rdmsr(MSR_IA32_BBL_CR_CTL, lo, hi);
155 lo |= 0x200000;
156 wrmsr(MSR_IA32_BBL_CR_CTL, lo, hi);
157 printk(KERN_NOTICE "CPU serial number disabled.\n");
158 clear_cpu_cap(c, X86_FEATURE_PN);
159
160 /* Disabling the serial number may affect the cpuid level */
161 c->cpuid_level = cpuid_eax(0);
162 }
163}
164
165static int __init x86_serial_nr_setup(char *s)
166{
167 disable_x86_serial_nr = 0;
168 return 1;
169}
170__setup("serialnumber", x86_serial_nr_setup);
171#else
172static inline int flag_is_changeable_p(u32 flag)
173{
174 return 1;
175}
176/* Probe for the CPUID instruction */
177static inline int have_cpuid_p(void)
178{
179 return 1;
180}
181static inline void squash_the_stupid_serial_number(struct cpuinfo_x86 *c)
182{
183}
184#endif
185
186/*
187 * Naming convention should be: <Name> [(<Codename>)]
188 * This table only is used unless init_<vendor>() below doesn't set it;
189 * in particular, if CPUID levels 0x80000002..4 are supported, this isn't used
190 *
191 */
192
193/* Look up CPU names by table lookup. */
194static char __cpuinit *table_lookup_model(struct cpuinfo_x86 *c)
195{
196 struct cpu_model_info *info;
197
198 if (c->x86_model >= 16)
199 return NULL; /* Range check */
200
201 if (!this_cpu)
202 return NULL;
203
204 info = this_cpu->c_models;
205
206 while (info && info->family) {
207 if (info->family == c->x86)
208 return info->model_names[c->x86_model];
209 info++;
210 }
211 return NULL; /* Not found */
212}
213
214__u32 cleared_cpu_caps[NCAPINTS] __cpuinitdata;
215
216/* Current gdt points %fs at the "master" per-cpu area: after this,
217 * it's on the real one. */
218void switch_to_new_gdt(void)
219{
220 struct desc_ptr gdt_descr;
221
222 gdt_descr.address = (long)get_cpu_gdt_table(smp_processor_id());
223 gdt_descr.size = GDT_SIZE - 1;
224 load_gdt(&gdt_descr);
225#ifdef CONFIG_X86_32
226 asm("mov %0, %%fs" : : "r" (__KERNEL_PERCPU) : "memory");
227#endif
228}
229
230static struct cpu_dev *cpu_devs[X86_VENDOR_NUM] = {};
67 231
68static void __cpuinit default_init(struct cpuinfo_x86 *c) 232static void __cpuinit default_init(struct cpuinfo_x86 *c)
69{ 233{
234#ifdef CONFIG_X86_64
235 display_cacheinfo(c);
236#else
70 /* Not much we can do here... */ 237 /* Not much we can do here... */
71 /* Check if at least it has cpuid */ 238 /* Check if at least it has cpuid */
72 if (c->cpuid_level == -1) { 239 if (c->cpuid_level == -1) {
@@ -76,28 +243,22 @@ static void __cpuinit default_init(struct cpuinfo_x86 *c)
76 else if (c->x86 == 3) 243 else if (c->x86 == 3)
77 strcpy(c->x86_model_id, "386"); 244 strcpy(c->x86_model_id, "386");
78 } 245 }
246#endif
79} 247}
80 248
81static struct cpu_dev __cpuinitdata default_cpu = { 249static struct cpu_dev __cpuinitdata default_cpu = {
82 .c_init = default_init, 250 .c_init = default_init,
83 .c_vendor = "Unknown", 251 .c_vendor = "Unknown",
252 .c_x86_vendor = X86_VENDOR_UNKNOWN,
84}; 253};
85static struct cpu_dev *this_cpu __cpuinitdata = &default_cpu;
86 254
87static int __init cachesize_setup(char *str) 255static void __cpuinit get_model_name(struct cpuinfo_x86 *c)
88{
89 get_option(&str, &cachesize_override);
90 return 1;
91}
92__setup("cachesize=", cachesize_setup);
93
94int __cpuinit get_model_name(struct cpuinfo_x86 *c)
95{ 256{
96 unsigned int *v; 257 unsigned int *v;
97 char *p, *q; 258 char *p, *q;
98 259
99 if (cpuid_eax(0x80000000) < 0x80000004) 260 if (c->extended_cpuid_level < 0x80000004)
100 return 0; 261 return;
101 262
102 v = (unsigned int *) c->x86_model_id; 263 v = (unsigned int *) c->x86_model_id;
103 cpuid(0x80000002, &v[0], &v[1], &v[2], &v[3]); 264 cpuid(0x80000002, &v[0], &v[1], &v[2], &v[3]);
@@ -116,30 +277,34 @@ int __cpuinit get_model_name(struct cpuinfo_x86 *c)
116 while (q <= &c->x86_model_id[48]) 277 while (q <= &c->x86_model_id[48])
117 *q++ = '\0'; /* Zero-pad the rest */ 278 *q++ = '\0'; /* Zero-pad the rest */
118 } 279 }
119
120 return 1;
121} 280}
122 281
123
124void __cpuinit display_cacheinfo(struct cpuinfo_x86 *c) 282void __cpuinit display_cacheinfo(struct cpuinfo_x86 *c)
125{ 283{
126 unsigned int n, dummy, ecx, edx, l2size; 284 unsigned int n, dummy, ebx, ecx, edx, l2size;
127 285
128 n = cpuid_eax(0x80000000); 286 n = c->extended_cpuid_level;
129 287
130 if (n >= 0x80000005) { 288 if (n >= 0x80000005) {
131 cpuid(0x80000005, &dummy, &dummy, &ecx, &edx); 289 cpuid(0x80000005, &dummy, &ebx, &ecx, &edx);
132 printk(KERN_INFO "CPU: L1 I Cache: %dK (%d bytes/line), D cache %dK (%d bytes/line)\n", 290 printk(KERN_INFO "CPU: L1 I Cache: %dK (%d bytes/line), D cache %dK (%d bytes/line)\n",
133 edx>>24, edx&0xFF, ecx>>24, ecx&0xFF); 291 edx>>24, edx&0xFF, ecx>>24, ecx&0xFF);
134 c->x86_cache_size = (ecx>>24)+(edx>>24); 292 c->x86_cache_size = (ecx>>24) + (edx>>24);
293#ifdef CONFIG_X86_64
294 /* On K8 L1 TLB is inclusive, so don't count it */
295 c->x86_tlbsize = 0;
296#endif
135 } 297 }
136 298
137 if (n < 0x80000006) /* Some chips just has a large L1. */ 299 if (n < 0x80000006) /* Some chips just has a large L1. */
138 return; 300 return;
139 301
140 ecx = cpuid_ecx(0x80000006); 302 cpuid(0x80000006, &dummy, &ebx, &ecx, &edx);
141 l2size = ecx >> 16; 303 l2size = ecx >> 16;
142 304
305#ifdef CONFIG_X86_64
306 c->x86_tlbsize += ((ebx >> 16) & 0xfff) + (ebx & 0xfff);
307#else
143 /* do processor-specific cache resizing */ 308 /* do processor-specific cache resizing */
144 if (this_cpu->c_size_cache) 309 if (this_cpu->c_size_cache)
145 l2size = this_cpu->c_size_cache(c, l2size); 310 l2size = this_cpu->c_size_cache(c, l2size);
@@ -150,116 +315,106 @@ void __cpuinit display_cacheinfo(struct cpuinfo_x86 *c)
150 315
151 if (l2size == 0) 316 if (l2size == 0)
152 return; /* Again, no L2 cache is possible */ 317 return; /* Again, no L2 cache is possible */
318#endif
153 319
154 c->x86_cache_size = l2size; 320 c->x86_cache_size = l2size;
155 321
156 printk(KERN_INFO "CPU: L2 Cache: %dK (%d bytes/line)\n", 322 printk(KERN_INFO "CPU: L2 Cache: %dK (%d bytes/line)\n",
157 l2size, ecx & 0xFF); 323 l2size, ecx & 0xFF);
158} 324}
159 325
160/* 326void __cpuinit detect_ht(struct cpuinfo_x86 *c)
161 * Naming convention should be: <Name> [(<Codename>)]
162 * This table only is used unless init_<vendor>() below doesn't set it;
163 * in particular, if CPUID levels 0x80000002..4 are supported, this isn't used
164 *
165 */
166
167/* Look up CPU names by table lookup. */
168static char __cpuinit *table_lookup_model(struct cpuinfo_x86 *c)
169{ 327{
170 struct cpu_model_info *info; 328#ifdef CONFIG_X86_HT
329 u32 eax, ebx, ecx, edx;
330 int index_msb, core_bits;
171 331
172 if (c->x86_model >= 16) 332 if (!cpu_has(c, X86_FEATURE_HT))
173 return NULL; /* Range check */ 333 return;
174 334
175 if (!this_cpu) 335 if (cpu_has(c, X86_FEATURE_CMP_LEGACY))
176 return NULL; 336 goto out;
177 337
178 info = this_cpu->c_models; 338 if (cpu_has(c, X86_FEATURE_XTOPOLOGY))
339 return;
179 340
180 while (info && info->family) { 341 cpuid(1, &eax, &ebx, &ecx, &edx);
181 if (info->family == c->x86) 342
182 return info->model_names[c->x86_model]; 343 smp_num_siblings = (ebx & 0xff0000) >> 16;
183 info++; 344
345 if (smp_num_siblings == 1) {
346 printk(KERN_INFO "CPU: Hyper-Threading is disabled\n");
347 } else if (smp_num_siblings > 1) {
348
349 if (smp_num_siblings > NR_CPUS) {
350 printk(KERN_WARNING "CPU: Unsupported number of siblings %d",
351 smp_num_siblings);
352 smp_num_siblings = 1;
353 return;
354 }
355
356 index_msb = get_count_order(smp_num_siblings);
357#ifdef CONFIG_X86_64
358 c->phys_proc_id = phys_pkg_id(index_msb);
359#else
360 c->phys_proc_id = phys_pkg_id(c->initial_apicid, index_msb);
361#endif
362
363 smp_num_siblings = smp_num_siblings / c->x86_max_cores;
364
365 index_msb = get_count_order(smp_num_siblings);
366
367 core_bits = get_count_order(c->x86_max_cores);
368
369#ifdef CONFIG_X86_64
370 c->cpu_core_id = phys_pkg_id(index_msb) &
371 ((1 << core_bits) - 1);
372#else
373 c->cpu_core_id = phys_pkg_id(c->initial_apicid, index_msb) &
374 ((1 << core_bits) - 1);
375#endif
184 } 376 }
185 return NULL; /* Not found */
186}
187 377
378out:
379 if ((c->x86_max_cores * smp_num_siblings) > 1) {
380 printk(KERN_INFO "CPU: Physical Processor ID: %d\n",
381 c->phys_proc_id);
382 printk(KERN_INFO "CPU: Processor Core ID: %d\n",
383 c->cpu_core_id);
384 }
385#endif
386}
188 387
189static void __cpuinit get_cpu_vendor(struct cpuinfo_x86 *c, int early) 388static void __cpuinit get_cpu_vendor(struct cpuinfo_x86 *c)
190{ 389{
191 char *v = c->x86_vendor_id; 390 char *v = c->x86_vendor_id;
192 int i; 391 int i;
193 static int printed; 392 static int printed;
194 393
195 for (i = 0; i < X86_VENDOR_NUM; i++) { 394 for (i = 0; i < X86_VENDOR_NUM; i++) {
196 if (cpu_devs[i]) { 395 if (!cpu_devs[i])
197 if (!strcmp(v, cpu_devs[i]->c_ident[0]) || 396 break;
198 (cpu_devs[i]->c_ident[1] && 397
199 !strcmp(v, cpu_devs[i]->c_ident[1]))) { 398 if (!strcmp(v, cpu_devs[i]->c_ident[0]) ||
200 c->x86_vendor = i; 399 (cpu_devs[i]->c_ident[1] &&
201 if (!early) 400 !strcmp(v, cpu_devs[i]->c_ident[1]))) {
202 this_cpu = cpu_devs[i]; 401 this_cpu = cpu_devs[i];
203 return; 402 c->x86_vendor = this_cpu->c_x86_vendor;
204 } 403 return;
205 } 404 }
206 } 405 }
406
207 if (!printed) { 407 if (!printed) {
208 printed++; 408 printed++;
209 printk(KERN_ERR "CPU: Vendor unknown, using generic init.\n"); 409 printk(KERN_ERR "CPU: vendor_id '%s' unknown, using generic init.\n", v);
210 printk(KERN_ERR "CPU: Your system may be unstable.\n"); 410 printk(KERN_ERR "CPU: Your system may be unstable.\n");
211 } 411 }
412
212 c->x86_vendor = X86_VENDOR_UNKNOWN; 413 c->x86_vendor = X86_VENDOR_UNKNOWN;
213 this_cpu = &default_cpu; 414 this_cpu = &default_cpu;
214} 415}
215 416
216 417void __cpuinit cpu_detect(struct cpuinfo_x86 *c)
217static int __init x86_fxsr_setup(char *s)
218{
219 setup_clear_cpu_cap(X86_FEATURE_FXSR);
220 setup_clear_cpu_cap(X86_FEATURE_XMM);
221 return 1;
222}
223__setup("nofxsr", x86_fxsr_setup);
224
225
226static int __init x86_sep_setup(char *s)
227{
228 setup_clear_cpu_cap(X86_FEATURE_SEP);
229 return 1;
230}
231__setup("nosep", x86_sep_setup);
232
233
234/* Standard macro to see if a specific flag is changeable */
235static inline int flag_is_changeable_p(u32 flag)
236{
237 u32 f1, f2;
238
239 asm("pushfl\n\t"
240 "pushfl\n\t"
241 "popl %0\n\t"
242 "movl %0,%1\n\t"
243 "xorl %2,%0\n\t"
244 "pushl %0\n\t"
245 "popfl\n\t"
246 "pushfl\n\t"
247 "popl %0\n\t"
248 "popfl\n\t"
249 : "=&r" (f1), "=&r" (f2)
250 : "ir" (flag));
251
252 return ((f1^f2) & flag) != 0;
253}
254
255
256/* Probe for the CPUID instruction */
257static int __cpuinit have_cpuid_p(void)
258{
259 return flag_is_changeable_p(X86_EFLAGS_ID);
260}
261
262void __init cpu_detect(struct cpuinfo_x86 *c)
263{ 418{
264 /* Get vendor name */ 419 /* Get vendor name */
265 cpuid(0x00000000, (unsigned int *)&c->cpuid_level, 420 cpuid(0x00000000, (unsigned int *)&c->cpuid_level,
@@ -268,50 +423,87 @@ void __init cpu_detect(struct cpuinfo_x86 *c)
268 (unsigned int *)&c->x86_vendor_id[4]); 423 (unsigned int *)&c->x86_vendor_id[4]);
269 424
270 c->x86 = 4; 425 c->x86 = 4;
426 /* Intel-defined flags: level 0x00000001 */
271 if (c->cpuid_level >= 0x00000001) { 427 if (c->cpuid_level >= 0x00000001) {
272 u32 junk, tfms, cap0, misc; 428 u32 junk, tfms, cap0, misc;
273 cpuid(0x00000001, &tfms, &misc, &junk, &cap0); 429 cpuid(0x00000001, &tfms, &misc, &junk, &cap0);
274 c->x86 = (tfms >> 8) & 15; 430 c->x86 = (tfms >> 8) & 0xf;
275 c->x86_model = (tfms >> 4) & 15; 431 c->x86_model = (tfms >> 4) & 0xf;
432 c->x86_mask = tfms & 0xf;
276 if (c->x86 == 0xf) 433 if (c->x86 == 0xf)
277 c->x86 += (tfms >> 20) & 0xff; 434 c->x86 += (tfms >> 20) & 0xff;
278 if (c->x86 >= 0x6) 435 if (c->x86 >= 0x6)
279 c->x86_model += ((tfms >> 16) & 0xF) << 4; 436 c->x86_model += ((tfms >> 16) & 0xf) << 4;
280 c->x86_mask = tfms & 15;
281 if (cap0 & (1<<19)) { 437 if (cap0 & (1<<19)) {
282 c->x86_cache_alignment = ((misc >> 8) & 0xff) * 8;
283 c->x86_clflush_size = ((misc >> 8) & 0xff) * 8; 438 c->x86_clflush_size = ((misc >> 8) & 0xff) * 8;
439 c->x86_cache_alignment = c->x86_clflush_size;
284 } 440 }
285 } 441 }
286} 442}
287static void __cpuinit early_get_cap(struct cpuinfo_x86 *c) 443
444static void __cpuinit get_cpu_cap(struct cpuinfo_x86 *c)
288{ 445{
289 u32 tfms, xlvl; 446 u32 tfms, xlvl;
290 unsigned int ebx; 447 u32 ebx;
291 448
292 memset(&c->x86_capability, 0, sizeof c->x86_capability); 449 /* Intel-defined flags: level 0x00000001 */
293 if (have_cpuid_p()) { 450 if (c->cpuid_level >= 0x00000001) {
294 /* Intel-defined flags: level 0x00000001 */ 451 u32 capability, excap;
295 if (c->cpuid_level >= 0x00000001) { 452 cpuid(0x00000001, &tfms, &ebx, &excap, &capability);
296 u32 capability, excap; 453 c->x86_capability[0] = capability;
297 cpuid(0x00000001, &tfms, &ebx, &excap, &capability); 454 c->x86_capability[4] = excap;
298 c->x86_capability[0] = capability; 455 }
299 c->x86_capability[4] = excap;
300 }
301 456
302 /* AMD-defined flags: level 0x80000001 */ 457 /* AMD-defined flags: level 0x80000001 */
303 xlvl = cpuid_eax(0x80000000); 458 xlvl = cpuid_eax(0x80000000);
304 if ((xlvl & 0xffff0000) == 0x80000000) { 459 c->extended_cpuid_level = xlvl;
305 if (xlvl >= 0x80000001) { 460 if ((xlvl & 0xffff0000) == 0x80000000) {
306 c->x86_capability[1] = cpuid_edx(0x80000001); 461 if (xlvl >= 0x80000001) {
307 c->x86_capability[6] = cpuid_ecx(0x80000001); 462 c->x86_capability[1] = cpuid_edx(0x80000001);
308 } 463 c->x86_capability[6] = cpuid_ecx(0x80000001);
309 } 464 }
465 }
310 466
467#ifdef CONFIG_X86_64
468 if (c->extended_cpuid_level >= 0x80000008) {
469 u32 eax = cpuid_eax(0x80000008);
470
471 c->x86_virt_bits = (eax >> 8) & 0xff;
472 c->x86_phys_bits = eax & 0xff;
311 } 473 }
474#endif
475
476 if (c->extended_cpuid_level >= 0x80000007)
477 c->x86_power = cpuid_edx(0x80000007);
312 478
313} 479}
314 480
481static void __cpuinit identify_cpu_without_cpuid(struct cpuinfo_x86 *c)
482{
483#ifdef CONFIG_X86_32
484 int i;
485
486 /*
487 * First of all, decide if this is a 486 or higher
488 * It's a 486 if we can modify the AC flag
489 */
490 if (flag_is_changeable_p(X86_EFLAGS_AC))
491 c->x86 = 4;
492 else
493 c->x86 = 3;
494
495 for (i = 0; i < X86_VENDOR_NUM; i++)
496 if (cpu_devs[i] && cpu_devs[i]->c_identify) {
497 c->x86_vendor_id[0] = 0;
498 cpu_devs[i]->c_identify(c);
499 if (c->x86_vendor_id[0]) {
500 get_cpu_vendor(c);
501 break;
502 }
503 }
504#endif
505}
506
315/* 507/*
316 * Do minimum CPU detection early. 508 * Do minimum CPU detection early.
317 * Fields really needed: vendor, cpuid_level, family, model, mask, 509 * Fields really needed: vendor, cpuid_level, family, model, mask,
@@ -321,25 +513,61 @@ static void __cpuinit early_get_cap(struct cpuinfo_x86 *c)
321 * WARNING: this function is only called on the BP. Don't add code here 513 * WARNING: this function is only called on the BP. Don't add code here
322 * that is supposed to run on all CPUs. 514 * that is supposed to run on all CPUs.
323 */ 515 */
324static void __init early_cpu_detect(void) 516static void __init early_identify_cpu(struct cpuinfo_x86 *c)
325{ 517{
326 struct cpuinfo_x86 *c = &boot_cpu_data; 518#ifdef CONFIG_X86_64
327 519 c->x86_clflush_size = 64;
328 c->x86_cache_alignment = 32; 520#else
329 c->x86_clflush_size = 32; 521 c->x86_clflush_size = 32;
522#endif
523 c->x86_cache_alignment = c->x86_clflush_size;
524
525 memset(&c->x86_capability, 0, sizeof c->x86_capability);
526 c->extended_cpuid_level = 0;
330 527
331 if (!have_cpuid_p()) 528 if (!have_cpuid_p())
529 identify_cpu_without_cpuid(c);
530
531 /* cyrix could have cpuid enabled via c_identify()*/
532 if (!have_cpuid_p())
332 return; 533 return;
333 534
334 cpu_detect(c); 535 cpu_detect(c);
335 536
336 get_cpu_vendor(c, 1); 537 get_cpu_vendor(c);
337 538
338 early_get_cap(c); 539 get_cpu_cap(c);
339 540
340 if (c->x86_vendor != X86_VENDOR_UNKNOWN && 541 if (this_cpu->c_early_init)
341 cpu_devs[c->x86_vendor]->c_early_init) 542 this_cpu->c_early_init(c);
342 cpu_devs[c->x86_vendor]->c_early_init(c); 543
544 validate_pat_support(c);
545}
546
547void __init early_cpu_init(void)
548{
549 struct cpu_dev **cdev;
550 int count = 0;
551
552 printk("KERNEL supported cpus:\n");
553 for (cdev = __x86_cpu_dev_start; cdev < __x86_cpu_dev_end; cdev++) {
554 struct cpu_dev *cpudev = *cdev;
555 unsigned int j;
556
557 if (count >= X86_VENDOR_NUM)
558 break;
559 cpu_devs[count] = cpudev;
560 count++;
561
562 for (j = 0; j < 2; j++) {
563 if (!cpudev->c_ident[j])
564 continue;
565 printk(" %s %s\n", cpudev->c_vendor,
566 cpudev->c_ident[j]);
567 }
568 }
569
570 early_identify_cpu(&boot_cpu_data);
343} 571}
344 572
345/* 573/*
@@ -357,86 +585,41 @@ static void __cpuinit detect_nopl(struct cpuinfo_x86 *c)
357 585
358static void __cpuinit generic_identify(struct cpuinfo_x86 *c) 586static void __cpuinit generic_identify(struct cpuinfo_x86 *c)
359{ 587{
360 u32 tfms, xlvl; 588 c->extended_cpuid_level = 0;
361 unsigned int ebx;
362
363 if (have_cpuid_p()) {
364 /* Get vendor name */
365 cpuid(0x00000000, (unsigned int *)&c->cpuid_level,
366 (unsigned int *)&c->x86_vendor_id[0],
367 (unsigned int *)&c->x86_vendor_id[8],
368 (unsigned int *)&c->x86_vendor_id[4]);
369
370 get_cpu_vendor(c, 0);
371 /* Initialize the standard set of capabilities */
372 /* Note that the vendor-specific code below might override */
373 /* Intel-defined flags: level 0x00000001 */
374 if (c->cpuid_level >= 0x00000001) {
375 u32 capability, excap;
376 cpuid(0x00000001, &tfms, &ebx, &excap, &capability);
377 c->x86_capability[0] = capability;
378 c->x86_capability[4] = excap;
379 c->x86 = (tfms >> 8) & 15;
380 c->x86_model = (tfms >> 4) & 15;
381 if (c->x86 == 0xf)
382 c->x86 += (tfms >> 20) & 0xff;
383 if (c->x86 >= 0x6)
384 c->x86_model += ((tfms >> 16) & 0xF) << 4;
385 c->x86_mask = tfms & 15;
386 c->initial_apicid = (ebx >> 24) & 0xFF;
387#ifdef CONFIG_X86_HT
388 c->apicid = phys_pkg_id(c->initial_apicid, 0);
389 c->phys_proc_id = c->initial_apicid;
390#else
391 c->apicid = c->initial_apicid;
392#endif
393 if (test_cpu_cap(c, X86_FEATURE_CLFLSH))
394 c->x86_clflush_size = ((ebx >> 8) & 0xff) * 8;
395 } else {
396 /* Have CPUID level 0 only - unheard of */
397 c->x86 = 4;
398 }
399 589
400 /* AMD-defined flags: level 0x80000001 */ 590 if (!have_cpuid_p())
401 xlvl = cpuid_eax(0x80000000); 591 identify_cpu_without_cpuid(c);
402 if ((xlvl & 0xffff0000) == 0x80000000) {
403 if (xlvl >= 0x80000001) {
404 c->x86_capability[1] = cpuid_edx(0x80000001);
405 c->x86_capability[6] = cpuid_ecx(0x80000001);
406 }
407 if (xlvl >= 0x80000004)
408 get_model_name(c); /* Default name */
409 }
410 592
411 init_scattered_cpuid_features(c); 593 /* cyrix could have cpuid enabled via c_identify()*/
412 detect_nopl(c); 594 if (!have_cpuid_p())
413 } 595 return;
414}
415 596
416static void __cpuinit squash_the_stupid_serial_number(struct cpuinfo_x86 *c) 597 cpu_detect(c);
417{
418 if (cpu_has(c, X86_FEATURE_PN) && disable_x86_serial_nr) {
419 /* Disable processor serial number */
420 unsigned long lo, hi;
421 rdmsr(MSR_IA32_BBL_CR_CTL, lo, hi);
422 lo |= 0x200000;
423 wrmsr(MSR_IA32_BBL_CR_CTL, lo, hi);
424 printk(KERN_NOTICE "CPU serial number disabled.\n");
425 clear_cpu_cap(c, X86_FEATURE_PN);
426 598
427 /* Disabling the serial number may affect the cpuid level */ 599 get_cpu_vendor(c);
428 c->cpuid_level = cpuid_eax(0);
429 }
430}
431 600
432static int __init x86_serial_nr_setup(char *s) 601 get_cpu_cap(c);
433{
434 disable_x86_serial_nr = 0;
435 return 1;
436}
437__setup("serialnumber", x86_serial_nr_setup);
438 602
603 if (c->cpuid_level >= 0x00000001) {
604 c->initial_apicid = (cpuid_ebx(1) >> 24) & 0xFF;
605#ifdef CONFIG_X86_32
606# ifdef CONFIG_X86_HT
607 c->apicid = phys_pkg_id(c->initial_apicid, 0);
608# else
609 c->apicid = c->initial_apicid;
610# endif
611#endif
439 612
613#ifdef CONFIG_X86_HT
614 c->phys_proc_id = c->initial_apicid;
615#endif
616 }
617
618 get_model_name(c); /* Default name */
619
620 init_scattered_cpuid_features(c);
621 detect_nopl(c);
622}
440 623
441/* 624/*
442 * This does the hard work of actually picking apart the CPU stuff... 625 * This does the hard work of actually picking apart the CPU stuff...
@@ -448,30 +631,29 @@ static void __cpuinit identify_cpu(struct cpuinfo_x86 *c)
448 c->loops_per_jiffy = loops_per_jiffy; 631 c->loops_per_jiffy = loops_per_jiffy;
449 c->x86_cache_size = -1; 632 c->x86_cache_size = -1;
450 c->x86_vendor = X86_VENDOR_UNKNOWN; 633 c->x86_vendor = X86_VENDOR_UNKNOWN;
451 c->cpuid_level = -1; /* CPUID not detected */
452 c->x86_model = c->x86_mask = 0; /* So far unknown... */ 634 c->x86_model = c->x86_mask = 0; /* So far unknown... */
453 c->x86_vendor_id[0] = '\0'; /* Unset */ 635 c->x86_vendor_id[0] = '\0'; /* Unset */
454 c->x86_model_id[0] = '\0'; /* Unset */ 636 c->x86_model_id[0] = '\0'; /* Unset */
455 c->x86_max_cores = 1; 637 c->x86_max_cores = 1;
638 c->x86_coreid_bits = 0;
639#ifdef CONFIG_X86_64
640 c->x86_clflush_size = 64;
641#else
642 c->cpuid_level = -1; /* CPUID not detected */
456 c->x86_clflush_size = 32; 643 c->x86_clflush_size = 32;
644#endif
645 c->x86_cache_alignment = c->x86_clflush_size;
457 memset(&c->x86_capability, 0, sizeof c->x86_capability); 646 memset(&c->x86_capability, 0, sizeof c->x86_capability);
458 647
459 if (!have_cpuid_p()) {
460 /*
461 * First of all, decide if this is a 486 or higher
462 * It's a 486 if we can modify the AC flag
463 */
464 if (flag_is_changeable_p(X86_EFLAGS_AC))
465 c->x86 = 4;
466 else
467 c->x86 = 3;
468 }
469
470 generic_identify(c); 648 generic_identify(c);
471 649
472 if (this_cpu->c_identify) 650 if (this_cpu->c_identify)
473 this_cpu->c_identify(c); 651 this_cpu->c_identify(c);
474 652
653#ifdef CONFIG_X86_64
654 c->apicid = phys_pkg_id(0);
655#endif
656
475 /* 657 /*
476 * Vendor-specific initialization. In this section we 658 * Vendor-specific initialization. In this section we
477 * canonicalize the feature flags, meaning if there are 659 * canonicalize the feature flags, meaning if there are
@@ -505,6 +687,10 @@ static void __cpuinit identify_cpu(struct cpuinfo_x86 *c)
505 c->x86, c->x86_model); 687 c->x86, c->x86_model);
506 } 688 }
507 689
690#ifdef CONFIG_X86_64
691 detect_ht(c);
692#endif
693
508 /* 694 /*
509 * On SMP, boot_cpu_data holds the common feature set between 695 * On SMP, boot_cpu_data holds the common feature set between
510 * all CPUs; so make sure that we indicate which features are 696 * all CPUs; so make sure that we indicate which features are
@@ -513,7 +699,7 @@ static void __cpuinit identify_cpu(struct cpuinfo_x86 *c)
513 */ 699 */
514 if (c != &boot_cpu_data) { 700 if (c != &boot_cpu_data) {
515 /* AND the already accumulated flags with these */ 701 /* AND the already accumulated flags with these */
516 for (i = 0 ; i < NCAPINTS ; i++) 702 for (i = 0; i < NCAPINTS; i++)
517 boot_cpu_data.x86_capability[i] &= c->x86_capability[i]; 703 boot_cpu_data.x86_capability[i] &= c->x86_capability[i];
518 } 704 }
519 705
@@ -521,72 +707,79 @@ static void __cpuinit identify_cpu(struct cpuinfo_x86 *c)
521 for (i = 0; i < NCAPINTS; i++) 707 for (i = 0; i < NCAPINTS; i++)
522 c->x86_capability[i] &= ~cleared_cpu_caps[i]; 708 c->x86_capability[i] &= ~cleared_cpu_caps[i];
523 709
710#ifdef CONFIG_X86_MCE
524 /* Init Machine Check Exception if available. */ 711 /* Init Machine Check Exception if available. */
525 mcheck_init(c); 712 mcheck_init(c);
713#endif
526 714
527 select_idle_routine(c); 715 select_idle_routine(c);
716
717#if defined(CONFIG_NUMA) && defined(CONFIG_X86_64)
718 numa_add_cpu(smp_processor_id());
719#endif
528} 720}
529 721
530void __init identify_boot_cpu(void) 722void __init identify_boot_cpu(void)
531{ 723{
532 identify_cpu(&boot_cpu_data); 724 identify_cpu(&boot_cpu_data);
725#ifdef CONFIG_X86_32
533 sysenter_setup(); 726 sysenter_setup();
534 enable_sep_cpu(); 727 enable_sep_cpu();
728#endif
535} 729}
536 730
537void __cpuinit identify_secondary_cpu(struct cpuinfo_x86 *c) 731void __cpuinit identify_secondary_cpu(struct cpuinfo_x86 *c)
538{ 732{
539 BUG_ON(c == &boot_cpu_data); 733 BUG_ON(c == &boot_cpu_data);
540 identify_cpu(c); 734 identify_cpu(c);
735#ifdef CONFIG_X86_32
541 enable_sep_cpu(); 736 enable_sep_cpu();
737#endif
542 mtrr_ap_init(); 738 mtrr_ap_init();
543} 739}
544 740
545#ifdef CONFIG_X86_HT 741struct msr_range {
546void __cpuinit detect_ht(struct cpuinfo_x86 *c) 742 unsigned min;
547{ 743 unsigned max;
548 u32 eax, ebx, ecx, edx; 744};
549 int index_msb, core_bits;
550
551 cpuid(1, &eax, &ebx, &ecx, &edx);
552
553 if (!cpu_has(c, X86_FEATURE_HT) || cpu_has(c, X86_FEATURE_CMP_LEGACY))
554 return;
555
556 smp_num_siblings = (ebx & 0xff0000) >> 16;
557 745
558 if (smp_num_siblings == 1) { 746static struct msr_range msr_range_array[] __cpuinitdata = {
559 printk(KERN_INFO "CPU: Hyper-Threading is disabled\n"); 747 { 0x00000000, 0x00000418},
560 } else if (smp_num_siblings > 1) { 748 { 0xc0000000, 0xc000040b},
749 { 0xc0010000, 0xc0010142},
750 { 0xc0011000, 0xc001103b},
751};
561 752
562 if (smp_num_siblings > NR_CPUS) { 753static void __cpuinit print_cpu_msr(void)
563 printk(KERN_WARNING "CPU: Unsupported number of the " 754{
564 "siblings %d", smp_num_siblings); 755 unsigned index;
565 smp_num_siblings = 1; 756 u64 val;
566 return; 757 int i;
758 unsigned index_min, index_max;
759
760 for (i = 0; i < ARRAY_SIZE(msr_range_array); i++) {
761 index_min = msr_range_array[i].min;
762 index_max = msr_range_array[i].max;
763 for (index = index_min; index < index_max; index++) {
764 if (rdmsrl_amd_safe(index, &val))
765 continue;
766 printk(KERN_INFO " MSR%08x: %016llx\n", index, val);
567 } 767 }
768 }
769}
568 770
569 index_msb = get_count_order(smp_num_siblings); 771static int show_msr __cpuinitdata;
570 c->phys_proc_id = phys_pkg_id(c->initial_apicid, index_msb); 772static __init int setup_show_msr(char *arg)
571 773{
572 printk(KERN_INFO "CPU: Physical Processor ID: %d\n", 774 int num;
573 c->phys_proc_id);
574
575 smp_num_siblings = smp_num_siblings / c->x86_max_cores;
576
577 index_msb = get_count_order(smp_num_siblings) ;
578 775
579 core_bits = get_count_order(c->x86_max_cores); 776 get_option(&arg, &num);
580 777
581 c->cpu_core_id = phys_pkg_id(c->initial_apicid, index_msb) & 778 if (num > 0)
582 ((1 << core_bits) - 1); 779 show_msr = num;
583 780 return 1;
584 if (c->x86_max_cores > 1)
585 printk(KERN_INFO "CPU: Processor Core ID: %d\n",
586 c->cpu_core_id);
587 }
588} 781}
589#endif 782__setup("show_msr=", setup_show_msr);
590 783
591static __init int setup_noclflush(char *arg) 784static __init int setup_noclflush(char *arg)
592{ 785{
@@ -605,17 +798,25 @@ void __cpuinit print_cpu_info(struct cpuinfo_x86 *c)
605 vendor = c->x86_vendor_id; 798 vendor = c->x86_vendor_id;
606 799
607 if (vendor && strncmp(c->x86_model_id, vendor, strlen(vendor))) 800 if (vendor && strncmp(c->x86_model_id, vendor, strlen(vendor)))
608 printk("%s ", vendor); 801 printk(KERN_CONT "%s ", vendor);
609 802
610 if (!c->x86_model_id[0]) 803 if (c->x86_model_id[0])
611 printk("%d86", c->x86); 804 printk(KERN_CONT "%s", c->x86_model_id);
612 else 805 else
613 printk("%s", c->x86_model_id); 806 printk(KERN_CONT "%d86", c->x86);
614 807
615 if (c->x86_mask || c->cpuid_level >= 0) 808 if (c->x86_mask || c->cpuid_level >= 0)
616 printk(" stepping %02x\n", c->x86_mask); 809 printk(KERN_CONT " stepping %02x\n", c->x86_mask);
617 else 810 else
618 printk("\n"); 811 printk(KERN_CONT "\n");
812
813#ifdef CONFIG_SMP
814 if (c->cpu_index < show_msr)
815 print_cpu_msr();
816#else
817 if (show_msr)
818 print_cpu_msr();
819#endif
619} 820}
620 821
621static __init int setup_disablecpuid(char *arg) 822static __init int setup_disablecpuid(char *arg)
@@ -631,19 +832,89 @@ __setup("clearcpuid=", setup_disablecpuid);
631 832
632cpumask_t cpu_initialized __cpuinitdata = CPU_MASK_NONE; 833cpumask_t cpu_initialized __cpuinitdata = CPU_MASK_NONE;
633 834
634void __init early_cpu_init(void) 835#ifdef CONFIG_X86_64
836struct x8664_pda **_cpu_pda __read_mostly;
837EXPORT_SYMBOL(_cpu_pda);
838
839struct desc_ptr idt_descr = { 256 * 16 - 1, (unsigned long) idt_table };
840
841char boot_cpu_stack[IRQSTACKSIZE] __page_aligned_bss;
842
843void __cpuinit pda_init(int cpu)
844{
845 struct x8664_pda *pda = cpu_pda(cpu);
846
847 /* Setup up data that may be needed in __get_free_pages early */
848 loadsegment(fs, 0);
849 loadsegment(gs, 0);
850 /* Memory clobbers used to order PDA accessed */
851 mb();
852 wrmsrl(MSR_GS_BASE, pda);
853 mb();
854
855 pda->cpunumber = cpu;
856 pda->irqcount = -1;
857 pda->kernelstack = (unsigned long)stack_thread_info() -
858 PDA_STACKOFFSET + THREAD_SIZE;
859 pda->active_mm = &init_mm;
860 pda->mmu_state = 0;
861
862 if (cpu == 0) {
863 /* others are initialized in smpboot.c */
864 pda->pcurrent = &init_task;
865 pda->irqstackptr = boot_cpu_stack;
866 pda->irqstackptr += IRQSTACKSIZE - 64;
867 } else {
868 if (!pda->irqstackptr) {
869 pda->irqstackptr = (char *)
870 __get_free_pages(GFP_ATOMIC, IRQSTACK_ORDER);
871 if (!pda->irqstackptr)
872 panic("cannot allocate irqstack for cpu %d",
873 cpu);
874 pda->irqstackptr += IRQSTACKSIZE - 64;
875 }
876
877 if (pda->nodenumber == 0 && cpu_to_node(cpu) != NUMA_NO_NODE)
878 pda->nodenumber = cpu_to_node(cpu);
879 }
880}
881
882char boot_exception_stacks[(N_EXCEPTION_STACKS - 1) * EXCEPTION_STKSZ +
883 DEBUG_STKSZ] __page_aligned_bss;
884
885extern asmlinkage void ignore_sysret(void);
886
887/* May not be marked __init: used by software suspend */
888void syscall_init(void)
635{ 889{
636 struct cpu_vendor_dev *cvdev; 890 /*
891 * LSTAR and STAR live in a bit strange symbiosis.
892 * They both write to the same internal register. STAR allows to
893 * set CS/DS but only a 32bit target. LSTAR sets the 64bit rip.
894 */
895 wrmsrl(MSR_STAR, ((u64)__USER32_CS)<<48 | ((u64)__KERNEL_CS)<<32);
896 wrmsrl(MSR_LSTAR, system_call);
897 wrmsrl(MSR_CSTAR, ignore_sysret);
637 898
638 for (cvdev = __x86cpuvendor_start ; 899#ifdef CONFIG_IA32_EMULATION
639 cvdev < __x86cpuvendor_end ; 900 syscall32_cpu_init();
640 cvdev++) 901#endif
641 cpu_devs[cvdev->vendor] = cvdev->cpu_dev;
642 902
643 early_cpu_detect(); 903 /* Flags to clear on syscall */
644 validate_pat_support(&boot_cpu_data); 904 wrmsrl(MSR_SYSCALL_MASK,
905 X86_EFLAGS_TF|X86_EFLAGS_DF|X86_EFLAGS_IF|X86_EFLAGS_IOPL);
645} 906}
646 907
908unsigned long kernel_eflags;
909
910/*
911 * Copies of the original ist values from the tss are only accessed during
912 * debugging, no special alignment required.
913 */
914DEFINE_PER_CPU(struct orig_ist, orig_ist);
915
916#else
917
647/* Make sure %fs is initialized properly in idle threads */ 918/* Make sure %fs is initialized properly in idle threads */
648struct pt_regs * __cpuinit idle_regs(struct pt_regs *regs) 919struct pt_regs * __cpuinit idle_regs(struct pt_regs *regs)
649{ 920{
@@ -651,25 +922,136 @@ struct pt_regs * __cpuinit idle_regs(struct pt_regs *regs)
651 regs->fs = __KERNEL_PERCPU; 922 regs->fs = __KERNEL_PERCPU;
652 return regs; 923 return regs;
653} 924}
654 925#endif
655/* Current gdt points %fs at the "master" per-cpu area: after this,
656 * it's on the real one. */
657void switch_to_new_gdt(void)
658{
659 struct desc_ptr gdt_descr;
660
661 gdt_descr.address = (long)get_cpu_gdt_table(smp_processor_id());
662 gdt_descr.size = GDT_SIZE - 1;
663 load_gdt(&gdt_descr);
664 asm("mov %0, %%fs" : : "r" (__KERNEL_PERCPU) : "memory");
665}
666 926
667/* 927/*
668 * cpu_init() initializes state that is per-CPU. Some data is already 928 * cpu_init() initializes state that is per-CPU. Some data is already
669 * initialized (naturally) in the bootstrap process, such as the GDT 929 * initialized (naturally) in the bootstrap process, such as the GDT
670 * and IDT. We reload them nevertheless, this function acts as a 930 * and IDT. We reload them nevertheless, this function acts as a
671 * 'CPU state barrier', nothing should get across. 931 * 'CPU state barrier', nothing should get across.
932 * A lot of state is already set up in PDA init for 64 bit
672 */ 933 */
934#ifdef CONFIG_X86_64
935void __cpuinit cpu_init(void)
936{
937 int cpu = stack_smp_processor_id();
938 struct tss_struct *t = &per_cpu(init_tss, cpu);
939 struct orig_ist *orig_ist = &per_cpu(orig_ist, cpu);
940 unsigned long v;
941 char *estacks = NULL;
942 struct task_struct *me;
943 int i;
944
945 /* CPU 0 is initialised in head64.c */
946 if (cpu != 0)
947 pda_init(cpu);
948 else
949 estacks = boot_exception_stacks;
950
951 me = current;
952
953 if (cpu_test_and_set(cpu, cpu_initialized))
954 panic("CPU#%d already initialized!\n", cpu);
955
956 printk(KERN_INFO "Initializing CPU#%d\n", cpu);
957
958 clear_in_cr4(X86_CR4_VME|X86_CR4_PVI|X86_CR4_TSD|X86_CR4_DE);
959
960 /*
961 * Initialize the per-CPU GDT with the boot GDT,
962 * and set up the GDT descriptor:
963 */
964
965 switch_to_new_gdt();
966 load_idt((const struct desc_ptr *)&idt_descr);
967
968 memset(me->thread.tls_array, 0, GDT_ENTRY_TLS_ENTRIES * 8);
969 syscall_init();
970
971 wrmsrl(MSR_FS_BASE, 0);
972 wrmsrl(MSR_KERNEL_GS_BASE, 0);
973 barrier();
974
975 check_efer();
976 if (cpu != 0 && x2apic)
977 enable_x2apic();
978
979 /*
980 * set up and load the per-CPU TSS
981 */
982 if (!orig_ist->ist[0]) {
983 static const unsigned int order[N_EXCEPTION_STACKS] = {
984 [0 ... N_EXCEPTION_STACKS - 1] = EXCEPTION_STACK_ORDER,
985 [DEBUG_STACK - 1] = DEBUG_STACK_ORDER
986 };
987 for (v = 0; v < N_EXCEPTION_STACKS; v++) {
988 if (cpu) {
989 estacks = (char *)__get_free_pages(GFP_ATOMIC, order[v]);
990 if (!estacks)
991 panic("Cannot allocate exception "
992 "stack %ld %d\n", v, cpu);
993 }
994 estacks += PAGE_SIZE << order[v];
995 orig_ist->ist[v] = t->x86_tss.ist[v] =
996 (unsigned long)estacks;
997 }
998 }
999
1000 t->x86_tss.io_bitmap_base = offsetof(struct tss_struct, io_bitmap);
1001 /*
1002 * <= is required because the CPU will access up to
1003 * 8 bits beyond the end of the IO permission bitmap.
1004 */
1005 for (i = 0; i <= IO_BITMAP_LONGS; i++)
1006 t->io_bitmap[i] = ~0UL;
1007
1008 atomic_inc(&init_mm.mm_count);
1009 me->active_mm = &init_mm;
1010 if (me->mm)
1011 BUG();
1012 enter_lazy_tlb(&init_mm, me);
1013
1014 load_sp0(t, &current->thread);
1015 set_tss_desc(cpu, t);
1016 load_TR_desc();
1017 load_LDT(&init_mm.context);
1018
1019#ifdef CONFIG_KGDB
1020 /*
1021 * If the kgdb is connected no debug regs should be altered. This
1022 * is only applicable when KGDB and a KGDB I/O module are built
1023 * into the kernel and you are using early debugging with
1024 * kgdbwait. KGDB will control the kernel HW breakpoint registers.
1025 */
1026 if (kgdb_connected && arch_kgdb_ops.correct_hw_break)
1027 arch_kgdb_ops.correct_hw_break();
1028 else {
1029#endif
1030 /*
1031 * Clear all 6 debug registers:
1032 */
1033
1034 set_debugreg(0UL, 0);
1035 set_debugreg(0UL, 1);
1036 set_debugreg(0UL, 2);
1037 set_debugreg(0UL, 3);
1038 set_debugreg(0UL, 6);
1039 set_debugreg(0UL, 7);
1040#ifdef CONFIG_KGDB
1041 /* If the kgdb is connected no debug regs should be altered. */
1042 }
1043#endif
1044
1045 fpu_init();
1046
1047 raw_local_save_flags(kernel_eflags);
1048
1049 if (is_uv_system())
1050 uv_cpu_init();
1051}
1052
1053#else
1054
673void __cpuinit cpu_init(void) 1055void __cpuinit cpu_init(void)
674{ 1056{
675 int cpu = smp_processor_id(); 1057 int cpu = smp_processor_id();
@@ -723,9 +1105,20 @@ void __cpuinit cpu_init(void)
723 /* 1105 /*
724 * Force FPU initialization: 1106 * Force FPU initialization:
725 */ 1107 */
726 current_thread_info()->status = 0; 1108 if (cpu_has_xsave)
1109 current_thread_info()->status = TS_XSAVE;
1110 else
1111 current_thread_info()->status = 0;
727 clear_used_math(); 1112 clear_used_math();
728 mxcsr_feature_mask_init(); 1113 mxcsr_feature_mask_init();
1114
1115 /*
1116 * Boot processor to setup the FP and extended state context info.
1117 */
1118 if (!smp_processor_id())
1119 init_thread_xstate();
1120
1121 xsave_init();
729} 1122}
730 1123
731#ifdef CONFIG_HOTPLUG_CPU 1124#ifdef CONFIG_HOTPLUG_CPU
@@ -739,3 +1132,5 @@ void __cpuinit cpu_uninit(void)
739 per_cpu(cpu_tlbstate, cpu).active_mm = &init_mm; 1132 per_cpu(cpu_tlbstate, cpu).active_mm = &init_mm;
740} 1133}
741#endif 1134#endif
1135
1136#endif
diff --git a/arch/x86/kernel/cpu/common_64.c b/arch/x86/kernel/cpu/common_64.c
deleted file mode 100644
index 43f1aa51da5d..000000000000
--- a/arch/x86/kernel/cpu/common_64.c
+++ /dev/null
@@ -1,765 +0,0 @@
1#include <linux/init.h>
2#include <linux/kernel.h>
3#include <linux/sched.h>
4#include <linux/string.h>
5#include <linux/bootmem.h>
6#include <linux/bitops.h>
7#include <linux/module.h>
8#include <linux/kgdb.h>
9#include <linux/topology.h>
10#include <linux/delay.h>
11#include <linux/smp.h>
12#include <linux/percpu.h>
13#include <asm/i387.h>
14#include <asm/msr.h>
15#include <asm/io.h>
16#include <asm/linkage.h>
17#include <asm/mmu_context.h>
18#include <asm/mtrr.h>
19#include <asm/mce.h>
20#include <asm/pat.h>
21#include <asm/asm.h>
22#include <asm/numa.h>
23#ifdef CONFIG_X86_LOCAL_APIC
24#include <asm/mpspec.h>
25#include <asm/apic.h>
26#include <mach_apic.h>
27#endif
28#include <asm/pda.h>
29#include <asm/pgtable.h>
30#include <asm/processor.h>
31#include <asm/desc.h>
32#include <asm/atomic.h>
33#include <asm/proto.h>
34#include <asm/sections.h>
35#include <asm/setup.h>
36#include <asm/genapic.h>
37
38#include "cpu.h"
39
40/* We need valid kernel segments for data and code in long mode too
41 * IRET will check the segment types kkeil 2000/10/28
42 * Also sysret mandates a special GDT layout
43 */
44/* The TLS descriptors are currently at a different place compared to i386.
45 Hopefully nobody expects them at a fixed place (Wine?) */
46DEFINE_PER_CPU(struct gdt_page, gdt_page) = { .gdt = {
47 [GDT_ENTRY_KERNEL32_CS] = { { { 0x0000ffff, 0x00cf9b00 } } },
48 [GDT_ENTRY_KERNEL_CS] = { { { 0x0000ffff, 0x00af9b00 } } },
49 [GDT_ENTRY_KERNEL_DS] = { { { 0x0000ffff, 0x00cf9300 } } },
50 [GDT_ENTRY_DEFAULT_USER32_CS] = { { { 0x0000ffff, 0x00cffb00 } } },
51 [GDT_ENTRY_DEFAULT_USER_DS] = { { { 0x0000ffff, 0x00cff300 } } },
52 [GDT_ENTRY_DEFAULT_USER_CS] = { { { 0x0000ffff, 0x00affb00 } } },
53} };
54EXPORT_PER_CPU_SYMBOL_GPL(gdt_page);
55
56__u32 cleared_cpu_caps[NCAPINTS] __cpuinitdata;
57
58/* Current gdt points %fs at the "master" per-cpu area: after this,
59 * it's on the real one. */
60void switch_to_new_gdt(void)
61{
62 struct desc_ptr gdt_descr;
63
64 gdt_descr.address = (long)get_cpu_gdt_table(smp_processor_id());
65 gdt_descr.size = GDT_SIZE - 1;
66 load_gdt(&gdt_descr);
67}
68
69struct cpu_dev *cpu_devs[X86_VENDOR_NUM] = {};
70
71static void __cpuinit default_init(struct cpuinfo_x86 *c)
72{
73 display_cacheinfo(c);
74}
75
76static struct cpu_dev __cpuinitdata default_cpu = {
77 .c_init = default_init,
78 .c_vendor = "Unknown",
79};
80static struct cpu_dev *this_cpu __cpuinitdata = &default_cpu;
81
82int __cpuinit get_model_name(struct cpuinfo_x86 *c)
83{
84 unsigned int *v;
85
86 if (c->extended_cpuid_level < 0x80000004)
87 return 0;
88
89 v = (unsigned int *) c->x86_model_id;
90 cpuid(0x80000002, &v[0], &v[1], &v[2], &v[3]);
91 cpuid(0x80000003, &v[4], &v[5], &v[6], &v[7]);
92 cpuid(0x80000004, &v[8], &v[9], &v[10], &v[11]);
93 c->x86_model_id[48] = 0;
94 return 1;
95}
96
97
98void __cpuinit display_cacheinfo(struct cpuinfo_x86 *c)
99{
100 unsigned int n, dummy, ebx, ecx, edx;
101
102 n = c->extended_cpuid_level;
103
104 if (n >= 0x80000005) {
105 cpuid(0x80000005, &dummy, &ebx, &ecx, &edx);
106 printk(KERN_INFO "CPU: L1 I Cache: %dK (%d bytes/line), "
107 "D cache %dK (%d bytes/line)\n",
108 edx>>24, edx&0xFF, ecx>>24, ecx&0xFF);
109 c->x86_cache_size = (ecx>>24) + (edx>>24);
110 /* On K8 L1 TLB is inclusive, so don't count it */
111 c->x86_tlbsize = 0;
112 }
113
114 if (n >= 0x80000006) {
115 cpuid(0x80000006, &dummy, &ebx, &ecx, &edx);
116 ecx = cpuid_ecx(0x80000006);
117 c->x86_cache_size = ecx >> 16;
118 c->x86_tlbsize += ((ebx >> 16) & 0xfff) + (ebx & 0xfff);
119
120 printk(KERN_INFO "CPU: L2 Cache: %dK (%d bytes/line)\n",
121 c->x86_cache_size, ecx & 0xFF);
122 }
123}
124
125void __cpuinit detect_ht(struct cpuinfo_x86 *c)
126{
127#ifdef CONFIG_SMP
128 u32 eax, ebx, ecx, edx;
129 int index_msb, core_bits;
130
131 cpuid(1, &eax, &ebx, &ecx, &edx);
132
133
134 if (!cpu_has(c, X86_FEATURE_HT))
135 return;
136 if (cpu_has(c, X86_FEATURE_CMP_LEGACY))
137 goto out;
138
139 smp_num_siblings = (ebx & 0xff0000) >> 16;
140
141 if (smp_num_siblings == 1) {
142 printk(KERN_INFO "CPU: Hyper-Threading is disabled\n");
143 } else if (smp_num_siblings > 1) {
144
145 if (smp_num_siblings > NR_CPUS) {
146 printk(KERN_WARNING "CPU: Unsupported number of "
147 "siblings %d", smp_num_siblings);
148 smp_num_siblings = 1;
149 return;
150 }
151
152 index_msb = get_count_order(smp_num_siblings);
153 c->phys_proc_id = phys_pkg_id(index_msb);
154
155 smp_num_siblings = smp_num_siblings / c->x86_max_cores;
156
157 index_msb = get_count_order(smp_num_siblings);
158
159 core_bits = get_count_order(c->x86_max_cores);
160
161 c->cpu_core_id = phys_pkg_id(index_msb) &
162 ((1 << core_bits) - 1);
163 }
164out:
165 if ((c->x86_max_cores * smp_num_siblings) > 1) {
166 printk(KERN_INFO "CPU: Physical Processor ID: %d\n",
167 c->phys_proc_id);
168 printk(KERN_INFO "CPU: Processor Core ID: %d\n",
169 c->cpu_core_id);
170 }
171
172#endif
173}
174
175static void __cpuinit get_cpu_vendor(struct cpuinfo_x86 *c)
176{
177 char *v = c->x86_vendor_id;
178 int i;
179 static int printed;
180
181 for (i = 0; i < X86_VENDOR_NUM; i++) {
182 if (cpu_devs[i]) {
183 if (!strcmp(v, cpu_devs[i]->c_ident[0]) ||
184 (cpu_devs[i]->c_ident[1] &&
185 !strcmp(v, cpu_devs[i]->c_ident[1]))) {
186 c->x86_vendor = i;
187 this_cpu = cpu_devs[i];
188 return;
189 }
190 }
191 }
192 if (!printed) {
193 printed++;
194 printk(KERN_ERR "CPU: Vendor unknown, using generic init.\n");
195 printk(KERN_ERR "CPU: Your system may be unstable.\n");
196 }
197 c->x86_vendor = X86_VENDOR_UNKNOWN;
198}
199
200static void __init early_cpu_support_print(void)
201{
202 int i,j;
203 struct cpu_dev *cpu_devx;
204
205 printk("KERNEL supported cpus:\n");
206 for (i = 0; i < X86_VENDOR_NUM; i++) {
207 cpu_devx = cpu_devs[i];
208 if (!cpu_devx)
209 continue;
210 for (j = 0; j < 2; j++) {
211 if (!cpu_devx->c_ident[j])
212 continue;
213 printk(" %s %s\n", cpu_devx->c_vendor,
214 cpu_devx->c_ident[j]);
215 }
216 }
217}
218
219/*
220 * The NOPL instruction is supposed to exist on all CPUs with
221 * family >= 6, unfortunately, that's not true in practice because
222 * of early VIA chips and (more importantly) broken virtualizers that
223 * are not easy to detect. Hence, probe for it based on first
224 * principles.
225 *
226 * Note: no 64-bit chip is known to lack these, but put the code here
227 * for consistency with 32 bits, and to make it utterly trivial to
228 * diagnose the problem should it ever surface.
229 */
230static void __cpuinit detect_nopl(struct cpuinfo_x86 *c)
231{
232 const u32 nopl_signature = 0x888c53b1; /* Random number */
233 u32 has_nopl = nopl_signature;
234
235 clear_cpu_cap(c, X86_FEATURE_NOPL);
236 if (c->x86 >= 6) {
237 asm volatile("\n"
238 "1: .byte 0x0f,0x1f,0xc0\n" /* nopl %eax */
239 "2:\n"
240 " .section .fixup,\"ax\"\n"
241 "3: xor %0,%0\n"
242 " jmp 2b\n"
243 " .previous\n"
244 _ASM_EXTABLE(1b,3b)
245 : "+a" (has_nopl));
246
247 if (has_nopl == nopl_signature)
248 set_cpu_cap(c, X86_FEATURE_NOPL);
249 }
250}
251
252static void __cpuinit early_identify_cpu(struct cpuinfo_x86 *c);
253
254void __init early_cpu_init(void)
255{
256 struct cpu_vendor_dev *cvdev;
257
258 for (cvdev = __x86cpuvendor_start ;
259 cvdev < __x86cpuvendor_end ;
260 cvdev++)
261 cpu_devs[cvdev->vendor] = cvdev->cpu_dev;
262 early_cpu_support_print();
263 early_identify_cpu(&boot_cpu_data);
264}
265
266/* Do some early cpuid on the boot CPU to get some parameter that are
267 needed before check_bugs. Everything advanced is in identify_cpu
268 below. */
269static void __cpuinit early_identify_cpu(struct cpuinfo_x86 *c)
270{
271 u32 tfms, xlvl;
272
273 c->loops_per_jiffy = loops_per_jiffy;
274 c->x86_cache_size = -1;
275 c->x86_vendor = X86_VENDOR_UNKNOWN;
276 c->x86_model = c->x86_mask = 0; /* So far unknown... */
277 c->x86_vendor_id[0] = '\0'; /* Unset */
278 c->x86_model_id[0] = '\0'; /* Unset */
279 c->x86_clflush_size = 64;
280 c->x86_cache_alignment = c->x86_clflush_size;
281 c->x86_max_cores = 1;
282 c->x86_coreid_bits = 0;
283 c->extended_cpuid_level = 0;
284 memset(&c->x86_capability, 0, sizeof c->x86_capability);
285
286 /* Get vendor name */
287 cpuid(0x00000000, (unsigned int *)&c->cpuid_level,
288 (unsigned int *)&c->x86_vendor_id[0],
289 (unsigned int *)&c->x86_vendor_id[8],
290 (unsigned int *)&c->x86_vendor_id[4]);
291
292 get_cpu_vendor(c);
293
294 /* Initialize the standard set of capabilities */
295 /* Note that the vendor-specific code below might override */
296
297 /* Intel-defined flags: level 0x00000001 */
298 if (c->cpuid_level >= 0x00000001) {
299 __u32 misc;
300 cpuid(0x00000001, &tfms, &misc, &c->x86_capability[4],
301 &c->x86_capability[0]);
302 c->x86 = (tfms >> 8) & 0xf;
303 c->x86_model = (tfms >> 4) & 0xf;
304 c->x86_mask = tfms & 0xf;
305 if (c->x86 == 0xf)
306 c->x86 += (tfms >> 20) & 0xff;
307 if (c->x86 >= 0x6)
308 c->x86_model += ((tfms >> 16) & 0xF) << 4;
309 if (test_cpu_cap(c, X86_FEATURE_CLFLSH))
310 c->x86_clflush_size = ((misc >> 8) & 0xff) * 8;
311 } else {
312 /* Have CPUID level 0 only - unheard of */
313 c->x86 = 4;
314 }
315
316 c->initial_apicid = (cpuid_ebx(1) >> 24) & 0xff;
317#ifdef CONFIG_SMP
318 c->phys_proc_id = c->initial_apicid;
319#endif
320 /* AMD-defined flags: level 0x80000001 */
321 xlvl = cpuid_eax(0x80000000);
322 c->extended_cpuid_level = xlvl;
323 if ((xlvl & 0xffff0000) == 0x80000000) {
324 if (xlvl >= 0x80000001) {
325 c->x86_capability[1] = cpuid_edx(0x80000001);
326 c->x86_capability[6] = cpuid_ecx(0x80000001);
327 }
328 if (xlvl >= 0x80000004)
329 get_model_name(c); /* Default name */
330 }
331
332 /* Transmeta-defined flags: level 0x80860001 */
333 xlvl = cpuid_eax(0x80860000);
334 if ((xlvl & 0xffff0000) == 0x80860000) {
335 /* Don't set x86_cpuid_level here for now to not confuse. */
336 if (xlvl >= 0x80860001)
337 c->x86_capability[2] = cpuid_edx(0x80860001);
338 }
339
340 if (c->extended_cpuid_level >= 0x80000007)
341 c->x86_power = cpuid_edx(0x80000007);
342
343 if (c->extended_cpuid_level >= 0x80000008) {
344 u32 eax = cpuid_eax(0x80000008);
345
346 c->x86_virt_bits = (eax >> 8) & 0xff;
347 c->x86_phys_bits = eax & 0xff;
348 }
349
350 detect_nopl(c);
351
352 if (c->x86_vendor != X86_VENDOR_UNKNOWN &&
353 cpu_devs[c->x86_vendor]->c_early_init)
354 cpu_devs[c->x86_vendor]->c_early_init(c);
355
356 validate_pat_support(c);
357}
358
359/*
360 * This does the hard work of actually picking apart the CPU stuff...
361 */
362static void __cpuinit identify_cpu(struct cpuinfo_x86 *c)
363{
364 int i;
365
366 early_identify_cpu(c);
367
368 init_scattered_cpuid_features(c);
369
370 c->apicid = phys_pkg_id(0);
371
372 /*
373 * Vendor-specific initialization. In this section we
374 * canonicalize the feature flags, meaning if there are
375 * features a certain CPU supports which CPUID doesn't
376 * tell us, CPUID claiming incorrect flags, or other bugs,
377 * we handle them here.
378 *
379 * At the end of this section, c->x86_capability better
380 * indicate the features this CPU genuinely supports!
381 */
382 if (this_cpu->c_init)
383 this_cpu->c_init(c);
384
385 detect_ht(c);
386
387 /*
388 * On SMP, boot_cpu_data holds the common feature set between
389 * all CPUs; so make sure that we indicate which features are
390 * common between the CPUs. The first time this routine gets
391 * executed, c == &boot_cpu_data.
392 */
393 if (c != &boot_cpu_data) {
394 /* AND the already accumulated flags with these */
395 for (i = 0; i < NCAPINTS; i++)
396 boot_cpu_data.x86_capability[i] &= c->x86_capability[i];
397 }
398
399 /* Clear all flags overriden by options */
400 for (i = 0; i < NCAPINTS; i++)
401 c->x86_capability[i] &= ~cleared_cpu_caps[i];
402
403#ifdef CONFIG_X86_MCE
404 mcheck_init(c);
405#endif
406 select_idle_routine(c);
407
408#ifdef CONFIG_NUMA
409 numa_add_cpu(smp_processor_id());
410#endif
411
412}
413
414void __cpuinit identify_boot_cpu(void)
415{
416 identify_cpu(&boot_cpu_data);
417}
418
419void __cpuinit identify_secondary_cpu(struct cpuinfo_x86 *c)
420{
421 BUG_ON(c == &boot_cpu_data);
422 identify_cpu(c);
423 mtrr_ap_init();
424}
425
426static __init int setup_noclflush(char *arg)
427{
428 setup_clear_cpu_cap(X86_FEATURE_CLFLSH);
429 return 1;
430}
431__setup("noclflush", setup_noclflush);
432
433struct msr_range {
434 unsigned min;
435 unsigned max;
436};
437
438static struct msr_range msr_range_array[] __cpuinitdata = {
439 { 0x00000000, 0x00000418},
440 { 0xc0000000, 0xc000040b},
441 { 0xc0010000, 0xc0010142},
442 { 0xc0011000, 0xc001103b},
443};
444
445static void __cpuinit print_cpu_msr(void)
446{
447 unsigned index;
448 u64 val;
449 int i;
450 unsigned index_min, index_max;
451
452 for (i = 0; i < ARRAY_SIZE(msr_range_array); i++) {
453 index_min = msr_range_array[i].min;
454 index_max = msr_range_array[i].max;
455 for (index = index_min; index < index_max; index++) {
456 if (rdmsrl_amd_safe(index, &val))
457 continue;
458 printk(KERN_INFO " MSR%08x: %016llx\n", index, val);
459 }
460 }
461}
462
463static int show_msr __cpuinitdata;
464static __init int setup_show_msr(char *arg)
465{
466 int num;
467
468 get_option(&arg, &num);
469
470 if (num > 0)
471 show_msr = num;
472 return 1;
473}
474__setup("show_msr=", setup_show_msr);
475
476void __cpuinit print_cpu_info(struct cpuinfo_x86 *c)
477{
478 if (c->x86_model_id[0])
479 printk(KERN_CONT "%s", c->x86_model_id);
480
481 if (c->x86_mask || c->cpuid_level >= 0)
482 printk(KERN_CONT " stepping %02x\n", c->x86_mask);
483 else
484 printk(KERN_CONT "\n");
485
486#ifdef CONFIG_SMP
487 if (c->cpu_index < show_msr)
488 print_cpu_msr();
489#else
490 if (show_msr)
491 print_cpu_msr();
492#endif
493}
494
495static __init int setup_disablecpuid(char *arg)
496{
497 int bit;
498 if (get_option(&arg, &bit) && bit < NCAPINTS*32)
499 setup_clear_cpu_cap(bit);
500 else
501 return 0;
502 return 1;
503}
504__setup("clearcpuid=", setup_disablecpuid);
505
506cpumask_t cpu_initialized __cpuinitdata = CPU_MASK_NONE;
507
508struct x8664_pda **_cpu_pda __read_mostly;
509EXPORT_SYMBOL(_cpu_pda);
510
511struct desc_ptr idt_descr = { 256 * 16 - 1, (unsigned long) idt_table };
512
513char boot_cpu_stack[IRQSTACKSIZE] __page_aligned_bss;
514
515unsigned long __supported_pte_mask __read_mostly = ~0UL;
516EXPORT_SYMBOL_GPL(__supported_pte_mask);
517
518static int do_not_nx __cpuinitdata;
519
520/* noexec=on|off
521Control non executable mappings for 64bit processes.
522
523on Enable(default)
524off Disable
525*/
526static int __init nonx_setup(char *str)
527{
528 if (!str)
529 return -EINVAL;
530 if (!strncmp(str, "on", 2)) {
531 __supported_pte_mask |= _PAGE_NX;
532 do_not_nx = 0;
533 } else if (!strncmp(str, "off", 3)) {
534 do_not_nx = 1;
535 __supported_pte_mask &= ~_PAGE_NX;
536 }
537 return 0;
538}
539early_param("noexec", nonx_setup);
540
541int force_personality32;
542
543/* noexec32=on|off
544Control non executable heap for 32bit processes.
545To control the stack too use noexec=off
546
547on PROT_READ does not imply PROT_EXEC for 32bit processes (default)
548off PROT_READ implies PROT_EXEC
549*/
550static int __init nonx32_setup(char *str)
551{
552 if (!strcmp(str, "on"))
553 force_personality32 &= ~READ_IMPLIES_EXEC;
554 else if (!strcmp(str, "off"))
555 force_personality32 |= READ_IMPLIES_EXEC;
556 return 1;
557}
558__setup("noexec32=", nonx32_setup);
559
560void pda_init(int cpu)
561{
562 struct x8664_pda *pda = cpu_pda(cpu);
563
564 /* Setup up data that may be needed in __get_free_pages early */
565 loadsegment(fs, 0);
566 loadsegment(gs, 0);
567 /* Memory clobbers used to order PDA accessed */
568 mb();
569 wrmsrl(MSR_GS_BASE, pda);
570 mb();
571
572 pda->cpunumber = cpu;
573 pda->irqcount = -1;
574 pda->kernelstack = (unsigned long)stack_thread_info() -
575 PDA_STACKOFFSET + THREAD_SIZE;
576 pda->active_mm = &init_mm;
577 pda->mmu_state = 0;
578
579 if (cpu == 0) {
580 /* others are initialized in smpboot.c */
581 pda->pcurrent = &init_task;
582 pda->irqstackptr = boot_cpu_stack;
583 pda->irqstackptr += IRQSTACKSIZE - 64;
584 } else {
585 if (!pda->irqstackptr) {
586 pda->irqstackptr = (char *)
587 __get_free_pages(GFP_ATOMIC, IRQSTACK_ORDER);
588 if (!pda->irqstackptr)
589 panic("cannot allocate irqstack for cpu %d",
590 cpu);
591 pda->irqstackptr += IRQSTACKSIZE - 64;
592 }
593
594 if (pda->nodenumber == 0 && cpu_to_node(cpu) != NUMA_NO_NODE)
595 pda->nodenumber = cpu_to_node(cpu);
596 }
597}
598
599char boot_exception_stacks[(N_EXCEPTION_STACKS - 1) * EXCEPTION_STKSZ +
600 DEBUG_STKSZ] __page_aligned_bss;
601
602extern asmlinkage void ignore_sysret(void);
603
604/* May not be marked __init: used by software suspend */
605void syscall_init(void)
606{
607 /*
608 * LSTAR and STAR live in a bit strange symbiosis.
609 * They both write to the same internal register. STAR allows to
610 * set CS/DS but only a 32bit target. LSTAR sets the 64bit rip.
611 */
612 wrmsrl(MSR_STAR, ((u64)__USER32_CS)<<48 | ((u64)__KERNEL_CS)<<32);
613 wrmsrl(MSR_LSTAR, system_call);
614 wrmsrl(MSR_CSTAR, ignore_sysret);
615
616#ifdef CONFIG_IA32_EMULATION
617 syscall32_cpu_init();
618#endif
619
620 /* Flags to clear on syscall */
621 wrmsrl(MSR_SYSCALL_MASK,
622 X86_EFLAGS_TF|X86_EFLAGS_DF|X86_EFLAGS_IF|X86_EFLAGS_IOPL);
623}
624
625void __cpuinit check_efer(void)
626{
627 unsigned long efer;
628
629 rdmsrl(MSR_EFER, efer);
630 if (!(efer & EFER_NX) || do_not_nx)
631 __supported_pte_mask &= ~_PAGE_NX;
632}
633
634unsigned long kernel_eflags;
635
636/*
637 * Copies of the original ist values from the tss are only accessed during
638 * debugging, no special alignment required.
639 */
640DEFINE_PER_CPU(struct orig_ist, orig_ist);
641
642/*
643 * cpu_init() initializes state that is per-CPU. Some data is already
644 * initialized (naturally) in the bootstrap process, such as the GDT
645 * and IDT. We reload them nevertheless, this function acts as a
646 * 'CPU state barrier', nothing should get across.
647 * A lot of state is already set up in PDA init.
648 */
649void __cpuinit cpu_init(void)
650{
651 int cpu = stack_smp_processor_id();
652 struct tss_struct *t = &per_cpu(init_tss, cpu);
653 struct orig_ist *orig_ist = &per_cpu(orig_ist, cpu);
654 unsigned long v;
655 char *estacks = NULL;
656 struct task_struct *me;
657 int i;
658
659 /* CPU 0 is initialised in head64.c */
660 if (cpu != 0)
661 pda_init(cpu);
662 else
663 estacks = boot_exception_stacks;
664
665 me = current;
666
667 if (cpu_test_and_set(cpu, cpu_initialized))
668 panic("CPU#%d already initialized!\n", cpu);
669
670 printk(KERN_INFO "Initializing CPU#%d\n", cpu);
671
672 clear_in_cr4(X86_CR4_VME|X86_CR4_PVI|X86_CR4_TSD|X86_CR4_DE);
673
674 /*
675 * Initialize the per-CPU GDT with the boot GDT,
676 * and set up the GDT descriptor:
677 */
678
679 switch_to_new_gdt();
680 load_idt((const struct desc_ptr *)&idt_descr);
681
682 memset(me->thread.tls_array, 0, GDT_ENTRY_TLS_ENTRIES * 8);
683 syscall_init();
684
685 wrmsrl(MSR_FS_BASE, 0);
686 wrmsrl(MSR_KERNEL_GS_BASE, 0);
687 barrier();
688
689 check_efer();
690 if (cpu != 0 && x2apic)
691 enable_x2apic();
692
693 /*
694 * set up and load the per-CPU TSS
695 */
696 if (!orig_ist->ist[0]) {
697 static const unsigned int order[N_EXCEPTION_STACKS] = {
698 [0 ... N_EXCEPTION_STACKS - 1] = EXCEPTION_STACK_ORDER,
699 [DEBUG_STACK - 1] = DEBUG_STACK_ORDER
700 };
701 for (v = 0; v < N_EXCEPTION_STACKS; v++) {
702 if (cpu) {
703 estacks = (char *)__get_free_pages(GFP_ATOMIC, order[v]);
704 if (!estacks)
705 panic("Cannot allocate exception "
706 "stack %ld %d\n", v, cpu);
707 }
708 estacks += PAGE_SIZE << order[v];
709 orig_ist->ist[v] = t->x86_tss.ist[v] =
710 (unsigned long)estacks;
711 }
712 }
713
714 t->x86_tss.io_bitmap_base = offsetof(struct tss_struct, io_bitmap);
715 /*
716 * <= is required because the CPU will access up to
717 * 8 bits beyond the end of the IO permission bitmap.
718 */
719 for (i = 0; i <= IO_BITMAP_LONGS; i++)
720 t->io_bitmap[i] = ~0UL;
721
722 atomic_inc(&init_mm.mm_count);
723 me->active_mm = &init_mm;
724 if (me->mm)
725 BUG();
726 enter_lazy_tlb(&init_mm, me);
727
728 load_sp0(t, &current->thread);
729 set_tss_desc(cpu, t);
730 load_TR_desc();
731 load_LDT(&init_mm.context);
732
733#ifdef CONFIG_KGDB
734 /*
735 * If the kgdb is connected no debug regs should be altered. This
736 * is only applicable when KGDB and a KGDB I/O module are built
737 * into the kernel and you are using early debugging with
738 * kgdbwait. KGDB will control the kernel HW breakpoint registers.
739 */
740 if (kgdb_connected && arch_kgdb_ops.correct_hw_break)
741 arch_kgdb_ops.correct_hw_break();
742 else {
743#endif
744 /*
745 * Clear all 6 debug registers:
746 */
747
748 set_debugreg(0UL, 0);
749 set_debugreg(0UL, 1);
750 set_debugreg(0UL, 2);
751 set_debugreg(0UL, 3);
752 set_debugreg(0UL, 6);
753 set_debugreg(0UL, 7);
754#ifdef CONFIG_KGDB
755 /* If the kgdb is connected no debug regs should be altered. */
756 }
757#endif
758
759 fpu_init();
760
761 raw_local_save_flags(kernel_eflags);
762
763 if (is_uv_system())
764 uv_cpu_init();
765}
diff --git a/arch/x86/kernel/cpu/cpu.h b/arch/x86/kernel/cpu/cpu.h
index 4d894e8565fe..de4094a39210 100644
--- a/arch/x86/kernel/cpu/cpu.h
+++ b/arch/x86/kernel/cpu/cpu.h
@@ -21,23 +21,16 @@ struct cpu_dev {
21 void (*c_init)(struct cpuinfo_x86 * c); 21 void (*c_init)(struct cpuinfo_x86 * c);
22 void (*c_identify)(struct cpuinfo_x86 * c); 22 void (*c_identify)(struct cpuinfo_x86 * c);
23 unsigned int (*c_size_cache)(struct cpuinfo_x86 * c, unsigned int size); 23 unsigned int (*c_size_cache)(struct cpuinfo_x86 * c, unsigned int size);
24 int c_x86_vendor;
24}; 25};
25 26
26extern struct cpu_dev * cpu_devs [X86_VENDOR_NUM]; 27#define cpu_dev_register(cpu_devX) \
28 static struct cpu_dev *__cpu_dev_##cpu_devX __used \
29 __attribute__((__section__(".x86_cpu_dev.init"))) = \
30 &cpu_devX;
27 31
28struct cpu_vendor_dev { 32extern struct cpu_dev *__x86_cpu_dev_start[], *__x86_cpu_dev_end[];
29 int vendor;
30 struct cpu_dev *cpu_dev;
31};
32
33#define cpu_vendor_dev_register(cpu_vendor_id, cpu_dev) \
34 static struct cpu_vendor_dev __cpu_vendor_dev_##cpu_vendor_id __used \
35 __attribute__((__section__(".x86cpuvendor.init"))) = \
36 { cpu_vendor_id, cpu_dev }
37
38extern struct cpu_vendor_dev __x86cpuvendor_start[], __x86cpuvendor_end[];
39 33
40extern int get_model_name(struct cpuinfo_x86 *c);
41extern void display_cacheinfo(struct cpuinfo_x86 *c); 34extern void display_cacheinfo(struct cpuinfo_x86 *c);
42 35
43#endif 36#endif
diff --git a/arch/x86/kernel/cpu/cyrix.c b/arch/x86/kernel/cpu/cyrix.c
index 898a5a2002ed..ffd0f5ed071a 100644
--- a/arch/x86/kernel/cpu/cyrix.c
+++ b/arch/x86/kernel/cpu/cyrix.c
@@ -121,7 +121,7 @@ static void __cpuinit set_cx86_reorder(void)
121 setCx86(CX86_CCR3, (ccr3 & 0x0f) | 0x10); /* enable MAPEN */ 121 setCx86(CX86_CCR3, (ccr3 & 0x0f) | 0x10); /* enable MAPEN */
122 122
123 /* Load/Store Serialize to mem access disable (=reorder it) */ 123 /* Load/Store Serialize to mem access disable (=reorder it) */
124 setCx86(CX86_PCR0, getCx86(CX86_PCR0) & ~0x80); 124 setCx86_old(CX86_PCR0, getCx86_old(CX86_PCR0) & ~0x80);
125 /* set load/store serialize from 1GB to 4GB */ 125 /* set load/store serialize from 1GB to 4GB */
126 ccr3 |= 0xe0; 126 ccr3 |= 0xe0;
127 setCx86(CX86_CCR3, ccr3); 127 setCx86(CX86_CCR3, ccr3);
@@ -132,11 +132,11 @@ static void __cpuinit set_cx86_memwb(void)
132 printk(KERN_INFO "Enable Memory-Write-back mode on Cyrix/NSC processor.\n"); 132 printk(KERN_INFO "Enable Memory-Write-back mode on Cyrix/NSC processor.\n");
133 133
134 /* CCR2 bit 2: unlock NW bit */ 134 /* CCR2 bit 2: unlock NW bit */
135 setCx86(CX86_CCR2, getCx86(CX86_CCR2) & ~0x04); 135 setCx86_old(CX86_CCR2, getCx86_old(CX86_CCR2) & ~0x04);
136 /* set 'Not Write-through' */ 136 /* set 'Not Write-through' */
137 write_cr0(read_cr0() | X86_CR0_NW); 137 write_cr0(read_cr0() | X86_CR0_NW);
138 /* CCR2 bit 2: lock NW bit and set WT1 */ 138 /* CCR2 bit 2: lock NW bit and set WT1 */
139 setCx86(CX86_CCR2, getCx86(CX86_CCR2) | 0x14); 139 setCx86_old(CX86_CCR2, getCx86_old(CX86_CCR2) | 0x14);
140} 140}
141 141
142/* 142/*
@@ -150,14 +150,14 @@ static void __cpuinit geode_configure(void)
150 local_irq_save(flags); 150 local_irq_save(flags);
151 151
152 /* Suspend on halt power saving and enable #SUSP pin */ 152 /* Suspend on halt power saving and enable #SUSP pin */
153 setCx86(CX86_CCR2, getCx86(CX86_CCR2) | 0x88); 153 setCx86_old(CX86_CCR2, getCx86_old(CX86_CCR2) | 0x88);
154 154
155 ccr3 = getCx86(CX86_CCR3); 155 ccr3 = getCx86(CX86_CCR3);
156 setCx86(CX86_CCR3, (ccr3 & 0x0f) | 0x10); /* enable MAPEN */ 156 setCx86(CX86_CCR3, (ccr3 & 0x0f) | 0x10); /* enable MAPEN */
157 157
158 158
159 /* FPU fast, DTE cache, Mem bypass */ 159 /* FPU fast, DTE cache, Mem bypass */
160 setCx86(CX86_CCR4, getCx86(CX86_CCR4) | 0x38); 160 setCx86_old(CX86_CCR4, getCx86_old(CX86_CCR4) | 0x38);
161 setCx86(CX86_CCR3, ccr3); /* disable MAPEN */ 161 setCx86(CX86_CCR3, ccr3); /* disable MAPEN */
162 162
163 set_cx86_memwb(); 163 set_cx86_memwb();
@@ -291,7 +291,7 @@ static void __cpuinit init_cyrix(struct cpuinfo_x86 *c)
291 /* GXm supports extended cpuid levels 'ala' AMD */ 291 /* GXm supports extended cpuid levels 'ala' AMD */
292 if (c->cpuid_level == 2) { 292 if (c->cpuid_level == 2) {
293 /* Enable cxMMX extensions (GX1 Datasheet 54) */ 293 /* Enable cxMMX extensions (GX1 Datasheet 54) */
294 setCx86(CX86_CCR7, getCx86(CX86_CCR7) | 1); 294 setCx86_old(CX86_CCR7, getCx86_old(CX86_CCR7) | 1);
295 295
296 /* 296 /*
297 * GXm : 0x30 ... 0x5f GXm datasheet 51 297 * GXm : 0x30 ... 0x5f GXm datasheet 51
@@ -301,7 +301,6 @@ static void __cpuinit init_cyrix(struct cpuinfo_x86 *c)
301 */ 301 */
302 if ((0x30 <= dir1 && dir1 <= 0x6f) || (0x80 <= dir1 && dir1 <= 0x8f)) 302 if ((0x30 <= dir1 && dir1 <= 0x6f) || (0x80 <= dir1 && dir1 <= 0x8f))
303 geode_configure(); 303 geode_configure();
304 get_model_name(c); /* get CPU marketing name */
305 return; 304 return;
306 } else { /* MediaGX */ 305 } else { /* MediaGX */
307 Cx86_cb[2] = (dir0_lsn & 1) ? '3' : '4'; 306 Cx86_cb[2] = (dir0_lsn & 1) ? '3' : '4';
@@ -314,7 +313,7 @@ static void __cpuinit init_cyrix(struct cpuinfo_x86 *c)
314 if (dir1 > 7) { 313 if (dir1 > 7) {
315 dir0_msn++; /* M II */ 314 dir0_msn++; /* M II */
316 /* Enable MMX extensions (App note 108) */ 315 /* Enable MMX extensions (App note 108) */
317 setCx86(CX86_CCR7, getCx86(CX86_CCR7)|1); 316 setCx86_old(CX86_CCR7, getCx86_old(CX86_CCR7)|1);
318 } else { 317 } else {
319 c->coma_bug = 1; /* 6x86MX, it has the bug. */ 318 c->coma_bug = 1; /* 6x86MX, it has the bug. */
320 } 319 }
@@ -429,7 +428,7 @@ static void __cpuinit cyrix_identify(struct cpuinfo_x86 *c)
429 local_irq_save(flags); 428 local_irq_save(flags);
430 ccr3 = getCx86(CX86_CCR3); 429 ccr3 = getCx86(CX86_CCR3);
431 setCx86(CX86_CCR3, (ccr3 & 0x0f) | 0x10); /* enable MAPEN */ 430 setCx86(CX86_CCR3, (ccr3 & 0x0f) | 0x10); /* enable MAPEN */
432 setCx86(CX86_CCR4, getCx86(CX86_CCR4) | 0x80); /* enable cpuid */ 431 setCx86_old(CX86_CCR4, getCx86_old(CX86_CCR4) | 0x80); /* enable cpuid */
433 setCx86(CX86_CCR3, ccr3); /* disable MAPEN */ 432 setCx86(CX86_CCR3, ccr3); /* disable MAPEN */
434 local_irq_restore(flags); 433 local_irq_restore(flags);
435 } 434 }
@@ -442,14 +441,16 @@ static struct cpu_dev cyrix_cpu_dev __cpuinitdata = {
442 .c_early_init = early_init_cyrix, 441 .c_early_init = early_init_cyrix,
443 .c_init = init_cyrix, 442 .c_init = init_cyrix,
444 .c_identify = cyrix_identify, 443 .c_identify = cyrix_identify,
444 .c_x86_vendor = X86_VENDOR_CYRIX,
445}; 445};
446 446
447cpu_vendor_dev_register(X86_VENDOR_CYRIX, &cyrix_cpu_dev); 447cpu_dev_register(cyrix_cpu_dev);
448 448
449static struct cpu_dev nsc_cpu_dev __cpuinitdata = { 449static struct cpu_dev nsc_cpu_dev __cpuinitdata = {
450 .c_vendor = "NSC", 450 .c_vendor = "NSC",
451 .c_ident = { "Geode by NSC" }, 451 .c_ident = { "Geode by NSC" },
452 .c_init = init_nsc, 452 .c_init = init_nsc,
453 .c_x86_vendor = X86_VENDOR_NSC,
453}; 454};
454 455
455cpu_vendor_dev_register(X86_VENDOR_NSC, &nsc_cpu_dev); 456cpu_dev_register(nsc_cpu_dev);
diff --git a/arch/x86/kernel/cpu/feature_names.c b/arch/x86/kernel/cpu/feature_names.c
deleted file mode 100644
index b96b69545fbf..000000000000
--- a/arch/x86/kernel/cpu/feature_names.c
+++ /dev/null
@@ -1,84 +0,0 @@
1/*
2 * Strings for the various x86 capability flags.
3 *
4 * This file must not contain any executable code.
5 */
6
7#include <asm/cpufeature.h>
8
9/*
10 * These flag bits must match the definitions in <asm/cpufeature.h>.
11 * NULL means this bit is undefined or reserved; either way it doesn't
12 * have meaning as far as Linux is concerned. Note that it's important
13 * to realize there is a difference between this table and CPUID -- if
14 * applications want to get the raw CPUID data, they should access
15 * /dev/cpu/<cpu_nr>/cpuid instead.
16 */
17const char * const x86_cap_flags[NCAPINTS*32] = {
18 /* Intel-defined */
19 "fpu", "vme", "de", "pse", "tsc", "msr", "pae", "mce",
20 "cx8", "apic", NULL, "sep", "mtrr", "pge", "mca", "cmov",
21 "pat", "pse36", "pn", "clflush", NULL, "dts", "acpi", "mmx",
22 "fxsr", "sse", "sse2", "ss", "ht", "tm", "ia64", "pbe",
23
24 /* AMD-defined */
25 NULL, NULL, NULL, NULL, NULL, NULL, NULL, NULL,
26 NULL, NULL, NULL, "syscall", NULL, NULL, NULL, NULL,
27 NULL, NULL, NULL, "mp", "nx", NULL, "mmxext", NULL,
28 NULL, "fxsr_opt", "pdpe1gb", "rdtscp", NULL, "lm",
29 "3dnowext", "3dnow",
30
31 /* Transmeta-defined */
32 "recovery", "longrun", NULL, "lrti", NULL, NULL, NULL, NULL,
33 NULL, NULL, NULL, NULL, NULL, NULL, NULL, NULL,
34 NULL, NULL, NULL, NULL, NULL, NULL, NULL, NULL,
35 NULL, NULL, NULL, NULL, NULL, NULL, NULL, NULL,
36
37 /* Other (Linux-defined) */
38 "cxmmx", "k6_mtrr", "cyrix_arr", "centaur_mcr",
39 NULL, NULL, NULL, NULL,
40 "constant_tsc", "up", NULL, "arch_perfmon",
41 "pebs", "bts", NULL, NULL,
42 "rep_good", NULL, NULL, NULL,
43 "nopl", NULL, NULL, NULL,
44 NULL, NULL, NULL, NULL, NULL, NULL, NULL, NULL,
45
46 /* Intel-defined (#2) */
47 "pni", NULL, NULL, "monitor", "ds_cpl", "vmx", "smx", "est",
48 "tm2", "ssse3", "cid", NULL, NULL, "cx16", "xtpr", NULL,
49 NULL, NULL, "dca", "sse4_1", "sse4_2", "x2apic", NULL, "popcnt",
50 NULL, NULL, NULL, NULL, NULL, NULL, NULL, NULL,
51
52 /* VIA/Cyrix/Centaur-defined */
53 NULL, NULL, "rng", "rng_en", NULL, NULL, "ace", "ace_en",
54 "ace2", "ace2_en", "phe", "phe_en", "pmm", "pmm_en", NULL, NULL,
55 NULL, NULL, NULL, NULL, NULL, NULL, NULL, NULL,
56 NULL, NULL, NULL, NULL, NULL, NULL, NULL, NULL,
57
58 /* AMD-defined (#2) */
59 "lahf_lm", "cmp_legacy", "svm", "extapic",
60 "cr8_legacy", "abm", "sse4a", "misalignsse",
61 "3dnowprefetch", "osvw", "ibs", "sse5",
62 "skinit", "wdt", NULL, NULL,
63 NULL, NULL, NULL, NULL, NULL, NULL, NULL, NULL,
64 NULL, NULL, NULL, NULL, NULL, NULL, NULL, NULL,
65
66 /* Auxiliary (Linux-defined) */
67 "ida", NULL, NULL, NULL, NULL, NULL, NULL, NULL,
68 NULL, NULL, NULL, NULL, NULL, NULL, NULL, NULL,
69 NULL, NULL, NULL, NULL, NULL, NULL, NULL, NULL,
70 NULL, NULL, NULL, NULL, NULL, NULL, NULL, NULL,
71};
72
73const char *const x86_power_flags[32] = {
74 "ts", /* temperature sensor */
75 "fid", /* frequency id control */
76 "vid", /* voltage id control */
77 "ttp", /* thermal trip */
78 "tm",
79 "stc",
80 "100mhzsteps",
81 "hwpstate",
82 "", /* tsc invariant mapped to constant_tsc */
83 /* nothing */
84};
diff --git a/arch/x86/kernel/cpu/intel.c b/arch/x86/kernel/cpu/intel.c
index f113ef4595f6..99468dbd08da 100644
--- a/arch/x86/kernel/cpu/intel.c
+++ b/arch/x86/kernel/cpu/intel.c
@@ -15,6 +15,11 @@
15#include <asm/ds.h> 15#include <asm/ds.h>
16#include <asm/bugs.h> 16#include <asm/bugs.h>
17 17
18#ifdef CONFIG_X86_64
19#include <asm/topology.h>
20#include <asm/numa_64.h>
21#endif
22
18#include "cpu.h" 23#include "cpu.h"
19 24
20#ifdef CONFIG_X86_LOCAL_APIC 25#ifdef CONFIG_X86_LOCAL_APIC
@@ -23,23 +28,22 @@
23#include <mach_apic.h> 28#include <mach_apic.h>
24#endif 29#endif
25 30
26#ifdef CONFIG_X86_INTEL_USERCOPY
27/*
28 * Alignment at which movsl is preferred for bulk memory copies.
29 */
30struct movsl_mask movsl_mask __read_mostly;
31#endif
32
33static void __cpuinit early_init_intel(struct cpuinfo_x86 *c) 31static void __cpuinit early_init_intel(struct cpuinfo_x86 *c)
34{ 32{
35 /* Netburst reports 64 bytes clflush size, but does IO in 128 bytes */
36 if (c->x86 == 15 && c->x86_cache_alignment == 64)
37 c->x86_cache_alignment = 128;
38 if ((c->x86 == 0xf && c->x86_model >= 0x03) || 33 if ((c->x86 == 0xf && c->x86_model >= 0x03) ||
39 (c->x86 == 0x6 && c->x86_model >= 0x0e)) 34 (c->x86 == 0x6 && c->x86_model >= 0x0e))
40 set_cpu_cap(c, X86_FEATURE_CONSTANT_TSC); 35 set_cpu_cap(c, X86_FEATURE_CONSTANT_TSC);
36
37#ifdef CONFIG_X86_64
38 set_cpu_cap(c, X86_FEATURE_SYSENTER32);
39#else
40 /* Netburst reports 64 bytes clflush size, but does IO in 128 bytes */
41 if (c->x86 == 15 && c->x86_cache_alignment == 64)
42 c->x86_cache_alignment = 128;
43#endif
41} 44}
42 45
46#ifdef CONFIG_X86_32
43/* 47/*
44 * Early probe support logic for ppro memory erratum #50 48 * Early probe support logic for ppro memory erratum #50
45 * 49 *
@@ -59,15 +63,54 @@ int __cpuinit ppro_with_ram_bug(void)
59 return 0; 63 return 0;
60} 64}
61 65
66#ifdef CONFIG_X86_F00F_BUG
67static void __cpuinit trap_init_f00f_bug(void)
68{
69 __set_fixmap(FIX_F00F_IDT, __pa(&idt_table), PAGE_KERNEL_RO);
62 70
63/* 71 /*
64 * P4 Xeon errata 037 workaround. 72 * Update the IDT descriptor and reload the IDT so that
65 * Hardware prefetcher may cause stale data to be loaded into the cache. 73 * it uses the read-only mapped virtual address.
66 */ 74 */
67static void __cpuinit Intel_errata_workarounds(struct cpuinfo_x86 *c) 75 idt_descr.address = fix_to_virt(FIX_F00F_IDT);
76 load_idt(&idt_descr);
77}
78#endif
79
80static void __cpuinit intel_workarounds(struct cpuinfo_x86 *c)
68{ 81{
69 unsigned long lo, hi; 82 unsigned long lo, hi;
70 83
84#ifdef CONFIG_X86_F00F_BUG
85 /*
86 * All current models of Pentium and Pentium with MMX technology CPUs
87 * have the F0 0F bug, which lets nonprivileged users lock up the system.
88 * Note that the workaround only should be initialized once...
89 */
90 c->f00f_bug = 0;
91 if (!paravirt_enabled() && c->x86 == 5) {
92 static int f00f_workaround_enabled;
93
94 c->f00f_bug = 1;
95 if (!f00f_workaround_enabled) {
96 trap_init_f00f_bug();
97 printk(KERN_NOTICE "Intel Pentium with F0 0F bug - workaround enabled.\n");
98 f00f_workaround_enabled = 1;
99 }
100 }
101#endif
102
103 /*
104 * SEP CPUID bug: Pentium Pro reports SEP but doesn't have it until
105 * model 3 mask 3
106 */
107 if ((c->x86<<8 | c->x86_model<<4 | c->x86_mask) < 0x633)
108 clear_cpu_cap(c, X86_FEATURE_SEP);
109
110 /*
111 * P4 Xeon errata 037 workaround.
112 * Hardware prefetcher may cause stale data to be loaded into the cache.
113 */
71 if ((c->x86 == 15) && (c->x86_model == 1) && (c->x86_mask == 1)) { 114 if ((c->x86 == 15) && (c->x86_model == 1) && (c->x86_mask == 1)) {
72 rdmsr(MSR_IA32_MISC_ENABLE, lo, hi); 115 rdmsr(MSR_IA32_MISC_ENABLE, lo, hi);
73 if ((lo & (1<<9)) == 0) { 116 if ((lo & (1<<9)) == 0) {
@@ -77,13 +120,68 @@ static void __cpuinit Intel_errata_workarounds(struct cpuinfo_x86 *c)
77 wrmsr (MSR_IA32_MISC_ENABLE, lo, hi); 120 wrmsr (MSR_IA32_MISC_ENABLE, lo, hi);
78 } 121 }
79 } 122 }
123
124 /*
125 * See if we have a good local APIC by checking for buggy Pentia,
126 * i.e. all B steppings and the C2 stepping of P54C when using their
127 * integrated APIC (see 11AP erratum in "Pentium Processor
128 * Specification Update").
129 */
130 if (cpu_has_apic && (c->x86<<8 | c->x86_model<<4) == 0x520 &&
131 (c->x86_mask < 0x6 || c->x86_mask == 0xb))
132 set_cpu_cap(c, X86_FEATURE_11AP);
133
134
135#ifdef CONFIG_X86_INTEL_USERCOPY
136 /*
137 * Set up the preferred alignment for movsl bulk memory moves
138 */
139 switch (c->x86) {
140 case 4: /* 486: untested */
141 break;
142 case 5: /* Old Pentia: untested */
143 break;
144 case 6: /* PII/PIII only like movsl with 8-byte alignment */
145 movsl_mask.mask = 7;
146 break;
147 case 15: /* P4 is OK down to 8-byte alignment */
148 movsl_mask.mask = 7;
149 break;
150 }
151#endif
152
153#ifdef CONFIG_X86_NUMAQ
154 numaq_tsc_disable();
155#endif
80} 156}
157#else
158static void __cpuinit intel_workarounds(struct cpuinfo_x86 *c)
159{
160}
161#endif
81 162
163static void __cpuinit srat_detect_node(void)
164{
165#if defined(CONFIG_NUMA) && defined(CONFIG_X86_64)
166 unsigned node;
167 int cpu = smp_processor_id();
168 int apicid = hard_smp_processor_id();
169
170 /* Don't do the funky fallback heuristics the AMD version employs
171 for now. */
172 node = apicid_to_node[apicid];
173 if (node == NUMA_NO_NODE || !node_online(node))
174 node = first_node(node_online_map);
175 numa_set_node(cpu, node);
176
177 printk(KERN_INFO "CPU %d/%x -> Node %d\n", cpu, apicid, node);
178#endif
179}
82 180
83/* 181/*
84 * find out the number of processor cores on the die 182 * find out the number of processor cores on the die
85 */ 183 */
86static int __cpuinit num_cpu_cores(struct cpuinfo_x86 *c) 184static int __cpuinit intel_num_cpu_cores(struct cpuinfo_x86 *c)
87{ 185{
88 unsigned int eax, ebx, ecx, edx; 186 unsigned int eax, ebx, ecx, edx;
89 187
@@ -98,45 +196,51 @@ static int __cpuinit num_cpu_cores(struct cpuinfo_x86 *c)
98 return 1; 196 return 1;
99} 197}
100 198
101#ifdef CONFIG_X86_F00F_BUG 199static void __cpuinit detect_vmx_virtcap(struct cpuinfo_x86 *c)
102static void __cpuinit trap_init_f00f_bug(void)
103{ 200{
104 __set_fixmap(FIX_F00F_IDT, __pa(&idt_table), PAGE_KERNEL_RO); 201 /* Intel VMX MSR indicated features */
105 202#define X86_VMX_FEATURE_PROC_CTLS_TPR_SHADOW 0x00200000
106 /* 203#define X86_VMX_FEATURE_PROC_CTLS_VNMI 0x00400000
107 * Update the IDT descriptor and reload the IDT so that 204#define X86_VMX_FEATURE_PROC_CTLS_2ND_CTLS 0x80000000
108 * it uses the read-only mapped virtual address. 205#define X86_VMX_FEATURE_PROC_CTLS2_VIRT_APIC 0x00000001
109 */ 206#define X86_VMX_FEATURE_PROC_CTLS2_EPT 0x00000002
110 idt_descr.address = fix_to_virt(FIX_F00F_IDT); 207#define X86_VMX_FEATURE_PROC_CTLS2_VPID 0x00000020
111 load_idt(&idt_descr); 208
209 u32 vmx_msr_low, vmx_msr_high, msr_ctl, msr_ctl2;
210
211 clear_cpu_cap(c, X86_FEATURE_TPR_SHADOW);
212 clear_cpu_cap(c, X86_FEATURE_VNMI);
213 clear_cpu_cap(c, X86_FEATURE_FLEXPRIORITY);
214 clear_cpu_cap(c, X86_FEATURE_EPT);
215 clear_cpu_cap(c, X86_FEATURE_VPID);
216
217 rdmsr(MSR_IA32_VMX_PROCBASED_CTLS, vmx_msr_low, vmx_msr_high);
218 msr_ctl = vmx_msr_high | vmx_msr_low;
219 if (msr_ctl & X86_VMX_FEATURE_PROC_CTLS_TPR_SHADOW)
220 set_cpu_cap(c, X86_FEATURE_TPR_SHADOW);
221 if (msr_ctl & X86_VMX_FEATURE_PROC_CTLS_VNMI)
222 set_cpu_cap(c, X86_FEATURE_VNMI);
223 if (msr_ctl & X86_VMX_FEATURE_PROC_CTLS_2ND_CTLS) {
224 rdmsr(MSR_IA32_VMX_PROCBASED_CTLS2,
225 vmx_msr_low, vmx_msr_high);
226 msr_ctl2 = vmx_msr_high | vmx_msr_low;
227 if ((msr_ctl2 & X86_VMX_FEATURE_PROC_CTLS2_VIRT_APIC) &&
228 (msr_ctl & X86_VMX_FEATURE_PROC_CTLS_TPR_SHADOW))
229 set_cpu_cap(c, X86_FEATURE_FLEXPRIORITY);
230 if (msr_ctl2 & X86_VMX_FEATURE_PROC_CTLS2_EPT)
231 set_cpu_cap(c, X86_FEATURE_EPT);
232 if (msr_ctl2 & X86_VMX_FEATURE_PROC_CTLS2_VPID)
233 set_cpu_cap(c, X86_FEATURE_VPID);
234 }
112} 235}
113#endif
114 236
115static void __cpuinit init_intel(struct cpuinfo_x86 *c) 237static void __cpuinit init_intel(struct cpuinfo_x86 *c)
116{ 238{
117 unsigned int l2 = 0; 239 unsigned int l2 = 0;
118 char *p = NULL;
119 240
120 early_init_intel(c); 241 early_init_intel(c);
121 242
122#ifdef CONFIG_X86_F00F_BUG 243 intel_workarounds(c);
123 /*
124 * All current models of Pentium and Pentium with MMX technology CPUs
125 * have the F0 0F bug, which lets nonprivileged users lock up the system.
126 * Note that the workaround only should be initialized once...
127 */
128 c->f00f_bug = 0;
129 if (!paravirt_enabled() && c->x86 == 5) {
130 static int f00f_workaround_enabled;
131
132 c->f00f_bug = 1;
133 if (!f00f_workaround_enabled) {
134 trap_init_f00f_bug();
135 printk(KERN_NOTICE "Intel Pentium with F0 0F bug - workaround enabled.\n");
136 f00f_workaround_enabled = 1;
137 }
138 }
139#endif
140 244
141 l2 = init_intel_cacheinfo(c); 245 l2 = init_intel_cacheinfo(c);
142 if (c->cpuid_level > 9) { 246 if (c->cpuid_level > 9) {
@@ -146,16 +250,32 @@ static void __cpuinit init_intel(struct cpuinfo_x86 *c)
146 set_cpu_cap(c, X86_FEATURE_ARCH_PERFMON); 250 set_cpu_cap(c, X86_FEATURE_ARCH_PERFMON);
147 } 251 }
148 252
149 /* SEP CPUID bug: Pentium Pro reports SEP but doesn't have it until model 3 mask 3 */ 253 if (cpu_has_xmm2)
150 if ((c->x86<<8 | c->x86_model<<4 | c->x86_mask) < 0x633) 254 set_cpu_cap(c, X86_FEATURE_LFENCE_RDTSC);
151 clear_cpu_cap(c, X86_FEATURE_SEP); 255 if (cpu_has_ds) {
256 unsigned int l1;
257 rdmsr(MSR_IA32_MISC_ENABLE, l1, l2);
258 if (!(l1 & (1<<11)))
259 set_cpu_cap(c, X86_FEATURE_BTS);
260 if (!(l1 & (1<<12)))
261 set_cpu_cap(c, X86_FEATURE_PEBS);
262 ds_init_intel(c);
263 }
152 264
265#ifdef CONFIG_X86_64
266 if (c->x86 == 15)
267 c->x86_cache_alignment = c->x86_clflush_size * 2;
268 if (c->x86 == 6)
269 set_cpu_cap(c, X86_FEATURE_REP_GOOD);
270#else
153 /* 271 /*
154 * Names for the Pentium II/Celeron processors 272 * Names for the Pentium II/Celeron processors
155 * detectable only by also checking the cache size. 273 * detectable only by also checking the cache size.
156 * Dixon is NOT a Celeron. 274 * Dixon is NOT a Celeron.
157 */ 275 */
158 if (c->x86 == 6) { 276 if (c->x86 == 6) {
277 char *p = NULL;
278
159 switch (c->x86_model) { 279 switch (c->x86_model) {
160 case 5: 280 case 5:
161 if (c->x86_mask == 0) { 281 if (c->x86_mask == 0) {
@@ -178,71 +298,41 @@ static void __cpuinit init_intel(struct cpuinfo_x86 *c)
178 p = "Celeron (Coppermine)"; 298 p = "Celeron (Coppermine)";
179 break; 299 break;
180 } 300 }
181 }
182
183 if (p)
184 strcpy(c->x86_model_id, p);
185
186 c->x86_max_cores = num_cpu_cores(c);
187
188 detect_ht(c);
189 301
190 /* Work around errata */ 302 if (p)
191 Intel_errata_workarounds(c); 303 strcpy(c->x86_model_id, p);
192
193#ifdef CONFIG_X86_INTEL_USERCOPY
194 /*
195 * Set up the preferred alignment for movsl bulk memory moves
196 */
197 switch (c->x86) {
198 case 4: /* 486: untested */
199 break;
200 case 5: /* Old Pentia: untested */
201 break;
202 case 6: /* PII/PIII only like movsl with 8-byte alignment */
203 movsl_mask.mask = 7;
204 break;
205 case 15: /* P4 is OK down to 8-byte alignment */
206 movsl_mask.mask = 7;
207 break;
208 } 304 }
209#endif
210 305
211 if (cpu_has_xmm2) 306 if (c->x86 == 15)
212 set_cpu_cap(c, X86_FEATURE_LFENCE_RDTSC);
213 if (c->x86 == 15) {
214 set_cpu_cap(c, X86_FEATURE_P4); 307 set_cpu_cap(c, X86_FEATURE_P4);
215 }
216 if (c->x86 == 6) 308 if (c->x86 == 6)
217 set_cpu_cap(c, X86_FEATURE_P3); 309 set_cpu_cap(c, X86_FEATURE_P3);
218 if (cpu_has_ds) {
219 unsigned int l1;
220 rdmsr(MSR_IA32_MISC_ENABLE, l1, l2);
221 if (!(l1 & (1<<11)))
222 set_cpu_cap(c, X86_FEATURE_BTS);
223 if (!(l1 & (1<<12)))
224 set_cpu_cap(c, X86_FEATURE_PEBS);
225 ds_init_intel(c);
226 }
227 310
228 if (cpu_has_bts) 311 if (cpu_has_bts)
229 ptrace_bts_init_intel(c); 312 ptrace_bts_init_intel(c);
230 313
231 /* 314#endif
232 * See if we have a good local APIC by checking for buggy Pentia,
233 * i.e. all B steppings and the C2 stepping of P54C when using their
234 * integrated APIC (see 11AP erratum in "Pentium Processor
235 * Specification Update").
236 */
237 if (cpu_has_apic && (c->x86<<8 | c->x86_model<<4) == 0x520 &&
238 (c->x86_mask < 0x6 || c->x86_mask == 0xb))
239 set_cpu_cap(c, X86_FEATURE_11AP);
240 315
241#ifdef CONFIG_X86_NUMAQ 316 detect_extended_topology(c);
242 numaq_tsc_disable(); 317 if (!cpu_has(c, X86_FEATURE_XTOPOLOGY)) {
318 /*
319 * let's use the legacy cpuid vector 0x1 and 0x4 for topology
320 * detection.
321 */
322 c->x86_max_cores = intel_num_cpu_cores(c);
323#ifdef CONFIG_X86_32
324 detect_ht(c);
243#endif 325#endif
326 }
327
328 /* Work around errata */
329 srat_detect_node();
330
331 if (cpu_has(c, X86_FEATURE_VMX))
332 detect_vmx_virtcap(c);
244} 333}
245 334
335#ifdef CONFIG_X86_32
246static unsigned int __cpuinit intel_size_cache(struct cpuinfo_x86 *c, unsigned int size) 336static unsigned int __cpuinit intel_size_cache(struct cpuinfo_x86 *c, unsigned int size)
247{ 337{
248 /* 338 /*
@@ -255,10 +345,12 @@ static unsigned int __cpuinit intel_size_cache(struct cpuinfo_x86 *c, unsigned i
255 size = 256; 345 size = 256;
256 return size; 346 return size;
257} 347}
348#endif
258 349
259static struct cpu_dev intel_cpu_dev __cpuinitdata = { 350static struct cpu_dev intel_cpu_dev __cpuinitdata = {
260 .c_vendor = "Intel", 351 .c_vendor = "Intel",
261 .c_ident = { "GenuineIntel" }, 352 .c_ident = { "GenuineIntel" },
353#ifdef CONFIG_X86_32
262 .c_models = { 354 .c_models = {
263 { .vendor = X86_VENDOR_INTEL, .family = 4, .model_names = 355 { .vendor = X86_VENDOR_INTEL, .family = 4, .model_names =
264 { 356 {
@@ -308,76 +400,12 @@ static struct cpu_dev intel_cpu_dev __cpuinitdata = {
308 } 400 }
309 }, 401 },
310 }, 402 },
403 .c_size_cache = intel_size_cache,
404#endif
311 .c_early_init = early_init_intel, 405 .c_early_init = early_init_intel,
312 .c_init = init_intel, 406 .c_init = init_intel,
313 .c_size_cache = intel_size_cache, 407 .c_x86_vendor = X86_VENDOR_INTEL,
314}; 408};
315 409
316cpu_vendor_dev_register(X86_VENDOR_INTEL, &intel_cpu_dev); 410cpu_dev_register(intel_cpu_dev);
317
318#ifndef CONFIG_X86_CMPXCHG
319unsigned long cmpxchg_386_u8(volatile void *ptr, u8 old, u8 new)
320{
321 u8 prev;
322 unsigned long flags;
323
324 /* Poor man's cmpxchg for 386. Unsuitable for SMP */
325 local_irq_save(flags);
326 prev = *(u8 *)ptr;
327 if (prev == old)
328 *(u8 *)ptr = new;
329 local_irq_restore(flags);
330 return prev;
331}
332EXPORT_SYMBOL(cmpxchg_386_u8);
333
334unsigned long cmpxchg_386_u16(volatile void *ptr, u16 old, u16 new)
335{
336 u16 prev;
337 unsigned long flags;
338
339 /* Poor man's cmpxchg for 386. Unsuitable for SMP */
340 local_irq_save(flags);
341 prev = *(u16 *)ptr;
342 if (prev == old)
343 *(u16 *)ptr = new;
344 local_irq_restore(flags);
345 return prev;
346}
347EXPORT_SYMBOL(cmpxchg_386_u16);
348
349unsigned long cmpxchg_386_u32(volatile void *ptr, u32 old, u32 new)
350{
351 u32 prev;
352 unsigned long flags;
353
354 /* Poor man's cmpxchg for 386. Unsuitable for SMP */
355 local_irq_save(flags);
356 prev = *(u32 *)ptr;
357 if (prev == old)
358 *(u32 *)ptr = new;
359 local_irq_restore(flags);
360 return prev;
361}
362EXPORT_SYMBOL(cmpxchg_386_u32);
363#endif
364
365#ifndef CONFIG_X86_CMPXCHG64
366unsigned long long cmpxchg_486_u64(volatile void *ptr, u64 old, u64 new)
367{
368 u64 prev;
369 unsigned long flags;
370
371 /* Poor man's cmpxchg8b for 386 and 486. Unsuitable for SMP */
372 local_irq_save(flags);
373 prev = *(u64 *)ptr;
374 if (prev == old)
375 *(u64 *)ptr = new;
376 local_irq_restore(flags);
377 return prev;
378}
379EXPORT_SYMBOL(cmpxchg_486_u64);
380#endif
381
382/* arch_initcall(intel_cpu_init); */
383 411
diff --git a/arch/x86/kernel/cpu/intel_64.c b/arch/x86/kernel/cpu/intel_64.c
deleted file mode 100644
index 1019c58d39f0..000000000000
--- a/arch/x86/kernel/cpu/intel_64.c
+++ /dev/null
@@ -1,95 +0,0 @@
1#include <linux/init.h>
2#include <linux/smp.h>
3#include <asm/processor.h>
4#include <asm/ptrace.h>
5#include <asm/topology.h>
6#include <asm/numa_64.h>
7
8#include "cpu.h"
9
10static void __cpuinit early_init_intel(struct cpuinfo_x86 *c)
11{
12 if ((c->x86 == 0xf && c->x86_model >= 0x03) ||
13 (c->x86 == 0x6 && c->x86_model >= 0x0e))
14 set_cpu_cap(c, X86_FEATURE_CONSTANT_TSC);
15
16 set_cpu_cap(c, X86_FEATURE_SYSENTER32);
17}
18
19/*
20 * find out the number of processor cores on the die
21 */
22static int __cpuinit intel_num_cpu_cores(struct cpuinfo_x86 *c)
23{
24 unsigned int eax, t;
25
26 if (c->cpuid_level < 4)
27 return 1;
28
29 cpuid_count(4, 0, &eax, &t, &t, &t);
30
31 if (eax & 0x1f)
32 return ((eax >> 26) + 1);
33 else
34 return 1;
35}
36
37static void __cpuinit srat_detect_node(void)
38{
39#ifdef CONFIG_NUMA
40 unsigned node;
41 int cpu = smp_processor_id();
42 int apicid = hard_smp_processor_id();
43
44 /* Don't do the funky fallback heuristics the AMD version employs
45 for now. */
46 node = apicid_to_node[apicid];
47 if (node == NUMA_NO_NODE || !node_online(node))
48 node = first_node(node_online_map);
49 numa_set_node(cpu, node);
50
51 printk(KERN_INFO "CPU %d/%x -> Node %d\n", cpu, apicid, node);
52#endif
53}
54
55static void __cpuinit init_intel(struct cpuinfo_x86 *c)
56{
57 init_intel_cacheinfo(c);
58 if (c->cpuid_level > 9) {
59 unsigned eax = cpuid_eax(10);
60 /* Check for version and the number of counters */
61 if ((eax & 0xff) && (((eax>>8) & 0xff) > 1))
62 set_cpu_cap(c, X86_FEATURE_ARCH_PERFMON);
63 }
64
65 if (cpu_has_ds) {
66 unsigned int l1, l2;
67 rdmsr(MSR_IA32_MISC_ENABLE, l1, l2);
68 if (!(l1 & (1<<11)))
69 set_cpu_cap(c, X86_FEATURE_BTS);
70 if (!(l1 & (1<<12)))
71 set_cpu_cap(c, X86_FEATURE_PEBS);
72 }
73
74
75 if (cpu_has_bts)
76 ds_init_intel(c);
77
78 if (c->x86 == 15)
79 c->x86_cache_alignment = c->x86_clflush_size * 2;
80 if (c->x86 == 6)
81 set_cpu_cap(c, X86_FEATURE_REP_GOOD);
82 set_cpu_cap(c, X86_FEATURE_LFENCE_RDTSC);
83 c->x86_max_cores = intel_num_cpu_cores(c);
84
85 srat_detect_node();
86}
87
88static struct cpu_dev intel_cpu_dev __cpuinitdata = {
89 .c_vendor = "Intel",
90 .c_ident = { "GenuineIntel" },
91 .c_early_init = early_init_intel,
92 .c_init = init_intel,
93};
94cpu_vendor_dev_register(X86_VENDOR_INTEL, &intel_cpu_dev);
95
diff --git a/arch/x86/kernel/cpu/intel_cacheinfo.c b/arch/x86/kernel/cpu/intel_cacheinfo.c
index 6b0a10b002f1..3f46afbb1cf1 100644
--- a/arch/x86/kernel/cpu/intel_cacheinfo.c
+++ b/arch/x86/kernel/cpu/intel_cacheinfo.c
@@ -1,8 +1,8 @@
1/* 1/*
2 * Routines to indentify caches on Intel CPU. 2 * Routines to indentify caches on Intel CPU.
3 * 3 *
4 * Changes: 4 * Changes:
5 * Venkatesh Pallipadi : Adding cache identification through cpuid(4) 5 * Venkatesh Pallipadi : Adding cache identification through cpuid(4)
6 * Ashok Raj <ashok.raj@intel.com>: Work with CPU hotplug infrastructure. 6 * Ashok Raj <ashok.raj@intel.com>: Work with CPU hotplug infrastructure.
7 * Andi Kleen / Andreas Herrmann : CPUID4 emulation on AMD. 7 * Andi Kleen / Andreas Herrmann : CPUID4 emulation on AMD.
8 */ 8 */
@@ -13,6 +13,7 @@
13#include <linux/compiler.h> 13#include <linux/compiler.h>
14#include <linux/cpu.h> 14#include <linux/cpu.h>
15#include <linux/sched.h> 15#include <linux/sched.h>
16#include <linux/pci.h>
16 17
17#include <asm/processor.h> 18#include <asm/processor.h>
18#include <asm/smp.h> 19#include <asm/smp.h>
@@ -130,9 +131,18 @@ struct _cpuid4_info {
130 union _cpuid4_leaf_ebx ebx; 131 union _cpuid4_leaf_ebx ebx;
131 union _cpuid4_leaf_ecx ecx; 132 union _cpuid4_leaf_ecx ecx;
132 unsigned long size; 133 unsigned long size;
134 unsigned long can_disable;
133 cpumask_t shared_cpu_map; /* future?: only cpus/node is needed */ 135 cpumask_t shared_cpu_map; /* future?: only cpus/node is needed */
134}; 136};
135 137
138#ifdef CONFIG_PCI
139static struct pci_device_id k8_nb_id[] = {
140 { PCI_DEVICE(PCI_VENDOR_ID_AMD, 0x1103) },
141 { PCI_DEVICE(PCI_VENDOR_ID_AMD, 0x1203) },
142 {}
143};
144#endif
145
136unsigned short num_cache_leaves; 146unsigned short num_cache_leaves;
137 147
138/* AMD doesn't have CPUID4. Emulate it here to report the same 148/* AMD doesn't have CPUID4. Emulate it here to report the same
@@ -182,9 +192,10 @@ static unsigned short assocs[] __cpuinitdata = {
182static unsigned char levels[] __cpuinitdata = { 1, 1, 2, 3 }; 192static unsigned char levels[] __cpuinitdata = { 1, 1, 2, 3 };
183static unsigned char types[] __cpuinitdata = { 1, 2, 3, 3 }; 193static unsigned char types[] __cpuinitdata = { 1, 2, 3, 3 };
184 194
185static void __cpuinit amd_cpuid4(int leaf, union _cpuid4_leaf_eax *eax, 195static void __cpuinit
186 union _cpuid4_leaf_ebx *ebx, 196amd_cpuid4(int leaf, union _cpuid4_leaf_eax *eax,
187 union _cpuid4_leaf_ecx *ecx) 197 union _cpuid4_leaf_ebx *ebx,
198 union _cpuid4_leaf_ecx *ecx)
188{ 199{
189 unsigned dummy; 200 unsigned dummy;
190 unsigned line_size, lines_per_tag, assoc, size_in_kb; 201 unsigned line_size, lines_per_tag, assoc, size_in_kb;
@@ -251,27 +262,40 @@ static void __cpuinit amd_cpuid4(int leaf, union _cpuid4_leaf_eax *eax,
251 (ebx->split.ways_of_associativity + 1) - 1; 262 (ebx->split.ways_of_associativity + 1) - 1;
252} 263}
253 264
254static int __cpuinit cpuid4_cache_lookup(int index, struct _cpuid4_info *this_leaf) 265static void __cpuinit
266amd_check_l3_disable(int index, struct _cpuid4_info *this_leaf)
267{
268 if (index < 3)
269 return;
270 this_leaf->can_disable = 1;
271}
272
273static int
274__cpuinit cpuid4_cache_lookup(int index, struct _cpuid4_info *this_leaf)
255{ 275{
256 union _cpuid4_leaf_eax eax; 276 union _cpuid4_leaf_eax eax;
257 union _cpuid4_leaf_ebx ebx; 277 union _cpuid4_leaf_ebx ebx;
258 union _cpuid4_leaf_ecx ecx; 278 union _cpuid4_leaf_ecx ecx;
259 unsigned edx; 279 unsigned edx;
260 280
261 if (boot_cpu_data.x86_vendor == X86_VENDOR_AMD) 281 if (boot_cpu_data.x86_vendor == X86_VENDOR_AMD) {
262 amd_cpuid4(index, &eax, &ebx, &ecx); 282 amd_cpuid4(index, &eax, &ebx, &ecx);
263 else 283 if (boot_cpu_data.x86 >= 0x10)
264 cpuid_count(4, index, &eax.full, &ebx.full, &ecx.full, &edx); 284 amd_check_l3_disable(index, this_leaf);
285 } else {
286 cpuid_count(4, index, &eax.full, &ebx.full, &ecx.full, &edx);
287 }
288
265 if (eax.split.type == CACHE_TYPE_NULL) 289 if (eax.split.type == CACHE_TYPE_NULL)
266 return -EIO; /* better error ? */ 290 return -EIO; /* better error ? */
267 291
268 this_leaf->eax = eax; 292 this_leaf->eax = eax;
269 this_leaf->ebx = ebx; 293 this_leaf->ebx = ebx;
270 this_leaf->ecx = ecx; 294 this_leaf->ecx = ecx;
271 this_leaf->size = (ecx.split.number_of_sets + 1) * 295 this_leaf->size = (ecx.split.number_of_sets + 1) *
272 (ebx.split.coherency_line_size + 1) * 296 (ebx.split.coherency_line_size + 1) *
273 (ebx.split.physical_line_partition + 1) * 297 (ebx.split.physical_line_partition + 1) *
274 (ebx.split.ways_of_associativity + 1); 298 (ebx.split.ways_of_associativity + 1);
275 return 0; 299 return 0;
276} 300}
277 301
@@ -453,7 +477,7 @@ unsigned int __cpuinit init_intel_cacheinfo(struct cpuinfo_x86 *c)
453 477
454/* pointer to _cpuid4_info array (for each cache leaf) */ 478/* pointer to _cpuid4_info array (for each cache leaf) */
455static DEFINE_PER_CPU(struct _cpuid4_info *, cpuid4_info); 479static DEFINE_PER_CPU(struct _cpuid4_info *, cpuid4_info);
456#define CPUID4_INFO_IDX(x, y) (&((per_cpu(cpuid4_info, x))[y])) 480#define CPUID4_INFO_IDX(x, y) (&((per_cpu(cpuid4_info, x))[y]))
457 481
458#ifdef CONFIG_SMP 482#ifdef CONFIG_SMP
459static void __cpuinit cache_shared_cpu_map_setup(unsigned int cpu, int index) 483static void __cpuinit cache_shared_cpu_map_setup(unsigned int cpu, int index)
@@ -490,7 +514,7 @@ static void __cpuinit cache_remove_shared_cpu_map(unsigned int cpu, int index)
490 514
491 this_leaf = CPUID4_INFO_IDX(cpu, index); 515 this_leaf = CPUID4_INFO_IDX(cpu, index);
492 for_each_cpu_mask_nr(sibling, this_leaf->shared_cpu_map) { 516 for_each_cpu_mask_nr(sibling, this_leaf->shared_cpu_map) {
493 sibling_leaf = CPUID4_INFO_IDX(sibling, index); 517 sibling_leaf = CPUID4_INFO_IDX(sibling, index);
494 cpu_clear(cpu, sibling_leaf->shared_cpu_map); 518 cpu_clear(cpu, sibling_leaf->shared_cpu_map);
495 } 519 }
496} 520}
@@ -572,7 +596,7 @@ struct _index_kobject {
572 596
573/* pointer to array of kobjects for cpuX/cache/indexY */ 597/* pointer to array of kobjects for cpuX/cache/indexY */
574static DEFINE_PER_CPU(struct _index_kobject *, index_kobject); 598static DEFINE_PER_CPU(struct _index_kobject *, index_kobject);
575#define INDEX_KOBJECT_PTR(x, y) (&((per_cpu(index_kobject, x))[y])) 599#define INDEX_KOBJECT_PTR(x, y) (&((per_cpu(index_kobject, x))[y]))
576 600
577#define show_one_plus(file_name, object, val) \ 601#define show_one_plus(file_name, object, val) \
578static ssize_t show_##file_name \ 602static ssize_t show_##file_name \
@@ -637,6 +661,99 @@ static ssize_t show_type(struct _cpuid4_info *this_leaf, char *buf) {
637 } 661 }
638} 662}
639 663
664#define to_object(k) container_of(k, struct _index_kobject, kobj)
665#define to_attr(a) container_of(a, struct _cache_attr, attr)
666
667#ifdef CONFIG_PCI
668static struct pci_dev *get_k8_northbridge(int node)
669{
670 struct pci_dev *dev = NULL;
671 int i;
672
673 for (i = 0; i <= node; i++) {
674 do {
675 dev = pci_get_device(PCI_ANY_ID, PCI_ANY_ID, dev);
676 if (!dev)
677 break;
678 } while (!pci_match_id(&k8_nb_id[0], dev));
679 if (!dev)
680 break;
681 }
682 return dev;
683}
684#else
685static struct pci_dev *get_k8_northbridge(int node)
686{
687 return NULL;
688}
689#endif
690
691static ssize_t show_cache_disable(struct _cpuid4_info *this_leaf, char *buf)
692{
693 int node = cpu_to_node(first_cpu(this_leaf->shared_cpu_map));
694 struct pci_dev *dev = NULL;
695 ssize_t ret = 0;
696 int i;
697
698 if (!this_leaf->can_disable)
699 return sprintf(buf, "Feature not enabled\n");
700
701 dev = get_k8_northbridge(node);
702 if (!dev) {
703 printk(KERN_ERR "Attempting AMD northbridge operation on a system with no northbridge\n");
704 return -EINVAL;
705 }
706
707 for (i = 0; i < 2; i++) {
708 unsigned int reg;
709
710 pci_read_config_dword(dev, 0x1BC + i * 4, &reg);
711
712 ret += sprintf(buf, "%sEntry: %d\n", buf, i);
713 ret += sprintf(buf, "%sReads: %s\tNew Entries: %s\n",
714 buf,
715 reg & 0x80000000 ? "Disabled" : "Allowed",
716 reg & 0x40000000 ? "Disabled" : "Allowed");
717 ret += sprintf(buf, "%sSubCache: %x\tIndex: %x\n",
718 buf, (reg & 0x30000) >> 16, reg & 0xfff);
719 }
720 return ret;
721}
722
723static ssize_t
724store_cache_disable(struct _cpuid4_info *this_leaf, const char *buf,
725 size_t count)
726{
727 int node = cpu_to_node(first_cpu(this_leaf->shared_cpu_map));
728 struct pci_dev *dev = NULL;
729 unsigned int ret, index, val;
730
731 if (!this_leaf->can_disable)
732 return 0;
733
734 if (strlen(buf) > 15)
735 return -EINVAL;
736
737 ret = sscanf(buf, "%x %x", &index, &val);
738 if (ret != 2)
739 return -EINVAL;
740 if (index > 1)
741 return -EINVAL;
742
743 val |= 0xc0000000;
744 dev = get_k8_northbridge(node);
745 if (!dev) {
746 printk(KERN_ERR "Attempting AMD northbridge operation on a system with no northbridge\n");
747 return -EINVAL;
748 }
749
750 pci_write_config_dword(dev, 0x1BC + index * 4, val & ~0x40000000);
751 wbinvd();
752 pci_write_config_dword(dev, 0x1BC + index * 4, val);
753
754 return 1;
755}
756
640struct _cache_attr { 757struct _cache_attr {
641 struct attribute attr; 758 struct attribute attr;
642 ssize_t (*show)(struct _cpuid4_info *, char *); 759 ssize_t (*show)(struct _cpuid4_info *, char *);
@@ -657,6 +774,8 @@ define_one_ro(size);
657define_one_ro(shared_cpu_map); 774define_one_ro(shared_cpu_map);
658define_one_ro(shared_cpu_list); 775define_one_ro(shared_cpu_list);
659 776
777static struct _cache_attr cache_disable = __ATTR(cache_disable, 0644, show_cache_disable, store_cache_disable);
778
660static struct attribute * default_attrs[] = { 779static struct attribute * default_attrs[] = {
661 &type.attr, 780 &type.attr,
662 &level.attr, 781 &level.attr,
@@ -667,12 +786,10 @@ static struct attribute * default_attrs[] = {
667 &size.attr, 786 &size.attr,
668 &shared_cpu_map.attr, 787 &shared_cpu_map.attr,
669 &shared_cpu_list.attr, 788 &shared_cpu_list.attr,
789 &cache_disable.attr,
670 NULL 790 NULL
671}; 791};
672 792
673#define to_object(k) container_of(k, struct _index_kobject, kobj)
674#define to_attr(a) container_of(a, struct _cache_attr, attr)
675
676static ssize_t show(struct kobject * kobj, struct attribute * attr, char * buf) 793static ssize_t show(struct kobject * kobj, struct attribute * attr, char * buf)
677{ 794{
678 struct _cache_attr *fattr = to_attr(attr); 795 struct _cache_attr *fattr = to_attr(attr);
@@ -682,14 +799,22 @@ static ssize_t show(struct kobject * kobj, struct attribute * attr, char * buf)
682 ret = fattr->show ? 799 ret = fattr->show ?
683 fattr->show(CPUID4_INFO_IDX(this_leaf->cpu, this_leaf->index), 800 fattr->show(CPUID4_INFO_IDX(this_leaf->cpu, this_leaf->index),
684 buf) : 801 buf) :
685 0; 802 0;
686 return ret; 803 return ret;
687} 804}
688 805
689static ssize_t store(struct kobject * kobj, struct attribute * attr, 806static ssize_t store(struct kobject * kobj, struct attribute * attr,
690 const char * buf, size_t count) 807 const char * buf, size_t count)
691{ 808{
692 return 0; 809 struct _cache_attr *fattr = to_attr(attr);
810 struct _index_kobject *this_leaf = to_object(kobj);
811 ssize_t ret;
812
813 ret = fattr->store ?
814 fattr->store(CPUID4_INFO_IDX(this_leaf->cpu, this_leaf->index),
815 buf, count) :
816 0;
817 return ret;
693} 818}
694 819
695static struct sysfs_ops sysfs_ops = { 820static struct sysfs_ops sysfs_ops = {
diff --git a/arch/x86/kernel/cpu/mcheck/mce_64.c b/arch/x86/kernel/cpu/mcheck/mce_64.c
index 726a5fcdf341..4b031a4ac856 100644
--- a/arch/x86/kernel/cpu/mcheck/mce_64.c
+++ b/arch/x86/kernel/cpu/mcheck/mce_64.c
@@ -860,7 +860,7 @@ error:
860 return err; 860 return err;
861} 861}
862 862
863static void mce_remove_device(unsigned int cpu) 863static __cpuinit void mce_remove_device(unsigned int cpu)
864{ 864{
865 int i; 865 int i;
866 866
diff --git a/arch/x86/kernel/cpu/mkcapflags.pl b/arch/x86/kernel/cpu/mkcapflags.pl
new file mode 100644
index 000000000000..dfea390e1608
--- /dev/null
+++ b/arch/x86/kernel/cpu/mkcapflags.pl
@@ -0,0 +1,32 @@
1#!/usr/bin/perl
2#
3# Generate the x86_cap_flags[] array from include/asm-x86/cpufeature.h
4#
5
6($in, $out) = @ARGV;
7
8open(IN, "< $in\0") or die "$0: cannot open: $in: $!\n";
9open(OUT, "> $out\0") or die "$0: cannot create: $out: $!\n";
10
11print OUT "#include <asm/cpufeature.h>\n\n";
12print OUT "const char * const x86_cap_flags[NCAPINTS*32] = {\n";
13
14while (defined($line = <IN>)) {
15 if ($line =~ /^\s*\#\s*define\s+(X86_FEATURE_(\S+))\s+(.*)$/) {
16 $macro = $1;
17 $feature = $2;
18 $tail = $3;
19 if ($tail =~ /\/\*\s*\"([^"]*)\".*\*\//) {
20 $feature = $1;
21 }
22
23 if ($feature ne '') {
24 printf OUT "\t%-32s = \"%s\",\n",
25 "[$macro]", "\L$feature";
26 }
27 }
28}
29print OUT "};\n";
30
31close(IN);
32close(OUT);
diff --git a/arch/x86/kernel/cpu/powerflags.c b/arch/x86/kernel/cpu/powerflags.c
new file mode 100644
index 000000000000..5abbea297e0c
--- /dev/null
+++ b/arch/x86/kernel/cpu/powerflags.c
@@ -0,0 +1,20 @@
1/*
2 * Strings for the various x86 power flags
3 *
4 * This file must not contain any executable code.
5 */
6
7#include <asm/cpufeature.h>
8
9const char *const x86_power_flags[32] = {
10 "ts", /* temperature sensor */
11 "fid", /* frequency id control */
12 "vid", /* voltage id control */
13 "ttp", /* thermal trip */
14 "tm",
15 "stc",
16 "100mhzsteps",
17 "hwpstate",
18 "", /* tsc invariant mapped to constant_tsc */
19 /* nothing */
20};
diff --git a/arch/x86/kernel/cpu/transmeta.c b/arch/x86/kernel/cpu/transmeta.c
index b911a2c61b8f..52b3fefbd5af 100644
--- a/arch/x86/kernel/cpu/transmeta.c
+++ b/arch/x86/kernel/cpu/transmeta.c
@@ -5,6 +5,18 @@
5#include <asm/msr.h> 5#include <asm/msr.h>
6#include "cpu.h" 6#include "cpu.h"
7 7
8static void __cpuinit early_init_transmeta(struct cpuinfo_x86 *c)
9{
10 u32 xlvl;
11
12 /* Transmeta-defined flags: level 0x80860001 */
13 xlvl = cpuid_eax(0x80860000);
14 if ((xlvl & 0xffff0000) == 0x80860000) {
15 if (xlvl >= 0x80860001)
16 c->x86_capability[2] = cpuid_edx(0x80860001);
17 }
18}
19
8static void __cpuinit init_transmeta(struct cpuinfo_x86 *c) 20static void __cpuinit init_transmeta(struct cpuinfo_x86 *c)
9{ 21{
10 unsigned int cap_mask, uk, max, dummy; 22 unsigned int cap_mask, uk, max, dummy;
@@ -12,7 +24,8 @@ static void __cpuinit init_transmeta(struct cpuinfo_x86 *c)
12 unsigned int cpu_rev, cpu_freq = 0, cpu_flags, new_cpu_rev; 24 unsigned int cpu_rev, cpu_freq = 0, cpu_flags, new_cpu_rev;
13 char cpu_info[65]; 25 char cpu_info[65];
14 26
15 get_model_name(c); /* Same as AMD/Cyrix */ 27 early_init_transmeta(c);
28
16 display_cacheinfo(c); 29 display_cacheinfo(c);
17 30
18 /* Print CMS and CPU revision */ 31 /* Print CMS and CPU revision */
@@ -85,23 +98,12 @@ static void __cpuinit init_transmeta(struct cpuinfo_x86 *c)
85#endif 98#endif
86} 99}
87 100
88static void __cpuinit transmeta_identify(struct cpuinfo_x86 *c)
89{
90 u32 xlvl;
91
92 /* Transmeta-defined flags: level 0x80860001 */
93 xlvl = cpuid_eax(0x80860000);
94 if ((xlvl & 0xffff0000) == 0x80860000) {
95 if (xlvl >= 0x80860001)
96 c->x86_capability[2] = cpuid_edx(0x80860001);
97 }
98}
99
100static struct cpu_dev transmeta_cpu_dev __cpuinitdata = { 101static struct cpu_dev transmeta_cpu_dev __cpuinitdata = {
101 .c_vendor = "Transmeta", 102 .c_vendor = "Transmeta",
102 .c_ident = { "GenuineTMx86", "TransmetaCPU" }, 103 .c_ident = { "GenuineTMx86", "TransmetaCPU" },
104 .c_early_init = early_init_transmeta,
103 .c_init = init_transmeta, 105 .c_init = init_transmeta,
104 .c_identify = transmeta_identify, 106 .c_x86_vendor = X86_VENDOR_TRANSMETA,
105}; 107};
106 108
107cpu_vendor_dev_register(X86_VENDOR_TRANSMETA, &transmeta_cpu_dev); 109cpu_dev_register(transmeta_cpu_dev);
diff --git a/arch/x86/kernel/cpu/umc.c b/arch/x86/kernel/cpu/umc.c
index b1fc90989d75..e777f79e0960 100644
--- a/arch/x86/kernel/cpu/umc.c
+++ b/arch/x86/kernel/cpu/umc.c
@@ -19,7 +19,8 @@ static struct cpu_dev umc_cpu_dev __cpuinitdata = {
19 } 19 }
20 }, 20 },
21 }, 21 },
22 .c_x86_vendor = X86_VENDOR_UMC,
22}; 23};
23 24
24cpu_vendor_dev_register(X86_VENDOR_UMC, &umc_cpu_dev); 25cpu_dev_register(umc_cpu_dev);
25 26
diff --git a/arch/x86/kernel/e820.c b/arch/x86/kernel/e820.c
index 66e48aa2dd1b..78e642feac30 100644
--- a/arch/x86/kernel/e820.c
+++ b/arch/x86/kernel/e820.c
@@ -148,6 +148,9 @@ void __init e820_print_map(char *who)
148 case E820_NVS: 148 case E820_NVS:
149 printk(KERN_CONT "(ACPI NVS)\n"); 149 printk(KERN_CONT "(ACPI NVS)\n");
150 break; 150 break;
151 case E820_UNUSABLE:
152 printk("(unusable)\n");
153 break;
151 default: 154 default:
152 printk(KERN_CONT "type %u\n", e820.map[i].type); 155 printk(KERN_CONT "type %u\n", e820.map[i].type);
153 break; 156 break;
@@ -1260,6 +1263,7 @@ static inline const char *e820_type_to_string(int e820_type)
1260 case E820_RAM: return "System RAM"; 1263 case E820_RAM: return "System RAM";
1261 case E820_ACPI: return "ACPI Tables"; 1264 case E820_ACPI: return "ACPI Tables";
1262 case E820_NVS: return "ACPI Non-volatile Storage"; 1265 case E820_NVS: return "ACPI Non-volatile Storage";
1266 case E820_UNUSABLE: return "Unusable memory";
1263 default: return "reserved"; 1267 default: return "reserved";
1264 } 1268 }
1265} 1269}
@@ -1267,6 +1271,7 @@ static inline const char *e820_type_to_string(int e820_type)
1267/* 1271/*
1268 * Mark e820 reserved areas as busy for the resource manager. 1272 * Mark e820 reserved areas as busy for the resource manager.
1269 */ 1273 */
1274static struct resource __initdata *e820_res;
1270void __init e820_reserve_resources(void) 1275void __init e820_reserve_resources(void)
1271{ 1276{
1272 int i; 1277 int i;
@@ -1274,6 +1279,7 @@ void __init e820_reserve_resources(void)
1274 u64 end; 1279 u64 end;
1275 1280
1276 res = alloc_bootmem_low(sizeof(struct resource) * e820.nr_map); 1281 res = alloc_bootmem_low(sizeof(struct resource) * e820.nr_map);
1282 e820_res = res;
1277 for (i = 0; i < e820.nr_map; i++) { 1283 for (i = 0; i < e820.nr_map; i++) {
1278 end = e820.map[i].addr + e820.map[i].size - 1; 1284 end = e820.map[i].addr + e820.map[i].size - 1;
1279#ifndef CONFIG_RESOURCES_64BIT 1285#ifndef CONFIG_RESOURCES_64BIT
@@ -1287,7 +1293,14 @@ void __init e820_reserve_resources(void)
1287 res->end = end; 1293 res->end = end;
1288 1294
1289 res->flags = IORESOURCE_MEM | IORESOURCE_BUSY; 1295 res->flags = IORESOURCE_MEM | IORESOURCE_BUSY;
1290 insert_resource(&iomem_resource, res); 1296
1297 /*
1298 * don't register the region that could be conflicted with
1299 * pci device BAR resource and insert them later in
1300 * pcibios_resource_survey()
1301 */
1302 if (e820.map[i].type != E820_RESERVED || res->start < (1ULL<<20))
1303 insert_resource(&iomem_resource, res);
1291 res++; 1304 res++;
1292 } 1305 }
1293 1306
@@ -1299,6 +1312,19 @@ void __init e820_reserve_resources(void)
1299 } 1312 }
1300} 1313}
1301 1314
1315void __init e820_reserve_resources_late(void)
1316{
1317 int i;
1318 struct resource *res;
1319
1320 res = e820_res;
1321 for (i = 0; i < e820.nr_map; i++) {
1322 if (!res->parent && res->end)
1323 reserve_region_with_split(&iomem_resource, res->start, res->end, res->name);
1324 res++;
1325 }
1326}
1327
1302char *__init default_machine_specific_memory_setup(void) 1328char *__init default_machine_specific_memory_setup(void)
1303{ 1329{
1304 char *who = "BIOS-e820"; 1330 char *who = "BIOS-e820";
diff --git a/arch/x86/es7000/es7000plat.c b/arch/x86/kernel/es7000_32.c
index 7789fde13c3f..849e5cd485b8 100644
--- a/arch/x86/es7000/es7000plat.c
+++ b/arch/x86/kernel/es7000_32.c
@@ -39,10 +39,93 @@
39#include <asm/nmi.h> 39#include <asm/nmi.h>
40#include <asm/smp.h> 40#include <asm/smp.h>
41#include <asm/apicdef.h> 41#include <asm/apicdef.h>
42#include "es7000.h"
43#include <mach_mpparse.h> 42#include <mach_mpparse.h>
44 43
45/* 44/*
45 * ES7000 chipsets
46 */
47
48#define NON_UNISYS 0
49#define ES7000_CLASSIC 1
50#define ES7000_ZORRO 2
51
52
53#define MIP_REG 1
54#define MIP_PSAI_REG 4
55
56#define MIP_BUSY 1
57#define MIP_SPIN 0xf0000
58#define MIP_VALID 0x0100000000000000ULL
59#define MIP_PORT(VALUE) ((VALUE >> 32) & 0xffff)
60
61#define MIP_RD_LO(VALUE) (VALUE & 0xffffffff)
62
63struct mip_reg_info {
64 unsigned long long mip_info;
65 unsigned long long delivery_info;
66 unsigned long long host_reg;
67 unsigned long long mip_reg;
68};
69
70struct part_info {
71 unsigned char type;
72 unsigned char length;
73 unsigned char part_id;
74 unsigned char apic_mode;
75 unsigned long snum;
76 char ptype[16];
77 char sname[64];
78 char pname[64];
79};
80
81struct psai {
82 unsigned long long entry_type;
83 unsigned long long addr;
84 unsigned long long bep_addr;
85};
86
87struct es7000_mem_info {
88 unsigned char type;
89 unsigned char length;
90 unsigned char resv[6];
91 unsigned long long start;
92 unsigned long long size;
93};
94
95struct es7000_oem_table {
96 unsigned long long hdr;
97 struct mip_reg_info mip;
98 struct part_info pif;
99 struct es7000_mem_info shm;
100 struct psai psai;
101};
102
103#ifdef CONFIG_ACPI
104
105struct oem_table {
106 struct acpi_table_header Header;
107 u32 OEMTableAddr;
108 u32 OEMTableSize;
109};
110
111extern int find_unisys_acpi_oem_table(unsigned long *oem_addr);
112#endif
113
114struct mip_reg {
115 unsigned long long off_0;
116 unsigned long long off_8;
117 unsigned long long off_10;
118 unsigned long long off_18;
119 unsigned long long off_20;
120 unsigned long long off_28;
121 unsigned long long off_30;
122 unsigned long long off_38;
123};
124
125#define MIP_SW_APIC 0x1020b
126#define MIP_FUNC(VALUE) (VALUE & 0xff)
127
128/*
46 * ES7000 Globals 129 * ES7000 Globals
47 */ 130 */
48 131
diff --git a/arch/x86/kernel/genx2apic_cluster.c b/arch/x86/kernel/genx2apic_cluster.c
index fed9f68efd66..e4bf2cc0d743 100644
--- a/arch/x86/kernel/genx2apic_cluster.c
+++ b/arch/x86/kernel/genx2apic_cluster.c
@@ -120,14 +120,9 @@ static unsigned long set_apic_id(unsigned int id)
120 return x; 120 return x;
121} 121}
122 122
123static unsigned int x2apic_read_id(void)
124{
125 return apic_read(APIC_ID);
126}
127
128static unsigned int phys_pkg_id(int index_msb) 123static unsigned int phys_pkg_id(int index_msb)
129{ 124{
130 return x2apic_read_id() >> index_msb; 125 return current_cpu_data.initial_apicid >> index_msb;
131} 126}
132 127
133static void x2apic_send_IPI_self(int vector) 128static void x2apic_send_IPI_self(int vector)
diff --git a/arch/x86/kernel/genx2apic_phys.c b/arch/x86/kernel/genx2apic_phys.c
index 958d537b4cc9..8f1343df2627 100644
--- a/arch/x86/kernel/genx2apic_phys.c
+++ b/arch/x86/kernel/genx2apic_phys.c
@@ -118,14 +118,9 @@ static unsigned long set_apic_id(unsigned int id)
118 return x; 118 return x;
119} 119}
120 120
121static unsigned int x2apic_read_id(void)
122{
123 return apic_read(APIC_ID);
124}
125
126static unsigned int phys_pkg_id(int index_msb) 121static unsigned int phys_pkg_id(int index_msb)
127{ 122{
128 return x2apic_read_id() >> index_msb; 123 return current_cpu_data.initial_apicid >> index_msb;
129} 124}
130 125
131void x2apic_send_IPI_self(int vector) 126void x2apic_send_IPI_self(int vector)
diff --git a/arch/x86/kernel/i387.c b/arch/x86/kernel/i387.c
index eb9ddd8efb82..45723f1fe198 100644
--- a/arch/x86/kernel/i387.c
+++ b/arch/x86/kernel/i387.c
@@ -21,9 +21,12 @@
21# include <asm/sigcontext32.h> 21# include <asm/sigcontext32.h>
22# include <asm/user32.h> 22# include <asm/user32.h>
23#else 23#else
24# define save_i387_ia32 save_i387 24# define save_i387_xstate_ia32 save_i387_xstate
25# define restore_i387_ia32 restore_i387 25# define restore_i387_xstate_ia32 restore_i387_xstate
26# define _fpstate_ia32 _fpstate 26# define _fpstate_ia32 _fpstate
27# define _xstate_ia32 _xstate
28# define sig_xstate_ia32_size sig_xstate_size
29# define fx_sw_reserved_ia32 fx_sw_reserved
27# define user_i387_ia32_struct user_i387_struct 30# define user_i387_ia32_struct user_i387_struct
28# define user32_fxsr_struct user_fxsr_struct 31# define user32_fxsr_struct user_fxsr_struct
29#endif 32#endif
@@ -36,6 +39,7 @@
36 39
37static unsigned int mxcsr_feature_mask __read_mostly = 0xffffffffu; 40static unsigned int mxcsr_feature_mask __read_mostly = 0xffffffffu;
38unsigned int xstate_size; 41unsigned int xstate_size;
42unsigned int sig_xstate_ia32_size = sizeof(struct _fpstate_ia32);
39static struct i387_fxsave_struct fx_scratch __cpuinitdata; 43static struct i387_fxsave_struct fx_scratch __cpuinitdata;
40 44
41void __cpuinit mxcsr_feature_mask_init(void) 45void __cpuinit mxcsr_feature_mask_init(void)
@@ -61,6 +65,11 @@ void __init init_thread_xstate(void)
61 return; 65 return;
62 } 66 }
63 67
68 if (cpu_has_xsave) {
69 xsave_cntxt_init();
70 return;
71 }
72
64 if (cpu_has_fxsr) 73 if (cpu_has_fxsr)
65 xstate_size = sizeof(struct i387_fxsave_struct); 74 xstate_size = sizeof(struct i387_fxsave_struct);
66#ifdef CONFIG_X86_32 75#ifdef CONFIG_X86_32
@@ -83,9 +92,19 @@ void __cpuinit fpu_init(void)
83 92
84 write_cr0(oldcr0 & ~(X86_CR0_TS|X86_CR0_EM)); /* clear TS and EM */ 93 write_cr0(oldcr0 & ~(X86_CR0_TS|X86_CR0_EM)); /* clear TS and EM */
85 94
95 /*
96 * Boot processor to setup the FP and extended state context info.
97 */
98 if (!smp_processor_id())
99 init_thread_xstate();
100 xsave_init();
101
86 mxcsr_feature_mask_init(); 102 mxcsr_feature_mask_init();
87 /* clean state in init */ 103 /* clean state in init */
88 current_thread_info()->status = 0; 104 if (cpu_has_xsave)
105 current_thread_info()->status = TS_XSAVE;
106 else
107 current_thread_info()->status = 0;
89 clear_used_math(); 108 clear_used_math();
90} 109}
91#endif /* CONFIG_X86_64 */ 110#endif /* CONFIG_X86_64 */
@@ -195,6 +214,13 @@ int xfpregs_set(struct task_struct *target, const struct user_regset *regset,
195 */ 214 */
196 target->thread.xstate->fxsave.mxcsr &= mxcsr_feature_mask; 215 target->thread.xstate->fxsave.mxcsr &= mxcsr_feature_mask;
197 216
217 /*
218 * update the header bits in the xsave header, indicating the
219 * presence of FP and SSE state.
220 */
221 if (cpu_has_xsave)
222 target->thread.xstate->xsave.xsave_hdr.xstate_bv |= XSTATE_FPSSE;
223
198 return ret; 224 return ret;
199} 225}
200 226
@@ -395,6 +421,12 @@ int fpregs_set(struct task_struct *target, const struct user_regset *regset,
395 if (!ret) 421 if (!ret)
396 convert_to_fxsr(target, &env); 422 convert_to_fxsr(target, &env);
397 423
424 /*
425 * update the header bit in the xsave header, indicating the
426 * presence of FP.
427 */
428 if (cpu_has_xsave)
429 target->thread.xstate->xsave.xsave_hdr.xstate_bv |= XSTATE_FP;
398 return ret; 430 return ret;
399} 431}
400 432
@@ -407,7 +439,6 @@ static inline int save_i387_fsave(struct _fpstate_ia32 __user *buf)
407 struct task_struct *tsk = current; 439 struct task_struct *tsk = current;
408 struct i387_fsave_struct *fp = &tsk->thread.xstate->fsave; 440 struct i387_fsave_struct *fp = &tsk->thread.xstate->fsave;
409 441
410 unlazy_fpu(tsk);
411 fp->status = fp->swd; 442 fp->status = fp->swd;
412 if (__copy_to_user(buf, fp, sizeof(struct i387_fsave_struct))) 443 if (__copy_to_user(buf, fp, sizeof(struct i387_fsave_struct)))
413 return -1; 444 return -1;
@@ -421,8 +452,6 @@ static int save_i387_fxsave(struct _fpstate_ia32 __user *buf)
421 struct user_i387_ia32_struct env; 452 struct user_i387_ia32_struct env;
422 int err = 0; 453 int err = 0;
423 454
424 unlazy_fpu(tsk);
425
426 convert_from_fxsr(&env, tsk); 455 convert_from_fxsr(&env, tsk);
427 if (__copy_to_user(buf, &env, sizeof(env))) 456 if (__copy_to_user(buf, &env, sizeof(env)))
428 return -1; 457 return -1;
@@ -432,16 +461,40 @@ static int save_i387_fxsave(struct _fpstate_ia32 __user *buf)
432 if (err) 461 if (err)
433 return -1; 462 return -1;
434 463
435 if (__copy_to_user(&buf->_fxsr_env[0], fx, 464 if (__copy_to_user(&buf->_fxsr_env[0], fx, xstate_size))
436 sizeof(struct i387_fxsave_struct)))
437 return -1; 465 return -1;
438 return 1; 466 return 1;
439} 467}
440 468
441int save_i387_ia32(struct _fpstate_ia32 __user *buf) 469static int save_i387_xsave(void __user *buf)
470{
471 struct _fpstate_ia32 __user *fx = buf;
472 int err = 0;
473
474 if (save_i387_fxsave(fx) < 0)
475 return -1;
476
477 err = __copy_to_user(&fx->sw_reserved, &fx_sw_reserved_ia32,
478 sizeof(struct _fpx_sw_bytes));
479 err |= __put_user(FP_XSTATE_MAGIC2,
480 (__u32 __user *) (buf + sig_xstate_ia32_size
481 - FP_XSTATE_MAGIC2_SIZE));
482 if (err)
483 return -1;
484
485 return 1;
486}
487
488int save_i387_xstate_ia32(void __user *buf)
442{ 489{
490 struct _fpstate_ia32 __user *fp = (struct _fpstate_ia32 __user *) buf;
491 struct task_struct *tsk = current;
492
443 if (!used_math()) 493 if (!used_math())
444 return 0; 494 return 0;
495
496 if (!access_ok(VERIFY_WRITE, buf, sig_xstate_ia32_size))
497 return -EACCES;
445 /* 498 /*
446 * This will cause a "finit" to be triggered by the next 499 * This will cause a "finit" to be triggered by the next
447 * attempted FPU operation by the 'current' process. 500 * attempted FPU operation by the 'current' process.
@@ -451,13 +504,17 @@ int save_i387_ia32(struct _fpstate_ia32 __user *buf)
451 if (!HAVE_HWFP) { 504 if (!HAVE_HWFP) {
452 return fpregs_soft_get(current, NULL, 505 return fpregs_soft_get(current, NULL,
453 0, sizeof(struct user_i387_ia32_struct), 506 0, sizeof(struct user_i387_ia32_struct),
454 NULL, buf) ? -1 : 1; 507 NULL, fp) ? -1 : 1;
455 } 508 }
456 509
510 unlazy_fpu(tsk);
511
512 if (cpu_has_xsave)
513 return save_i387_xsave(fp);
457 if (cpu_has_fxsr) 514 if (cpu_has_fxsr)
458 return save_i387_fxsave(buf); 515 return save_i387_fxsave(fp);
459 else 516 else
460 return save_i387_fsave(buf); 517 return save_i387_fsave(fp);
461} 518}
462 519
463static inline int restore_i387_fsave(struct _fpstate_ia32 __user *buf) 520static inline int restore_i387_fsave(struct _fpstate_ia32 __user *buf)
@@ -468,14 +525,15 @@ static inline int restore_i387_fsave(struct _fpstate_ia32 __user *buf)
468 sizeof(struct i387_fsave_struct)); 525 sizeof(struct i387_fsave_struct));
469} 526}
470 527
471static int restore_i387_fxsave(struct _fpstate_ia32 __user *buf) 528static int restore_i387_fxsave(struct _fpstate_ia32 __user *buf,
529 unsigned int size)
472{ 530{
473 struct task_struct *tsk = current; 531 struct task_struct *tsk = current;
474 struct user_i387_ia32_struct env; 532 struct user_i387_ia32_struct env;
475 int err; 533 int err;
476 534
477 err = __copy_from_user(&tsk->thread.xstate->fxsave, &buf->_fxsr_env[0], 535 err = __copy_from_user(&tsk->thread.xstate->fxsave, &buf->_fxsr_env[0],
478 sizeof(struct i387_fxsave_struct)); 536 size);
479 /* mxcsr reserved bits must be masked to zero for security reasons */ 537 /* mxcsr reserved bits must be masked to zero for security reasons */
480 tsk->thread.xstate->fxsave.mxcsr &= mxcsr_feature_mask; 538 tsk->thread.xstate->fxsave.mxcsr &= mxcsr_feature_mask;
481 if (err || __copy_from_user(&env, buf, sizeof(env))) 539 if (err || __copy_from_user(&env, buf, sizeof(env)))
@@ -485,14 +543,69 @@ static int restore_i387_fxsave(struct _fpstate_ia32 __user *buf)
485 return 0; 543 return 0;
486} 544}
487 545
488int restore_i387_ia32(struct _fpstate_ia32 __user *buf) 546static int restore_i387_xsave(void __user *buf)
547{
548 struct _fpx_sw_bytes fx_sw_user;
549 struct _fpstate_ia32 __user *fx_user =
550 ((struct _fpstate_ia32 __user *) buf);
551 struct i387_fxsave_struct __user *fx =
552 (struct i387_fxsave_struct __user *) &fx_user->_fxsr_env[0];
553 struct xsave_hdr_struct *xsave_hdr =
554 &current->thread.xstate->xsave.xsave_hdr;
555 u64 mask;
556 int err;
557
558 if (check_for_xstate(fx, buf, &fx_sw_user))
559 goto fx_only;
560
561 mask = fx_sw_user.xstate_bv;
562
563 err = restore_i387_fxsave(buf, fx_sw_user.xstate_size);
564
565 xsave_hdr->xstate_bv &= pcntxt_mask;
566 /*
567 * These bits must be zero.
568 */
569 xsave_hdr->reserved1[0] = xsave_hdr->reserved1[1] = 0;
570
571 /*
572 * Init the state that is not present in the memory layout
573 * and enabled by the OS.
574 */
575 mask = ~(pcntxt_mask & ~mask);
576 xsave_hdr->xstate_bv &= mask;
577
578 return err;
579fx_only:
580 /*
581 * Couldn't find the extended state information in the memory
582 * layout. Restore the FP/SSE and init the other extended state
583 * enabled by the OS.
584 */
585 xsave_hdr->xstate_bv = XSTATE_FPSSE;
586 return restore_i387_fxsave(buf, sizeof(struct i387_fxsave_struct));
587}
588
589int restore_i387_xstate_ia32(void __user *buf)
489{ 590{
490 int err; 591 int err;
491 struct task_struct *tsk = current; 592 struct task_struct *tsk = current;
593 struct _fpstate_ia32 __user *fp = (struct _fpstate_ia32 __user *) buf;
492 594
493 if (HAVE_HWFP) 595 if (HAVE_HWFP)
494 clear_fpu(tsk); 596 clear_fpu(tsk);
495 597
598 if (!buf) {
599 if (used_math()) {
600 clear_fpu(tsk);
601 clear_used_math();
602 }
603
604 return 0;
605 } else
606 if (!access_ok(VERIFY_READ, buf, sig_xstate_ia32_size))
607 return -EACCES;
608
496 if (!used_math()) { 609 if (!used_math()) {
497 err = init_fpu(tsk); 610 err = init_fpu(tsk);
498 if (err) 611 if (err)
@@ -500,14 +613,17 @@ int restore_i387_ia32(struct _fpstate_ia32 __user *buf)
500 } 613 }
501 614
502 if (HAVE_HWFP) { 615 if (HAVE_HWFP) {
503 if (cpu_has_fxsr) 616 if (cpu_has_xsave)
504 err = restore_i387_fxsave(buf); 617 err = restore_i387_xsave(buf);
618 else if (cpu_has_fxsr)
619 err = restore_i387_fxsave(fp, sizeof(struct
620 i387_fxsave_struct));
505 else 621 else
506 err = restore_i387_fsave(buf); 622 err = restore_i387_fsave(fp);
507 } else { 623 } else {
508 err = fpregs_soft_set(current, NULL, 624 err = fpregs_soft_set(current, NULL,
509 0, sizeof(struct user_i387_ia32_struct), 625 0, sizeof(struct user_i387_ia32_struct),
510 NULL, buf) != 0; 626 NULL, fp) != 0;
511 } 627 }
512 set_used_math(); 628 set_used_math();
513 629
diff --git a/arch/x86/kernel/process.c b/arch/x86/kernel/process.c
index ec7a2ba9bce8..c622772744d8 100644
--- a/arch/x86/kernel/process.c
+++ b/arch/x86/kernel/process.c
@@ -15,7 +15,6 @@ unsigned long idle_nomwait;
15EXPORT_SYMBOL(idle_nomwait); 15EXPORT_SYMBOL(idle_nomwait);
16 16
17struct kmem_cache *task_xstate_cachep; 17struct kmem_cache *task_xstate_cachep;
18static int force_mwait __cpuinitdata;
19 18
20int arch_dup_task_struct(struct task_struct *dst, struct task_struct *src) 19int arch_dup_task_struct(struct task_struct *dst, struct task_struct *src)
21{ 20{
diff --git a/arch/x86/kernel/sigframe.h b/arch/x86/kernel/sigframe.h
index 8b4956e800ac..cc673aa55ce4 100644
--- a/arch/x86/kernel/sigframe.h
+++ b/arch/x86/kernel/sigframe.h
@@ -3,9 +3,18 @@ struct sigframe {
3 char __user *pretcode; 3 char __user *pretcode;
4 int sig; 4 int sig;
5 struct sigcontext sc; 5 struct sigcontext sc;
6 struct _fpstate fpstate; 6 /*
7 * fpstate is unused. fpstate is moved/allocated after
8 * retcode[] below. This movement allows to have the FP state and the
9 * future state extensions (xsave) stay together.
10 * And at the same time retaining the unused fpstate, prevents changing
11 * the offset of extramask[] in the sigframe and thus prevent any
12 * legacy application accessing/modifying it.
13 */
14 struct _fpstate fpstate_unused;
7 unsigned long extramask[_NSIG_WORDS-1]; 15 unsigned long extramask[_NSIG_WORDS-1];
8 char retcode[8]; 16 char retcode[8];
17 /* fp state follows here */
9}; 18};
10 19
11struct rt_sigframe { 20struct rt_sigframe {
@@ -15,14 +24,15 @@ struct rt_sigframe {
15 void __user *puc; 24 void __user *puc;
16 struct siginfo info; 25 struct siginfo info;
17 struct ucontext uc; 26 struct ucontext uc;
18 struct _fpstate fpstate;
19 char retcode[8]; 27 char retcode[8];
28 /* fp state follows here */
20}; 29};
21#else 30#else
22struct rt_sigframe { 31struct rt_sigframe {
23 char __user *pretcode; 32 char __user *pretcode;
24 struct ucontext uc; 33 struct ucontext uc;
25 struct siginfo info; 34 struct siginfo info;
35 /* fp state follows here */
26}; 36};
27 37
28int ia32_setup_rt_frame(int sig, struct k_sigaction *ka, siginfo_t *info, 38int ia32_setup_rt_frame(int sig, struct k_sigaction *ka, siginfo_t *info,
diff --git a/arch/x86/kernel/signal_32.c b/arch/x86/kernel/signal_32.c
index 2a2435d3037d..b21070ea33a4 100644
--- a/arch/x86/kernel/signal_32.c
+++ b/arch/x86/kernel/signal_32.c
@@ -161,28 +161,14 @@ restore_sigcontext(struct pt_regs *regs, struct sigcontext __user *sc,
161 } 161 }
162 162
163 { 163 {
164 struct _fpstate __user *buf; 164 void __user *buf;
165 165
166 err |= __get_user(buf, &sc->fpstate); 166 err |= __get_user(buf, &sc->fpstate);
167 if (buf) { 167 err |= restore_i387_xstate(buf);
168 if (!access_ok(VERIFY_READ, buf, sizeof(*buf)))
169 goto badframe;
170 err |= restore_i387(buf);
171 } else {
172 struct task_struct *me = current;
173
174 if (used_math()) {
175 clear_fpu(me);
176 clear_used_math();
177 }
178 }
179 } 168 }
180 169
181 err |= __get_user(*pax, &sc->ax); 170 err |= __get_user(*pax, &sc->ax);
182 return err; 171 return err;
183
184badframe:
185 return 1;
186} 172}
187 173
188asmlinkage unsigned long sys_sigreturn(unsigned long __unused) 174asmlinkage unsigned long sys_sigreturn(unsigned long __unused)
@@ -264,7 +250,7 @@ badframe:
264 * Set up a signal frame. 250 * Set up a signal frame.
265 */ 251 */
266static int 252static int
267setup_sigcontext(struct sigcontext __user *sc, struct _fpstate __user *fpstate, 253setup_sigcontext(struct sigcontext __user *sc, void __user *fpstate,
268 struct pt_regs *regs, unsigned long mask) 254 struct pt_regs *regs, unsigned long mask)
269{ 255{
270 int tmp, err = 0; 256 int tmp, err = 0;
@@ -291,7 +277,7 @@ setup_sigcontext(struct sigcontext __user *sc, struct _fpstate __user *fpstate,
291 err |= __put_user(regs->sp, &sc->sp_at_signal); 277 err |= __put_user(regs->sp, &sc->sp_at_signal);
292 err |= __put_user(regs->ss, (unsigned int __user *)&sc->ss); 278 err |= __put_user(regs->ss, (unsigned int __user *)&sc->ss);
293 279
294 tmp = save_i387(fpstate); 280 tmp = save_i387_xstate(fpstate);
295 if (tmp < 0) 281 if (tmp < 0)
296 err = 1; 282 err = 1;
297 else 283 else
@@ -308,7 +294,8 @@ setup_sigcontext(struct sigcontext __user *sc, struct _fpstate __user *fpstate,
308 * Determine which stack to use.. 294 * Determine which stack to use..
309 */ 295 */
310static inline void __user * 296static inline void __user *
311get_sigframe(struct k_sigaction *ka, struct pt_regs *regs, size_t frame_size) 297get_sigframe(struct k_sigaction *ka, struct pt_regs *regs, size_t frame_size,
298 void **fpstate)
312{ 299{
313 unsigned long sp; 300 unsigned long sp;
314 301
@@ -334,6 +321,11 @@ get_sigframe(struct k_sigaction *ka, struct pt_regs *regs, size_t frame_size)
334 sp = (unsigned long) ka->sa.sa_restorer; 321 sp = (unsigned long) ka->sa.sa_restorer;
335 } 322 }
336 323
324 if (used_math()) {
325 sp = sp - sig_xstate_size;
326 *fpstate = (struct _fpstate *) sp;
327 }
328
337 sp -= frame_size; 329 sp -= frame_size;
338 /* 330 /*
339 * Align the stack pointer according to the i386 ABI, 331 * Align the stack pointer according to the i386 ABI,
@@ -352,8 +344,9 @@ setup_frame(int sig, struct k_sigaction *ka, sigset_t *set,
352 void __user *restorer; 344 void __user *restorer;
353 int err = 0; 345 int err = 0;
354 int usig; 346 int usig;
347 void __user *fpstate = NULL;
355 348
356 frame = get_sigframe(ka, regs, sizeof(*frame)); 349 frame = get_sigframe(ka, regs, sizeof(*frame), &fpstate);
357 350
358 if (!access_ok(VERIFY_WRITE, frame, sizeof(*frame))) 351 if (!access_ok(VERIFY_WRITE, frame, sizeof(*frame)))
359 goto give_sigsegv; 352 goto give_sigsegv;
@@ -368,7 +361,7 @@ setup_frame(int sig, struct k_sigaction *ka, sigset_t *set,
368 if (err) 361 if (err)
369 goto give_sigsegv; 362 goto give_sigsegv;
370 363
371 err = setup_sigcontext(&frame->sc, &frame->fpstate, regs, set->sig[0]); 364 err = setup_sigcontext(&frame->sc, fpstate, regs, set->sig[0]);
372 if (err) 365 if (err)
373 goto give_sigsegv; 366 goto give_sigsegv;
374 367
@@ -429,8 +422,9 @@ static int setup_rt_frame(int sig, struct k_sigaction *ka, siginfo_t *info,
429 void __user *restorer; 422 void __user *restorer;
430 int err = 0; 423 int err = 0;
431 int usig; 424 int usig;
425 void __user *fpstate = NULL;
432 426
433 frame = get_sigframe(ka, regs, sizeof(*frame)); 427 frame = get_sigframe(ka, regs, sizeof(*frame), &fpstate);
434 428
435 if (!access_ok(VERIFY_WRITE, frame, sizeof(*frame))) 429 if (!access_ok(VERIFY_WRITE, frame, sizeof(*frame)))
436 goto give_sigsegv; 430 goto give_sigsegv;
@@ -449,13 +443,16 @@ static int setup_rt_frame(int sig, struct k_sigaction *ka, siginfo_t *info,
449 goto give_sigsegv; 443 goto give_sigsegv;
450 444
451 /* Create the ucontext. */ 445 /* Create the ucontext. */
452 err |= __put_user(0, &frame->uc.uc_flags); 446 if (cpu_has_xsave)
447 err |= __put_user(UC_FP_XSTATE, &frame->uc.uc_flags);
448 else
449 err |= __put_user(0, &frame->uc.uc_flags);
453 err |= __put_user(0, &frame->uc.uc_link); 450 err |= __put_user(0, &frame->uc.uc_link);
454 err |= __put_user(current->sas_ss_sp, &frame->uc.uc_stack.ss_sp); 451 err |= __put_user(current->sas_ss_sp, &frame->uc.uc_stack.ss_sp);
455 err |= __put_user(sas_ss_flags(regs->sp), 452 err |= __put_user(sas_ss_flags(regs->sp),
456 &frame->uc.uc_stack.ss_flags); 453 &frame->uc.uc_stack.ss_flags);
457 err |= __put_user(current->sas_ss_size, &frame->uc.uc_stack.ss_size); 454 err |= __put_user(current->sas_ss_size, &frame->uc.uc_stack.ss_size);
458 err |= setup_sigcontext(&frame->uc.uc_mcontext, &frame->fpstate, 455 err |= setup_sigcontext(&frame->uc.uc_mcontext, fpstate,
459 regs, set->sig[0]); 456 regs, set->sig[0]);
460 err |= __copy_to_user(&frame->uc.uc_sigmask, set, sizeof(*set)); 457 err |= __copy_to_user(&frame->uc.uc_sigmask, set, sizeof(*set));
461 if (err) 458 if (err)
diff --git a/arch/x86/kernel/signal_64.c b/arch/x86/kernel/signal_64.c
index 694aa888bb19..823a55bf8c39 100644
--- a/arch/x86/kernel/signal_64.c
+++ b/arch/x86/kernel/signal_64.c
@@ -53,69 +53,6 @@ sys_sigaltstack(const stack_t __user *uss, stack_t __user *uoss,
53} 53}
54 54
55/* 55/*
56 * Signal frame handlers.
57 */
58
59static inline int save_i387(struct _fpstate __user *buf)
60{
61 struct task_struct *tsk = current;
62 int err = 0;
63
64 BUILD_BUG_ON(sizeof(struct user_i387_struct) !=
65 sizeof(tsk->thread.xstate->fxsave));
66
67 if ((unsigned long)buf % 16)
68 printk("save_i387: bad fpstate %p\n", buf);
69
70 if (!used_math())
71 return 0;
72 clear_used_math(); /* trigger finit */
73 if (task_thread_info(tsk)->status & TS_USEDFPU) {
74 err = save_i387_checking((struct i387_fxsave_struct __user *)
75 buf);
76 if (err)
77 return err;
78 task_thread_info(tsk)->status &= ~TS_USEDFPU;
79 stts();
80 } else {
81 if (__copy_to_user(buf, &tsk->thread.xstate->fxsave,
82 sizeof(struct i387_fxsave_struct)))
83 return -1;
84 }
85 return 1;
86}
87
88/*
89 * This restores directly out of user space. Exceptions are handled.
90 */
91static inline int restore_i387(struct _fpstate __user *buf)
92{
93 struct task_struct *tsk = current;
94 int err;
95
96 if (!used_math()) {
97 err = init_fpu(tsk);
98 if (err)
99 return err;
100 }
101
102 if (!(task_thread_info(current)->status & TS_USEDFPU)) {
103 clts();
104 task_thread_info(current)->status |= TS_USEDFPU;
105 }
106 err = restore_fpu_checking((__force struct i387_fxsave_struct *)buf);
107 if (unlikely(err)) {
108 /*
109 * Encountered an error while doing the restore from the
110 * user buffer, clear the fpu state.
111 */
112 clear_fpu(tsk);
113 clear_used_math();
114 }
115 return err;
116}
117
118/*
119 * Do a signal return; undo the signal stack. 56 * Do a signal return; undo the signal stack.
120 */ 57 */
121static int 58static int
@@ -159,25 +96,11 @@ restore_sigcontext(struct pt_regs *regs, struct sigcontext __user *sc,
159 { 96 {
160 struct _fpstate __user *buf; 97 struct _fpstate __user *buf;
161 err |= __get_user(buf, &sc->fpstate); 98 err |= __get_user(buf, &sc->fpstate);
162 99 err |= restore_i387_xstate(buf);
163 if (buf) {
164 if (!access_ok(VERIFY_READ, buf, sizeof(*buf)))
165 goto badframe;
166 err |= restore_i387(buf);
167 } else {
168 struct task_struct *me = current;
169 if (used_math()) {
170 clear_fpu(me);
171 clear_used_math();
172 }
173 }
174 } 100 }
175 101
176 err |= __get_user(*pax, &sc->ax); 102 err |= __get_user(*pax, &sc->ax);
177 return err; 103 return err;
178
179badframe:
180 return 1;
181} 104}
182 105
183asmlinkage long sys_rt_sigreturn(struct pt_regs *regs) 106asmlinkage long sys_rt_sigreturn(struct pt_regs *regs)
@@ -269,26 +192,23 @@ get_stack(struct k_sigaction *ka, struct pt_regs *regs, unsigned long size)
269 sp = current->sas_ss_sp + current->sas_ss_size; 192 sp = current->sas_ss_sp + current->sas_ss_size;
270 } 193 }
271 194
272 return (void __user *)round_down(sp - size, 16); 195 return (void __user *)round_down(sp - size, 64);
273} 196}
274 197
275static int setup_rt_frame(int sig, struct k_sigaction *ka, siginfo_t *info, 198static int setup_rt_frame(int sig, struct k_sigaction *ka, siginfo_t *info,
276 sigset_t *set, struct pt_regs *regs) 199 sigset_t *set, struct pt_regs *regs)
277{ 200{
278 struct rt_sigframe __user *frame; 201 struct rt_sigframe __user *frame;
279 struct _fpstate __user *fp = NULL; 202 void __user *fp = NULL;
280 int err = 0; 203 int err = 0;
281 struct task_struct *me = current; 204 struct task_struct *me = current;
282 205
283 if (used_math()) { 206 if (used_math()) {
284 fp = get_stack(ka, regs, sizeof(struct _fpstate)); 207 fp = get_stack(ka, regs, sig_xstate_size);
285 frame = (void __user *)round_down( 208 frame = (void __user *)round_down(
286 (unsigned long)fp - sizeof(struct rt_sigframe), 16) - 8; 209 (unsigned long)fp - sizeof(struct rt_sigframe), 16) - 8;
287 210
288 if (!access_ok(VERIFY_WRITE, fp, sizeof(struct _fpstate))) 211 if (save_i387_xstate(fp) < 0)
289 goto give_sigsegv;
290
291 if (save_i387(fp) < 0)
292 err |= -1; 212 err |= -1;
293 } else 213 } else
294 frame = get_stack(ka, regs, sizeof(struct rt_sigframe)) - 8; 214 frame = get_stack(ka, regs, sizeof(struct rt_sigframe)) - 8;
@@ -303,7 +223,10 @@ static int setup_rt_frame(int sig, struct k_sigaction *ka, siginfo_t *info,
303 } 223 }
304 224
305 /* Create the ucontext. */ 225 /* Create the ucontext. */
306 err |= __put_user(0, &frame->uc.uc_flags); 226 if (cpu_has_xsave)
227 err |= __put_user(UC_FP_XSTATE, &frame->uc.uc_flags);
228 else
229 err |= __put_user(0, &frame->uc.uc_flags);
307 err |= __put_user(0, &frame->uc.uc_link); 230 err |= __put_user(0, &frame->uc.uc_link);
308 err |= __put_user(me->sas_ss_sp, &frame->uc.uc_stack.ss_sp); 231 err |= __put_user(me->sas_ss_sp, &frame->uc.uc_stack.ss_sp);
309 err |= __put_user(sas_ss_flags(regs->sp), 232 err |= __put_user(sas_ss_flags(regs->sp),
diff --git a/arch/x86/kernel/traps_32.c b/arch/x86/kernel/traps_32.c
index 03df8e45e5a1..da5a5964fccb 100644
--- a/arch/x86/kernel/traps_32.c
+++ b/arch/x86/kernel/traps_32.c
@@ -1228,7 +1228,6 @@ void __init trap_init(void)
1228 1228
1229 set_bit(SYSCALL_VECTOR, used_vectors); 1229 set_bit(SYSCALL_VECTOR, used_vectors);
1230 1230
1231 init_thread_xstate();
1232 /* 1231 /*
1233 * Should be a barrier for any external CPU state: 1232 * Should be a barrier for any external CPU state:
1234 */ 1233 */
diff --git a/arch/x86/kernel/traps_64.c b/arch/x86/kernel/traps_64.c
index 7a31f104bef9..2887a789e38f 100644
--- a/arch/x86/kernel/traps_64.c
+++ b/arch/x86/kernel/traps_64.c
@@ -1138,7 +1138,7 @@ asmlinkage void math_state_restore(void)
1138 /* 1138 /*
1139 * Paranoid restore. send a SIGSEGV if we fail to restore the state. 1139 * Paranoid restore. send a SIGSEGV if we fail to restore the state.
1140 */ 1140 */
1141 if (unlikely(restore_fpu_checking(&me->thread.xstate->fxsave))) { 1141 if (unlikely(restore_fpu_checking(me))) {
1142 stts(); 1142 stts();
1143 force_sig(SIGSEGV, me); 1143 force_sig(SIGSEGV, me);
1144 return; 1144 return;
@@ -1179,10 +1179,6 @@ void __init trap_init(void)
1179 set_system_gate(IA32_SYSCALL_VECTOR, ia32_syscall); 1179 set_system_gate(IA32_SYSCALL_VECTOR, ia32_syscall);
1180#endif 1180#endif
1181 /* 1181 /*
1182 * initialize the per thread extended state:
1183 */
1184 init_thread_xstate();
1185 /*
1186 * Should be a barrier for any external CPU state: 1182 * Should be a barrier for any external CPU state:
1187 */ 1183 */
1188 cpu_init(); 1184 cpu_init();
diff --git a/arch/x86/kernel/vmlinux_32.lds.S b/arch/x86/kernel/vmlinux_32.lds.S
index af5bdad84604..a9b8560adbc2 100644
--- a/arch/x86/kernel/vmlinux_32.lds.S
+++ b/arch/x86/kernel/vmlinux_32.lds.S
@@ -140,10 +140,10 @@ SECTIONS
140 *(.con_initcall.init) 140 *(.con_initcall.init)
141 __con_initcall_end = .; 141 __con_initcall_end = .;
142 } 142 }
143 .x86cpuvendor.init : AT(ADDR(.x86cpuvendor.init) - LOAD_OFFSET) { 143 .x86_cpu_dev.init : AT(ADDR(.x86_cpu_dev.init) - LOAD_OFFSET) {
144 __x86cpuvendor_start = .; 144 __x86_cpu_dev_start = .;
145 *(.x86cpuvendor.init) 145 *(.x86_cpu_dev.init)
146 __x86cpuvendor_end = .; 146 __x86_cpu_dev_end = .;
147 } 147 }
148 SECURITY_INIT 148 SECURITY_INIT
149 . = ALIGN(4); 149 . = ALIGN(4);
@@ -180,6 +180,7 @@ SECTIONS
180 . = ALIGN(PAGE_SIZE); 180 . = ALIGN(PAGE_SIZE);
181 .data.percpu : AT(ADDR(.data.percpu) - LOAD_OFFSET) { 181 .data.percpu : AT(ADDR(.data.percpu) - LOAD_OFFSET) {
182 __per_cpu_start = .; 182 __per_cpu_start = .;
183 *(.data.percpu.page_aligned)
183 *(.data.percpu) 184 *(.data.percpu)
184 *(.data.percpu.shared_aligned) 185 *(.data.percpu.shared_aligned)
185 __per_cpu_end = .; 186 __per_cpu_end = .;
diff --git a/arch/x86/kernel/vmlinux_64.lds.S b/arch/x86/kernel/vmlinux_64.lds.S
index 63e5c1a22e88..201e81a91a95 100644
--- a/arch/x86/kernel/vmlinux_64.lds.S
+++ b/arch/x86/kernel/vmlinux_64.lds.S
@@ -168,13 +168,12 @@ SECTIONS
168 *(.con_initcall.init) 168 *(.con_initcall.init)
169 } 169 }
170 __con_initcall_end = .; 170 __con_initcall_end = .;
171 . = ALIGN(16); 171 __x86_cpu_dev_start = .;
172 __x86cpuvendor_start = .; 172 .x86_cpu_dev.init : AT(ADDR(.x86_cpu_dev.init) - LOAD_OFFSET) {
173 .x86cpuvendor.init : AT(ADDR(.x86cpuvendor.init) - LOAD_OFFSET) { 173 *(.x86_cpu_dev.init)
174 *(.x86cpuvendor.init)
175 } 174 }
176 __x86cpuvendor_end = .;
177 SECURITY_INIT 175 SECURITY_INIT
176 __x86_cpu_dev_end = .;
178 177
179 . = ALIGN(8); 178 . = ALIGN(8);
180 .parainstructions : AT(ADDR(.parainstructions) - LOAD_OFFSET) { 179 .parainstructions : AT(ADDR(.parainstructions) - LOAD_OFFSET) {
diff --git a/arch/x86/kernel/xsave.c b/arch/x86/kernel/xsave.c
new file mode 100644
index 000000000000..07713d64debe
--- /dev/null
+++ b/arch/x86/kernel/xsave.c
@@ -0,0 +1,316 @@
1/*
2 * xsave/xrstor support.
3 *
4 * Author: Suresh Siddha <suresh.b.siddha@intel.com>
5 */
6#include <linux/bootmem.h>
7#include <linux/compat.h>
8#include <asm/i387.h>
9#ifdef CONFIG_IA32_EMULATION
10#include <asm/sigcontext32.h>
11#endif
12#include <asm/xcr.h>
13
14/*
15 * Supported feature mask by the CPU and the kernel.
16 */
17u64 pcntxt_mask;
18
19struct _fpx_sw_bytes fx_sw_reserved;
20#ifdef CONFIG_IA32_EMULATION
21struct _fpx_sw_bytes fx_sw_reserved_ia32;
22#endif
23
24/*
25 * Check for the presence of extended state information in the
26 * user fpstate pointer in the sigcontext.
27 */
28int check_for_xstate(struct i387_fxsave_struct __user *buf,
29 void __user *fpstate,
30 struct _fpx_sw_bytes *fx_sw_user)
31{
32 int min_xstate_size = sizeof(struct i387_fxsave_struct) +
33 sizeof(struct xsave_hdr_struct);
34 unsigned int magic2;
35 int err;
36
37 err = __copy_from_user(fx_sw_user, &buf->sw_reserved[0],
38 sizeof(struct _fpx_sw_bytes));
39
40 if (err)
41 return err;
42
43 /*
44 * First Magic check failed.
45 */
46 if (fx_sw_user->magic1 != FP_XSTATE_MAGIC1)
47 return -1;
48
49 /*
50 * Check for error scenarios.
51 */
52 if (fx_sw_user->xstate_size < min_xstate_size ||
53 fx_sw_user->xstate_size > xstate_size ||
54 fx_sw_user->xstate_size > fx_sw_user->extended_size)
55 return -1;
56
57 err = __get_user(magic2, (__u32 *) (((void *)fpstate) +
58 fx_sw_user->extended_size -
59 FP_XSTATE_MAGIC2_SIZE));
60 /*
61 * Check for the presence of second magic word at the end of memory
62 * layout. This detects the case where the user just copied the legacy
63 * fpstate layout with out copying the extended state information
64 * in the memory layout.
65 */
66 if (err || magic2 != FP_XSTATE_MAGIC2)
67 return -1;
68
69 return 0;
70}
71
72#ifdef CONFIG_X86_64
73/*
74 * Signal frame handlers.
75 */
76
77int save_i387_xstate(void __user *buf)
78{
79 struct task_struct *tsk = current;
80 int err = 0;
81
82 if (!access_ok(VERIFY_WRITE, buf, sig_xstate_size))
83 return -EACCES;
84
85 BUG_ON(sig_xstate_size < xstate_size);
86
87 if ((unsigned long)buf % 64)
88 printk("save_i387_xstate: bad fpstate %p\n", buf);
89
90 if (!used_math())
91 return 0;
92 clear_used_math(); /* trigger finit */
93 if (task_thread_info(tsk)->status & TS_USEDFPU) {
94 /*
95 * Start with clearing the user buffer. This will present a
96 * clean context for the bytes not touched by the fxsave/xsave.
97 */
98 __clear_user(buf, sig_xstate_size);
99
100 if (task_thread_info(tsk)->status & TS_XSAVE)
101 err = xsave_user(buf);
102 else
103 err = fxsave_user(buf);
104
105 if (err)
106 return err;
107 task_thread_info(tsk)->status &= ~TS_USEDFPU;
108 stts();
109 } else {
110 if (__copy_to_user(buf, &tsk->thread.xstate->fxsave,
111 xstate_size))
112 return -1;
113 }
114
115 if (task_thread_info(tsk)->status & TS_XSAVE) {
116 struct _fpstate __user *fx = buf;
117
118 err = __copy_to_user(&fx->sw_reserved, &fx_sw_reserved,
119 sizeof(struct _fpx_sw_bytes));
120
121 err |= __put_user(FP_XSTATE_MAGIC2,
122 (__u32 __user *) (buf + sig_xstate_size
123 - FP_XSTATE_MAGIC2_SIZE));
124 }
125
126 return 1;
127}
128
129/*
130 * Restore the extended state if present. Otherwise, restore the FP/SSE
131 * state.
132 */
133int restore_user_xstate(void __user *buf)
134{
135 struct _fpx_sw_bytes fx_sw_user;
136 u64 mask;
137 int err;
138
139 if (((unsigned long)buf % 64) ||
140 check_for_xstate(buf, buf, &fx_sw_user))
141 goto fx_only;
142
143 mask = fx_sw_user.xstate_bv;
144
145 /*
146 * restore the state passed by the user.
147 */
148 err = xrestore_user(buf, mask);
149 if (err)
150 return err;
151
152 /*
153 * init the state skipped by the user.
154 */
155 mask = pcntxt_mask & ~mask;
156
157 xrstor_state(init_xstate_buf, mask);
158
159 return 0;
160
161fx_only:
162 /*
163 * couldn't find the extended state information in the
164 * memory layout. Restore just the FP/SSE and init all
165 * the other extended state.
166 */
167 xrstor_state(init_xstate_buf, pcntxt_mask & ~XSTATE_FPSSE);
168 return fxrstor_checking((__force struct i387_fxsave_struct *)buf);
169}
170
171/*
172 * This restores directly out of user space. Exceptions are handled.
173 */
174int restore_i387_xstate(void __user *buf)
175{
176 struct task_struct *tsk = current;
177 int err = 0;
178
179 if (!buf) {
180 if (used_math())
181 goto clear;
182 return 0;
183 } else
184 if (!access_ok(VERIFY_READ, buf, sig_xstate_size))
185 return -EACCES;
186
187 if (!used_math()) {
188 err = init_fpu(tsk);
189 if (err)
190 return err;
191 }
192
193 if (!(task_thread_info(current)->status & TS_USEDFPU)) {
194 clts();
195 task_thread_info(current)->status |= TS_USEDFPU;
196 }
197 if (task_thread_info(tsk)->status & TS_XSAVE)
198 err = restore_user_xstate(buf);
199 else
200 err = fxrstor_checking((__force struct i387_fxsave_struct *)
201 buf);
202 if (unlikely(err)) {
203 /*
204 * Encountered an error while doing the restore from the
205 * user buffer, clear the fpu state.
206 */
207clear:
208 clear_fpu(tsk);
209 clear_used_math();
210 }
211 return err;
212}
213#endif
214
215/*
216 * Prepare the SW reserved portion of the fxsave memory layout, indicating
217 * the presence of the extended state information in the memory layout
218 * pointed by the fpstate pointer in the sigcontext.
219 * This will be saved when ever the FP and extended state context is
220 * saved on the user stack during the signal handler delivery to the user.
221 */
222void prepare_fx_sw_frame(void)
223{
224 int size_extended = (xstate_size - sizeof(struct i387_fxsave_struct)) +
225 FP_XSTATE_MAGIC2_SIZE;
226
227 sig_xstate_size = sizeof(struct _fpstate) + size_extended;
228
229#ifdef CONFIG_IA32_EMULATION
230 sig_xstate_ia32_size = sizeof(struct _fpstate_ia32) + size_extended;
231#endif
232
233 memset(&fx_sw_reserved, 0, sizeof(fx_sw_reserved));
234
235 fx_sw_reserved.magic1 = FP_XSTATE_MAGIC1;
236 fx_sw_reserved.extended_size = sig_xstate_size;
237 fx_sw_reserved.xstate_bv = pcntxt_mask;
238 fx_sw_reserved.xstate_size = xstate_size;
239#ifdef CONFIG_IA32_EMULATION
240 memcpy(&fx_sw_reserved_ia32, &fx_sw_reserved,
241 sizeof(struct _fpx_sw_bytes));
242 fx_sw_reserved_ia32.extended_size = sig_xstate_ia32_size;
243#endif
244}
245
246/*
247 * Represents init state for the supported extended state.
248 */
249struct xsave_struct *init_xstate_buf;
250
251#ifdef CONFIG_X86_64
252unsigned int sig_xstate_size = sizeof(struct _fpstate);
253#endif
254
255/*
256 * Enable the extended processor state save/restore feature
257 */
258void __cpuinit xsave_init(void)
259{
260 if (!cpu_has_xsave)
261 return;
262
263 set_in_cr4(X86_CR4_OSXSAVE);
264
265 /*
266 * Enable all the features that the HW is capable of
267 * and the Linux kernel is aware of.
268 */
269 xsetbv(XCR_XFEATURE_ENABLED_MASK, pcntxt_mask);
270}
271
272/*
273 * setup the xstate image representing the init state
274 */
275void setup_xstate_init(void)
276{
277 init_xstate_buf = alloc_bootmem(xstate_size);
278 init_xstate_buf->i387.mxcsr = MXCSR_DEFAULT;
279}
280
281/*
282 * Enable and initialize the xsave feature.
283 */
284void __init xsave_cntxt_init(void)
285{
286 unsigned int eax, ebx, ecx, edx;
287
288 cpuid_count(0xd, 0, &eax, &ebx, &ecx, &edx);
289 pcntxt_mask = eax + ((u64)edx << 32);
290
291 if ((pcntxt_mask & XSTATE_FPSSE) != XSTATE_FPSSE) {
292 printk(KERN_ERR "FP/SSE not shown under xsave features 0x%llx\n",
293 pcntxt_mask);
294 BUG();
295 }
296
297 /*
298 * for now OS knows only about FP/SSE
299 */
300 pcntxt_mask = pcntxt_mask & XCNTXT_MASK;
301 xsave_init();
302
303 /*
304 * Recompute the context size for enabled features
305 */
306 cpuid_count(0xd, 0, &eax, &ebx, &ecx, &edx);
307 xstate_size = ebx;
308
309 prepare_fx_sw_frame();
310
311 setup_xstate_init();
312
313 printk(KERN_INFO "xsave/xrstor: enabled xstate_bv 0x%llx, "
314 "cntxt size 0x%x\n",
315 pcntxt_mask, xstate_size);
316}
diff --git a/arch/x86/kvm/vmx.h b/arch/x86/kvm/vmx.h
index 23e8373507ad..17e25995b65b 100644
--- a/arch/x86/kvm/vmx.h
+++ b/arch/x86/kvm/vmx.h
@@ -331,21 +331,6 @@ enum vmcs_field {
331 331
332#define AR_RESERVD_MASK 0xfffe0f00 332#define AR_RESERVD_MASK 0xfffe0f00
333 333
334#define MSR_IA32_VMX_BASIC 0x480
335#define MSR_IA32_VMX_PINBASED_CTLS 0x481
336#define MSR_IA32_VMX_PROCBASED_CTLS 0x482
337#define MSR_IA32_VMX_EXIT_CTLS 0x483
338#define MSR_IA32_VMX_ENTRY_CTLS 0x484
339#define MSR_IA32_VMX_MISC 0x485
340#define MSR_IA32_VMX_CR0_FIXED0 0x486
341#define MSR_IA32_VMX_CR0_FIXED1 0x487
342#define MSR_IA32_VMX_CR4_FIXED0 0x488
343#define MSR_IA32_VMX_CR4_FIXED1 0x489
344#define MSR_IA32_VMX_VMCS_ENUM 0x48a
345#define MSR_IA32_VMX_PROCBASED_CTLS2 0x48b
346#define MSR_IA32_VMX_EPT_VPID_CAP 0x48c
347
348#define MSR_IA32_FEATURE_CONTROL 0x3a
349#define MSR_IA32_FEATURE_CONTROL_LOCKED 0x1 334#define MSR_IA32_FEATURE_CONTROL_LOCKED 0x1
350#define MSR_IA32_FEATURE_CONTROL_VMXON_ENABLED 0x4 335#define MSR_IA32_FEATURE_CONTROL_VMXON_ENABLED 0x4
351 336
diff --git a/arch/x86/lib/Makefile b/arch/x86/lib/Makefile
index aa3fa4119424..55e11aa6d66c 100644
--- a/arch/x86/lib/Makefile
+++ b/arch/x86/lib/Makefile
@@ -17,9 +17,6 @@ ifeq ($(CONFIG_X86_32),y)
17 lib-$(CONFIG_X86_USE_3DNOW) += mmx_32.o 17 lib-$(CONFIG_X86_USE_3DNOW) += mmx_32.o
18else 18else
19 obj-y += io_64.o iomap_copy_64.o 19 obj-y += io_64.o iomap_copy_64.o
20
21 CFLAGS_csum-partial_64.o := -funroll-loops
22
23 lib-y += csum-partial_64.o csum-copy_64.o csum-wrappers_64.o 20 lib-y += csum-partial_64.o csum-copy_64.o csum-wrappers_64.o
24 lib-y += thunk_64.o clear_page_64.o copy_page_64.o 21 lib-y += thunk_64.o clear_page_64.o copy_page_64.o
25 lib-y += memmove_64.o memset_64.o 22 lib-y += memmove_64.o memset_64.o
diff --git a/arch/x86/lib/usercopy_32.c b/arch/x86/lib/usercopy_32.c
index 24e60944971a..9e68075544f6 100644
--- a/arch/x86/lib/usercopy_32.c
+++ b/arch/x86/lib/usercopy_32.c
@@ -14,6 +14,13 @@
14#include <asm/uaccess.h> 14#include <asm/uaccess.h>
15#include <asm/mmx.h> 15#include <asm/mmx.h>
16 16
17#ifdef CONFIG_X86_INTEL_USERCOPY
18/*
19 * Alignment at which movsl is preferred for bulk memory copies.
20 */
21struct movsl_mask movsl_mask __read_mostly;
22#endif
23
17static inline int __movsl_is_ok(unsigned long a1, unsigned long a2, unsigned long n) 24static inline int __movsl_is_ok(unsigned long a1, unsigned long a2, unsigned long n)
18{ 25{
19#ifdef CONFIG_X86_INTEL_USERCOPY 26#ifdef CONFIG_X86_INTEL_USERCOPY
diff --git a/arch/x86/mach-generic/Makefile b/arch/x86/mach-generic/Makefile
index 4706de7676b1..6730f4e7c744 100644
--- a/arch/x86/mach-generic/Makefile
+++ b/arch/x86/mach-generic/Makefile
@@ -9,4 +9,3 @@ obj-$(CONFIG_X86_NUMAQ) += numaq.o
9obj-$(CONFIG_X86_SUMMIT) += summit.o 9obj-$(CONFIG_X86_SUMMIT) += summit.o
10obj-$(CONFIG_X86_BIGSMP) += bigsmp.o 10obj-$(CONFIG_X86_BIGSMP) += bigsmp.o
11obj-$(CONFIG_X86_ES7000) += es7000.o 11obj-$(CONFIG_X86_ES7000) += es7000.o
12obj-$(CONFIG_X86_ES7000) += ../../x86/es7000/
diff --git a/arch/x86/mm/init_64.c b/arch/x86/mm/init_64.c
index fb30486c82f7..83e13f2d53d2 100644
--- a/arch/x86/mm/init_64.c
+++ b/arch/x86/mm/init_64.c
@@ -88,6 +88,62 @@ early_param("gbpages", parse_direct_gbpages_on);
88 88
89int after_bootmem; 89int after_bootmem;
90 90
91unsigned long __supported_pte_mask __read_mostly = ~0UL;
92EXPORT_SYMBOL_GPL(__supported_pte_mask);
93
94static int do_not_nx __cpuinitdata;
95
96/*
97 * noexec=on|off
98 * Control non-executable mappings for 64-bit processes.
99 *
100 * on Enable (default)
101 * off Disable
102 */
103static int __init nonx_setup(char *str)
104{
105 if (!str)
106 return -EINVAL;
107 if (!strncmp(str, "on", 2)) {
108 __supported_pte_mask |= _PAGE_NX;
109 do_not_nx = 0;
110 } else if (!strncmp(str, "off", 3)) {
111 do_not_nx = 1;
112 __supported_pte_mask &= ~_PAGE_NX;
113 }
114 return 0;
115}
116early_param("noexec", nonx_setup);
117
118void __cpuinit check_efer(void)
119{
120 unsigned long efer;
121
122 rdmsrl(MSR_EFER, efer);
123 if (!(efer & EFER_NX) || do_not_nx)
124 __supported_pte_mask &= ~_PAGE_NX;
125}
126
127int force_personality32;
128
129/*
130 * noexec32=on|off
131 * Control non executable heap for 32bit processes.
132 * To control the stack too use noexec=off
133 *
134 * on PROT_READ does not imply PROT_EXEC for 32-bit processes (default)
135 * off PROT_READ implies PROT_EXEC
136 */
137static int __init nonx32_setup(char *str)
138{
139 if (!strcmp(str, "on"))
140 force_personality32 &= ~READ_IMPLIES_EXEC;
141 else if (!strcmp(str, "off"))
142 force_personality32 |= READ_IMPLIES_EXEC;
143 return 1;
144}
145__setup("noexec32=", nonx32_setup);
146
91/* 147/*
92 * NOTE: This function is marked __ref because it calls __init function 148 * NOTE: This function is marked __ref because it calls __init function
93 * (alloc_bootmem_pages). It's safe to do it ONLY when after_bootmem == 0. 149 * (alloc_bootmem_pages). It's safe to do it ONLY when after_bootmem == 0.
diff --git a/arch/x86/pci/i386.c b/arch/x86/pci/i386.c
index 8791fc55e715..844df0cbbd3e 100644
--- a/arch/x86/pci/i386.c
+++ b/arch/x86/pci/i386.c
@@ -33,6 +33,7 @@
33#include <linux/bootmem.h> 33#include <linux/bootmem.h>
34 34
35#include <asm/pat.h> 35#include <asm/pat.h>
36#include <asm/e820.h>
36 37
37#include "pci.h" 38#include "pci.h"
38 39
@@ -227,6 +228,8 @@ void __init pcibios_resource_survey(void)
227 pcibios_allocate_bus_resources(&pci_root_buses); 228 pcibios_allocate_bus_resources(&pci_root_buses);
228 pcibios_allocate_resources(0); 229 pcibios_allocate_resources(0);
229 pcibios_allocate_resources(1); 230 pcibios_allocate_resources(1);
231
232 e820_reserve_resources_late();
230} 233}
231 234
232/** 235/**
diff --git a/arch/x86/pci/mmconfig-shared.c b/arch/x86/pci/mmconfig-shared.c
index d9635764ce3d..654a2234f8f3 100644
--- a/arch/x86/pci/mmconfig-shared.c
+++ b/arch/x86/pci/mmconfig-shared.c
@@ -209,7 +209,7 @@ static int __init pci_mmcfg_check_hostbridge(void)
209 return name != NULL; 209 return name != NULL;
210} 210}
211 211
212static void __init pci_mmcfg_insert_resources(unsigned long resource_flags) 212static void __init pci_mmcfg_insert_resources(void)
213{ 213{
214#define PCI_MMCFG_RESOURCE_NAME_LEN 19 214#define PCI_MMCFG_RESOURCE_NAME_LEN 19
215 int i; 215 int i;
@@ -233,7 +233,7 @@ static void __init pci_mmcfg_insert_resources(unsigned long resource_flags)
233 cfg->pci_segment); 233 cfg->pci_segment);
234 res->start = cfg->address; 234 res->start = cfg->address;
235 res->end = res->start + (num_buses << 20) - 1; 235 res->end = res->start + (num_buses << 20) - 1;
236 res->flags = IORESOURCE_MEM | resource_flags; 236 res->flags = IORESOURCE_MEM | IORESOURCE_BUSY;
237 insert_resource(&iomem_resource, res); 237 insert_resource(&iomem_resource, res);
238 names += PCI_MMCFG_RESOURCE_NAME_LEN; 238 names += PCI_MMCFG_RESOURCE_NAME_LEN;
239 } 239 }
@@ -434,11 +434,9 @@ static void __init __pci_mmcfg_init(int early)
434 (pci_mmcfg_config[0].address == 0)) 434 (pci_mmcfg_config[0].address == 0))
435 return; 435 return;
436 436
437 if (pci_mmcfg_arch_init()) { 437 if (pci_mmcfg_arch_init())
438 if (known_bridge)
439 pci_mmcfg_insert_resources(IORESOURCE_BUSY);
440 pci_probe = (pci_probe & ~PCI_PROBE_MASK) | PCI_PROBE_MMCONF; 438 pci_probe = (pci_probe & ~PCI_PROBE_MASK) | PCI_PROBE_MMCONF;
441 } else { 439 else {
442 /* 440 /*
443 * Signal not to attempt to insert mmcfg resources because 441 * Signal not to attempt to insert mmcfg resources because
444 * the architecture mmcfg setup could not initialize. 442 * the architecture mmcfg setup could not initialize.
@@ -475,7 +473,7 @@ static int __init pci_mmcfg_late_insert_resources(void)
475 * marked so it won't cause request errors when __request_region is 473 * marked so it won't cause request errors when __request_region is
476 * called. 474 * called.
477 */ 475 */
478 pci_mmcfg_insert_resources(0); 476 pci_mmcfg_insert_resources();
479 477
480 return 0; 478 return 0;
481} 479}
diff --git a/arch/x86/power/cpu_32.c b/arch/x86/power/cpu_32.c
index d3e083dea720..274d06082f48 100644
--- a/arch/x86/power/cpu_32.c
+++ b/arch/x86/power/cpu_32.c
@@ -11,6 +11,7 @@
11#include <linux/suspend.h> 11#include <linux/suspend.h>
12#include <asm/mtrr.h> 12#include <asm/mtrr.h>
13#include <asm/mce.h> 13#include <asm/mce.h>
14#include <asm/xcr.h>
14 15
15static struct saved_context saved_context; 16static struct saved_context saved_context;
16 17
@@ -126,6 +127,12 @@ static void __restore_processor_state(struct saved_context *ctxt)
126 if (boot_cpu_has(X86_FEATURE_SEP)) 127 if (boot_cpu_has(X86_FEATURE_SEP))
127 enable_sep_cpu(); 128 enable_sep_cpu();
128 129
130 /*
131 * restore XCR0 for xsave capable cpu's.
132 */
133 if (cpu_has_xsave)
134 xsetbv(XCR_XFEATURE_ENABLED_MASK, pcntxt_mask);
135
129 fix_processor_context(); 136 fix_processor_context();
130 do_fpu_end(); 137 do_fpu_end();
131 mtrr_ap_init(); 138 mtrr_ap_init();
diff --git a/arch/x86/power/cpu_64.c b/arch/x86/power/cpu_64.c
index 66bdfb591fd8..e3b6cf70d62c 100644
--- a/arch/x86/power/cpu_64.c
+++ b/arch/x86/power/cpu_64.c
@@ -14,6 +14,7 @@
14#include <asm/page.h> 14#include <asm/page.h>
15#include <asm/pgtable.h> 15#include <asm/pgtable.h>
16#include <asm/mtrr.h> 16#include <asm/mtrr.h>
17#include <asm/xcr.h>
17 18
18static void fix_processor_context(void); 19static void fix_processor_context(void);
19 20
@@ -122,6 +123,12 @@ static void __restore_processor_state(struct saved_context *ctxt)
122 wrmsrl(MSR_GS_BASE, ctxt->gs_base); 123 wrmsrl(MSR_GS_BASE, ctxt->gs_base);
123 wrmsrl(MSR_KERNEL_GS_BASE, ctxt->gs_kernel_base); 124 wrmsrl(MSR_KERNEL_GS_BASE, ctxt->gs_kernel_base);
124 125
126 /*
127 * restore XCR0 for xsave capable cpu's.
128 */
129 if (cpu_has_xsave)
130 xsetbv(XCR_XFEATURE_ENABLED_MASK, pcntxt_mask);
131
125 fix_processor_context(); 132 fix_processor_context();
126 133
127 do_fpu_end(); 134 do_fpu_end();
diff --git a/include/asm-generic/vmlinux.lds.h b/include/asm-generic/vmlinux.lds.h
index cb752ba72466..7440a0dceddb 100644
--- a/include/asm-generic/vmlinux.lds.h
+++ b/include/asm-generic/vmlinux.lds.h
@@ -385,6 +385,7 @@
385 . = ALIGN(align); \ 385 . = ALIGN(align); \
386 VMLINUX_SYMBOL(__per_cpu_start) = .; \ 386 VMLINUX_SYMBOL(__per_cpu_start) = .; \
387 .data.percpu : AT(ADDR(.data.percpu) - LOAD_OFFSET) { \ 387 .data.percpu : AT(ADDR(.data.percpu) - LOAD_OFFSET) { \
388 *(.data.percpu.page_aligned) \
388 *(.data.percpu) \ 389 *(.data.percpu) \
389 *(.data.percpu.shared_aligned) \ 390 *(.data.percpu.shared_aligned) \
390 } \ 391 } \
diff --git a/include/asm-x86/bugs.h b/include/asm-x86/bugs.h
index 4761c461d23a..dc604985f2ad 100644
--- a/include/asm-x86/bugs.h
+++ b/include/asm-x86/bugs.h
@@ -2,6 +2,11 @@
2#define ASM_X86__BUGS_H 2#define ASM_X86__BUGS_H
3 3
4extern void check_bugs(void); 4extern void check_bugs(void);
5
6#if defined(CONFIG_CPU_SUP_INTEL) && defined(CONFIG_X86_32)
5int ppro_with_ram_bug(void); 7int ppro_with_ram_bug(void);
8#else
9static inline int ppro_with_ram_bug(void) { return 0; }
10#endif
6 11
7#endif /* ASM_X86__BUGS_H */ 12#endif /* ASM_X86__BUGS_H */
diff --git a/include/asm-x86/cpufeature.h b/include/asm-x86/cpufeature.h
index 065c6a86ed80..adfeae6586e1 100644
--- a/include/asm-x86/cpufeature.h
+++ b/include/asm-x86/cpufeature.h
@@ -6,7 +6,13 @@
6 6
7#include <asm/required-features.h> 7#include <asm/required-features.h>
8 8
9#define NCAPINTS 8 /* N 32-bit words worth of info */ 9#define NCAPINTS 9 /* N 32-bit words worth of info */
10
11/*
12 * Note: If the comment begins with a quoted string, that string is used
13 * in /proc/cpuinfo instead of the macro name. If the string is "",
14 * this feature bit is not displayed in /proc/cpuinfo at all.
15 */
10 16
11/* Intel-defined CPU features, CPUID level 0x00000001 (edx), word 0 */ 17/* Intel-defined CPU features, CPUID level 0x00000001 (edx), word 0 */
12#define X86_FEATURE_FPU (0*32+ 0) /* Onboard FPU */ 18#define X86_FEATURE_FPU (0*32+ 0) /* Onboard FPU */
@@ -14,7 +20,7 @@
14#define X86_FEATURE_DE (0*32+ 2) /* Debugging Extensions */ 20#define X86_FEATURE_DE (0*32+ 2) /* Debugging Extensions */
15#define X86_FEATURE_PSE (0*32+ 3) /* Page Size Extensions */ 21#define X86_FEATURE_PSE (0*32+ 3) /* Page Size Extensions */
16#define X86_FEATURE_TSC (0*32+ 4) /* Time Stamp Counter */ 22#define X86_FEATURE_TSC (0*32+ 4) /* Time Stamp Counter */
17#define X86_FEATURE_MSR (0*32+ 5) /* Model-Specific Registers, RDMSR, WRMSR */ 23#define X86_FEATURE_MSR (0*32+ 5) /* Model-Specific Registers */
18#define X86_FEATURE_PAE (0*32+ 6) /* Physical Address Extensions */ 24#define X86_FEATURE_PAE (0*32+ 6) /* Physical Address Extensions */
19#define X86_FEATURE_MCE (0*32+ 7) /* Machine Check Architecture */ 25#define X86_FEATURE_MCE (0*32+ 7) /* Machine Check Architecture */
20#define X86_FEATURE_CX8 (0*32+ 8) /* CMPXCHG8 instruction */ 26#define X86_FEATURE_CX8 (0*32+ 8) /* CMPXCHG8 instruction */
@@ -23,22 +29,23 @@
23#define X86_FEATURE_MTRR (0*32+12) /* Memory Type Range Registers */ 29#define X86_FEATURE_MTRR (0*32+12) /* Memory Type Range Registers */
24#define X86_FEATURE_PGE (0*32+13) /* Page Global Enable */ 30#define X86_FEATURE_PGE (0*32+13) /* Page Global Enable */
25#define X86_FEATURE_MCA (0*32+14) /* Machine Check Architecture */ 31#define X86_FEATURE_MCA (0*32+14) /* Machine Check Architecture */
26#define X86_FEATURE_CMOV (0*32+15) /* CMOV instruction (FCMOVCC and FCOMI too if FPU present) */ 32#define X86_FEATURE_CMOV (0*32+15) /* CMOV instructions */
33 /* (plus FCMOVcc, FCOMI with FPU) */
27#define X86_FEATURE_PAT (0*32+16) /* Page Attribute Table */ 34#define X86_FEATURE_PAT (0*32+16) /* Page Attribute Table */
28#define X86_FEATURE_PSE36 (0*32+17) /* 36-bit PSEs */ 35#define X86_FEATURE_PSE36 (0*32+17) /* 36-bit PSEs */
29#define X86_FEATURE_PN (0*32+18) /* Processor serial number */ 36#define X86_FEATURE_PN (0*32+18) /* Processor serial number */
30#define X86_FEATURE_CLFLSH (0*32+19) /* Supports the CLFLUSH instruction */ 37#define X86_FEATURE_CLFLSH (0*32+19) /* "clflush" CLFLUSH instruction */
31#define X86_FEATURE_DS (0*32+21) /* Debug Store */ 38#define X86_FEATURE_DS (0*32+21) /* "dts" Debug Store */
32#define X86_FEATURE_ACPI (0*32+22) /* ACPI via MSR */ 39#define X86_FEATURE_ACPI (0*32+22) /* ACPI via MSR */
33#define X86_FEATURE_MMX (0*32+23) /* Multimedia Extensions */ 40#define X86_FEATURE_MMX (0*32+23) /* Multimedia Extensions */
34#define X86_FEATURE_FXSR (0*32+24) /* FXSAVE and FXRSTOR instructions (fast save and restore */ 41#define X86_FEATURE_FXSR (0*32+24) /* FXSAVE/FXRSTOR, CR4.OSFXSR */
35 /* of FPU context), and CR4.OSFXSR available */ 42#define X86_FEATURE_XMM (0*32+25) /* "sse" */
36#define X86_FEATURE_XMM (0*32+25) /* Streaming SIMD Extensions */ 43#define X86_FEATURE_XMM2 (0*32+26) /* "sse2" */
37#define X86_FEATURE_XMM2 (0*32+26) /* Streaming SIMD Extensions-2 */ 44#define X86_FEATURE_SELFSNOOP (0*32+27) /* "ss" CPU self snoop */
38#define X86_FEATURE_SELFSNOOP (0*32+27) /* CPU self snoop */
39#define X86_FEATURE_HT (0*32+28) /* Hyper-Threading */ 45#define X86_FEATURE_HT (0*32+28) /* Hyper-Threading */
40#define X86_FEATURE_ACC (0*32+29) /* Automatic clock control */ 46#define X86_FEATURE_ACC (0*32+29) /* "tm" Automatic clock control */
41#define X86_FEATURE_IA64 (0*32+30) /* IA-64 processor */ 47#define X86_FEATURE_IA64 (0*32+30) /* IA-64 processor */
48#define X86_FEATURE_PBE (0*32+31) /* Pending Break Enable */
42 49
43/* AMD-defined CPU features, CPUID level 0x80000001, word 1 */ 50/* AMD-defined CPU features, CPUID level 0x80000001, word 1 */
44/* Don't duplicate feature flags which are redundant with Intel! */ 51/* Don't duplicate feature flags which are redundant with Intel! */
@@ -46,7 +53,8 @@
46#define X86_FEATURE_MP (1*32+19) /* MP Capable. */ 53#define X86_FEATURE_MP (1*32+19) /* MP Capable. */
47#define X86_FEATURE_NX (1*32+20) /* Execute Disable */ 54#define X86_FEATURE_NX (1*32+20) /* Execute Disable */
48#define X86_FEATURE_MMXEXT (1*32+22) /* AMD MMX extensions */ 55#define X86_FEATURE_MMXEXT (1*32+22) /* AMD MMX extensions */
49#define X86_FEATURE_GBPAGES (1*32+26) /* GB pages */ 56#define X86_FEATURE_FXSR_OPT (1*32+25) /* FXSAVE/FXRSTOR optimizations */
57#define X86_FEATURE_GBPAGES (1*32+26) /* "pdpe1gb" GB pages */
50#define X86_FEATURE_RDTSCP (1*32+27) /* RDTSCP */ 58#define X86_FEATURE_RDTSCP (1*32+27) /* RDTSCP */
51#define X86_FEATURE_LM (1*32+29) /* Long Mode (x86-64) */ 59#define X86_FEATURE_LM (1*32+29) /* Long Mode (x86-64) */
52#define X86_FEATURE_3DNOWEXT (1*32+30) /* AMD 3DNow! extensions */ 60#define X86_FEATURE_3DNOWEXT (1*32+30) /* AMD 3DNow! extensions */
@@ -64,54 +72,79 @@
64#define X86_FEATURE_CYRIX_ARR (3*32+ 2) /* Cyrix ARRs (= MTRRs) */ 72#define X86_FEATURE_CYRIX_ARR (3*32+ 2) /* Cyrix ARRs (= MTRRs) */
65#define X86_FEATURE_CENTAUR_MCR (3*32+ 3) /* Centaur MCRs (= MTRRs) */ 73#define X86_FEATURE_CENTAUR_MCR (3*32+ 3) /* Centaur MCRs (= MTRRs) */
66/* cpu types for specific tunings: */ 74/* cpu types for specific tunings: */
67#define X86_FEATURE_K8 (3*32+ 4) /* Opteron, Athlon64 */ 75#define X86_FEATURE_K8 (3*32+ 4) /* "" Opteron, Athlon64 */
68#define X86_FEATURE_K7 (3*32+ 5) /* Athlon */ 76#define X86_FEATURE_K7 (3*32+ 5) /* "" Athlon */
69#define X86_FEATURE_P3 (3*32+ 6) /* P3 */ 77#define X86_FEATURE_P3 (3*32+ 6) /* "" P3 */
70#define X86_FEATURE_P4 (3*32+ 7) /* P4 */ 78#define X86_FEATURE_P4 (3*32+ 7) /* "" P4 */
71#define X86_FEATURE_CONSTANT_TSC (3*32+ 8) /* TSC ticks at a constant rate */ 79#define X86_FEATURE_CONSTANT_TSC (3*32+ 8) /* TSC ticks at a constant rate */
72#define X86_FEATURE_UP (3*32+ 9) /* smp kernel running on up */ 80#define X86_FEATURE_UP (3*32+ 9) /* smp kernel running on up */
73#define X86_FEATURE_FXSAVE_LEAK (3*32+10) /* FXSAVE leaks FOP/FIP/FOP */ 81#define X86_FEATURE_FXSAVE_LEAK (3*32+10) /* "" FXSAVE leaks FOP/FIP/FOP */
74#define X86_FEATURE_ARCH_PERFMON (3*32+11) /* Intel Architectural PerfMon */ 82#define X86_FEATURE_ARCH_PERFMON (3*32+11) /* Intel Architectural PerfMon */
83#define X86_FEATURE_NOPL (3*32+20) /* The NOPL (0F 1F) instructions */
75#define X86_FEATURE_PEBS (3*32+12) /* Precise-Event Based Sampling */ 84#define X86_FEATURE_PEBS (3*32+12) /* Precise-Event Based Sampling */
76#define X86_FEATURE_BTS (3*32+13) /* Branch Trace Store */ 85#define X86_FEATURE_BTS (3*32+13) /* Branch Trace Store */
77#define X86_FEATURE_SYSCALL32 (3*32+14) /* syscall in ia32 userspace */ 86#define X86_FEATURE_SYSCALL32 (3*32+14) /* "" syscall in ia32 userspace */
78#define X86_FEATURE_SYSENTER32 (3*32+15) /* sysenter in ia32 userspace */ 87#define X86_FEATURE_SYSENTER32 (3*32+15) /* "" sysenter in ia32 userspace */
79#define X86_FEATURE_REP_GOOD (3*32+16) /* rep microcode works well on this CPU */ 88#define X86_FEATURE_REP_GOOD (3*32+16) /* rep microcode works well */
80#define X86_FEATURE_MFENCE_RDTSC (3*32+17) /* Mfence synchronizes RDTSC */ 89#define X86_FEATURE_MFENCE_RDTSC (3*32+17) /* "" Mfence synchronizes RDTSC */
81#define X86_FEATURE_LFENCE_RDTSC (3*32+18) /* Lfence synchronizes RDTSC */ 90#define X86_FEATURE_LFENCE_RDTSC (3*32+18) /* "" Lfence synchronizes RDTSC */
82#define X86_FEATURE_11AP (3*32+19) /* Bad local APIC aka 11AP */ 91#define X86_FEATURE_11AP (3*32+19) /* "" Bad local APIC aka 11AP */
83#define X86_FEATURE_NOPL (3*32+20) /* The NOPL (0F 1F) instructions */ 92#define X86_FEATURE_NOPL (3*32+20) /* The NOPL (0F 1F) instructions */
84#define X86_FEATURE_AMDC1E (3*32+21) /* AMD C1E detected */ 93#define X86_FEATURE_AMDC1E (3*32+21) /* AMD C1E detected */
94#define X86_FEATURE_XTOPOLOGY (3*32+21) /* cpu topology enum extensions */
85 95
86/* Intel-defined CPU features, CPUID level 0x00000001 (ecx), word 4 */ 96/* Intel-defined CPU features, CPUID level 0x00000001 (ecx), word 4 */
87#define X86_FEATURE_XMM3 (4*32+ 0) /* Streaming SIMD Extensions-3 */ 97#define X86_FEATURE_XMM3 (4*32+ 0) /* "pni" SSE-3 */
88#define X86_FEATURE_MWAIT (4*32+ 3) /* Monitor/Mwait support */ 98#define X86_FEATURE_PCLMULQDQ (4*32+ 1) /* PCLMULQDQ instruction */
89#define X86_FEATURE_DSCPL (4*32+ 4) /* CPL Qualified Debug Store */ 99#define X86_FEATURE_DTES64 (4*32+ 2) /* 64-bit Debug Store */
100#define X86_FEATURE_MWAIT (4*32+ 3) /* "monitor" Monitor/Mwait support */
101#define X86_FEATURE_DSCPL (4*32+ 4) /* "ds_cpl" CPL Qual. Debug Store */
102#define X86_FEATURE_VMX (4*32+ 5) /* Hardware virtualization */
103#define X86_FEATURE_SMX (4*32+ 6) /* Safer mode */
90#define X86_FEATURE_EST (4*32+ 7) /* Enhanced SpeedStep */ 104#define X86_FEATURE_EST (4*32+ 7) /* Enhanced SpeedStep */
91#define X86_FEATURE_TM2 (4*32+ 8) /* Thermal Monitor 2 */ 105#define X86_FEATURE_TM2 (4*32+ 8) /* Thermal Monitor 2 */
106#define X86_FEATURE_SSSE3 (4*32+ 9) /* Supplemental SSE-3 */
92#define X86_FEATURE_CID (4*32+10) /* Context ID */ 107#define X86_FEATURE_CID (4*32+10) /* Context ID */
108#define X86_FEATURE_FMA (4*32+12) /* Fused multiply-add */
93#define X86_FEATURE_CX16 (4*32+13) /* CMPXCHG16B */ 109#define X86_FEATURE_CX16 (4*32+13) /* CMPXCHG16B */
94#define X86_FEATURE_XTPR (4*32+14) /* Send Task Priority Messages */ 110#define X86_FEATURE_XTPR (4*32+14) /* Send Task Priority Messages */
111#define X86_FEATURE_PDCM (4*32+15) /* Performance Capabilities */
95#define X86_FEATURE_DCA (4*32+18) /* Direct Cache Access */ 112#define X86_FEATURE_DCA (4*32+18) /* Direct Cache Access */
113#define X86_FEATURE_XMM4_1 (4*32+19) /* "sse4_1" SSE-4.1 */
114#define X86_FEATURE_XMM4_2 (4*32+20) /* "sse4_2" SSE-4.2 */
96#define X86_FEATURE_X2APIC (4*32+21) /* x2APIC */ 115#define X86_FEATURE_X2APIC (4*32+21) /* x2APIC */
97#define X86_FEATURE_XMM4_2 (4*32+20) /* Streaming SIMD Extensions-4.2 */ 116#define X86_FEATURE_AES (4*32+25) /* AES instructions */
117#define X86_FEATURE_XSAVE (4*32+26) /* XSAVE/XRSTOR/XSETBV/XGETBV */
118#define X86_FEATURE_OSXSAVE (4*32+27) /* "" XSAVE enabled in the OS */
119#define X86_FEATURE_AVX (4*32+28) /* Advanced Vector Extensions */
98 120
99/* VIA/Cyrix/Centaur-defined CPU features, CPUID level 0xC0000001, word 5 */ 121/* VIA/Cyrix/Centaur-defined CPU features, CPUID level 0xC0000001, word 5 */
100#define X86_FEATURE_XSTORE (5*32+ 2) /* on-CPU RNG present (xstore insn) */ 122#define X86_FEATURE_XSTORE (5*32+ 2) /* "rng" RNG present (xstore) */
101#define X86_FEATURE_XSTORE_EN (5*32+ 3) /* on-CPU RNG enabled */ 123#define X86_FEATURE_XSTORE_EN (5*32+ 3) /* "rng_en" RNG enabled */
102#define X86_FEATURE_XCRYPT (5*32+ 6) /* on-CPU crypto (xcrypt insn) */ 124#define X86_FEATURE_XCRYPT (5*32+ 6) /* "ace" on-CPU crypto (xcrypt) */
103#define X86_FEATURE_XCRYPT_EN (5*32+ 7) /* on-CPU crypto enabled */ 125#define X86_FEATURE_XCRYPT_EN (5*32+ 7) /* "ace_en" on-CPU crypto enabled */
104#define X86_FEATURE_ACE2 (5*32+ 8) /* Advanced Cryptography Engine v2 */ 126#define X86_FEATURE_ACE2 (5*32+ 8) /* Advanced Cryptography Engine v2 */
105#define X86_FEATURE_ACE2_EN (5*32+ 9) /* ACE v2 enabled */ 127#define X86_FEATURE_ACE2_EN (5*32+ 9) /* ACE v2 enabled */
106#define X86_FEATURE_PHE (5*32+ 10) /* PadLock Hash Engine */ 128#define X86_FEATURE_PHE (5*32+10) /* PadLock Hash Engine */
107#define X86_FEATURE_PHE_EN (5*32+ 11) /* PHE enabled */ 129#define X86_FEATURE_PHE_EN (5*32+11) /* PHE enabled */
108#define X86_FEATURE_PMM (5*32+ 12) /* PadLock Montgomery Multiplier */ 130#define X86_FEATURE_PMM (5*32+12) /* PadLock Montgomery Multiplier */
109#define X86_FEATURE_PMM_EN (5*32+ 13) /* PMM enabled */ 131#define X86_FEATURE_PMM_EN (5*32+13) /* PMM enabled */
110 132
111/* More extended AMD flags: CPUID level 0x80000001, ecx, word 6 */ 133/* More extended AMD flags: CPUID level 0x80000001, ecx, word 6 */
112#define X86_FEATURE_LAHF_LM (6*32+ 0) /* LAHF/SAHF in long mode */ 134#define X86_FEATURE_LAHF_LM (6*32+ 0) /* LAHF/SAHF in long mode */
113#define X86_FEATURE_CMP_LEGACY (6*32+ 1) /* If yes HyperThreading not valid */ 135#define X86_FEATURE_CMP_LEGACY (6*32+ 1) /* If yes HyperThreading not valid */
114#define X86_FEATURE_IBS (6*32+ 10) /* Instruction Based Sampling */ 136#define X86_FEATURE_SVM (6*32+ 2) /* Secure virtual machine */
137#define X86_FEATURE_EXTAPIC (6*32+ 3) /* Extended APIC space */
138#define X86_FEATURE_CR8_LEGACY (6*32+ 4) /* CR8 in 32-bit mode */
139#define X86_FEATURE_ABM (6*32+ 5) /* Advanced bit manipulation */
140#define X86_FEATURE_SSE4A (6*32+ 6) /* SSE-4A */
141#define X86_FEATURE_MISALIGNSSE (6*32+ 7) /* Misaligned SSE mode */
142#define X86_FEATURE_3DNOWPREFETCH (6*32+ 8) /* 3DNow prefetch instructions */
143#define X86_FEATURE_OSVW (6*32+ 9) /* OS Visible Workaround */
144#define X86_FEATURE_IBS (6*32+10) /* Instruction Based Sampling */
145#define X86_FEATURE_SSE5 (6*32+11) /* SSE-5 */
146#define X86_FEATURE_SKINIT (6*32+12) /* SKINIT/STGI instructions */
147#define X86_FEATURE_WDT (6*32+13) /* Watchdog timer */
115 148
116/* 149/*
117 * Auxiliary flags: Linux defined - For features scattered in various 150 * Auxiliary flags: Linux defined - For features scattered in various
@@ -119,6 +152,13 @@
119 */ 152 */
120#define X86_FEATURE_IDA (7*32+ 0) /* Intel Dynamic Acceleration */ 153#define X86_FEATURE_IDA (7*32+ 0) /* Intel Dynamic Acceleration */
121 154
155/* Virtualization flags: Linux defined */
156#define X86_FEATURE_TPR_SHADOW (8*32+ 0) /* Intel TPR Shadow */
157#define X86_FEATURE_VNMI (8*32+ 1) /* Intel Virtual NMI */
158#define X86_FEATURE_FLEXPRIORITY (8*32+ 2) /* Intel FlexPriority */
159#define X86_FEATURE_EPT (8*32+ 3) /* Intel Extended Page Table */
160#define X86_FEATURE_VPID (8*32+ 4) /* Intel Virtual Processor ID */
161
122#if defined(__KERNEL__) && !defined(__ASSEMBLY__) 162#if defined(__KERNEL__) && !defined(__ASSEMBLY__)
123 163
124#include <linux/bitops.h> 164#include <linux/bitops.h>
@@ -152,7 +192,7 @@ extern const char * const x86_power_flags[32];
152} while (0) 192} while (0)
153#define setup_force_cpu_cap(bit) do { \ 193#define setup_force_cpu_cap(bit) do { \
154 set_cpu_cap(&boot_cpu_data, bit); \ 194 set_cpu_cap(&boot_cpu_data, bit); \
155 clear_bit(bit, (unsigned long *)cleared_cpu_caps); \ 195 clear_bit(bit, (unsigned long *)cleared_cpu_caps); \
156} while (0) 196} while (0)
157 197
158#define cpu_has_fpu boot_cpu_has(X86_FEATURE_FPU) 198#define cpu_has_fpu boot_cpu_has(X86_FEATURE_FPU)
@@ -193,8 +233,10 @@ extern const char * const x86_power_flags[32];
193#define cpu_has_gbpages boot_cpu_has(X86_FEATURE_GBPAGES) 233#define cpu_has_gbpages boot_cpu_has(X86_FEATURE_GBPAGES)
194#define cpu_has_arch_perfmon boot_cpu_has(X86_FEATURE_ARCH_PERFMON) 234#define cpu_has_arch_perfmon boot_cpu_has(X86_FEATURE_ARCH_PERFMON)
195#define cpu_has_pat boot_cpu_has(X86_FEATURE_PAT) 235#define cpu_has_pat boot_cpu_has(X86_FEATURE_PAT)
196#define cpu_has_x2apic boot_cpu_has(X86_FEATURE_X2APIC) 236#define cpu_has_xmm4_1 boot_cpu_has(X86_FEATURE_XMM4_1)
197#define cpu_has_xmm4_2 boot_cpu_has(X86_FEATURE_XMM4_2) 237#define cpu_has_xmm4_2 boot_cpu_has(X86_FEATURE_XMM4_2)
238#define cpu_has_x2apic boot_cpu_has(X86_FEATURE_X2APIC)
239#define cpu_has_xsave boot_cpu_has(X86_FEATURE_XSAVE)
198 240
199#if defined(CONFIG_X86_INVLPG) || defined(CONFIG_X86_64) 241#if defined(CONFIG_X86_INVLPG) || defined(CONFIG_X86_64)
200# define cpu_has_invlpg 1 242# define cpu_has_invlpg 1
diff --git a/include/asm-x86/e820.h b/include/asm-x86/e820.h
index f52daf176bcb..5abbdec06bd2 100644
--- a/include/asm-x86/e820.h
+++ b/include/asm-x86/e820.h
@@ -43,6 +43,7 @@
43#define E820_RESERVED 2 43#define E820_RESERVED 2
44#define E820_ACPI 3 44#define E820_ACPI 3
45#define E820_NVS 4 45#define E820_NVS 4
46#define E820_UNUSABLE 5
46 47
47/* reserved RAM used by kernel itself */ 48/* reserved RAM used by kernel itself */
48#define E820_RESERVED_KERN 128 49#define E820_RESERVED_KERN 128
@@ -121,6 +122,7 @@ extern void e820_register_active_regions(int nid, unsigned long start_pfn,
121extern u64 e820_hole_size(u64 start, u64 end); 122extern u64 e820_hole_size(u64 start, u64 end);
122extern void finish_e820_parsing(void); 123extern void finish_e820_parsing(void);
123extern void e820_reserve_resources(void); 124extern void e820_reserve_resources(void);
125extern void e820_reserve_resources_late(void);
124extern void setup_memory_map(void); 126extern void setup_memory_map(void);
125extern char *default_machine_specific_memory_setup(void); 127extern char *default_machine_specific_memory_setup(void);
126extern char *machine_specific_memory_setup(void); 128extern char *machine_specific_memory_setup(void);
diff --git a/include/asm-x86/i387.h b/include/asm-x86/i387.h
index 1ecdc3ed96e4..9ba862a4eac0 100644
--- a/include/asm-x86/i387.h
+++ b/include/asm-x86/i387.h
@@ -19,7 +19,9 @@
19#include <asm/sigcontext.h> 19#include <asm/sigcontext.h>
20#include <asm/user.h> 20#include <asm/user.h>
21#include <asm/uaccess.h> 21#include <asm/uaccess.h>
22#include <asm/xsave.h>
22 23
24extern unsigned int sig_xstate_size;
23extern void fpu_init(void); 25extern void fpu_init(void);
24extern void mxcsr_feature_mask_init(void); 26extern void mxcsr_feature_mask_init(void);
25extern int init_fpu(struct task_struct *child); 27extern int init_fpu(struct task_struct *child);
@@ -31,12 +33,18 @@ extern user_regset_active_fn fpregs_active, xfpregs_active;
31extern user_regset_get_fn fpregs_get, xfpregs_get, fpregs_soft_get; 33extern user_regset_get_fn fpregs_get, xfpregs_get, fpregs_soft_get;
32extern user_regset_set_fn fpregs_set, xfpregs_set, fpregs_soft_set; 34extern user_regset_set_fn fpregs_set, xfpregs_set, fpregs_soft_set;
33 35
36extern struct _fpx_sw_bytes fx_sw_reserved;
34#ifdef CONFIG_IA32_EMULATION 37#ifdef CONFIG_IA32_EMULATION
38extern unsigned int sig_xstate_ia32_size;
39extern struct _fpx_sw_bytes fx_sw_reserved_ia32;
35struct _fpstate_ia32; 40struct _fpstate_ia32;
36extern int save_i387_ia32(struct _fpstate_ia32 __user *buf); 41struct _xstate_ia32;
37extern int restore_i387_ia32(struct _fpstate_ia32 __user *buf); 42extern int save_i387_xstate_ia32(void __user *buf);
43extern int restore_i387_xstate_ia32(void __user *buf);
38#endif 44#endif
39 45
46#define X87_FSW_ES (1 << 7) /* Exception Summary */
47
40#ifdef CONFIG_X86_64 48#ifdef CONFIG_X86_64
41 49
42/* Ignore delayed exceptions from user space */ 50/* Ignore delayed exceptions from user space */
@@ -47,7 +55,7 @@ static inline void tolerant_fwait(void)
47 _ASM_EXTABLE(1b, 2b)); 55 _ASM_EXTABLE(1b, 2b));
48} 56}
49 57
50static inline int restore_fpu_checking(struct i387_fxsave_struct *fx) 58static inline int fxrstor_checking(struct i387_fxsave_struct *fx)
51{ 59{
52 int err; 60 int err;
53 61
@@ -67,15 +75,31 @@ static inline int restore_fpu_checking(struct i387_fxsave_struct *fx)
67 return err; 75 return err;
68} 76}
69 77
70#define X87_FSW_ES (1 << 7) /* Exception Summary */ 78static inline int restore_fpu_checking(struct task_struct *tsk)
79{
80 if (task_thread_info(tsk)->status & TS_XSAVE)
81 return xrstor_checking(&tsk->thread.xstate->xsave);
82 else
83 return fxrstor_checking(&tsk->thread.xstate->fxsave);
84}
71 85
72/* AMD CPUs don't save/restore FDP/FIP/FOP unless an exception 86/* AMD CPUs don't save/restore FDP/FIP/FOP unless an exception
73 is pending. Clear the x87 state here by setting it to fixed 87 is pending. Clear the x87 state here by setting it to fixed
74 values. The kernel data segment can be sometimes 0 and sometimes 88 values. The kernel data segment can be sometimes 0 and sometimes
75 new user value. Both should be ok. 89 new user value. Both should be ok.
76 Use the PDA as safe address because it should be already in L1. */ 90 Use the PDA as safe address because it should be already in L1. */
77static inline void clear_fpu_state(struct i387_fxsave_struct *fx) 91static inline void clear_fpu_state(struct task_struct *tsk)
78{ 92{
93 struct xsave_struct *xstate = &tsk->thread.xstate->xsave;
94 struct i387_fxsave_struct *fx = &tsk->thread.xstate->fxsave;
95
96 /*
97 * xsave header may indicate the init state of the FP.
98 */
99 if ((task_thread_info(tsk)->status & TS_XSAVE) &&
100 !(xstate->xsave_hdr.xstate_bv & XSTATE_FP))
101 return;
102
79 if (unlikely(fx->swd & X87_FSW_ES)) 103 if (unlikely(fx->swd & X87_FSW_ES))
80 asm volatile("fnclex"); 104 asm volatile("fnclex");
81 alternative_input(ASM_NOP8 ASM_NOP2, 105 alternative_input(ASM_NOP8 ASM_NOP2,
@@ -84,7 +108,7 @@ static inline void clear_fpu_state(struct i387_fxsave_struct *fx)
84 X86_FEATURE_FXSAVE_LEAK); 108 X86_FEATURE_FXSAVE_LEAK);
85} 109}
86 110
87static inline int save_i387_checking(struct i387_fxsave_struct __user *fx) 111static inline int fxsave_user(struct i387_fxsave_struct __user *fx)
88{ 112{
89 int err; 113 int err;
90 114
@@ -108,7 +132,7 @@ static inline int save_i387_checking(struct i387_fxsave_struct __user *fx)
108 return err; 132 return err;
109} 133}
110 134
111static inline void __save_init_fpu(struct task_struct *tsk) 135static inline void fxsave(struct task_struct *tsk)
112{ 136{
113 /* Using "rex64; fxsave %0" is broken because, if the memory operand 137 /* Using "rex64; fxsave %0" is broken because, if the memory operand
114 uses any extended registers for addressing, a second REX prefix 138 uses any extended registers for addressing, a second REX prefix
@@ -133,7 +157,16 @@ static inline void __save_init_fpu(struct task_struct *tsk)
133 : "=m" (tsk->thread.xstate->fxsave) 157 : "=m" (tsk->thread.xstate->fxsave)
134 : "cdaSDb" (&tsk->thread.xstate->fxsave)); 158 : "cdaSDb" (&tsk->thread.xstate->fxsave));
135#endif 159#endif
136 clear_fpu_state(&tsk->thread.xstate->fxsave); 160}
161
162static inline void __save_init_fpu(struct task_struct *tsk)
163{
164 if (task_thread_info(tsk)->status & TS_XSAVE)
165 xsave(tsk);
166 else
167 fxsave(tsk);
168
169 clear_fpu_state(tsk);
137 task_thread_info(tsk)->status &= ~TS_USEDFPU; 170 task_thread_info(tsk)->status &= ~TS_USEDFPU;
138} 171}
139 172
@@ -148,6 +181,10 @@ static inline void tolerant_fwait(void)
148 181
149static inline void restore_fpu(struct task_struct *tsk) 182static inline void restore_fpu(struct task_struct *tsk)
150{ 183{
184 if (task_thread_info(tsk)->status & TS_XSAVE) {
185 xrstor_checking(&tsk->thread.xstate->xsave);
186 return;
187 }
151 /* 188 /*
152 * The "nop" is needed to make the instructions the same 189 * The "nop" is needed to make the instructions the same
153 * length. 190 * length.
@@ -173,6 +210,27 @@ static inline void restore_fpu(struct task_struct *tsk)
173 */ 210 */
174static inline void __save_init_fpu(struct task_struct *tsk) 211static inline void __save_init_fpu(struct task_struct *tsk)
175{ 212{
213 if (task_thread_info(tsk)->status & TS_XSAVE) {
214 struct xsave_struct *xstate = &tsk->thread.xstate->xsave;
215 struct i387_fxsave_struct *fx = &tsk->thread.xstate->fxsave;
216
217 xsave(tsk);
218
219 /*
220 * xsave header may indicate the init state of the FP.
221 */
222 if (!(xstate->xsave_hdr.xstate_bv & XSTATE_FP))
223 goto end;
224
225 if (unlikely(fx->swd & X87_FSW_ES))
226 asm volatile("fnclex");
227
228 /*
229 * we can do a simple return here or be paranoid :)
230 */
231 goto clear_state;
232 }
233
176 /* Use more nops than strictly needed in case the compiler 234 /* Use more nops than strictly needed in case the compiler
177 varies code */ 235 varies code */
178 alternative_input( 236 alternative_input(
@@ -182,6 +240,7 @@ static inline void __save_init_fpu(struct task_struct *tsk)
182 X86_FEATURE_FXSR, 240 X86_FEATURE_FXSR,
183 [fx] "m" (tsk->thread.xstate->fxsave), 241 [fx] "m" (tsk->thread.xstate->fxsave),
184 [fsw] "m" (tsk->thread.xstate->fxsave.swd) : "memory"); 242 [fsw] "m" (tsk->thread.xstate->fxsave.swd) : "memory");
243clear_state:
185 /* AMD K7/K8 CPUs don't save/restore FDP/FIP/FOP unless an exception 244 /* AMD K7/K8 CPUs don't save/restore FDP/FIP/FOP unless an exception
186 is pending. Clear the x87 state here by setting it to fixed 245 is pending. Clear the x87 state here by setting it to fixed
187 values. safe_address is a random variable that should be in L1 */ 246 values. safe_address is a random variable that should be in L1 */
@@ -191,16 +250,17 @@ static inline void __save_init_fpu(struct task_struct *tsk)
191 "fildl %[addr]", /* set F?P to defined value */ 250 "fildl %[addr]", /* set F?P to defined value */
192 X86_FEATURE_FXSAVE_LEAK, 251 X86_FEATURE_FXSAVE_LEAK,
193 [addr] "m" (safe_address)); 252 [addr] "m" (safe_address));
253end:
194 task_thread_info(tsk)->status &= ~TS_USEDFPU; 254 task_thread_info(tsk)->status &= ~TS_USEDFPU;
195} 255}
196 256
257#endif /* CONFIG_X86_64 */
258
197/* 259/*
198 * Signal frame handlers... 260 * Signal frame handlers...
199 */ 261 */
200extern int save_i387(struct _fpstate __user *buf); 262extern int save_i387_xstate(void __user *buf);
201extern int restore_i387(struct _fpstate __user *buf); 263extern int restore_i387_xstate(void __user *buf);
202
203#endif /* CONFIG_X86_64 */
204 264
205static inline void __unlazy_fpu(struct task_struct *tsk) 265static inline void __unlazy_fpu(struct task_struct *tsk)
206{ 266{
diff --git a/include/asm-x86/msr-index.h b/include/asm-x86/msr-index.h
index 3052f058ab06..0bb43301a202 100644
--- a/include/asm-x86/msr-index.h
+++ b/include/asm-x86/msr-index.h
@@ -176,6 +176,7 @@
176#define MSR_IA32_TSC 0x00000010 176#define MSR_IA32_TSC 0x00000010
177#define MSR_IA32_PLATFORM_ID 0x00000017 177#define MSR_IA32_PLATFORM_ID 0x00000017
178#define MSR_IA32_EBL_CR_POWERON 0x0000002a 178#define MSR_IA32_EBL_CR_POWERON 0x0000002a
179#define MSR_IA32_FEATURE_CONTROL 0x0000003a
179 180
180#define MSR_IA32_APICBASE 0x0000001b 181#define MSR_IA32_APICBASE 0x0000001b
181#define MSR_IA32_APICBASE_BSP (1<<8) 182#define MSR_IA32_APICBASE_BSP (1<<8)
@@ -310,4 +311,19 @@
310/* Geode defined MSRs */ 311/* Geode defined MSRs */
311#define MSR_GEODE_BUSCONT_CONF0 0x00001900 312#define MSR_GEODE_BUSCONT_CONF0 0x00001900
312 313
314/* Intel VT MSRs */
315#define MSR_IA32_VMX_BASIC 0x00000480
316#define MSR_IA32_VMX_PINBASED_CTLS 0x00000481
317#define MSR_IA32_VMX_PROCBASED_CTLS 0x00000482
318#define MSR_IA32_VMX_EXIT_CTLS 0x00000483
319#define MSR_IA32_VMX_ENTRY_CTLS 0x00000484
320#define MSR_IA32_VMX_MISC 0x00000485
321#define MSR_IA32_VMX_CR0_FIXED0 0x00000486
322#define MSR_IA32_VMX_CR0_FIXED1 0x00000487
323#define MSR_IA32_VMX_CR4_FIXED0 0x00000488
324#define MSR_IA32_VMX_CR4_FIXED1 0x00000489
325#define MSR_IA32_VMX_VMCS_ENUM 0x0000048a
326#define MSR_IA32_VMX_PROCBASED_CTLS2 0x0000048b
327#define MSR_IA32_VMX_EPT_VPID_CAP 0x0000048c
328
313#endif /* ASM_X86__MSR_INDEX_H */ 329#endif /* ASM_X86__MSR_INDEX_H */
diff --git a/include/asm-x86/processor-cyrix.h b/include/asm-x86/processor-cyrix.h
index 97568ada1f97..1198f2a0e42c 100644
--- a/include/asm-x86/processor-cyrix.h
+++ b/include/asm-x86/processor-cyrix.h
@@ -28,3 +28,11 @@ static inline void setCx86(u8 reg, u8 data)
28 outb(reg, 0x22); 28 outb(reg, 0x22);
29 outb(data, 0x23); 29 outb(data, 0x23);
30} 30}
31
32#define getCx86_old(reg) ({ outb((reg), 0x22); inb(0x23); })
33
34#define setCx86_old(reg, data) do { \
35 outb((reg), 0x22); \
36 outb((data), 0x23); \
37} while (0)
38
diff --git a/include/asm-x86/processor-flags.h b/include/asm-x86/processor-flags.h
index 5dd79774f693..dc5f0712f9fa 100644
--- a/include/asm-x86/processor-flags.h
+++ b/include/asm-x86/processor-flags.h
@@ -59,6 +59,7 @@
59#define X86_CR4_OSFXSR 0x00000200 /* enable fast FPU save and restore */ 59#define X86_CR4_OSFXSR 0x00000200 /* enable fast FPU save and restore */
60#define X86_CR4_OSXMMEXCPT 0x00000400 /* enable unmasked SSE exceptions */ 60#define X86_CR4_OSXMMEXCPT 0x00000400 /* enable unmasked SSE exceptions */
61#define X86_CR4_VMXE 0x00002000 /* enable VMX virtualization */ 61#define X86_CR4_VMXE 0x00002000 /* enable VMX virtualization */
62#define X86_CR4_OSXSAVE 0x00040000 /* enable xsave and xrestore */
62 63
63/* 64/*
64 * x86-64 Task Priority Register, CR8 65 * x86-64 Task Priority Register, CR8
diff --git a/include/asm-x86/processor.h b/include/asm-x86/processor.h
index 5eaf9bf0a623..c7d35464a4bb 100644
--- a/include/asm-x86/processor.h
+++ b/include/asm-x86/processor.h
@@ -76,11 +76,11 @@ struct cpuinfo_x86 {
76 int x86_tlbsize; 76 int x86_tlbsize;
77 __u8 x86_virt_bits; 77 __u8 x86_virt_bits;
78 __u8 x86_phys_bits; 78 __u8 x86_phys_bits;
79#endif
79 /* CPUID returned core id bits: */ 80 /* CPUID returned core id bits: */
80 __u8 x86_coreid_bits; 81 __u8 x86_coreid_bits;
81 /* Max extended CPUID function supported: */ 82 /* Max extended CPUID function supported: */
82 __u32 extended_cpuid_level; 83 __u32 extended_cpuid_level;
83#endif
84 /* Maximum supported CPUID level, -1=no CPUID: */ 84 /* Maximum supported CPUID level, -1=no CPUID: */
85 int cpuid_level; 85 int cpuid_level;
86 __u32 x86_capability[NCAPINTS]; 86 __u32 x86_capability[NCAPINTS];
@@ -166,11 +166,8 @@ extern void init_scattered_cpuid_features(struct cpuinfo_x86 *c);
166extern unsigned int init_intel_cacheinfo(struct cpuinfo_x86 *c); 166extern unsigned int init_intel_cacheinfo(struct cpuinfo_x86 *c);
167extern unsigned short num_cache_leaves; 167extern unsigned short num_cache_leaves;
168 168
169#if defined(CONFIG_X86_HT) || defined(CONFIG_X86_64) 169extern void detect_extended_topology(struct cpuinfo_x86 *c);
170extern void detect_ht(struct cpuinfo_x86 *c); 170extern void detect_ht(struct cpuinfo_x86 *c);
171#else
172static inline void detect_ht(struct cpuinfo_x86 *c) {}
173#endif
174 171
175static inline void native_cpuid(unsigned int *eax, unsigned int *ebx, 172static inline void native_cpuid(unsigned int *eax, unsigned int *ebx,
176 unsigned int *ecx, unsigned int *edx) 173 unsigned int *ecx, unsigned int *edx)
@@ -327,7 +324,12 @@ struct i387_fxsave_struct {
327 /* 16*16 bytes for each XMM-reg = 256 bytes: */ 324 /* 16*16 bytes for each XMM-reg = 256 bytes: */
328 u32 xmm_space[64]; 325 u32 xmm_space[64];
329 326
330 u32 padding[24]; 327 u32 padding[12];
328
329 union {
330 u32 padding1[12];
331 u32 sw_reserved[12];
332 };
331 333
332} __attribute__((aligned(16))); 334} __attribute__((aligned(16)));
333 335
@@ -351,10 +353,23 @@ struct i387_soft_struct {
351 u32 entry_eip; 353 u32 entry_eip;
352}; 354};
353 355
356struct xsave_hdr_struct {
357 u64 xstate_bv;
358 u64 reserved1[2];
359 u64 reserved2[5];
360} __attribute__((packed));
361
362struct xsave_struct {
363 struct i387_fxsave_struct i387;
364 struct xsave_hdr_struct xsave_hdr;
365 /* new processor state extensions will go here */
366} __attribute__ ((packed, aligned (64)));
367
354union thread_xstate { 368union thread_xstate {
355 struct i387_fsave_struct fsave; 369 struct i387_fsave_struct fsave;
356 struct i387_fxsave_struct fxsave; 370 struct i387_fxsave_struct fxsave;
357 struct i387_soft_struct soft; 371 struct i387_soft_struct soft;
372 struct xsave_struct xsave;
358}; 373};
359 374
360#ifdef CONFIG_X86_64 375#ifdef CONFIG_X86_64
diff --git a/include/asm-x86/sigcontext.h b/include/asm-x86/sigcontext.h
index 24879c85b291..ee813f4fe5d5 100644
--- a/include/asm-x86/sigcontext.h
+++ b/include/asm-x86/sigcontext.h
@@ -4,6 +4,40 @@
4#include <linux/compiler.h> 4#include <linux/compiler.h>
5#include <asm/types.h> 5#include <asm/types.h>
6 6
7#define FP_XSTATE_MAGIC1 0x46505853U
8#define FP_XSTATE_MAGIC2 0x46505845U
9#define FP_XSTATE_MAGIC2_SIZE sizeof(FP_XSTATE_MAGIC2)
10
11/*
12 * bytes 464..511 in the current 512byte layout of fxsave/fxrstor frame
13 * are reserved for SW usage. On cpu's supporting xsave/xrstor, these bytes
14 * are used to extended the fpstate pointer in the sigcontext, which now
15 * includes the extended state information along with fpstate information.
16 *
17 * Presence of FP_XSTATE_MAGIC1 at the beginning of this SW reserved
18 * area and FP_XSTATE_MAGIC2 at the end of memory layout
19 * (extended_size - FP_XSTATE_MAGIC2_SIZE) indicates the presence of the
20 * extended state information in the memory layout pointed by the fpstate
21 * pointer in sigcontext.
22 */
23struct _fpx_sw_bytes {
24 __u32 magic1; /* FP_XSTATE_MAGIC1 */
25 __u32 extended_size; /* total size of the layout referred by
26 * fpstate pointer in the sigcontext.
27 */
28 __u64 xstate_bv;
29 /* feature bit mask (including fp/sse/extended
30 * state) that is present in the memory
31 * layout.
32 */
33 __u32 xstate_size; /* actual xsave state size, based on the
34 * features saved in the layout.
35 * 'extended_size' will be greater than
36 * 'xstate_size'.
37 */
38 __u32 padding[7]; /* for future use. */
39};
40
7#ifdef __i386__ 41#ifdef __i386__
8/* 42/*
9 * As documented in the iBCS2 standard.. 43 * As documented in the iBCS2 standard..
@@ -53,7 +87,13 @@ struct _fpstate {
53 unsigned long reserved; 87 unsigned long reserved;
54 struct _fpxreg _fxsr_st[8]; /* FXSR FPU reg data is ignored */ 88 struct _fpxreg _fxsr_st[8]; /* FXSR FPU reg data is ignored */
55 struct _xmmreg _xmm[8]; 89 struct _xmmreg _xmm[8];
56 unsigned long padding[56]; 90 unsigned long padding1[44];
91
92 union {
93 unsigned long padding2[12];
94 struct _fpx_sw_bytes sw_reserved; /* represents the extended
95 * state info */
96 };
57}; 97};
58 98
59#define X86_FXSR_MAGIC 0x0000 99#define X86_FXSR_MAGIC 0x0000
@@ -79,7 +119,15 @@ struct sigcontext {
79 unsigned long flags; 119 unsigned long flags;
80 unsigned long sp_at_signal; 120 unsigned long sp_at_signal;
81 unsigned short ss, __ssh; 121 unsigned short ss, __ssh;
82 struct _fpstate __user *fpstate; 122
123 /*
124 * fpstate is really (struct _fpstate *) or (struct _xstate *)
125 * depending on the FP_XSTATE_MAGIC1 encoded in the SW reserved
126 * bytes of (struct _fpstate) and FP_XSTATE_MAGIC2 present at the end
127 * of extended memory layout. See comments at the defintion of
128 * (struct _fpx_sw_bytes)
129 */
130 void __user *fpstate; /* zero when no FPU/extended context */
83 unsigned long oldmask; 131 unsigned long oldmask;
84 unsigned long cr2; 132 unsigned long cr2;
85}; 133};
@@ -130,7 +178,12 @@ struct _fpstate {
130 __u32 mxcsr_mask; 178 __u32 mxcsr_mask;
131 __u32 st_space[32]; /* 8*16 bytes for each FP-reg */ 179 __u32 st_space[32]; /* 8*16 bytes for each FP-reg */
132 __u32 xmm_space[64]; /* 16*16 bytes for each XMM-reg */ 180 __u32 xmm_space[64]; /* 16*16 bytes for each XMM-reg */
133 __u32 reserved2[24]; 181 __u32 reserved2[12];
182 union {
183 __u32 reserved3[12];
184 struct _fpx_sw_bytes sw_reserved; /* represents the extended
185 * state information */
186 };
134}; 187};
135 188
136#ifdef __KERNEL__ 189#ifdef __KERNEL__
@@ -161,7 +214,15 @@ struct sigcontext {
161 unsigned long trapno; 214 unsigned long trapno;
162 unsigned long oldmask; 215 unsigned long oldmask;
163 unsigned long cr2; 216 unsigned long cr2;
164 struct _fpstate __user *fpstate; /* zero when no FPU context */ 217
218 /*
219 * fpstate is really (struct _fpstate *) or (struct _xstate *)
220 * depending on the FP_XSTATE_MAGIC1 encoded in the SW reserved
221 * bytes of (struct _fpstate) and FP_XSTATE_MAGIC2 present at the end
222 * of extended memory layout. See comments at the defintion of
223 * (struct _fpx_sw_bytes)
224 */
225 void __user *fpstate; /* zero when no FPU/extended context */
165 unsigned long reserved1[8]; 226 unsigned long reserved1[8];
166}; 227};
167#else /* __KERNEL__ */ 228#else /* __KERNEL__ */
@@ -202,4 +263,22 @@ struct sigcontext {
202 263
203#endif /* !__i386__ */ 264#endif /* !__i386__ */
204 265
266struct _xsave_hdr {
267 __u64 xstate_bv;
268 __u64 reserved1[2];
269 __u64 reserved2[5];
270};
271
272/*
273 * Extended state pointed by the fpstate pointer in the sigcontext.
274 * In addition to the fpstate, information encoded in the xstate_hdr
275 * indicates the presence of other extended state information
276 * supported by the processor and OS.
277 */
278struct _xstate {
279 struct _fpstate fpstate;
280 struct _xsave_hdr xstate_hdr;
281 /* new processor state extensions go here */
282};
283
205#endif /* ASM_X86__SIGCONTEXT_H */ 284#endif /* ASM_X86__SIGCONTEXT_H */
diff --git a/include/asm-x86/sigcontext32.h b/include/asm-x86/sigcontext32.h
index 4e2ec732dd01..8c347032c2f2 100644
--- a/include/asm-x86/sigcontext32.h
+++ b/include/asm-x86/sigcontext32.h
@@ -40,7 +40,11 @@ struct _fpstate_ia32 {
40 __u32 reserved; 40 __u32 reserved;
41 struct _fpxreg _fxsr_st[8]; 41 struct _fpxreg _fxsr_st[8];
42 struct _xmmreg _xmm[8]; /* It's actually 16 */ 42 struct _xmmreg _xmm[8]; /* It's actually 16 */
43 __u32 padding[56]; 43 __u32 padding[44];
44 union {
45 __u32 padding2[12];
46 struct _fpx_sw_bytes sw_reserved;
47 };
44}; 48};
45 49
46struct sigcontext_ia32 { 50struct sigcontext_ia32 {
diff --git a/include/asm-x86/thread_info.h b/include/asm-x86/thread_info.h
index 4db0066a3a35..3f4e52bb77f5 100644
--- a/include/asm-x86/thread_info.h
+++ b/include/asm-x86/thread_info.h
@@ -241,6 +241,7 @@ static inline struct thread_info *stack_thread_info(void)
241#define TS_POLLING 0x0004 /* true if in idle loop 241#define TS_POLLING 0x0004 /* true if in idle loop
242 and not sleeping */ 242 and not sleeping */
243#define TS_RESTORE_SIGMASK 0x0008 /* restore signal mask in do_signal() */ 243#define TS_RESTORE_SIGMASK 0x0008 /* restore signal mask in do_signal() */
244#define TS_XSAVE 0x0010 /* Use xsave/xrstor */
244 245
245#define tsk_is_polling(t) (task_thread_info(t)->status & TS_POLLING) 246#define tsk_is_polling(t) (task_thread_info(t)->status & TS_POLLING)
246 247
diff --git a/include/asm-x86/ucontext.h b/include/asm-x86/ucontext.h
index 9948dd328084..89eaa5456a7e 100644
--- a/include/asm-x86/ucontext.h
+++ b/include/asm-x86/ucontext.h
@@ -1,6 +1,12 @@
1#ifndef ASM_X86__UCONTEXT_H 1#ifndef ASM_X86__UCONTEXT_H
2#define ASM_X86__UCONTEXT_H 2#define ASM_X86__UCONTEXT_H
3 3
4#define UC_FP_XSTATE 0x1 /* indicates the presence of extended state
5 * information in the memory layout pointed
6 * by the fpstate pointer in the ucontext's
7 * sigcontext struct (uc_mcontext).
8 */
9
4struct ucontext { 10struct ucontext {
5 unsigned long uc_flags; 11 unsigned long uc_flags;
6 struct ucontext *uc_link; 12 struct ucontext *uc_link;
diff --git a/include/asm-x86/xcr.h b/include/asm-x86/xcr.h
new file mode 100644
index 000000000000..f2cba4e79a23
--- /dev/null
+++ b/include/asm-x86/xcr.h
@@ -0,0 +1,49 @@
1/* -*- linux-c -*- ------------------------------------------------------- *
2 *
3 * Copyright 2008 rPath, Inc. - All Rights Reserved
4 *
5 * This file is part of the Linux kernel, and is made available under
6 * the terms of the GNU General Public License version 2 or (at your
7 * option) any later version; incorporated herein by reference.
8 *
9 * ----------------------------------------------------------------------- */
10
11/*
12 * asm-x86/xcr.h
13 *
14 * Definitions for the eXtended Control Register instructions
15 */
16
17#ifndef _ASM_X86_XCR_H
18#define _ASM_X86_XCR_H
19
20#define XCR_XFEATURE_ENABLED_MASK 0x00000000
21
22#ifdef __KERNEL__
23# ifndef __ASSEMBLY__
24
25#include <linux/types.h>
26
27static inline u64 xgetbv(u32 index)
28{
29 u32 eax, edx;
30
31 asm volatile(".byte 0x0f,0x01,0xd0" /* xgetbv */
32 : "=a" (eax), "=d" (edx)
33 : "c" (index));
34 return eax + ((u64)edx << 32);
35}
36
37static inline void xsetbv(u32 index, u64 value)
38{
39 u32 eax = value;
40 u32 edx = value >> 32;
41
42 asm volatile(".byte 0x0f,0x01,0xd1" /* xsetbv */
43 : : "a" (eax), "d" (edx), "c" (index));
44}
45
46# endif /* __ASSEMBLY__ */
47#endif /* __KERNEL__ */
48
49#endif /* _ASM_X86_XCR_H */
diff --git a/include/asm-x86/xsave.h b/include/asm-x86/xsave.h
new file mode 100644
index 000000000000..08e9a1ac07a9
--- /dev/null
+++ b/include/asm-x86/xsave.h
@@ -0,0 +1,118 @@
1#ifndef __ASM_X86_XSAVE_H
2#define __ASM_X86_XSAVE_H
3
4#include <linux/types.h>
5#include <asm/processor.h>
6#include <asm/i387.h>
7
8#define XSTATE_FP 0x1
9#define XSTATE_SSE 0x2
10
11#define XSTATE_FPSSE (XSTATE_FP | XSTATE_SSE)
12
13#define FXSAVE_SIZE 512
14
15/*
16 * These are the features that the OS can handle currently.
17 */
18#define XCNTXT_MASK (XSTATE_FP | XSTATE_SSE)
19
20#ifdef CONFIG_X86_64
21#define REX_PREFIX "0x48, "
22#else
23#define REX_PREFIX
24#endif
25
26extern unsigned int xstate_size;
27extern u64 pcntxt_mask;
28extern struct xsave_struct *init_xstate_buf;
29
30extern void xsave_cntxt_init(void);
31extern void xsave_init(void);
32extern int init_fpu(struct task_struct *child);
33extern int check_for_xstate(struct i387_fxsave_struct __user *buf,
34 void __user *fpstate,
35 struct _fpx_sw_bytes *sw);
36
37static inline int xrstor_checking(struct xsave_struct *fx)
38{
39 int err;
40
41 asm volatile("1: .byte " REX_PREFIX "0x0f,0xae,0x2f\n\t"
42 "2:\n"
43 ".section .fixup,\"ax\"\n"
44 "3: movl $-1,%[err]\n"
45 " jmp 2b\n"
46 ".previous\n"
47 _ASM_EXTABLE(1b, 3b)
48 : [err] "=r" (err)
49 : "D" (fx), "m" (*fx), "a" (-1), "d" (-1), "0" (0)
50 : "memory");
51
52 return err;
53}
54
55static inline int xsave_user(struct xsave_struct __user *buf)
56{
57 int err;
58 __asm__ __volatile__("1: .byte " REX_PREFIX "0x0f,0xae,0x27\n"
59 "2:\n"
60 ".section .fixup,\"ax\"\n"
61 "3: movl $-1,%[err]\n"
62 " jmp 2b\n"
63 ".previous\n"
64 ".section __ex_table,\"a\"\n"
65 _ASM_ALIGN "\n"
66 _ASM_PTR "1b,3b\n"
67 ".previous"
68 : [err] "=r" (err)
69 : "D" (buf), "a" (-1), "d" (-1), "0" (0)
70 : "memory");
71 if (unlikely(err) && __clear_user(buf, xstate_size))
72 err = -EFAULT;
73 /* No need to clear here because the caller clears USED_MATH */
74 return err;
75}
76
77static inline int xrestore_user(struct xsave_struct __user *buf, u64 mask)
78{
79 int err;
80 struct xsave_struct *xstate = ((__force struct xsave_struct *)buf);
81 u32 lmask = mask;
82 u32 hmask = mask >> 32;
83
84 __asm__ __volatile__("1: .byte " REX_PREFIX "0x0f,0xae,0x2f\n"
85 "2:\n"
86 ".section .fixup,\"ax\"\n"
87 "3: movl $-1,%[err]\n"
88 " jmp 2b\n"
89 ".previous\n"
90 ".section __ex_table,\"a\"\n"
91 _ASM_ALIGN "\n"
92 _ASM_PTR "1b,3b\n"
93 ".previous"
94 : [err] "=r" (err)
95 : "D" (xstate), "a" (lmask), "d" (hmask), "0" (0)
96 : "memory"); /* memory required? */
97 return err;
98}
99
100static inline void xrstor_state(struct xsave_struct *fx, u64 mask)
101{
102 u32 lmask = mask;
103 u32 hmask = mask >> 32;
104
105 asm volatile(".byte " REX_PREFIX "0x0f,0xae,0x2f\n\t"
106 : : "D" (fx), "m" (*fx), "a" (lmask), "d" (hmask)
107 : "memory");
108}
109
110static inline void xsave(struct task_struct *tsk)
111{
112 /* This, however, we can work around by forcing the compiler to select
113 an addressing mode that doesn't require extended registers. */
114 __asm__ __volatile__(".byte " REX_PREFIX "0x0f,0xae,0x27"
115 : : "D" (&(tsk->thread.xstate->xsave)),
116 "a" (-1), "d"(-1) : "memory");
117}
118#endif
diff --git a/include/linux/ioport.h b/include/linux/ioport.h
index 350033e8f4e1..ee9bcc6f32b6 100644
--- a/include/linux/ioport.h
+++ b/include/linux/ioport.h
@@ -108,6 +108,9 @@ extern struct resource iomem_resource;
108 108
109extern int request_resource(struct resource *root, struct resource *new); 109extern int request_resource(struct resource *root, struct resource *new);
110extern int release_resource(struct resource *new); 110extern int release_resource(struct resource *new);
111extern void reserve_region_with_split(struct resource *root,
112 resource_size_t start, resource_size_t end,
113 const char *name);
111extern int insert_resource(struct resource *parent, struct resource *new); 114extern int insert_resource(struct resource *parent, struct resource *new);
112extern void insert_resource_expand_to_fit(struct resource *root, struct resource *new); 115extern void insert_resource_expand_to_fit(struct resource *root, struct resource *new);
113extern int allocate_resource(struct resource *root, struct resource *new, 116extern int allocate_resource(struct resource *root, struct resource *new,
diff --git a/include/linux/percpu.h b/include/linux/percpu.h
index fac3337547eb..9f2a3751873a 100644
--- a/include/linux/percpu.h
+++ b/include/linux/percpu.h
@@ -23,12 +23,19 @@
23 __attribute__((__section__(SHARED_ALIGNED_SECTION))) \ 23 __attribute__((__section__(SHARED_ALIGNED_SECTION))) \
24 PER_CPU_ATTRIBUTES __typeof__(type) per_cpu__##name \ 24 PER_CPU_ATTRIBUTES __typeof__(type) per_cpu__##name \
25 ____cacheline_aligned_in_smp 25 ____cacheline_aligned_in_smp
26
27#define DEFINE_PER_CPU_PAGE_ALIGNED(type, name) \
28 __attribute__((__section__(".data.percpu.page_aligned"))) \
29 PER_CPU_ATTRIBUTES __typeof__(type) per_cpu__##name
26#else 30#else
27#define DEFINE_PER_CPU(type, name) \ 31#define DEFINE_PER_CPU(type, name) \
28 PER_CPU_ATTRIBUTES __typeof__(type) per_cpu__##name 32 PER_CPU_ATTRIBUTES __typeof__(type) per_cpu__##name
29 33
30#define DEFINE_PER_CPU_SHARED_ALIGNED(type, name) \ 34#define DEFINE_PER_CPU_SHARED_ALIGNED(type, name) \
31 DEFINE_PER_CPU(type, name) 35 DEFINE_PER_CPU(type, name)
36
37#define DEFINE_PER_CPU_PAGE_ALIGNED(type, name) \
38 DEFINE_PER_CPU(type, name)
32#endif 39#endif
33 40
34#define EXPORT_PER_CPU_SYMBOL(var) EXPORT_SYMBOL(per_cpu__##var) 41#define EXPORT_PER_CPU_SYMBOL(var) EXPORT_SYMBOL(per_cpu__##var)
diff --git a/kernel/resource.c b/kernel/resource.c
index 03d796c1b2e9..414d6fc9131e 100644
--- a/kernel/resource.c
+++ b/kernel/resource.c
@@ -516,6 +516,74 @@ int adjust_resource(struct resource *res, resource_size_t start, resource_size_t
516 return result; 516 return result;
517} 517}
518 518
519static void __init __reserve_region_with_split(struct resource *root,
520 resource_size_t start, resource_size_t end,
521 const char *name)
522{
523 struct resource *parent = root;
524 struct resource *conflict;
525 struct resource *res = kzalloc(sizeof(*res), GFP_KERNEL);
526
527 if (!res)
528 return;
529
530 res->name = name;
531 res->start = start;
532 res->end = end;
533 res->flags = IORESOURCE_BUSY;
534
535 for (;;) {
536 conflict = __request_resource(parent, res);
537 if (!conflict)
538 break;
539 if (conflict != parent) {
540 parent = conflict;
541 if (!(conflict->flags & IORESOURCE_BUSY))
542 continue;
543 }
544
545 /* Uhhuh, that didn't work out.. */
546 kfree(res);
547 res = NULL;
548 break;
549 }
550
551 if (!res) {
552 printk(KERN_DEBUG " __reserve_region_with_split: (%s) [%llx, %llx], res: (%s) [%llx, %llx]\n",
553 conflict->name, conflict->start, conflict->end,
554 name, start, end);
555
556 /* failed, split and try again */
557
558 /* conflict coverred whole area */
559 if (conflict->start <= start && conflict->end >= end)
560 return;
561
562 if (conflict->start > start)
563 __reserve_region_with_split(root, start, conflict->start-1, name);
564 if (!(conflict->flags & IORESOURCE_BUSY)) {
565 resource_size_t common_start, common_end;
566
567 common_start = max(conflict->start, start);
568 common_end = min(conflict->end, end);
569 if (common_start < common_end)
570 __reserve_region_with_split(root, common_start, common_end, name);
571 }
572 if (conflict->end < end)
573 __reserve_region_with_split(root, conflict->end+1, end, name);
574 }
575
576}
577
578void reserve_region_with_split(struct resource *root,
579 resource_size_t start, resource_size_t end,
580 const char *name)
581{
582 write_lock(&resource_lock);
583 __reserve_region_with_split(root, start, end, name);
584 write_unlock(&resource_lock);
585}
586
519EXPORT_SYMBOL(adjust_resource); 587EXPORT_SYMBOL(adjust_resource);
520 588
521/** 589/**
generated by cgit v1.2.2 (git 2.25.0) at 2026-01-02 00:25:41 -0500