aboutsummaryrefslogtreecommitdiffstats
diff options
context:
space:
mode:
authorNagarathnam Muthusamy <nagarathnam.muthusamy@oracle.com>2017-09-21 11:05:31 -0400
committerDavid S. Miller <davem@davemloft.net>2017-11-15 00:21:03 -0500
commit9a08862a5d2e266ecea1865547463da2745fc687 (patch)
treec651ef6b7c2d7dc75e0ec73fcc6d2e2f15f72a9e
parent23198ddffb6cddb5d5824230af4dd4b46e4046a4 (diff)
vDSO for sparc
Following patch is based on work done by Nick Alcock on 64-bit vDSO for sparc in Oracle linux. I have extended it to include support for 32-bit vDSO for sparc on 64-bit kernel. vDSO for sparc is based on the X86 implementation. This patch provides vDSO support for both 64-bit and 32-bit programs on 64-bit kernel. vDSO will be disabled on 32-bit linux kernel on sparc. *) vclock_gettime.c contains all the vdso functions. Since data page is mapped before the vdso code page, the pointer to data page is got by subracting offset from an address in the vdso code page. The return address stored in %i7 is used for this purpose. *) During compilation, both 32-bit and 64-bit vdso images are compiled and are converted into raw bytes by vdso2c program to be ready for mapping into the process. 32-bit images are compiled only if CONFIG_COMPAT is enabled. vdso2c generates two files vdso-image-64.c and vdso-image-32.c which contains the respective vDSO image in C structure. *) During vdso initialization, required number of vdso pages are allocated and raw bytes are copied into the pages. *) During every exec, these pages are mapped into the process through arch_setup_additional_pages and the location of mapping is passed on to the process through aux vector AT_SYSINFO_EHDR which is used by glibc. *) A new update_vsyscall routine for sparc is added to keep the data page in vdso updated. *) As vDSO cannot contain dynamically relocatable references, a new version of cpu_relax is added for the use of vDSO. This change also requires a putback to glibc to use vDSO. For testing, programs planning to try vDSO can be compiled against the generated vdso(64/32).so in the source. Testing: ======== [root@localhost ~]# cat vdso_test.c int main() { struct timespec tv_start, tv_end; struct timeval tv_tmp; int i; int count = 1 * 1000 * 10000; long long diff; clock_gettime(0, &tv_start); for (i = 0; i < count; i++) gettimeofday(&tv_tmp, NULL); clock_gettime(0, &tv_end); diff = (long long)(tv_end.tv_sec - tv_start.tv_sec)*(1*1000*1000*1000); diff += (tv_end.tv_nsec - tv_start.tv_nsec); printf("Start sec: %d\n", tv_start.tv_sec); printf("End sec : %d\n", tv_end.tv_sec); printf("%d cycles in %lld ns = %f ns/cycle\n", count, diff, (double)diff / (double)count); return 0; } [root@localhost ~]# cc vdso_test.c -o t32_without_fix -m32 -lrt [root@localhost ~]# ./t32_without_fix Start sec: 1502396130 End sec : 1502396140 10000000 cycles in 9565148528 ns = 956.514853 ns/cycle [root@localhost ~]# cc vdso_test.c -o t32_with_fix -m32 ./vdso32.so.dbg [root@localhost ~]# ./t32_with_fix Start sec: 1502396168 End sec : 1502396169 10000000 cycles in 798141262 ns = 79.814126 ns/cycle [root@localhost ~]# cc vdso_test.c -o t64_without_fix -m64 -lrt [root@localhost ~]# ./t64_without_fix Start sec: 1502396208 End sec : 1502396218 10000000 cycles in 9846091800 ns = 984.609180 ns/cycle [root@localhost ~]# cc vdso_test.c -o t64_with_fix -m64 ./vdso64.so.dbg [root@localhost ~]# ./t64_with_fix Start sec: 1502396257 End sec : 1502396257 10000000 cycles in 380984048 ns = 38.098405 ns/cycle V1 to V2 Changes: ================= Added hot patching code to switch the read stick instruction to read tick instruction based on the hardware. V2 to V3 Changes: ================= Merged latest changes from sparc-next and moved the initialization of clocksource_tick.archdata.vclock_mode to time_init_early. Disabled queued spinlock and rwlock configuration when simulating 32-bit config to compile 32-bit VDSO. V3 to V4 Changes: ================= Hardcoded the page size as 8192 in linker script for both 64-bit and 32-bit binaries. Removed unused variables in vdso2c.h. Added -mv8plus flag to Makefile to prevent the generation of relocation entries for __lshrdi3 in 32-bit vdso binary. Signed-off-by: Nick Alcock <nick.alcock@oracle.com> Signed-off-by: Nagarathnam Muthusamy <nagarathnam.muthusamy@oracle.com> Reviewed-by: Shannon Nelson <shannon.nelson@oracle.com> Signed-off-by: David S. Miller <davem@davemloft.net>
Diffstat
-rw-r--r--arch/sparc/Kbuild1
-rw-r--r--arch/sparc/Kconfig2
-rw-r--r--arch/sparc/Makefile4
-rw-r--r--arch/sparc/include/asm/clocksource.h17
-rw-r--r--arch/sparc/include/asm/elf_64.h14
-rw-r--r--arch/sparc/include/asm/mmu_64.h1
-rw-r--r--arch/sparc/include/asm/processor_64.h8
-rw-r--r--arch/sparc/include/asm/vdso.h24
-rw-r--r--arch/sparc/include/asm/vvar.h74
-rw-r--r--arch/sparc/include/uapi/asm/auxvec.h4
-rw-r--r--arch/sparc/kernel/Makefile1
-rw-r--r--arch/sparc/kernel/time_64.c11
-rw-r--r--arch/sparc/kernel/vdso.c70
-rw-r--r--arch/sparc/vdso/.gitignore3
-rw-r--r--arch/sparc/vdso/Makefile149
-rw-r--r--arch/sparc/vdso/vclock_gettime.c264
-rw-r--r--arch/sparc/vdso/vdso-layout.lds.S104
-rw-r--r--arch/sparc/vdso/vdso-note.S12
-rw-r--r--arch/sparc/vdso/vdso.lds.S25
-rw-r--r--arch/sparc/vdso/vdso2c.c234
-rw-r--r--arch/sparc/vdso/vdso2c.h143
-rw-r--r--arch/sparc/vdso/vdso32/.gitignore1
-rw-r--r--arch/sparc/vdso/vdso32/vclock_gettime.c26
-rw-r--r--arch/sparc/vdso/vdso32/vdso-note.S12
-rw-r--r--arch/sparc/vdso/vdso32/vdso32.lds.S24
-rw-r--r--arch/sparc/vdso/vma.c268
26 files changed, 1494 insertions, 2 deletions
diff --git a/arch/sparc/Kbuild b/arch/sparc/Kbuild
index 675afa285ddb..b4d0f570cc00 100644
--- a/arch/sparc/Kbuild
+++ b/arch/sparc/Kbuild
@@ -7,3 +7,4 @@ obj-y += mm/
7obj-y += math-emu/ 7obj-y += math-emu/
8obj-y += net/ 8obj-y += net/
9obj-y += crypto/ 9obj-y += crypto/
10obj-$(CONFIG_SPARC64) += vdso/
diff --git a/arch/sparc/Kconfig b/arch/sparc/Kconfig
index 0be3828752e5..cd28b6cd95a2 100644
--- a/arch/sparc/Kconfig
+++ b/arch/sparc/Kconfig
@@ -85,6 +85,8 @@ config SPARC64
85 select HAVE_REGS_AND_STACK_ACCESS_API 85 select HAVE_REGS_AND_STACK_ACCESS_API
86 select ARCH_USE_QUEUED_RWLOCKS 86 select ARCH_USE_QUEUED_RWLOCKS
87 select ARCH_USE_QUEUED_SPINLOCKS 87 select ARCH_USE_QUEUED_SPINLOCKS
88 select GENERIC_TIME_VSYSCALL
89 select ARCH_CLOCKSOURCE_DATA
88 90
89config ARCH_DEFCONFIG 91config ARCH_DEFCONFIG
90 string 92 string
diff --git a/arch/sparc/Makefile b/arch/sparc/Makefile
index 8496a074bd0e..0395b0de6cb4 100644
--- a/arch/sparc/Makefile
+++ b/arch/sparc/Makefile
@@ -80,6 +80,10 @@ install:
80archclean: 80archclean:
81 $(Q)$(MAKE) $(clean)=$(boot) 81 $(Q)$(MAKE) $(clean)=$(boot)
82 82
83PHONY += vdso_install
84vdso_install:
85 $(Q)$(MAKE) $(build)=arch/sparc/vdso $@
86
83# This is the image used for packaging 87# This is the image used for packaging
84KBUILD_IMAGE := $(boot)/zImage 88KBUILD_IMAGE := $(boot)/zImage
85 89
diff --git a/arch/sparc/include/asm/clocksource.h b/arch/sparc/include/asm/clocksource.h
new file mode 100644
index 000000000000..d63ef224befe
--- /dev/null
+++ b/arch/sparc/include/asm/clocksource.h
@@ -0,0 +1,17 @@
1/*
2 * Copyright (c) 2017 Oracle and/or its affiliates. All rights reserved.
3 */
4
5#ifndef _ASM_SPARC_CLOCKSOURCE_H
6#define _ASM_SPARC_CLOCKSOURCE_H
7
8/* VDSO clocksources */
9#define VCLOCK_NONE 0 /* Nothing userspace can do. */
10#define VCLOCK_TICK 1 /* Use %tick. */
11#define VCLOCK_STICK 2 /* Use %stick. */
12
13struct arch_clocksource_data {
14 int vclock_mode;
15};
16
17#endif /* _ASM_SPARC_CLOCKSOURCE_H */
diff --git a/arch/sparc/include/asm/elf_64.h b/arch/sparc/include/asm/elf_64.h
index 3f2d403873bd..1aa35a334a22 100644
--- a/arch/sparc/include/asm/elf_64.h
+++ b/arch/sparc/include/asm/elf_64.h
@@ -210,4 +210,18 @@ do { if ((ex).e_ident[EI_CLASS] == ELFCLASS32) \
210 (current->personality & (~PER_MASK))); \ 210 (current->personality & (~PER_MASK))); \
211} while (0) 211} while (0)
212 212
213extern unsigned int vdso_enabled;
214
215#define ARCH_DLINFO \
216do { \
217 if (vdso_enabled) \
218 NEW_AUX_ENT(AT_SYSINFO_EHDR, \
219 (unsigned long)current->mm->context.vdso); \
220} while (0)
221
222struct linux_binprm;
223
224#define ARCH_HAS_SETUP_ADDITIONAL_PAGES 1
225extern int arch_setup_additional_pages(struct linux_binprm *bprm,
226 int uses_interp);
213#endif /* !(__ASM_SPARC64_ELF_H) */ 227#endif /* !(__ASM_SPARC64_ELF_H) */
diff --git a/arch/sparc/include/asm/mmu_64.h b/arch/sparc/include/asm/mmu_64.h
index 83b36a5371ff..4142132dcb6b 100644
--- a/arch/sparc/include/asm/mmu_64.h
+++ b/arch/sparc/include/asm/mmu_64.h
@@ -96,6 +96,7 @@ typedef struct {
96 unsigned long thp_pte_count; 96 unsigned long thp_pte_count;
97 struct tsb_config tsb_block[MM_NUM_TSBS]; 97 struct tsb_config tsb_block[MM_NUM_TSBS];
98 struct hv_tsb_descr tsb_descr[MM_NUM_TSBS]; 98 struct hv_tsb_descr tsb_descr[MM_NUM_TSBS];
99 void *vdso;
99} mm_context_t; 100} mm_context_t;
100 101
101#endif /* !__ASSEMBLY__ */ 102#endif /* !__ASSEMBLY__ */
diff --git a/arch/sparc/include/asm/processor_64.h b/arch/sparc/include/asm/processor_64.h
index f04dc5a43062..bcb3172abeda 100644
--- a/arch/sparc/include/asm/processor_64.h
+++ b/arch/sparc/include/asm/processor_64.h
@@ -199,6 +199,13 @@ unsigned long get_wchan(struct task_struct *task);
199 * To make a long story short, we are trying to yield the current cpu 199 * To make a long story short, we are trying to yield the current cpu
200 * strand during busy loops. 200 * strand during busy loops.
201 */ 201 */
202#ifdef BUILD_VDSO
203#define cpu_relax() asm volatile("\n99:\n\t" \
204 "rd %%ccr, %%g0\n\t" \
205 "rd %%ccr, %%g0\n\t" \
206 "rd %%ccr, %%g0\n\t" \
207 ::: "memory")
208#else /* ! BUILD_VDSO */
202#define cpu_relax() asm volatile("\n99:\n\t" \ 209#define cpu_relax() asm volatile("\n99:\n\t" \
203 "rd %%ccr, %%g0\n\t" \ 210 "rd %%ccr, %%g0\n\t" \
204 "rd %%ccr, %%g0\n\t" \ 211 "rd %%ccr, %%g0\n\t" \
@@ -210,6 +217,7 @@ unsigned long get_wchan(struct task_struct *task);
210 "nop\n\t" \ 217 "nop\n\t" \
211 ".previous" \ 218 ".previous" \
212 ::: "memory") 219 ::: "memory")
220#endif
213 221
214/* Prefetch support. This is tuned for UltraSPARC-III and later. 222/* Prefetch support. This is tuned for UltraSPARC-III and later.
215 * UltraSPARC-I will treat these as nops, and UltraSPARC-II has 223 * UltraSPARC-I will treat these as nops, and UltraSPARC-II has
diff --git a/arch/sparc/include/asm/vdso.h b/arch/sparc/include/asm/vdso.h
new file mode 100644
index 000000000000..93b628731a5e
--- /dev/null
+++ b/arch/sparc/include/asm/vdso.h
@@ -0,0 +1,24 @@
1/*
2 * Copyright (c) 2017 Oracle and/or its affiliates. All rights reserved.
3 */
4
5#ifndef _ASM_SPARC_VDSO_H
6#define _ASM_SPARC_VDSO_H
7
8struct vdso_image {
9 void *data;
10 unsigned long size; /* Always a multiple of PAGE_SIZE */
11 long sym_vvar_start; /* Negative offset to the vvar area */
12 long sym_vread_tick; /* Start of vread_tick section */
13 long sym_vread_tick_patch_start; /* Start of tick read */
14 long sym_vread_tick_patch_end; /* End of tick read */
15};
16
17#ifdef CONFIG_SPARC64
18extern const struct vdso_image vdso_image_64_builtin;
19#endif
20#ifdef CONFIG_COMPAT
21extern const struct vdso_image vdso_image_32_builtin;
22#endif
23
24#endif /* _ASM_SPARC_VDSO_H */
diff --git a/arch/sparc/include/asm/vvar.h b/arch/sparc/include/asm/vvar.h
new file mode 100644
index 000000000000..0289503d1cb0
--- /dev/null
+++ b/arch/sparc/include/asm/vvar.h
@@ -0,0 +1,74 @@
1/*
2 * Copyright (c) 2017 Oracle and/or its affiliates. All rights reserved.
3 */
4
5#ifndef _ASM_SPARC_VVAR_DATA_H
6#define _ASM_SPARC_VVAR_DATA_H
7
8#include <asm/clocksource.h>
9#include <linux/seqlock.h>
10#include <linux/time.h>
11#include <linux/types.h>
12
13struct vvar_data {
14 unsigned int seq;
15
16 int vclock_mode;
17 struct { /* extract of a clocksource struct */
18 u64 cycle_last;
19 u64 mask;
20 int mult;
21 int shift;
22 } clock;
23 /* open coded 'struct timespec' */
24 u64 wall_time_sec;
25 u64 wall_time_snsec;
26 u64 monotonic_time_snsec;
27 u64 monotonic_time_sec;
28 u64 monotonic_time_coarse_sec;
29 u64 monotonic_time_coarse_nsec;
30 u64 wall_time_coarse_sec;
31 u64 wall_time_coarse_nsec;
32
33 int tz_minuteswest;
34 int tz_dsttime;
35};
36
37extern struct vvar_data *vvar_data;
38extern int vdso_fix_stick;
39
40static inline unsigned int vvar_read_begin(const struct vvar_data *s)
41{
42 unsigned int ret;
43
44repeat:
45 ret = READ_ONCE(s->seq);
46 if (unlikely(ret & 1)) {
47 cpu_relax();
48 goto repeat;
49 }
50 smp_rmb(); /* Finish all reads before we return seq */
51 return ret;
52}
53
54static inline int vvar_read_retry(const struct vvar_data *s,
55 unsigned int start)
56{
57 smp_rmb(); /* Finish all reads before checking the value of seq */
58 return unlikely(s->seq != start);
59}
60
61static inline void vvar_write_begin(struct vvar_data *s)
62{
63 ++s->seq;
64 smp_wmb(); /* Makes sure that increment of seq is reflected */
65}
66
67static inline void vvar_write_end(struct vvar_data *s)
68{
69 smp_wmb(); /* Makes the value of seq current before we increment */
70 ++s->seq;
71}
72
73
74#endif /* _ASM_SPARC_VVAR_DATA_H */
diff --git a/arch/sparc/include/uapi/asm/auxvec.h b/arch/sparc/include/uapi/asm/auxvec.h
index ad6f360261f6..5f80a70cc901 100644
--- a/arch/sparc/include/uapi/asm/auxvec.h
+++ b/arch/sparc/include/uapi/asm/auxvec.h
@@ -1,4 +1,8 @@
1#ifndef __ASMSPARC_AUXVEC_H 1#ifndef __ASMSPARC_AUXVEC_H
2#define __ASMSPARC_AUXVEC_H 2#define __ASMSPARC_AUXVEC_H
3 3
4#define AT_SYSINFO_EHDR 33
5
6#define AT_VECTOR_SIZE_ARCH 1
7
4#endif /* !(__ASMSPARC_AUXVEC_H) */ 8#endif /* !(__ASMSPARC_AUXVEC_H) */
diff --git a/arch/sparc/kernel/Makefile b/arch/sparc/kernel/Makefile
index aac609889ee4..408071e675e8 100644
--- a/arch/sparc/kernel/Makefile
+++ b/arch/sparc/kernel/Makefile
@@ -42,6 +42,7 @@ obj-$(CONFIG_SPARC32) += systbls_32.o
42obj-y += time_$(BITS).o 42obj-y += time_$(BITS).o
43obj-$(CONFIG_SPARC32) += windows.o 43obj-$(CONFIG_SPARC32) += windows.o
44obj-y += cpu.o 44obj-y += cpu.o
45obj-$(CONFIG_SPARC64) += vdso.o
45obj-$(CONFIG_SPARC32) += devices.o 46obj-$(CONFIG_SPARC32) += devices.o
46obj-y += ptrace_$(BITS).o 47obj-y += ptrace_$(BITS).o
47obj-y += unaligned_$(BITS).o 48obj-y += unaligned_$(BITS).o
diff --git a/arch/sparc/kernel/time_64.c b/arch/sparc/kernel/time_64.c
index 5c8524ff09d9..79fdf576dab0 100644
--- a/arch/sparc/kernel/time_64.c
+++ b/arch/sparc/kernel/time_64.c
@@ -52,6 +52,8 @@
52 52
53DEFINE_SPINLOCK(rtc_lock); 53DEFINE_SPINLOCK(rtc_lock);
54 54
55unsigned int __read_mostly vdso_fix_stick;
56
55#ifdef CONFIG_SMP 57#ifdef CONFIG_SMP
56unsigned long profile_pc(struct pt_regs *regs) 58unsigned long profile_pc(struct pt_regs *regs)
57{ 59{
@@ -829,12 +831,17 @@ static void init_tick_ops(struct sparc64_tick_ops *ops)
829void __init time_init_early(void) 831void __init time_init_early(void)
830{ 832{
831 if (tlb_type == spitfire) { 833 if (tlb_type == spitfire) {
832 if (is_hummingbird()) 834 if (is_hummingbird()) {
833 init_tick_ops(&hbtick_operations); 835 init_tick_ops(&hbtick_operations);
834 else 836 clocksource_tick.archdata.vclock_mode = VCLOCK_NONE;
837 } else {
835 init_tick_ops(&tick_operations); 838 init_tick_ops(&tick_operations);
839 clocksource_tick.archdata.vclock_mode = VCLOCK_TICK;
840 vdso_fix_stick = 1;
841 }
836 } else { 842 } else {
837 init_tick_ops(&stick_operations); 843 init_tick_ops(&stick_operations);
844 clocksource_tick.archdata.vclock_mode = VCLOCK_STICK;
838 } 845 }
839} 846}
840 847
diff --git a/arch/sparc/kernel/vdso.c b/arch/sparc/kernel/vdso.c
new file mode 100644
index 000000000000..58880662b271
--- /dev/null
+++ b/arch/sparc/kernel/vdso.c
@@ -0,0 +1,70 @@
1/*
2 * Copyright (C) 2001 Andrea Arcangeli <andrea@suse.de> SuSE
3 * Copyright 2003 Andi Kleen, SuSE Labs.
4 *
5 * Thanks to hpa@transmeta.com for some useful hint.
6 * Special thanks to Ingo Molnar for his early experience with
7 * a different vsyscall implementation for Linux/IA32 and for the name.
8 */
9
10#include <linux/seqlock.h>
11#include <linux/time.h>
12#include <linux/timekeeper_internal.h>
13
14#include <asm/vvar.h>
15
16void update_vsyscall_tz(void)
17{
18 if (unlikely(vvar_data == NULL))
19 return;
20
21 vvar_data->tz_minuteswest = sys_tz.tz_minuteswest;
22 vvar_data->tz_dsttime = sys_tz.tz_dsttime;
23}
24
25void update_vsyscall(struct timekeeper *tk)
26{
27 struct vvar_data *vdata = vvar_data;
28
29 if (unlikely(vdata == NULL))
30 return;
31
32 vvar_write_begin(vdata);
33 vdata->vclock_mode = tk->tkr_mono.clock->archdata.vclock_mode;
34 vdata->clock.cycle_last = tk->tkr_mono.cycle_last;
35 vdata->clock.mask = tk->tkr_mono.mask;
36 vdata->clock.mult = tk->tkr_mono.mult;
37 vdata->clock.shift = tk->tkr_mono.shift;
38
39 vdata->wall_time_sec = tk->xtime_sec;
40 vdata->wall_time_snsec = tk->tkr_mono.xtime_nsec;
41
42 vdata->monotonic_time_sec = tk->xtime_sec +
43 tk->wall_to_monotonic.tv_sec;
44 vdata->monotonic_time_snsec = tk->tkr_mono.xtime_nsec +
45 (tk->wall_to_monotonic.tv_nsec <<
46 tk->tkr_mono.shift);
47
48 while (vdata->monotonic_time_snsec >=
49 (((u64)NSEC_PER_SEC) << tk->tkr_mono.shift)) {
50 vdata->monotonic_time_snsec -=
51 ((u64)NSEC_PER_SEC) << tk->tkr_mono.shift;
52 vdata->monotonic_time_sec++;
53 }
54
55 vdata->wall_time_coarse_sec = tk->xtime_sec;
56 vdata->wall_time_coarse_nsec =
57 (long)(tk->tkr_mono.xtime_nsec >> tk->tkr_mono.shift);
58
59 vdata->monotonic_time_coarse_sec =
60 vdata->wall_time_coarse_sec + tk->wall_to_monotonic.tv_sec;
61 vdata->monotonic_time_coarse_nsec =
62 vdata->wall_time_coarse_nsec + tk->wall_to_monotonic.tv_nsec;
63
64 while (vdata->monotonic_time_coarse_nsec >= NSEC_PER_SEC) {
65 vdata->monotonic_time_coarse_nsec -= NSEC_PER_SEC;
66 vdata->monotonic_time_coarse_sec++;
67 }
68
69 vvar_write_end(vdata);
70}
diff --git a/arch/sparc/vdso/.gitignore b/arch/sparc/vdso/.gitignore
new file mode 100644
index 000000000000..ef925b998222
--- /dev/null
+++ b/arch/sparc/vdso/.gitignore
@@ -0,0 +1,3 @@
1vdso.lds
2vdso-image-*.c
3vdso2c
diff --git a/arch/sparc/vdso/Makefile b/arch/sparc/vdso/Makefile
new file mode 100644
index 000000000000..a6615d8864f7
--- /dev/null
+++ b/arch/sparc/vdso/Makefile
@@ -0,0 +1,149 @@
1#
2# Building vDSO images for sparc.
3#
4
5KBUILD_CFLAGS += $(DISABLE_LTO)
6
7VDSO64-$(CONFIG_SPARC64) := y
8VDSOCOMPAT-$(CONFIG_COMPAT) := y
9
10# files to link into the vdso
11vobjs-y := vdso-note.o vclock_gettime.o
12
13# files to link into kernel
14obj-y += vma.o
15
16# vDSO images to build
17vdso_img-$(VDSO64-y) += 64
18vdso_img-$(VDSOCOMPAT-y) += 32
19
20vobjs := $(foreach F,$(vobjs-y),$(obj)/$F)
21
22$(obj)/vdso.o: $(obj)/vdso.so
23
24targets += vdso.lds $(vobjs-y)
25
26# Build the vDSO image C files and link them in.
27vdso_img_objs := $(vdso_img-y:%=vdso-image-%.o)
28vdso_img_cfiles := $(vdso_img-y:%=vdso-image-%.c)
29vdso_img_sodbg := $(vdso_img-y:%=vdso%.so.dbg)
30obj-y += $(vdso_img_objs)
31targets += $(vdso_img_cfiles)
32targets += $(vdso_img_sodbg)
33.SECONDARY: $(vdso_img-y:%=$(obj)/vdso-image-%.c) \
34 $(vdso_img-y:%=$(obj)/vdso%.so)
35
36export CPPFLAGS_vdso.lds += -P -C
37
38VDSO_LDFLAGS_vdso.lds = -m64 -Wl,-soname=linux-vdso.so.1 \
39 -Wl,--no-undefined \
40 -Wl,-z,max-page-size=8192 -Wl,-z,common-page-size=8192 \
41 $(DISABLE_LTO)
42
43$(obj)/vdso64.so.dbg: $(src)/vdso.lds $(vobjs) FORCE
44 $(call if_changed,vdso)
45
46HOST_EXTRACFLAGS += -I$(srctree)/tools/include
47hostprogs-y += vdso2c
48
49quiet_cmd_vdso2c = VDSO2C $@
50define cmd_vdso2c
51 $(obj)/vdso2c $< $(<:%.dbg=%) $@
52endef
53
54$(obj)/vdso-image-%.c: $(obj)/vdso%.so.dbg $(obj)/vdso%.so $(obj)/vdso2c FORCE
55 $(call if_changed,vdso2c)
56
57#
58# Don't omit frame pointers for ease of userspace debugging, but do
59# optimize sibling calls.
60#
61CFL := $(PROFILING) -mcmodel=medlow -fPIC -O2 -fasynchronous-unwind-tables \
62 -m64 -ffixed-g2 -ffixed-g3 -fcall-used-g4 -fcall-used-g5 -ffixed-g6 \
63 -ffixed-g7 $(filter -g%,$(KBUILD_CFLAGS)) \
64 $(call cc-option, -fno-stack-protector) -fno-omit-frame-pointer \
65 -foptimize-sibling-calls -DBUILD_VDSO
66
67$(vobjs): KBUILD_CFLAGS += $(CFL)
68
69#
70# vDSO code runs in userspace and -pg doesn't help with profiling anyway.
71#
72CFLAGS_REMOVE_vdso-note.o = -pg
73CFLAGS_REMOVE_vclock_gettime.o = -pg
74
75$(obj)/%.so: OBJCOPYFLAGS := -S
76$(obj)/%.so: $(obj)/%.so.dbg
77 $(call if_changed,objcopy)
78
79CPPFLAGS_vdso32.lds = $(CPPFLAGS_vdso.lds)
80VDSO_LDFLAGS_vdso32.lds = -m32 -Wl,-m,elf32_sparc,-soname=linux-gate.so.1
81
82#This makes sure the $(obj) subdirectory exists even though vdso32/
83#is not a kbuild sub-make subdirectory
84override obj-dirs = $(dir $(obj)) $(obj)/vdso32/
85
86targets += vdso32/vdso32.lds
87targets += vdso32/vdso-note.o
88targets += vdso32/vclock_gettime.o
89
90KBUILD_AFLAGS_32 := $(filter-out -m64,$(KBUILD_AFLAGS)) -DBUILD_VDSO
91$(obj)/vdso32.so.dbg: KBUILD_AFLAGS = $(KBUILD_AFLAGS_32)
92$(obj)/vdso32.so.dbg: asflags-$(CONFIG_SPARC64) += -m32
93
94KBUILD_CFLAGS_32 := $(filter-out -m64,$(KBUILD_CFLAGS))
95KBUILD_CFLAGS_32 := $(filter-out -mcmodel=medlow,$(KBUILD_CFLAGS_32))
96KBUILD_CFLAGS_32 := $(filter-out -fno-pic,$(KBUILD_CFLAGS_32))
97KBUILD_CFLAGS_32 := $(filter-out $(GCC_PLUGINS_CFLAGS),$(KBUILD_CFLAGS_32))
98KBUILD_CFLAGS_32 += -m32 -msoft-float -fpic -mno-app-regs -ffixed-g7
99KBUILD_CFLAGS_32 += $(call cc-option, -fno-stack-protector)
100KBUILD_CFLAGS_32 += $(call cc-option, -foptimize-sibling-calls)
101KBUILD_CFLAGS_32 += -fno-omit-frame-pointer
102KBUILD_CFLAGS_32 += -DDISABLE_BRANCH_PROFILING
103KBUILD_CFLAGS_32 += -mv8plus
104$(obj)/vdso32.so.dbg: KBUILD_CFLAGS = $(KBUILD_CFLAGS_32)
105
106$(obj)/vdso32.so.dbg: FORCE \
107 $(obj)/vdso32/vdso32.lds \
108 $(obj)/vdso32/vclock_gettime.o \
109 $(obj)/vdso32/vdso-note.o
110 $(call if_changed,vdso)
111
112#
113# The DSO images are built using a special linker script.
114#
115quiet_cmd_vdso = VDSO $@
116 cmd_vdso = $(CC) -nostdlib -o $@ \
117 $(VDSO_LDFLAGS) $(VDSO_LDFLAGS_$(filter %.lds,$(^F))) \
118 -Wl,-T,$(filter %.lds,$^) $(filter %.o,$^)
119
120VDSO_LDFLAGS = -fPIC -shared $(call cc-ldoption, -Wl$(comma)--hash-style=sysv) \
121 $(call cc-ldoption, -Wl$(comma)--build-id) -Wl,-Bsymbolic
122GCOV_PROFILE := n
123
124#
125# Install the unstripped copies of vdso*.so. If our toolchain supports
126# build-id, install .build-id links as well.
127#
128quiet_cmd_vdso_install = INSTALL $(@:install_%=%)
129define cmd_vdso_install
130 cp $< "$(MODLIB)/vdso/$(@:install_%=%)"; \
131 if readelf -n $< |grep -q 'Build ID'; then \
132 buildid=`readelf -n $< |grep 'Build ID' |sed -e 's/^.*Build ID: \(.*\)$$/\1/'`; \
133 first=`echo $$buildid | cut -b-2`; \
134 last=`echo $$buildid | cut -b3-`; \
135 mkdir -p "$(MODLIB)/vdso/.build-id/$$first"; \
136 ln -sf "../../$(@:install_%=%)" "$(MODLIB)/vdso/.build-id/$$first/$$last.debug"; \
137 fi
138endef
139
140vdso_img_insttargets := $(vdso_img_sodbg:%.dbg=install_%)
141
142$(MODLIB)/vdso: FORCE
143 @mkdir -p $(MODLIB)/vdso
144
145$(vdso_img_insttargets): install_%: $(obj)/%.dbg $(MODLIB)/vdso FORCE
146 $(call cmd,vdso_install)
147
148PHONY += vdso_install $(vdso_img_insttargets)
149vdso_install: $(vdso_img_insttargets) FORCE
diff --git a/arch/sparc/vdso/vclock_gettime.c b/arch/sparc/vdso/vclock_gettime.c
new file mode 100644
index 000000000000..3feb3d960ca5
--- /dev/null
+++ b/arch/sparc/vdso/vclock_gettime.c
@@ -0,0 +1,264 @@
1/*
2 * Copyright 2006 Andi Kleen, SUSE Labs.
3 * Subject to the GNU Public License, v.2
4 *
5 * Fast user context implementation of clock_gettime, gettimeofday, and time.
6 *
7 * The code should have no internal unresolved relocations.
8 * Check with readelf after changing.
9 * Also alternative() doesn't work.
10 */
11/*
12 * Copyright (c) 2017 Oracle and/or its affiliates. All rights reserved.
13 */
14
15/* Disable profiling for userspace code: */
16#ifndef DISABLE_BRANCH_PROFILING
17#define DISABLE_BRANCH_PROFILING
18#endif
19
20#include <linux/kernel.h>
21#include <linux/time.h>
22#include <linux/string.h>
23#include <asm/io.h>
24#include <asm/unistd.h>
25#include <asm/timex.h>
26#include <asm/clocksource.h>
27#include <asm/vvar.h>
28
29#undef TICK_PRIV_BIT
30#ifdef CONFIG_SPARC64
31#define TICK_PRIV_BIT (1UL << 63)
32#else
33#define TICK_PRIV_BIT (1ULL << 63)
34#endif
35
36#define SYSCALL_STRING \
37 "ta 0x6d;" \
38 "sub %%g0, %%o0, %%o0;" \
39
40#define SYSCALL_CLOBBERS \
41 "f0", "f1", "f2", "f3", "f4", "f5", "f6", "f7", \
42 "f8", "f9", "f10", "f11", "f12", "f13", "f14", "f15", \
43 "f16", "f17", "f18", "f19", "f20", "f21", "f22", "f23", \
44 "f24", "f25", "f26", "f27", "f28", "f29", "f30", "f31", \
45 "f32", "f34", "f36", "f38", "f40", "f42", "f44", "f46", \
46 "f48", "f50", "f52", "f54", "f56", "f58", "f60", "f62", \
47 "cc", "memory"
48
49/*
50 * Compute the vvar page's address in the process address space, and return it
51 * as a pointer to the vvar_data.
52 */
53static notrace noinline struct vvar_data *
54get_vvar_data(void)
55{
56 unsigned long ret;
57
58 /*
59 * vdso data page is the first vDSO page so grab the return address
60 * and move up a page to get to the data page.
61 */
62 ret = (unsigned long)__builtin_return_address(0);
63 ret &= ~(8192 - 1);
64 ret -= 8192;
65
66 return (struct vvar_data *) ret;
67}
68
69static notrace long
70vdso_fallback_gettime(long clock, struct timespec *ts)
71{
72 register long num __asm__("g1") = __NR_clock_gettime;
73 register long o0 __asm__("o0") = clock;
74 register long o1 __asm__("o1") = (long) ts;
75
76 __asm__ __volatile__(SYSCALL_STRING : "=r" (o0) : "r" (num),
77 "0" (o0), "r" (o1) : SYSCALL_CLOBBERS);
78 return o0;
79}
80
81static notrace __always_inline long
82vdso_fallback_gettimeofday(struct timeval *tv, struct timezone *tz)
83{
84 register long num __asm__("g1") = __NR_gettimeofday;
85 register long o0 __asm__("o0") = (long) tv;
86 register long o1 __asm__("o1") = (long) tz;
87
88 __asm__ __volatile__(SYSCALL_STRING : "=r" (o0) : "r" (num),
89 "0" (o0), "r" (o1) : SYSCALL_CLOBBERS);
90 return o0;
91}
92
93#ifdef CONFIG_SPARC64
94static notrace noinline u64
95vread_tick(void) {
96 u64 ret;
97
98 __asm__ __volatile__("rd %%asr24, %0 \n"
99 ".section .vread_tick_patch, \"ax\" \n"
100 "rd %%tick, %0 \n"
101 ".previous \n"
102 : "=&r" (ret));
103 return ret & ~TICK_PRIV_BIT;
104}
105#else
106static notrace noinline u64
107vread_tick(void)
108{
109 unsigned int lo, hi;
110
111 __asm__ __volatile__("rd %%asr24, %%g1\n\t"
112 "srlx %%g1, 32, %1\n\t"
113 "srl %%g1, 0, %0\n"
114 ".section .vread_tick_patch, \"ax\" \n"
115 "rd %%tick, %%g1\n"
116 ".previous \n"
117 : "=&r" (lo), "=&r" (hi)
118 :
119 : "g1");
120 return lo | ((u64)hi << 32);
121}
122#endif
123
124static notrace inline u64
125vgetsns(struct vvar_data *vvar)
126{
127 u64 v;
128 u64 cycles;
129
130 cycles = vread_tick();
131 v = (cycles - vvar->clock.cycle_last) & vvar->clock.mask;
132 return v * vvar->clock.mult;
133}
134
135static notrace noinline int
136do_realtime(struct vvar_data *vvar, struct timespec *ts)
137{
138 unsigned long seq;
139 u64 ns;
140
141 ts->tv_nsec = 0;
142 do {
143 seq = vvar_read_begin(vvar);
144 ts->tv_sec = vvar->wall_time_sec;
145 ns = vvar->wall_time_snsec;
146 ns += vgetsns(vvar);
147 ns >>= vvar->clock.shift;
148 } while (unlikely(vvar_read_retry(vvar, seq)));
149
150 timespec_add_ns(ts, ns);
151
152 return 0;
153}
154
155static notrace noinline int
156do_monotonic(struct vvar_data *vvar, struct timespec *ts)
157{
158 unsigned long seq;
159 u64 ns;
160
161 ts->tv_nsec = 0;
162 do {
163 seq = vvar_read_begin(vvar);
164 ts->tv_sec = vvar->monotonic_time_sec;
165 ns = vvar->monotonic_time_snsec;
166 ns += vgetsns(vvar);
167 ns >>= vvar->clock.shift;
168 } while (unlikely(vvar_read_retry(vvar, seq)));
169
170 timespec_add_ns(ts, ns);
171
172 return 0;
173}
174
175static notrace noinline int
176do_realtime_coarse(struct vvar_data *vvar, struct timespec *ts)
177{
178 unsigned long seq;
179
180 do {
181 seq = vvar_read_begin(vvar);
182 ts->tv_sec = vvar->wall_time_coarse_sec;
183 ts->tv_nsec = vvar->wall_time_coarse_nsec;
184 } while (unlikely(vvar_read_retry(vvar, seq)));
185 return 0;
186}
187
188static notrace noinline int
189do_monotonic_coarse(struct vvar_data *vvar, struct timespec *ts)
190{
191 unsigned long seq;
192
193 do {
194 seq = vvar_read_begin(vvar);
195 ts->tv_sec = vvar->monotonic_time_coarse_sec;
196 ts->tv_nsec = vvar->monotonic_time_coarse_nsec;
197 } while (unlikely(vvar_read_retry(vvar, seq)));
198
199 return 0;
200}
201
202notrace int
203__vdso_clock_gettime(clockid_t clock, struct timespec *ts)
204{
205 struct vvar_data *vvd = get_vvar_data();
206
207 switch (clock) {
208 case CLOCK_REALTIME:
209 if (unlikely(vvd->vclock_mode == VCLOCK_NONE))
210 break;
211 return do_realtime(vvd, ts);
212 case CLOCK_MONOTONIC:
213 if (unlikely(vvd->vclock_mode == VCLOCK_NONE))
214 break;
215 return do_monotonic(vvd, ts);
216 case CLOCK_REALTIME_COARSE:
217 return do_realtime_coarse(vvd, ts);
218 case CLOCK_MONOTONIC_COARSE:
219 return do_monotonic_coarse(vvd, ts);
220 }
221 /*
222 * Unknown clock ID ? Fall back to the syscall.
223 */
224 return vdso_fallback_gettime(clock, ts);
225}
226int
227clock_gettime(clockid_t, struct timespec *)
228 __attribute__((weak, alias("__vdso_clock_gettime")));
229
230notrace int
231__vdso_gettimeofday(struct timeval *tv, struct timezone *tz)
232{
233 struct vvar_data *vvd = get_vvar_data();
234
235 if (likely(vvd->vclock_mode != VCLOCK_NONE)) {
236 if (likely(tv != NULL)) {
237 union tstv_t {
238 struct timespec ts;
239 struct timeval tv;
240 } *tstv = (union tstv_t *) tv;
241 do_realtime(vvd, &tstv->ts);
242 /*
243 * Assign before dividing to ensure that the division is
244 * done in the type of tv_usec, not tv_nsec.
245 *
246 * There cannot be > 1 billion usec in a second:
247 * do_realtime() has already distributed such overflow
248 * into tv_sec. So we can assign it to an int safely.
249 */
250 tstv->tv.tv_usec = tstv->ts.tv_nsec;
251 tstv->tv.tv_usec /= 1000;
252 }
253 if (unlikely(tz != NULL)) {
254 /* Avoid memcpy. Some old compilers fail to inline it */
255 tz->tz_minuteswest = vvd->tz_minuteswest;
256 tz->tz_dsttime = vvd->tz_dsttime;
257 }
258 return 0;
259 }
260 return vdso_fallback_gettimeofday(tv, tz);
261}
262int
263gettimeofday(struct timeval *, struct timezone *)
264 __attribute__((weak, alias("__vdso_gettimeofday")));
diff --git a/arch/sparc/vdso/vdso-layout.lds.S b/arch/sparc/vdso/vdso-layout.lds.S
new file mode 100644
index 000000000000..f2c83abaca12
--- /dev/null
+++ b/arch/sparc/vdso/vdso-layout.lds.S
@@ -0,0 +1,104 @@
1/*
2 * Linker script for vDSO. This is an ELF shared object prelinked to
3 * its virtual address, and with only one read-only segment.
4 * This script controls its layout.
5 */
6
7#if defined(BUILD_VDSO64)
8# define SHDR_SIZE 64
9#elif defined(BUILD_VDSO32)
10# define SHDR_SIZE 40
11#else
12# error unknown VDSO target
13#endif
14
15#define NUM_FAKE_SHDRS 7
16
17SECTIONS
18{
19 /*
20 * User/kernel shared data is before the vDSO. This may be a little
21 * uglier than putting it after the vDSO, but it avoids issues with
22 * non-allocatable things that dangle past the end of the PT_LOAD
23 * segment. Page size is 8192 for both 64-bit and 32-bit vdso binaries
24 */
25
26 vvar_start = . -8192;
27 vvar_data = vvar_start;
28
29 . = SIZEOF_HEADERS;
30
31 .hash : { *(.hash) } :text
32 .gnu.hash : { *(.gnu.hash) }
33 .dynsym : { *(.dynsym) }
34 .dynstr : { *(.dynstr) }
35 .gnu.version : { *(.gnu.version) }
36 .gnu.version_d : { *(.gnu.version_d) }
37 .gnu.version_r : { *(.gnu.version_r) }
38
39 .dynamic : { *(.dynamic) } :text :dynamic
40
41 .rodata : {
42 *(.rodata*)
43 *(.data*)
44 *(.sdata*)
45 *(.got.plt) *(.got)
46 *(.gnu.linkonce.d.*)
47 *(.bss*)
48 *(.dynbss*)
49 *(.gnu.linkonce.b.*)
50
51 /*
52 * Ideally this would live in a C file: kept in here for
53 * compatibility with x86-64.
54 */
55 VDSO_FAKE_SECTION_TABLE_START = .;
56 . = . + NUM_FAKE_SHDRS * SHDR_SIZE;
57 VDSO_FAKE_SECTION_TABLE_END = .;
58 } :text
59
60 .fake_shstrtab : { *(.fake_shstrtab) } :text
61
62
63 .note : { *(.note.*) } :text :note
64
65 .eh_frame_hdr : { *(.eh_frame_hdr) } :text :eh_frame_hdr
66 .eh_frame : { KEEP (*(.eh_frame)) } :text
67
68
69 /*
70 * Text is well-separated from actual data: there's plenty of
71 * stuff that isn't used at runtime in between.
72 */
73
74 .text : { *(.text*) } :text =0x90909090,
75
76 .vread_tick_patch : {
77 vread_tick_patch_start = .;
78 *(.vread_tick_patch)
79 vread_tick_patch_end = .;
80 }
81
82 /DISCARD/ : {
83 *(.discard)
84 *(.discard.*)
85 *(__bug_table)
86 }
87}
88
89/*
90 * Very old versions of ld do not recognize this name token; use the constant.
91 */
92#define PT_GNU_EH_FRAME 0x6474e550
93
94/*
95 * We must supply the ELF program headers explicitly to get just one
96 * PT_LOAD segment, and set the flags explicitly to make segments read-only.
97 */
98PHDRS
99{
100 text PT_LOAD FLAGS(5) FILEHDR PHDRS; /* PF_R|PF_X */
101 dynamic PT_DYNAMIC FLAGS(4); /* PF_R */
102 note PT_NOTE FLAGS(4); /* PF_R */
103 eh_frame_hdr PT_GNU_EH_FRAME;
104}
diff --git a/arch/sparc/vdso/vdso-note.S b/arch/sparc/vdso/vdso-note.S
new file mode 100644
index 000000000000..79a071e4357e
--- /dev/null
+++ b/arch/sparc/vdso/vdso-note.S
@@ -0,0 +1,12 @@
1/*
2 * This supplies .note.* sections to go into the PT_NOTE inside the vDSO text.
3 * Here we can supply some information useful to userland.
4 */
5
6#include <linux/uts.h>
7#include <linux/version.h>
8#include <linux/elfnote.h>
9
10ELFNOTE_START(Linux, 0, "a")
11 .long LINUX_VERSION_CODE
12ELFNOTE_END
diff --git a/arch/sparc/vdso/vdso.lds.S b/arch/sparc/vdso/vdso.lds.S
new file mode 100644
index 000000000000..f3caa29a331c
--- /dev/null
+++ b/arch/sparc/vdso/vdso.lds.S
@@ -0,0 +1,25 @@
1/*
2 * Linker script for 64-bit vDSO.
3 * We #include the file to define the layout details.
4 *
5 * This file defines the version script giving the user-exported symbols in
6 * the DSO.
7 */
8
9#define BUILD_VDSO64
10
11#include "vdso-layout.lds.S"
12
13/*
14 * This controls what userland symbols we export from the vDSO.
15 */
16VERSION {
17 LINUX_2.6 {
18 global:
19 clock_gettime;
20 __vdso_clock_gettime;
21 gettimeofday;
22 __vdso_gettimeofday;
23 local: *;
24 };
25}
diff --git a/arch/sparc/vdso/vdso2c.c b/arch/sparc/vdso/vdso2c.c
new file mode 100644
index 000000000000..9f5b1cd6d51d
--- /dev/null
+++ b/arch/sparc/vdso/vdso2c.c
@@ -0,0 +1,234 @@
1/*
2 * vdso2c - A vdso image preparation tool
3 * Copyright (c) 2014 Andy Lutomirski and others
4 * Licensed under the GPL v2
5 *
6 * vdso2c requires stripped and unstripped input. It would be trivial
7 * to fully strip the input in here, but, for reasons described below,
8 * we need to write a section table. Doing this is more or less
9 * equivalent to dropping all non-allocatable sections, but it's
10 * easier to let objcopy handle that instead of doing it ourselves.
11 * If we ever need to do something fancier than what objcopy provides,
12 * it would be straightforward to add here.
13 *
14 * We keep a section table for a few reasons:
15 *
16 * Binutils has issues debugging the vDSO: it reads the section table to
17 * find SHT_NOTE; it won't look at PT_NOTE for the in-memory vDSO, which
18 * would break build-id if we removed the section table. Binutils
19 * also requires that shstrndx != 0. See:
20 * https://sourceware.org/bugzilla/show_bug.cgi?id=17064
21 *
22 * elfutils might not look for PT_NOTE if there is a section table at
23 * all. I don't know whether this matters for any practical purpose.
24 *
25 * For simplicity, rather than hacking up a partial section table, we
26 * just write a mostly complete one. We omit non-dynamic symbols,
27 * though, since they're rather large.
28 *
29 * Once binutils gets fixed, we might be able to drop this for all but
30 * the 64-bit vdso, since build-id only works in kernel RPMs, and
31 * systems that update to new enough kernel RPMs will likely update
32 * binutils in sync. build-id has never worked for home-built kernel
33 * RPMs without manual symlinking, and I suspect that no one ever does
34 * that.
35 */
36
37/*
38 * Copyright (c) 2017 Oracle and/or its affiliates. All rights reserved.
39 */
40
41#include <inttypes.h>
42#include <stdint.h>
43#include <unistd.h>
44#include <stdarg.h>
45#include <stdlib.h>
46#include <stdio.h>
47#include <string.h>
48#include <fcntl.h>
49#include <err.h>
50
51#include <sys/mman.h>
52#include <sys/types.h>
53#include <tools/be_byteshift.h>
54
55#include <linux/elf.h>
56#include <linux/types.h>
57#include <linux/kernel.h>
58
59const char *outfilename;
60
61/* Symbols that we need in vdso2c. */
62enum {
63 sym_vvar_start,
64 sym_VDSO_FAKE_SECTION_TABLE_START,
65 sym_VDSO_FAKE_SECTION_TABLE_END,
66 sym_vread_tick,
67 sym_vread_tick_patch_start,
68 sym_vread_tick_patch_end
69};
70
71struct vdso_sym {
72 const char *name;
73 int export;
74};
75
76struct vdso_sym required_syms[] = {
77 [sym_vvar_start] = {"vvar_start", 1},
78 [sym_VDSO_FAKE_SECTION_TABLE_START] = {
79 "VDSO_FAKE_SECTION_TABLE_START", 0
80 },
81 [sym_VDSO_FAKE_SECTION_TABLE_END] = {
82 "VDSO_FAKE_SECTION_TABLE_END", 0
83 },
84 [sym_vread_tick] = {"vread_tick", 1},
85 [sym_vread_tick_patch_start] = {"vread_tick_patch_start", 1},
86 [sym_vread_tick_patch_end] = {"vread_tick_patch_end", 1}
87};
88
89__attribute__((format(printf, 1, 2))) __attribute__((noreturn))
90static void fail(const char *format, ...)
91{
92 va_list ap;
93
94 va_start(ap, format);
95 fprintf(stderr, "Error: ");
96 vfprintf(stderr, format, ap);
97 if (outfilename)
98 unlink(outfilename);
99 exit(1);
100 va_end(ap);
101}
102
103/*
104 * Evil macros for big-endian reads and writes
105 */
106#define GBE(x, bits, ifnot) \
107 __builtin_choose_expr( \
108 (sizeof(*(x)) == bits/8), \
109 (__typeof__(*(x)))get_unaligned_be##bits(x), ifnot)
110
111#define LAST_GBE(x) \
112 __builtin_choose_expr(sizeof(*(x)) == 1, *(x), (void)(0))
113
114#define GET_BE(x) \
115 GBE(x, 64, GBE(x, 32, GBE(x, 16, LAST_GBE(x))))
116
117#define PBE(x, val, bits, ifnot) \
118 __builtin_choose_expr( \
119 (sizeof(*(x)) == bits/8), \
120 put_unaligned_be##bits((val), (x)), ifnot)
121
122#define LAST_PBE(x, val) \
123 __builtin_choose_expr(sizeof(*(x)) == 1, *(x) = (val), (void)(0))
124
125#define PUT_BE(x, val) \
126 PBE(x, val, 64, PBE(x, val, 32, PBE(x, val, 16, LAST_PBE(x, val))))
127
128#define NSYMS ARRAY_SIZE(required_syms)
129
130#define BITSFUNC3(name, bits, suffix) name##bits##suffix
131#define BITSFUNC2(name, bits, suffix) BITSFUNC3(name, bits, suffix)
132#define BITSFUNC(name) BITSFUNC2(name, ELF_BITS, )
133
134#define INT_BITS BITSFUNC2(int, ELF_BITS, _t)
135
136#define ELF_BITS_XFORM2(bits, x) Elf##bits##_##x
137#define ELF_BITS_XFORM(bits, x) ELF_BITS_XFORM2(bits, x)
138#define ELF(x) ELF_BITS_XFORM(ELF_BITS, x)
139
140#define ELF_BITS 64
141#include "vdso2c.h"
142#undef ELF_BITS
143
144#define ELF_BITS 32
145#include "vdso2c.h"
146#undef ELF_BITS
147
148static void go(void *raw_addr, size_t raw_len,
149 void *stripped_addr, size_t stripped_len,
150 FILE *outfile, const char *name)
151{
152 Elf64_Ehdr *hdr = (Elf64_Ehdr *)raw_addr;
153
154 if (hdr->e_ident[EI_CLASS] == ELFCLASS64) {
155 go64(raw_addr, raw_len, stripped_addr, stripped_len,
156 outfile, name);
157 } else if (hdr->e_ident[EI_CLASS] == ELFCLASS32) {
158 go32(raw_addr, raw_len, stripped_addr, stripped_len,
159 outfile, name);
160 } else {
161 fail("unknown ELF class\n");
162 }
163}
164
165static void map_input(const char *name, void **addr, size_t *len, int prot)
166{
167 off_t tmp_len;
168
169 int fd = open(name, O_RDONLY);
170
171 if (fd == -1)
172 err(1, "%s", name);
173
174 tmp_len = lseek(fd, 0, SEEK_END);
175 if (tmp_len == (off_t)-1)
176 err(1, "lseek");
177 *len = (size_t)tmp_len;
178
179 *addr = mmap(NULL, tmp_len, prot, MAP_PRIVATE, fd, 0);
180 if (*addr == MAP_FAILED)
181 err(1, "mmap");
182
183 close(fd);
184}
185
186int main(int argc, char **argv)
187{
188 size_t raw_len, stripped_len;
189 void *raw_addr, *stripped_addr;
190 FILE *outfile;
191 char *name, *tmp;
192 int namelen;
193
194 if (argc != 4) {
195 printf("Usage: vdso2c RAW_INPUT STRIPPED_INPUT OUTPUT\n");
196 return 1;
197 }
198
199 /*
200 * Figure out the struct name. If we're writing to a .so file,
201 * generate raw output insted.
202 */
203 name = strdup(argv[3]);
204 namelen = strlen(name);
205 if (namelen >= 3 && !strcmp(name + namelen - 3, ".so")) {
206 name = NULL;
207 } else {
208 tmp = strrchr(name, '/');
209 if (tmp)
210 name = tmp + 1;
211 tmp = strchr(name, '.');
212 if (tmp)
213 *tmp = '\0';
214 for (tmp = name; *tmp; tmp++)
215 if (*tmp == '-')
216 *tmp = '_';
217 }
218
219 map_input(argv[1], &raw_addr, &raw_len, PROT_READ);
220 map_input(argv[2], &stripped_addr, &stripped_len, PROT_READ);
221
222 outfilename = argv[3];
223 outfile = fopen(outfilename, "w");
224 if (!outfile)
225 err(1, "%s", argv[2]);
226
227 go(raw_addr, raw_len, stripped_addr, stripped_len, outfile, name);
228
229 munmap(raw_addr, raw_len);
230 munmap(stripped_addr, stripped_len);
231 fclose(outfile);
232
233 return 0;
234}
diff --git a/arch/sparc/vdso/vdso2c.h b/arch/sparc/vdso/vdso2c.h
new file mode 100644
index 000000000000..808decb0f7be
--- /dev/null
+++ b/arch/sparc/vdso/vdso2c.h
@@ -0,0 +1,143 @@
1/*
2 * Copyright (c) 2017 Oracle and/or its affiliates. All rights reserved.
3 */
4
5/*
6 * This file is included up to twice from vdso2c.c. It generates code for
7 * 32-bit and 64-bit vDSOs. We will eventually need both for 64-bit builds,
8 * since 32-bit vDSOs will then be built for 32-bit userspace.
9 */
10
11static void BITSFUNC(go)(void *raw_addr, size_t raw_len,
12 void *stripped_addr, size_t stripped_len,
13 FILE *outfile, const char *name)
14{
15 int found_load = 0;
16 unsigned long load_size = -1; /* Work around bogus warning */
17 unsigned long mapping_size;
18 int i;
19 unsigned long j;
20
21 ELF(Shdr) *symtab_hdr = NULL, *strtab_hdr;
22 ELF(Ehdr) *hdr = (ELF(Ehdr) *)raw_addr;
23 ELF(Dyn) *dyn = 0, *dyn_end = 0;
24 INT_BITS syms[NSYMS] = {};
25
26 ELF(Phdr) *pt = (ELF(Phdr) *)(raw_addr + GET_BE(&hdr->e_phoff));
27
28 /* Walk the segment table. */
29 for (i = 0; i < GET_BE(&hdr->e_phnum); i++) {
30 if (GET_BE(&pt[i].p_type) == PT_LOAD) {
31 if (found_load)
32 fail("multiple PT_LOAD segs\n");
33
34 if (GET_BE(&pt[i].p_offset) != 0 ||
35 GET_BE(&pt[i].p_vaddr) != 0)
36 fail("PT_LOAD in wrong place\n");
37
38 if (GET_BE(&pt[i].p_memsz) != GET_BE(&pt[i].p_filesz))
39 fail("cannot handle memsz != filesz\n");
40
41 load_size = GET_BE(&pt[i].p_memsz);
42 found_load = 1;
43 } else if (GET_BE(&pt[i].p_type) == PT_DYNAMIC) {
44 dyn = raw_addr + GET_BE(&pt[i].p_offset);
45 dyn_end = raw_addr + GET_BE(&pt[i].p_offset) +
46 GET_BE(&pt[i].p_memsz);
47 }
48 }
49 if (!found_load)
50 fail("no PT_LOAD seg\n");
51
52 if (stripped_len < load_size)
53 fail("stripped input is too short\n");
54
55 /* Walk the dynamic table */
56 for (i = 0; dyn + i < dyn_end &&
57 GET_BE(&dyn[i].d_tag) != DT_NULL; i++) {
58 typeof(dyn[i].d_tag) tag = GET_BE(&dyn[i].d_tag);
59 typeof(dyn[i].d_un.d_val) val = GET_BE(&dyn[i].d_un.d_val);
60
61 if ((tag == DT_RELSZ || tag == DT_RELASZ) && (val != 0))
62 fail("vdso image contains dynamic relocations\n");
63 }
64
65 /* Walk the section table */
66 for (i = 0; i < GET_BE(&hdr->e_shnum); i++) {
67 ELF(Shdr) *sh = raw_addr + GET_BE(&hdr->e_shoff) +
68 GET_BE(&hdr->e_shentsize) * i;
69 if (GET_BE(&sh->sh_type) == SHT_SYMTAB)
70 symtab_hdr = sh;
71 }
72
73 if (!symtab_hdr)
74 fail("no symbol table\n");
75
76 strtab_hdr = raw_addr + GET_BE(&hdr->e_shoff) +
77 GET_BE(&hdr->e_shentsize) * GET_BE(&symtab_hdr->sh_link);
78
79 /* Walk the symbol table */
80 for (i = 0;
81 i < GET_BE(&symtab_hdr->sh_size) / GET_BE(&symtab_hdr->sh_entsize);
82 i++) {
83 int k;
84
85 ELF(Sym) *sym = raw_addr + GET_BE(&symtab_hdr->sh_offset) +
86 GET_BE(&symtab_hdr->sh_entsize) * i;
87 const char *name = raw_addr + GET_BE(&strtab_hdr->sh_offset) +
88 GET_BE(&sym->st_name);
89
90 for (k = 0; k < NSYMS; k++) {
91 if (!strcmp(name, required_syms[k].name)) {
92 if (syms[k]) {
93 fail("duplicate symbol %s\n",
94 required_syms[k].name);
95 }
96
97 /*
98 * Careful: we use negative addresses, but
99 * st_value is unsigned, so we rely
100 * on syms[k] being a signed type of the
101 * correct width.
102 */
103 syms[k] = GET_BE(&sym->st_value);
104 }
105 }
106 }
107
108 /* Validate mapping addresses. */
109 if (syms[sym_vvar_start] % 8192)
110 fail("vvar_begin must be a multiple of 8192\n");
111
112 if (!name) {
113 fwrite(stripped_addr, stripped_len, 1, outfile);
114 return;
115 }
116
117 mapping_size = (stripped_len + 8191) / 8192 * 8192;
118
119 fprintf(outfile, "/* AUTOMATICALLY GENERATED -- DO NOT EDIT */\n\n");
120 fprintf(outfile, "#include <linux/cache.h>\n");
121 fprintf(outfile, "#include <asm/vdso.h>\n");
122 fprintf(outfile, "\n");
123 fprintf(outfile,
124 "static unsigned char raw_data[%lu] __ro_after_init __aligned(8192)= {",
125 mapping_size);
126 for (j = 0; j < stripped_len; j++) {
127 if (j % 10 == 0)
128 fprintf(outfile, "\n\t");
129 fprintf(outfile, "0x%02X, ",
130 (int)((unsigned char *)stripped_addr)[j]);
131 }
132 fprintf(outfile, "\n};\n\n");
133
134 fprintf(outfile, "const struct vdso_image %s_builtin = {\n", name);
135 fprintf(outfile, "\t.data = raw_data,\n");
136 fprintf(outfile, "\t.size = %lu,\n", mapping_size);
137 for (i = 0; i < NSYMS; i++) {
138 if (required_syms[i].export && syms[i])
139 fprintf(outfile, "\t.sym_%s = %" PRIi64 ",\n",
140 required_syms[i].name, (int64_t)syms[i]);
141 }
142 fprintf(outfile, "};\n");
143}
diff --git a/arch/sparc/vdso/vdso32/.gitignore b/arch/sparc/vdso/vdso32/.gitignore
new file mode 100644
index 000000000000..e45fba9d0ced
--- /dev/null
+++ b/arch/sparc/vdso/vdso32/.gitignore
@@ -0,0 +1 @@
vdso32.lds
diff --git a/arch/sparc/vdso/vdso32/vclock_gettime.c b/arch/sparc/vdso/vdso32/vclock_gettime.c
new file mode 100644
index 000000000000..026abb3b826c
--- /dev/null
+++ b/arch/sparc/vdso/vdso32/vclock_gettime.c
@@ -0,0 +1,26 @@
1/*
2 * Copyright (c) 2017 Oracle and/or its affiliates. All rights reserved.
3 */
4
5#define BUILD_VDSO32
6
7#ifndef CONFIG_CC_OPTIMIZE_FOR_SIZE
8#undef CONFIG_OPTIMIZE_INLINING
9#endif
10
11#ifdef CONFIG_SPARC64
12
13/*
14 * in case of a 32 bit VDSO for a 64 bit kernel fake a 32 bit kernel
15 * configuration
16 */
17#undef CONFIG_64BIT
18#undef CONFIG_SPARC64
19#define BUILD_VDSO32_64
20#define CONFIG_32BIT
21#undef CONFIG_QUEUED_RWLOCKS
22#undef CONFIG_QUEUED_SPINLOCKS
23
24#endif
25
26#include "../vclock_gettime.c"
diff --git a/arch/sparc/vdso/vdso32/vdso-note.S b/arch/sparc/vdso/vdso32/vdso-note.S
new file mode 100644
index 000000000000..e234983cf0d8
--- /dev/null
+++ b/arch/sparc/vdso/vdso32/vdso-note.S
@@ -0,0 +1,12 @@
1/*
2 * This supplies .note.* sections to go into the PT_NOTE inside the vDSO
3 * text. Here we can supply some information useful to userland.
4 */
5
6#include <linux/uts.h>
7#include <linux/version.h>
8#include <linux/elfnote.h>
9
10ELFNOTE_START(Linux, 0, "a")
11 .long LINUX_VERSION_CODE
12ELFNOTE_END
diff --git a/arch/sparc/vdso/vdso32/vdso32.lds.S b/arch/sparc/vdso/vdso32/vdso32.lds.S
new file mode 100644
index 000000000000..53575ee154c4
--- /dev/null
+++ b/arch/sparc/vdso/vdso32/vdso32.lds.S
@@ -0,0 +1,24 @@
1/*
2 * Linker script for sparc32 vDSO
3 * We #include the file to define the layout details.
4 *
5 * This file defines the version script giving the user-exported symbols in
6 * the DSO.
7 */
8
9#define BUILD_VDSO32
10#include "../vdso-layout.lds.S"
11
12/*
13 * This controls what userland symbols we export from the vDSO.
14 */
15VERSION {
16 LINUX_2.6 {
17 global:
18 clock_gettime;
19 __vdso_clock_gettime;
20 gettimeofday;
21 __vdso_gettimeofday;
22 local: *;
23 };
24}
diff --git a/arch/sparc/vdso/vma.c b/arch/sparc/vdso/vma.c
new file mode 100644
index 000000000000..0a6f50098e23
--- /dev/null
+++ b/arch/sparc/vdso/vma.c
@@ -0,0 +1,268 @@
1/*
2 * Set up the VMAs to tell the VM about the vDSO.
3 * Copyright 2007 Andi Kleen, SUSE Labs.
4 * Subject to the GPL, v.2
5 */
6
7/*
8 * Copyright (c) 2017 Oracle and/or its affiliates. All rights reserved.
9 */
10
11#include <linux/mm.h>
12#include <linux/err.h>
13#include <linux/sched.h>
14#include <linux/slab.h>
15#include <linux/init.h>
16#include <linux/linkage.h>
17#include <linux/random.h>
18#include <linux/elf.h>
19#include <asm/vdso.h>
20#include <asm/vvar.h>
21#include <asm/page.h>
22
23unsigned int __read_mostly vdso_enabled = 1;
24
25static struct vm_special_mapping vvar_mapping = {
26 .name = "[vvar]"
27};
28
29#ifdef CONFIG_SPARC64
30static struct vm_special_mapping vdso_mapping64 = {
31 .name = "[vdso]"
32};
33#endif
34
35#ifdef CONFIG_COMPAT
36static struct vm_special_mapping vdso_mapping32 = {
37 .name = "[vdso]"
38};
39#endif
40
41struct vvar_data *vvar_data;
42
43#define SAVE_INSTR_SIZE 4
44
45/*
46 * Allocate pages for the vdso and vvar, and copy in the vdso text from the
47 * kernel image.
48 */
49int __init init_vdso_image(const struct vdso_image *image,
50 struct vm_special_mapping *vdso_mapping)
51{
52 int i;
53 struct page *dp, **dpp = NULL;
54 int dnpages = 0;
55 struct page *cp, **cpp = NULL;
56 int cnpages = (image->size) / PAGE_SIZE;
57
58 /*
59 * First, the vdso text. This is initialied data, an integral number of
60 * pages long.
61 */
62 if (WARN_ON(image->size % PAGE_SIZE != 0))
63 goto oom;
64
65 cpp = kcalloc(cnpages, sizeof(struct page *), GFP_KERNEL);
66 vdso_mapping->pages = cpp;
67
68 if (!cpp)
69 goto oom;
70
71 if (vdso_fix_stick) {
72 /*
73 * If the system uses %tick instead of %stick, patch the VDSO
74 * with instruction reading %tick instead of %stick.
75 */
76 unsigned int j, k = SAVE_INSTR_SIZE;
77 unsigned char *data = image->data;
78
79 for (j = image->sym_vread_tick_patch_start;
80 j < image->sym_vread_tick_patch_end; j++) {
81
82 data[image->sym_vread_tick + k] = data[j];
83 k++;
84 }
85 }
86
87 for (i = 0; i < cnpages; i++) {
88 cp = alloc_page(GFP_KERNEL);
89 if (!cp)
90 goto oom;
91 cpp[i] = cp;
92 copy_page(page_address(cp), image->data + i * PAGE_SIZE);
93 }
94
95 /*
96 * Now the vvar page. This is uninitialized data.
97 */
98
99 if (vvar_data == NULL) {
100 dnpages = (sizeof(struct vvar_data) / PAGE_SIZE) + 1;
101 if (WARN_ON(dnpages != 1))
102 goto oom;
103 dpp = kcalloc(dnpages, sizeof(struct page *), GFP_KERNEL);
104 vvar_mapping.pages = dpp;
105
106 if (!dpp)
107 goto oom;
108
109 dp = alloc_page(GFP_KERNEL);
110 if (!dp)
111 goto oom;
112
113 dpp[0] = dp;
114 vvar_data = page_address(dp);
115 memset(vvar_data, 0, PAGE_SIZE);
116
117 vvar_data->seq = 0;
118 }
119
120 return 0;
121 oom:
122 if (cpp != NULL) {
123 for (i = 0; i < cnpages; i++) {
124 if (cpp[i] != NULL)
125 __free_page(cpp[i]);
126 }
127 kfree(cpp);
128 vdso_mapping->pages = NULL;
129 }
130
131 if (dpp != NULL) {
132 for (i = 0; i < dnpages; i++) {
133 if (dpp[i] != NULL)
134 __free_page(dpp[i]);
135 }
136 kfree(dpp);
137 vvar_mapping.pages = NULL;
138 }
139
140 pr_warn("Cannot allocate vdso\n");
141 vdso_enabled = 0;
142 return -ENOMEM;
143}
144
145static int __init init_vdso(void)
146{
147 int err = 0;
148#ifdef CONFIG_SPARC64
149 err = init_vdso_image(&vdso_image_64_builtin, &vdso_mapping64);
150 if (err)
151 return err;
152#endif
153
154#ifdef CONFIG_COMPAT
155 err = init_vdso_image(&vdso_image_32_builtin, &vdso_mapping32);
156#endif
157 return err;
158
159}
160subsys_initcall(init_vdso);
161
162struct linux_binprm;
163
164/* Shuffle the vdso up a bit, randomly. */
165static unsigned long vdso_addr(unsigned long start, unsigned int len)
166{
167 unsigned int offset;
168
169 /* This loses some more bits than a modulo, but is cheaper */
170 offset = get_random_int() & (PTRS_PER_PTE - 1);
171 return start + (offset << PAGE_SHIFT);
172}
173
174static int map_vdso(const struct vdso_image *image,
175 struct vm_special_mapping *vdso_mapping)
176{
177 struct mm_struct *mm = current->mm;
178 struct vm_area_struct *vma;
179 unsigned long text_start, addr = 0;
180 int ret = 0;
181
182 down_write(&mm->mmap_sem);
183
184 /*
185 * First, get an unmapped region: then randomize it, and make sure that
186 * region is free.
187 */
188 if (current->flags & PF_RANDOMIZE) {
189 addr = get_unmapped_area(NULL, 0,
190 image->size - image->sym_vvar_start,
191 0, 0);
192 if (IS_ERR_VALUE(addr)) {
193 ret = addr;
194 goto up_fail;
195 }
196 addr = vdso_addr(addr, image->size - image->sym_vvar_start);
197 }
198 addr = get_unmapped_area(NULL, addr,
199 image->size - image->sym_vvar_start, 0, 0);
200 if (IS_ERR_VALUE(addr)) {
201 ret = addr;
202 goto up_fail;
203 }
204
205 text_start = addr - image->sym_vvar_start;
206 current->mm->context.vdso = (void __user *)text_start;
207
208 /*
209 * MAYWRITE to allow gdb to COW and set breakpoints
210 */
211 vma = _install_special_mapping(mm,
212 text_start,
213 image->size,
214 VM_READ|VM_EXEC|
215 VM_MAYREAD|VM_MAYWRITE|VM_MAYEXEC,
216 vdso_mapping);
217
218 if (IS_ERR(vma)) {
219 ret = PTR_ERR(vma);
220 goto up_fail;
221 }
222
223 vma = _install_special_mapping(mm,
224 addr,
225 -image->sym_vvar_start,
226 VM_READ|VM_MAYREAD,
227 &vvar_mapping);
228
229 if (IS_ERR(vma)) {
230 ret = PTR_ERR(vma);
231 do_munmap(mm, text_start, image->size, NULL);
232 }
233
234up_fail:
235 if (ret)
236 current->mm->context.vdso = NULL;
237
238 up_write(&mm->mmap_sem);
239 return ret;
240}
241
242int arch_setup_additional_pages(struct linux_binprm *bprm, int uses_interp)
243{
244
245 if (!vdso_enabled)
246 return 0;
247
248#if defined CONFIG_COMPAT
249 if (!(is_32bit_task()))
250 return map_vdso(&vdso_image_64_builtin, &vdso_mapping64);
251 else
252 return map_vdso(&vdso_image_32_builtin, &vdso_mapping32);
253#else
254 return map_vdso(&vdso_image_64_builtin, &vdso_mapping64);
255#endif
256
257}
258
259static __init int vdso_setup(char *s)
260{
261 int err;
262 unsigned long val;
263
264 err = kstrtoul(s, 10, &val);
265 vdso_enabled = val;
266 return err;
267}
268__setup("vdso=", vdso_setup);
generated by cgit v1.2.2 (git 2.25.0) at 2025-08-15 22:30:59 -0400