aboutsummaryrefslogtreecommitdiffstats
path: root/arch/x86/kernel/cpu
diff options
context:
space:
mode:
authorIngo Molnar <mingo@kernel.org>2012-09-19 11:01:25 -0400
committerIngo Molnar <mingo@kernel.org>2012-09-19 11:01:25 -0400
commitf1f652447694f92beff8a534d94b36ea441c939a (patch)
tree177f029e418d517e39dfa2495a9a67a45381b833 /arch/x86/kernel/cpu
parent55babd8f41f122f5f4c7cebf520c766c983282c6 (diff)
parent5698bd757d55b1bb87edd1a9744ab09c142abfc2 (diff)
Merge tag 'v3.6-rc6' into x86/mce
Merge Linux v3.6-rc6, to refresh this tree. Signed-off-by: Ingo Molnar <mingo@kernel.org>
Diffstat (limited to 'arch/x86/kernel/cpu')
-rw-r--r--arch/x86/kernel/cpu/common.c2
-rw-r--r--arch/x86/kernel/cpu/mcheck/mce-severity.c7
-rw-r--r--arch/x86/kernel/cpu/mcheck/mce.c43
-rw-r--r--arch/x86/kernel/cpu/perf_event.c89
-rw-r--r--arch/x86/kernel/cpu/perf_event.h20
-rw-r--r--arch/x86/kernel/cpu/perf_event_amd_ibs.c4
-rw-r--r--arch/x86/kernel/cpu/perf_event_intel.c11
-rw-r--r--arch/x86/kernel/cpu/perf_event_intel_ds.c7
-rw-r--r--arch/x86/kernel/cpu/perf_event_intel_lbr.c3
-rw-r--r--arch/x86/kernel/cpu/perf_event_intel_uncore.c253
-rw-r--r--arch/x86/kernel/cpu/perf_event_intel_uncore.h48
11 files changed, 317 insertions, 170 deletions
diff --git a/arch/x86/kernel/cpu/common.c b/arch/x86/kernel/cpu/common.c
index 46d8786d655e..a5fbc3c5fccc 100644
--- a/arch/x86/kernel/cpu/common.c
+++ b/arch/x86/kernel/cpu/common.c
@@ -144,6 +144,8 @@ static int __init x86_xsave_setup(char *s)
144{ 144{
145 setup_clear_cpu_cap(X86_FEATURE_XSAVE); 145 setup_clear_cpu_cap(X86_FEATURE_XSAVE);
146 setup_clear_cpu_cap(X86_FEATURE_XSAVEOPT); 146 setup_clear_cpu_cap(X86_FEATURE_XSAVEOPT);
147 setup_clear_cpu_cap(X86_FEATURE_AVX);
148 setup_clear_cpu_cap(X86_FEATURE_AVX2);
147 return 1; 149 return 1;
148} 150}
149__setup("noxsave", x86_xsave_setup); 151__setup("noxsave", x86_xsave_setup);
diff --git a/arch/x86/kernel/cpu/mcheck/mce-severity.c b/arch/x86/kernel/cpu/mcheck/mce-severity.c
index 413c2ced887c..13017626f9a8 100644
--- a/arch/x86/kernel/cpu/mcheck/mce-severity.c
+++ b/arch/x86/kernel/cpu/mcheck/mce-severity.c
@@ -55,13 +55,6 @@ static struct severity {
55#define MCI_UC_S (MCI_STATUS_UC|MCI_STATUS_S) 55#define MCI_UC_S (MCI_STATUS_UC|MCI_STATUS_S)
56#define MCI_UC_SAR (MCI_STATUS_UC|MCI_STATUS_S|MCI_STATUS_AR) 56#define MCI_UC_SAR (MCI_STATUS_UC|MCI_STATUS_S|MCI_STATUS_AR)
57#define MCI_ADDR (MCI_STATUS_ADDRV|MCI_STATUS_MISCV) 57#define MCI_ADDR (MCI_STATUS_ADDRV|MCI_STATUS_MISCV)
58#define MCACOD 0xffff
59/* Architecturally defined codes from SDM Vol. 3B Chapter 15 */
60#define MCACOD_SCRUB 0x00C0 /* 0xC0-0xCF Memory Scrubbing */
61#define MCACOD_SCRUBMSK 0xfff0
62#define MCACOD_L3WB 0x017A /* L3 Explicit Writeback */
63#define MCACOD_DATA 0x0134 /* Data Load */
64#define MCACOD_INSTR 0x0150 /* Instruction Fetch */
65 58
66 MCESEV( 59 MCESEV(
67 NO, "Invalid", 60 NO, "Invalid",
diff --git a/arch/x86/kernel/cpu/mcheck/mce.c b/arch/x86/kernel/cpu/mcheck/mce.c
index 8c1beea6cabf..c311122ea838 100644
--- a/arch/x86/kernel/cpu/mcheck/mce.c
+++ b/arch/x86/kernel/cpu/mcheck/mce.c
@@ -103,6 +103,8 @@ DEFINE_PER_CPU(mce_banks_t, mce_poll_banks) = {
103 103
104static DEFINE_PER_CPU(struct work_struct, mce_work); 104static DEFINE_PER_CPU(struct work_struct, mce_work);
105 105
106static void (*quirk_no_way_out)(int bank, struct mce *m, struct pt_regs *regs);
107
106/* 108/*
107 * CPU/chipset specific EDAC code can register a notifier call here to print 109 * CPU/chipset specific EDAC code can register a notifier call here to print
108 * MCE errors in a human-readable form. 110 * MCE errors in a human-readable form.
@@ -650,14 +652,18 @@ EXPORT_SYMBOL_GPL(machine_check_poll);
650 * Do a quick check if any of the events requires a panic. 652 * Do a quick check if any of the events requires a panic.
651 * This decides if we keep the events around or clear them. 653 * This decides if we keep the events around or clear them.
652 */ 654 */
653static int mce_no_way_out(struct mce *m, char **msg, unsigned long *validp) 655static int mce_no_way_out(struct mce *m, char **msg, unsigned long *validp,
656 struct pt_regs *regs)
654{ 657{
655 int i, ret = 0; 658 int i, ret = 0;
656 659
657 for (i = 0; i < banks; i++) { 660 for (i = 0; i < banks; i++) {
658 m->status = mce_rdmsrl(MSR_IA32_MCx_STATUS(i)); 661 m->status = mce_rdmsrl(MSR_IA32_MCx_STATUS(i));
659 if (m->status & MCI_STATUS_VAL) 662 if (m->status & MCI_STATUS_VAL) {
660 __set_bit(i, validp); 663 __set_bit(i, validp);
664 if (quirk_no_way_out)
665 quirk_no_way_out(i, m, regs);
666 }
661 if (mce_severity(m, tolerant, msg) >= MCE_PANIC_SEVERITY) 667 if (mce_severity(m, tolerant, msg) >= MCE_PANIC_SEVERITY)
662 ret = 1; 668 ret = 1;
663 } 669 }
@@ -1040,7 +1046,7 @@ void do_machine_check(struct pt_regs *regs, long error_code)
1040 *final = m; 1046 *final = m;
1041 1047
1042 memset(valid_banks, 0, sizeof(valid_banks)); 1048 memset(valid_banks, 0, sizeof(valid_banks));
1043 no_way_out = mce_no_way_out(&m, &msg, valid_banks); 1049 no_way_out = mce_no_way_out(&m, &msg, valid_banks, regs);
1044 1050
1045 barrier(); 1051 barrier();
1046 1052
@@ -1451,6 +1457,34 @@ static void __mcheck_cpu_init_generic(void)
1451 } 1457 }
1452} 1458}
1453 1459
1460/*
1461 * During IFU recovery Sandy Bridge -EP4S processors set the RIPV and
1462 * EIPV bits in MCG_STATUS to zero on the affected logical processor (SDM
1463 * Vol 3B Table 15-20). But this confuses both the code that determines
1464 * whether the machine check occurred in kernel or user mode, and also
1465 * the severity assessment code. Pretend that EIPV was set, and take the
1466 * ip/cs values from the pt_regs that mce_gather_info() ignored earlier.
1467 */
1468static void quirk_sandybridge_ifu(int bank, struct mce *m, struct pt_regs *regs)
1469{
1470 if (bank != 0)
1471 return;
1472 if ((m->mcgstatus & (MCG_STATUS_EIPV|MCG_STATUS_RIPV)) != 0)
1473 return;
1474 if ((m->status & (MCI_STATUS_OVER|MCI_STATUS_UC|
1475 MCI_STATUS_EN|MCI_STATUS_MISCV|MCI_STATUS_ADDRV|
1476 MCI_STATUS_PCC|MCI_STATUS_S|MCI_STATUS_AR|
1477 MCACOD)) !=
1478 (MCI_STATUS_UC|MCI_STATUS_EN|
1479 MCI_STATUS_MISCV|MCI_STATUS_ADDRV|MCI_STATUS_S|
1480 MCI_STATUS_AR|MCACOD_INSTR))
1481 return;
1482
1483 m->mcgstatus |= MCG_STATUS_EIPV;
1484 m->ip = regs->ip;
1485 m->cs = regs->cs;
1486}
1487
1454/* Add per CPU specific workarounds here */ 1488/* Add per CPU specific workarounds here */
1455static int __cpuinit __mcheck_cpu_apply_quirks(struct cpuinfo_x86 *c) 1489static int __cpuinit __mcheck_cpu_apply_quirks(struct cpuinfo_x86 *c)
1456{ 1490{
@@ -1548,6 +1582,9 @@ static int __cpuinit __mcheck_cpu_apply_quirks(struct cpuinfo_x86 *c)
1548 */ 1582 */
1549 if (c->x86 == 6 && c->x86_model <= 13 && mce_bootlog < 0) 1583 if (c->x86 == 6 && c->x86_model <= 13 && mce_bootlog < 0)
1550 mce_bootlog = 0; 1584 mce_bootlog = 0;
1585
1586 if (c->x86 == 6 && c->x86_model == 45)
1587 quirk_no_way_out = quirk_sandybridge_ifu;
1551 } 1588 }
1552 if (monarch_timeout < 0) 1589 if (monarch_timeout < 0)
1553 monarch_timeout = 0; 1590 monarch_timeout = 0;
diff --git a/arch/x86/kernel/cpu/perf_event.c b/arch/x86/kernel/cpu/perf_event.c
index 29557aa06dda..915b876edd1e 100644
--- a/arch/x86/kernel/cpu/perf_event.c
+++ b/arch/x86/kernel/cpu/perf_event.c
@@ -32,6 +32,8 @@
32#include <asm/smp.h> 32#include <asm/smp.h>
33#include <asm/alternative.h> 33#include <asm/alternative.h>
34#include <asm/timer.h> 34#include <asm/timer.h>
35#include <asm/desc.h>
36#include <asm/ldt.h>
35 37
36#include "perf_event.h" 38#include "perf_event.h"
37 39
@@ -1738,6 +1740,29 @@ valid_user_frame(const void __user *fp, unsigned long size)
1738 return (__range_not_ok(fp, size, TASK_SIZE) == 0); 1740 return (__range_not_ok(fp, size, TASK_SIZE) == 0);
1739} 1741}
1740 1742
1743static unsigned long get_segment_base(unsigned int segment)
1744{
1745 struct desc_struct *desc;
1746 int idx = segment >> 3;
1747
1748 if ((segment & SEGMENT_TI_MASK) == SEGMENT_LDT) {
1749 if (idx > LDT_ENTRIES)
1750 return 0;
1751
1752 if (idx > current->active_mm->context.size)
1753 return 0;
1754
1755 desc = current->active_mm->context.ldt;
1756 } else {
1757 if (idx > GDT_ENTRIES)
1758 return 0;
1759
1760 desc = __this_cpu_ptr(&gdt_page.gdt[0]);
1761 }
1762
1763 return get_desc_base(desc + idx);
1764}
1765
1741#ifdef CONFIG_COMPAT 1766#ifdef CONFIG_COMPAT
1742 1767
1743#include <asm/compat.h> 1768#include <asm/compat.h>
@@ -1746,13 +1771,17 @@ static inline int
1746perf_callchain_user32(struct pt_regs *regs, struct perf_callchain_entry *entry) 1771perf_callchain_user32(struct pt_regs *regs, struct perf_callchain_entry *entry)
1747{ 1772{
1748 /* 32-bit process in 64-bit kernel. */ 1773 /* 32-bit process in 64-bit kernel. */
1774 unsigned long ss_base, cs_base;
1749 struct stack_frame_ia32 frame; 1775 struct stack_frame_ia32 frame;
1750 const void __user *fp; 1776 const void __user *fp;
1751 1777
1752 if (!test_thread_flag(TIF_IA32)) 1778 if (!test_thread_flag(TIF_IA32))
1753 return 0; 1779 return 0;
1754 1780
1755 fp = compat_ptr(regs->bp); 1781 cs_base = get_segment_base(regs->cs);
1782 ss_base = get_segment_base(regs->ss);
1783
1784 fp = compat_ptr(ss_base + regs->bp);
1756 while (entry->nr < PERF_MAX_STACK_DEPTH) { 1785 while (entry->nr < PERF_MAX_STACK_DEPTH) {
1757 unsigned long bytes; 1786 unsigned long bytes;
1758 frame.next_frame = 0; 1787 frame.next_frame = 0;
@@ -1765,8 +1794,8 @@ perf_callchain_user32(struct pt_regs *regs, struct perf_callchain_entry *entry)
1765 if (!valid_user_frame(fp, sizeof(frame))) 1794 if (!valid_user_frame(fp, sizeof(frame)))
1766 break; 1795 break;
1767 1796
1768 perf_callchain_store(entry, frame.return_address); 1797 perf_callchain_store(entry, cs_base + frame.return_address);
1769 fp = compat_ptr(frame.next_frame); 1798 fp = compat_ptr(ss_base + frame.next_frame);
1770 } 1799 }
1771 return 1; 1800 return 1;
1772} 1801}
@@ -1789,6 +1818,12 @@ perf_callchain_user(struct perf_callchain_entry *entry, struct pt_regs *regs)
1789 return; 1818 return;
1790 } 1819 }
1791 1820
1821 /*
1822 * We don't know what to do with VM86 stacks.. ignore them for now.
1823 */
1824 if (regs->flags & (X86_VM_MASK | PERF_EFLAGS_VM))
1825 return;
1826
1792 fp = (void __user *)regs->bp; 1827 fp = (void __user *)regs->bp;
1793 1828
1794 perf_callchain_store(entry, regs->ip); 1829 perf_callchain_store(entry, regs->ip);
@@ -1816,16 +1851,50 @@ perf_callchain_user(struct perf_callchain_entry *entry, struct pt_regs *regs)
1816 } 1851 }
1817} 1852}
1818 1853
1819unsigned long perf_instruction_pointer(struct pt_regs *regs) 1854/*
1855 * Deal with code segment offsets for the various execution modes:
1856 *
1857 * VM86 - the good olde 16 bit days, where the linear address is
1858 * 20 bits and we use regs->ip + 0x10 * regs->cs.
1859 *
1860 * IA32 - Where we need to look at GDT/LDT segment descriptor tables
1861 * to figure out what the 32bit base address is.
1862 *
1863 * X32 - has TIF_X32 set, but is running in x86_64
1864 *
1865 * X86_64 - CS,DS,SS,ES are all zero based.
1866 */
1867static unsigned long code_segment_base(struct pt_regs *regs)
1820{ 1868{
1821 unsigned long ip; 1869 /*
1870 * If we are in VM86 mode, add the segment offset to convert to a
1871 * linear address.
1872 */
1873 if (regs->flags & X86_VM_MASK)
1874 return 0x10 * regs->cs;
1875
1876 /*
1877 * For IA32 we look at the GDT/LDT segment base to convert the
1878 * effective IP to a linear address.
1879 */
1880#ifdef CONFIG_X86_32
1881 if (user_mode(regs) && regs->cs != __USER_CS)
1882 return get_segment_base(regs->cs);
1883#else
1884 if (test_thread_flag(TIF_IA32)) {
1885 if (user_mode(regs) && regs->cs != __USER32_CS)
1886 return get_segment_base(regs->cs);
1887 }
1888#endif
1889 return 0;
1890}
1822 1891
1892unsigned long perf_instruction_pointer(struct pt_regs *regs)
1893{
1823 if (perf_guest_cbs && perf_guest_cbs->is_in_guest()) 1894 if (perf_guest_cbs && perf_guest_cbs->is_in_guest())
1824 ip = perf_guest_cbs->get_guest_ip(); 1895 return perf_guest_cbs->get_guest_ip();
1825 else
1826 ip = instruction_pointer(regs);
1827 1896
1828 return ip; 1897 return regs->ip + code_segment_base(regs);
1829} 1898}
1830 1899
1831unsigned long perf_misc_flags(struct pt_regs *regs) 1900unsigned long perf_misc_flags(struct pt_regs *regs)
@@ -1838,7 +1907,7 @@ unsigned long perf_misc_flags(struct pt_regs *regs)
1838 else 1907 else
1839 misc |= PERF_RECORD_MISC_GUEST_KERNEL; 1908 misc |= PERF_RECORD_MISC_GUEST_KERNEL;
1840 } else { 1909 } else {
1841 if (!kernel_ip(regs->ip)) 1910 if (user_mode(regs))
1842 misc |= PERF_RECORD_MISC_USER; 1911 misc |= PERF_RECORD_MISC_USER;
1843 else 1912 else
1844 misc |= PERF_RECORD_MISC_KERNEL; 1913 misc |= PERF_RECORD_MISC_KERNEL;
diff --git a/arch/x86/kernel/cpu/perf_event.h b/arch/x86/kernel/cpu/perf_event.h
index 821d53b696d1..6605a81ba339 100644
--- a/arch/x86/kernel/cpu/perf_event.h
+++ b/arch/x86/kernel/cpu/perf_event.h
@@ -516,6 +516,26 @@ static inline bool kernel_ip(unsigned long ip)
516#endif 516#endif
517} 517}
518 518
519/*
520 * Not all PMUs provide the right context information to place the reported IP
521 * into full context. Specifically segment registers are typically not
522 * supplied.
523 *
524 * Assuming the address is a linear address (it is for IBS), we fake the CS and
525 * vm86 mode using the known zero-based code segment and 'fix up' the registers
526 * to reflect this.
527 *
528 * Intel PEBS/LBR appear to typically provide the effective address, nothing
529 * much we can do about that but pray and treat it like a linear address.
530 */
531static inline void set_linear_ip(struct pt_regs *regs, unsigned long ip)
532{
533 regs->cs = kernel_ip(ip) ? __KERNEL_CS : __USER_CS;
534 if (regs->flags & X86_VM_MASK)
535 regs->flags ^= (PERF_EFLAGS_VM | X86_VM_MASK);
536 regs->ip = ip;
537}
538
519#ifdef CONFIG_CPU_SUP_AMD 539#ifdef CONFIG_CPU_SUP_AMD
520 540
521int amd_pmu_init(void); 541int amd_pmu_init(void);
diff --git a/arch/x86/kernel/cpu/perf_event_amd_ibs.c b/arch/x86/kernel/cpu/perf_event_amd_ibs.c
index da9bcdcd9856..7bfb5bec8630 100644
--- a/arch/x86/kernel/cpu/perf_event_amd_ibs.c
+++ b/arch/x86/kernel/cpu/perf_event_amd_ibs.c
@@ -13,6 +13,8 @@
13 13
14#include <asm/apic.h> 14#include <asm/apic.h>
15 15
16#include "perf_event.h"
17
16static u32 ibs_caps; 18static u32 ibs_caps;
17 19
18#if defined(CONFIG_PERF_EVENTS) && defined(CONFIG_CPU_SUP_AMD) 20#if defined(CONFIG_PERF_EVENTS) && defined(CONFIG_CPU_SUP_AMD)
@@ -536,7 +538,7 @@ static int perf_ibs_handle_irq(struct perf_ibs *perf_ibs, struct pt_regs *iregs)
536 if (check_rip && (ibs_data.regs[2] & IBS_RIP_INVALID)) { 538 if (check_rip && (ibs_data.regs[2] & IBS_RIP_INVALID)) {
537 regs.flags &= ~PERF_EFLAGS_EXACT; 539 regs.flags &= ~PERF_EFLAGS_EXACT;
538 } else { 540 } else {
539 instruction_pointer_set(&regs, ibs_data.regs[1]); 541 set_linear_ip(&regs, ibs_data.regs[1]);
540 regs.flags |= PERF_EFLAGS_EXACT; 542 regs.flags |= PERF_EFLAGS_EXACT;
541 } 543 }
542 544
diff --git a/arch/x86/kernel/cpu/perf_event_intel.c b/arch/x86/kernel/cpu/perf_event_intel.c
index 382366977d4c..0d3d63afa76a 100644
--- a/arch/x86/kernel/cpu/perf_event_intel.c
+++ b/arch/x86/kernel/cpu/perf_event_intel.c
@@ -1522,8 +1522,16 @@ static struct perf_guest_switch_msr *intel_guest_get_msrs(int *nr)
1522 arr[0].msr = MSR_CORE_PERF_GLOBAL_CTRL; 1522 arr[0].msr = MSR_CORE_PERF_GLOBAL_CTRL;
1523 arr[0].host = x86_pmu.intel_ctrl & ~cpuc->intel_ctrl_guest_mask; 1523 arr[0].host = x86_pmu.intel_ctrl & ~cpuc->intel_ctrl_guest_mask;
1524 arr[0].guest = x86_pmu.intel_ctrl & ~cpuc->intel_ctrl_host_mask; 1524 arr[0].guest = x86_pmu.intel_ctrl & ~cpuc->intel_ctrl_host_mask;
1525 /*
1526 * If PMU counter has PEBS enabled it is not enough to disable counter
1527 * on a guest entry since PEBS memory write can overshoot guest entry
1528 * and corrupt guest memory. Disabling PEBS solves the problem.
1529 */
1530 arr[1].msr = MSR_IA32_PEBS_ENABLE;
1531 arr[1].host = cpuc->pebs_enabled;
1532 arr[1].guest = 0;
1525 1533
1526 *nr = 1; 1534 *nr = 2;
1527 return arr; 1535 return arr;
1528} 1536}
1529 1537
@@ -2000,6 +2008,7 @@ __init int intel_pmu_init(void)
2000 break; 2008 break;
2001 2009
2002 case 28: /* Atom */ 2010 case 28: /* Atom */
2011 case 54: /* Cedariew */
2003 memcpy(hw_cache_event_ids, atom_hw_cache_event_ids, 2012 memcpy(hw_cache_event_ids, atom_hw_cache_event_ids,
2004 sizeof(hw_cache_event_ids)); 2013 sizeof(hw_cache_event_ids));
2005 2014
diff --git a/arch/x86/kernel/cpu/perf_event_intel_ds.c b/arch/x86/kernel/cpu/perf_event_intel_ds.c
index 629ae0b7ad90..e38d97bf4259 100644
--- a/arch/x86/kernel/cpu/perf_event_intel_ds.c
+++ b/arch/x86/kernel/cpu/perf_event_intel_ds.c
@@ -499,7 +499,7 @@ static int intel_pmu_pebs_fixup_ip(struct pt_regs *regs)
499 * We sampled a branch insn, rewind using the LBR stack 499 * We sampled a branch insn, rewind using the LBR stack
500 */ 500 */
501 if (ip == to) { 501 if (ip == to) {
502 regs->ip = from; 502 set_linear_ip(regs, from);
503 return 1; 503 return 1;
504 } 504 }
505 505
@@ -529,7 +529,7 @@ static int intel_pmu_pebs_fixup_ip(struct pt_regs *regs)
529 } while (to < ip); 529 } while (to < ip);
530 530
531 if (to == ip) { 531 if (to == ip) {
532 regs->ip = old_to; 532 set_linear_ip(regs, old_to);
533 return 1; 533 return 1;
534 } 534 }
535 535
@@ -569,7 +569,8 @@ static void __intel_pmu_pebs_event(struct perf_event *event,
569 * A possible PERF_SAMPLE_REGS will have to transfer all regs. 569 * A possible PERF_SAMPLE_REGS will have to transfer all regs.
570 */ 570 */
571 regs = *iregs; 571 regs = *iregs;
572 regs.ip = pebs->ip; 572 regs.flags = pebs->flags;
573 set_linear_ip(&regs, pebs->ip);
573 regs.bp = pebs->bp; 574 regs.bp = pebs->bp;
574 regs.sp = pebs->sp; 575 regs.sp = pebs->sp;
575 576
diff --git a/arch/x86/kernel/cpu/perf_event_intel_lbr.c b/arch/x86/kernel/cpu/perf_event_intel_lbr.c
index 520b4265fcd2..da02e9cc3754 100644
--- a/arch/x86/kernel/cpu/perf_event_intel_lbr.c
+++ b/arch/x86/kernel/cpu/perf_event_intel_lbr.c
@@ -686,7 +686,8 @@ void intel_pmu_lbr_init_atom(void)
686 * to have an operational LBR which can freeze 686 * to have an operational LBR which can freeze
687 * on PMU interrupt 687 * on PMU interrupt
688 */ 688 */
689 if (boot_cpu_data.x86_mask < 10) { 689 if (boot_cpu_data.x86_model == 28
690 && boot_cpu_data.x86_mask < 10) {
690 pr_cont("LBR disabled due to erratum"); 691 pr_cont("LBR disabled due to erratum");
691 return; 692 return;
692 } 693 }
diff --git a/arch/x86/kernel/cpu/perf_event_intel_uncore.c b/arch/x86/kernel/cpu/perf_event_intel_uncore.c
index 7563fda9f033..0a5571080e74 100644
--- a/arch/x86/kernel/cpu/perf_event_intel_uncore.c
+++ b/arch/x86/kernel/cpu/perf_event_intel_uncore.c
@@ -796,7 +796,6 @@ static struct intel_uncore_type *nhm_msr_uncores[] = {
796 796
797DEFINE_UNCORE_FORMAT_ATTR(event5, event, "config:1-5"); 797DEFINE_UNCORE_FORMAT_ATTR(event5, event, "config:1-5");
798DEFINE_UNCORE_FORMAT_ATTR(counter, counter, "config:6-7"); 798DEFINE_UNCORE_FORMAT_ATTR(counter, counter, "config:6-7");
799DEFINE_UNCORE_FORMAT_ATTR(mm_cfg, mm_cfg, "config:63");
800DEFINE_UNCORE_FORMAT_ATTR(match, match, "config1:0-63"); 799DEFINE_UNCORE_FORMAT_ATTR(match, match, "config1:0-63");
801DEFINE_UNCORE_FORMAT_ATTR(mask, mask, "config2:0-63"); 800DEFINE_UNCORE_FORMAT_ATTR(mask, mask, "config2:0-63");
802 801
@@ -902,16 +901,21 @@ static struct attribute_group nhmex_uncore_cbox_format_group = {
902 .attrs = nhmex_uncore_cbox_formats_attr, 901 .attrs = nhmex_uncore_cbox_formats_attr,
903}; 902};
904 903
904/* msr offset for each instance of cbox */
905static unsigned nhmex_cbox_msr_offsets[] = {
906 0x0, 0x80, 0x40, 0xc0, 0x20, 0xa0, 0x60, 0xe0, 0x240, 0x2c0,
907};
908
905static struct intel_uncore_type nhmex_uncore_cbox = { 909static struct intel_uncore_type nhmex_uncore_cbox = {
906 .name = "cbox", 910 .name = "cbox",
907 .num_counters = 6, 911 .num_counters = 6,
908 .num_boxes = 8, 912 .num_boxes = 10,
909 .perf_ctr_bits = 48, 913 .perf_ctr_bits = 48,
910 .event_ctl = NHMEX_C0_MSR_PMON_EV_SEL0, 914 .event_ctl = NHMEX_C0_MSR_PMON_EV_SEL0,
911 .perf_ctr = NHMEX_C0_MSR_PMON_CTR0, 915 .perf_ctr = NHMEX_C0_MSR_PMON_CTR0,
912 .event_mask = NHMEX_PMON_RAW_EVENT_MASK, 916 .event_mask = NHMEX_PMON_RAW_EVENT_MASK,
913 .box_ctl = NHMEX_C0_MSR_PMON_GLOBAL_CTL, 917 .box_ctl = NHMEX_C0_MSR_PMON_GLOBAL_CTL,
914 .msr_offset = NHMEX_C_MSR_OFFSET, 918 .msr_offsets = nhmex_cbox_msr_offsets,
915 .pair_ctr_ctl = 1, 919 .pair_ctr_ctl = 1,
916 .ops = &nhmex_uncore_ops, 920 .ops = &nhmex_uncore_ops,
917 .format_group = &nhmex_uncore_cbox_format_group 921 .format_group = &nhmex_uncore_cbox_format_group
@@ -1032,24 +1036,22 @@ static struct intel_uncore_type nhmex_uncore_bbox = {
1032 1036
1033static int nhmex_sbox_hw_config(struct intel_uncore_box *box, struct perf_event *event) 1037static int nhmex_sbox_hw_config(struct intel_uncore_box *box, struct perf_event *event)
1034{ 1038{
1035 struct hw_perf_event_extra *reg1 = &event->hw.extra_reg; 1039 struct hw_perf_event *hwc = &event->hw;
1036 struct hw_perf_event_extra *reg2 = &event->hw.branch_reg; 1040 struct hw_perf_event_extra *reg1 = &hwc->extra_reg;
1041 struct hw_perf_event_extra *reg2 = &hwc->branch_reg;
1037 1042
1038 if (event->attr.config & NHMEX_S_PMON_MM_CFG_EN) { 1043 /* only TO_R_PROG_EV event uses the match/mask register */
1039 reg1->config = event->attr.config1; 1044 if ((hwc->config & NHMEX_PMON_CTL_EV_SEL_MASK) !=
1040 reg2->config = event->attr.config2; 1045 NHMEX_S_EVENT_TO_R_PROG_EV)
1041 } else { 1046 return 0;
1042 reg1->config = ~0ULL;
1043 reg2->config = ~0ULL;
1044 }
1045 1047
1046 if (box->pmu->pmu_idx == 0) 1048 if (box->pmu->pmu_idx == 0)
1047 reg1->reg = NHMEX_S0_MSR_MM_CFG; 1049 reg1->reg = NHMEX_S0_MSR_MM_CFG;
1048 else 1050 else
1049 reg1->reg = NHMEX_S1_MSR_MM_CFG; 1051 reg1->reg = NHMEX_S1_MSR_MM_CFG;
1050
1051 reg1->idx = 0; 1052 reg1->idx = 0;
1052 1053 reg1->config = event->attr.config1;
1054 reg2->config = event->attr.config2;
1053 return 0; 1055 return 0;
1054} 1056}
1055 1057
@@ -1059,8 +1061,8 @@ static void nhmex_sbox_msr_enable_event(struct intel_uncore_box *box, struct per
1059 struct hw_perf_event_extra *reg1 = &hwc->extra_reg; 1061 struct hw_perf_event_extra *reg1 = &hwc->extra_reg;
1060 struct hw_perf_event_extra *reg2 = &hwc->branch_reg; 1062 struct hw_perf_event_extra *reg2 = &hwc->branch_reg;
1061 1063
1062 wrmsrl(reg1->reg, 0); 1064 if (reg1->idx != EXTRA_REG_NONE) {
1063 if (reg1->config != ~0ULL || reg2->config != ~0ULL) { 1065 wrmsrl(reg1->reg, 0);
1064 wrmsrl(reg1->reg + 1, reg1->config); 1066 wrmsrl(reg1->reg + 1, reg1->config);
1065 wrmsrl(reg1->reg + 2, reg2->config); 1067 wrmsrl(reg1->reg + 2, reg2->config);
1066 wrmsrl(reg1->reg, NHMEX_S_PMON_MM_CFG_EN); 1068 wrmsrl(reg1->reg, NHMEX_S_PMON_MM_CFG_EN);
@@ -1074,7 +1076,6 @@ static struct attribute *nhmex_uncore_sbox_formats_attr[] = {
1074 &format_attr_edge.attr, 1076 &format_attr_edge.attr,
1075 &format_attr_inv.attr, 1077 &format_attr_inv.attr,
1076 &format_attr_thresh8.attr, 1078 &format_attr_thresh8.attr,
1077 &format_attr_mm_cfg.attr,
1078 &format_attr_match.attr, 1079 &format_attr_match.attr,
1079 &format_attr_mask.attr, 1080 &format_attr_mask.attr,
1080 NULL, 1081 NULL,
@@ -1142,6 +1143,9 @@ static struct extra_reg nhmex_uncore_mbox_extra_regs[] = {
1142 EVENT_EXTRA_END 1143 EVENT_EXTRA_END
1143}; 1144};
1144 1145
1146/* Nehalem-EX or Westmere-EX ? */
1147bool uncore_nhmex;
1148
1145static bool nhmex_mbox_get_shared_reg(struct intel_uncore_box *box, int idx, u64 config) 1149static bool nhmex_mbox_get_shared_reg(struct intel_uncore_box *box, int idx, u64 config)
1146{ 1150{
1147 struct intel_uncore_extra_reg *er; 1151 struct intel_uncore_extra_reg *er;
@@ -1171,18 +1175,29 @@ static bool nhmex_mbox_get_shared_reg(struct intel_uncore_box *box, int idx, u64
1171 return false; 1175 return false;
1172 1176
1173 /* mask of the shared fields */ 1177 /* mask of the shared fields */
1174 mask = NHMEX_M_PMON_ZDP_CTL_FVC_MASK; 1178 if (uncore_nhmex)
1179 mask = NHMEX_M_PMON_ZDP_CTL_FVC_MASK;
1180 else
1181 mask = WSMEX_M_PMON_ZDP_CTL_FVC_MASK;
1175 er = &box->shared_regs[EXTRA_REG_NHMEX_M_ZDP_CTL_FVC]; 1182 er = &box->shared_regs[EXTRA_REG_NHMEX_M_ZDP_CTL_FVC];
1176 1183
1177 raw_spin_lock_irqsave(&er->lock, flags); 1184 raw_spin_lock_irqsave(&er->lock, flags);
1178 /* add mask of the non-shared field if it's in use */ 1185 /* add mask of the non-shared field if it's in use */
1179 if (__BITS_VALUE(atomic_read(&er->ref), idx, 8)) 1186 if (__BITS_VALUE(atomic_read(&er->ref), idx, 8)) {
1180 mask |= NHMEX_M_PMON_ZDP_CTL_FVC_EVENT_MASK(idx); 1187 if (uncore_nhmex)
1188 mask |= NHMEX_M_PMON_ZDP_CTL_FVC_EVENT_MASK(idx);
1189 else
1190 mask |= WSMEX_M_PMON_ZDP_CTL_FVC_EVENT_MASK(idx);
1191 }
1181 1192
1182 if (!atomic_read(&er->ref) || !((er->config ^ config) & mask)) { 1193 if (!atomic_read(&er->ref) || !((er->config ^ config) & mask)) {
1183 atomic_add(1 << (idx * 8), &er->ref); 1194 atomic_add(1 << (idx * 8), &er->ref);
1184 mask = NHMEX_M_PMON_ZDP_CTL_FVC_MASK | 1195 if (uncore_nhmex)
1185 NHMEX_M_PMON_ZDP_CTL_FVC_EVENT_MASK(idx); 1196 mask = NHMEX_M_PMON_ZDP_CTL_FVC_MASK |
1197 NHMEX_M_PMON_ZDP_CTL_FVC_EVENT_MASK(idx);
1198 else
1199 mask = WSMEX_M_PMON_ZDP_CTL_FVC_MASK |
1200 WSMEX_M_PMON_ZDP_CTL_FVC_EVENT_MASK(idx);
1186 er->config &= ~mask; 1201 er->config &= ~mask;
1187 er->config |= (config & mask); 1202 er->config |= (config & mask);
1188 ret = true; 1203 ret = true;
@@ -1216,7 +1231,10 @@ u64 nhmex_mbox_alter_er(struct perf_event *event, int new_idx, bool modify)
1216 1231
1217 /* get the non-shared control bits and shift them */ 1232 /* get the non-shared control bits and shift them */
1218 idx = orig_idx - EXTRA_REG_NHMEX_M_ZDP_CTL_FVC; 1233 idx = orig_idx - EXTRA_REG_NHMEX_M_ZDP_CTL_FVC;
1219 config &= NHMEX_M_PMON_ZDP_CTL_FVC_EVENT_MASK(idx); 1234 if (uncore_nhmex)
1235 config &= NHMEX_M_PMON_ZDP_CTL_FVC_EVENT_MASK(idx);
1236 else
1237 config &= WSMEX_M_PMON_ZDP_CTL_FVC_EVENT_MASK(idx);
1220 if (new_idx > orig_idx) { 1238 if (new_idx > orig_idx) {
1221 idx = new_idx - orig_idx; 1239 idx = new_idx - orig_idx;
1222 config <<= 3 * idx; 1240 config <<= 3 * idx;
@@ -1226,6 +1244,10 @@ u64 nhmex_mbox_alter_er(struct perf_event *event, int new_idx, bool modify)
1226 } 1244 }
1227 1245
1228 /* add the shared control bits back */ 1246 /* add the shared control bits back */
1247 if (uncore_nhmex)
1248 config |= NHMEX_M_PMON_ZDP_CTL_FVC_MASK & reg1->config;
1249 else
1250 config |= WSMEX_M_PMON_ZDP_CTL_FVC_MASK & reg1->config;
1229 config |= NHMEX_M_PMON_ZDP_CTL_FVC_MASK & reg1->config; 1251 config |= NHMEX_M_PMON_ZDP_CTL_FVC_MASK & reg1->config;
1230 if (modify) { 1252 if (modify) {
1231 /* adjust the main event selector */ 1253 /* adjust the main event selector */
@@ -1264,7 +1286,8 @@ again:
1264 } 1286 }
1265 1287
1266 /* for the match/mask registers */ 1288 /* for the match/mask registers */
1267 if ((uncore_box_is_fake(box) || !reg2->alloc) && 1289 if (reg2->idx != EXTRA_REG_NONE &&
1290 (uncore_box_is_fake(box) || !reg2->alloc) &&
1268 !nhmex_mbox_get_shared_reg(box, reg2->idx, reg2->config)) 1291 !nhmex_mbox_get_shared_reg(box, reg2->idx, reg2->config))
1269 goto fail; 1292 goto fail;
1270 1293
@@ -1278,7 +1301,8 @@ again:
1278 if (idx[0] != 0xff && idx[0] != __BITS_VALUE(reg1->idx, 0, 8)) 1301 if (idx[0] != 0xff && idx[0] != __BITS_VALUE(reg1->idx, 0, 8))
1279 nhmex_mbox_alter_er(event, idx[0], true); 1302 nhmex_mbox_alter_er(event, idx[0], true);
1280 reg1->alloc |= alloc; 1303 reg1->alloc |= alloc;
1281 reg2->alloc = 1; 1304 if (reg2->idx != EXTRA_REG_NONE)
1305 reg2->alloc = 1;
1282 } 1306 }
1283 return NULL; 1307 return NULL;
1284fail: 1308fail:
@@ -1342,9 +1366,6 @@ static int nhmex_mbox_hw_config(struct intel_uncore_box *box, struct perf_event
1342 struct extra_reg *er; 1366 struct extra_reg *er;
1343 unsigned msr; 1367 unsigned msr;
1344 int reg_idx = 0; 1368 int reg_idx = 0;
1345
1346 if (WARN_ON_ONCE(reg1->idx != -1))
1347 return -EINVAL;
1348 /* 1369 /*
1349 * The mbox events may require 2 extra MSRs at the most. But only 1370 * The mbox events may require 2 extra MSRs at the most. But only
1350 * the lower 32 bits in these MSRs are significant, so we can use 1371 * the lower 32 bits in these MSRs are significant, so we can use
@@ -1355,11 +1376,6 @@ static int nhmex_mbox_hw_config(struct intel_uncore_box *box, struct perf_event
1355 continue; 1376 continue;
1356 if (event->attr.config1 & ~er->valid_mask) 1377 if (event->attr.config1 & ~er->valid_mask)
1357 return -EINVAL; 1378 return -EINVAL;
1358 if (er->idx == __BITS_VALUE(reg1->idx, 0, 8) ||
1359 er->idx == __BITS_VALUE(reg1->idx, 1, 8))
1360 continue;
1361 if (WARN_ON_ONCE(reg_idx >= 2))
1362 return -EINVAL;
1363 1379
1364 msr = er->msr + type->msr_offset * box->pmu->pmu_idx; 1380 msr = er->msr + type->msr_offset * box->pmu->pmu_idx;
1365 if (WARN_ON_ONCE(msr >= 0xffff || er->idx >= 0xff)) 1381 if (WARN_ON_ONCE(msr >= 0xffff || er->idx >= 0xff))
@@ -1368,6 +1384,8 @@ static int nhmex_mbox_hw_config(struct intel_uncore_box *box, struct perf_event
1368 /* always use the 32~63 bits to pass the PLD config */ 1384 /* always use the 32~63 bits to pass the PLD config */
1369 if (er->idx == EXTRA_REG_NHMEX_M_PLD) 1385 if (er->idx == EXTRA_REG_NHMEX_M_PLD)
1370 reg_idx = 1; 1386 reg_idx = 1;
1387 else if (WARN_ON_ONCE(reg_idx > 0))
1388 return -EINVAL;
1371 1389
1372 reg1->idx &= ~(0xff << (reg_idx * 8)); 1390 reg1->idx &= ~(0xff << (reg_idx * 8));
1373 reg1->reg &= ~(0xffff << (reg_idx * 16)); 1391 reg1->reg &= ~(0xffff << (reg_idx * 16));
@@ -1376,17 +1394,21 @@ static int nhmex_mbox_hw_config(struct intel_uncore_box *box, struct perf_event
1376 reg1->config = event->attr.config1; 1394 reg1->config = event->attr.config1;
1377 reg_idx++; 1395 reg_idx++;
1378 } 1396 }
1379 /* use config2 to pass the filter config */ 1397 /*
1380 reg2->idx = EXTRA_REG_NHMEX_M_FILTER; 1398 * The mbox only provides ability to perform address matching
1381 if (event->attr.config2 & NHMEX_M_PMON_MM_CFG_EN) 1399 * for the PLD events.
1382 reg2->config = event->attr.config2; 1400 */
1383 else 1401 if (reg_idx == 2) {
1384 reg2->config = ~0ULL; 1402 reg2->idx = EXTRA_REG_NHMEX_M_FILTER;
1385 if (box->pmu->pmu_idx == 0) 1403 if (event->attr.config2 & NHMEX_M_PMON_MM_CFG_EN)
1386 reg2->reg = NHMEX_M0_MSR_PMU_MM_CFG; 1404 reg2->config = event->attr.config2;
1387 else 1405 else
1388 reg2->reg = NHMEX_M1_MSR_PMU_MM_CFG; 1406 reg2->config = ~0ULL;
1389 1407 if (box->pmu->pmu_idx == 0)
1408 reg2->reg = NHMEX_M0_MSR_PMU_MM_CFG;
1409 else
1410 reg2->reg = NHMEX_M1_MSR_PMU_MM_CFG;
1411 }
1390 return 0; 1412 return 0;
1391} 1413}
1392 1414
@@ -1422,34 +1444,36 @@ static void nhmex_mbox_msr_enable_event(struct intel_uncore_box *box, struct per
1422 wrmsrl(__BITS_VALUE(reg1->reg, 1, 16), 1444 wrmsrl(__BITS_VALUE(reg1->reg, 1, 16),
1423 nhmex_mbox_shared_reg_config(box, idx)); 1445 nhmex_mbox_shared_reg_config(box, idx));
1424 1446
1425 wrmsrl(reg2->reg, 0); 1447 if (reg2->idx != EXTRA_REG_NONE) {
1426 if (reg2->config != ~0ULL) { 1448 wrmsrl(reg2->reg, 0);
1427 wrmsrl(reg2->reg + 1, 1449 if (reg2->config != ~0ULL) {
1428 reg2->config & NHMEX_M_PMON_ADDR_MATCH_MASK); 1450 wrmsrl(reg2->reg + 1,
1429 wrmsrl(reg2->reg + 2, NHMEX_M_PMON_ADDR_MASK_MASK & 1451 reg2->config & NHMEX_M_PMON_ADDR_MATCH_MASK);
1430 (reg2->config >> NHMEX_M_PMON_ADDR_MASK_SHIFT)); 1452 wrmsrl(reg2->reg + 2, NHMEX_M_PMON_ADDR_MASK_MASK &
1431 wrmsrl(reg2->reg, NHMEX_M_PMON_MM_CFG_EN); 1453 (reg2->config >> NHMEX_M_PMON_ADDR_MASK_SHIFT));
1454 wrmsrl(reg2->reg, NHMEX_M_PMON_MM_CFG_EN);
1455 }
1432 } 1456 }
1433 1457
1434 wrmsrl(hwc->config_base, hwc->config | NHMEX_PMON_CTL_EN_BIT0); 1458 wrmsrl(hwc->config_base, hwc->config | NHMEX_PMON_CTL_EN_BIT0);
1435} 1459}
1436 1460
1437DEFINE_UNCORE_FORMAT_ATTR(count_mode, count_mode, "config:2-3"); 1461DEFINE_UNCORE_FORMAT_ATTR(count_mode, count_mode, "config:2-3");
1438DEFINE_UNCORE_FORMAT_ATTR(storage_mode, storage_mode, "config:4-5"); 1462DEFINE_UNCORE_FORMAT_ATTR(storage_mode, storage_mode, "config:4-5");
1439DEFINE_UNCORE_FORMAT_ATTR(wrap_mode, wrap_mode, "config:6"); 1463DEFINE_UNCORE_FORMAT_ATTR(wrap_mode, wrap_mode, "config:6");
1440DEFINE_UNCORE_FORMAT_ATTR(flag_mode, flag_mode, "config:7"); 1464DEFINE_UNCORE_FORMAT_ATTR(flag_mode, flag_mode, "config:7");
1441DEFINE_UNCORE_FORMAT_ATTR(inc_sel, inc_sel, "config:9-13"); 1465DEFINE_UNCORE_FORMAT_ATTR(inc_sel, inc_sel, "config:9-13");
1442DEFINE_UNCORE_FORMAT_ATTR(set_flag_sel, set_flag_sel, "config:19-21"); 1466DEFINE_UNCORE_FORMAT_ATTR(set_flag_sel, set_flag_sel, "config:19-21");
1443DEFINE_UNCORE_FORMAT_ATTR(filter_cfg, filter_cfg, "config2:63"); 1467DEFINE_UNCORE_FORMAT_ATTR(filter_cfg_en, filter_cfg_en, "config2:63");
1444DEFINE_UNCORE_FORMAT_ATTR(filter_match, filter_match, "config2:0-33"); 1468DEFINE_UNCORE_FORMAT_ATTR(filter_match, filter_match, "config2:0-33");
1445DEFINE_UNCORE_FORMAT_ATTR(filter_mask, filter_mask, "config2:34-61"); 1469DEFINE_UNCORE_FORMAT_ATTR(filter_mask, filter_mask, "config2:34-61");
1446DEFINE_UNCORE_FORMAT_ATTR(dsp, dsp, "config1:0-31"); 1470DEFINE_UNCORE_FORMAT_ATTR(dsp, dsp, "config1:0-31");
1447DEFINE_UNCORE_FORMAT_ATTR(thr, thr, "config1:0-31"); 1471DEFINE_UNCORE_FORMAT_ATTR(thr, thr, "config1:0-31");
1448DEFINE_UNCORE_FORMAT_ATTR(fvc, fvc, "config1:0-31"); 1472DEFINE_UNCORE_FORMAT_ATTR(fvc, fvc, "config1:0-31");
1449DEFINE_UNCORE_FORMAT_ATTR(pgt, pgt, "config1:0-31"); 1473DEFINE_UNCORE_FORMAT_ATTR(pgt, pgt, "config1:0-31");
1450DEFINE_UNCORE_FORMAT_ATTR(map, map, "config1:0-31"); 1474DEFINE_UNCORE_FORMAT_ATTR(map, map, "config1:0-31");
1451DEFINE_UNCORE_FORMAT_ATTR(iss, iss, "config1:0-31"); 1475DEFINE_UNCORE_FORMAT_ATTR(iss, iss, "config1:0-31");
1452DEFINE_UNCORE_FORMAT_ATTR(pld, pld, "config1:32-63"); 1476DEFINE_UNCORE_FORMAT_ATTR(pld, pld, "config1:32-63");
1453 1477
1454static struct attribute *nhmex_uncore_mbox_formats_attr[] = { 1478static struct attribute *nhmex_uncore_mbox_formats_attr[] = {
1455 &format_attr_count_mode.attr, 1479 &format_attr_count_mode.attr,
@@ -1458,7 +1482,7 @@ static struct attribute *nhmex_uncore_mbox_formats_attr[] = {
1458 &format_attr_flag_mode.attr, 1482 &format_attr_flag_mode.attr,
1459 &format_attr_inc_sel.attr, 1483 &format_attr_inc_sel.attr,
1460 &format_attr_set_flag_sel.attr, 1484 &format_attr_set_flag_sel.attr,
1461 &format_attr_filter_cfg.attr, 1485 &format_attr_filter_cfg_en.attr,
1462 &format_attr_filter_match.attr, 1486 &format_attr_filter_match.attr,
1463 &format_attr_filter_mask.attr, 1487 &format_attr_filter_mask.attr,
1464 &format_attr_dsp.attr, 1488 &format_attr_dsp.attr,
@@ -1482,6 +1506,12 @@ static struct uncore_event_desc nhmex_uncore_mbox_events[] = {
1482 { /* end: all zeroes */ }, 1506 { /* end: all zeroes */ },
1483}; 1507};
1484 1508
1509static struct uncore_event_desc wsmex_uncore_mbox_events[] = {
1510 INTEL_UNCORE_EVENT_DESC(bbox_cmds_read, "inc_sel=0xd,fvc=0x5000"),
1511 INTEL_UNCORE_EVENT_DESC(bbox_cmds_write, "inc_sel=0xd,fvc=0x5040"),
1512 { /* end: all zeroes */ },
1513};
1514
1485static struct intel_uncore_ops nhmex_uncore_mbox_ops = { 1515static struct intel_uncore_ops nhmex_uncore_mbox_ops = {
1486 NHMEX_UNCORE_OPS_COMMON_INIT(), 1516 NHMEX_UNCORE_OPS_COMMON_INIT(),
1487 .enable_event = nhmex_mbox_msr_enable_event, 1517 .enable_event = nhmex_mbox_msr_enable_event,
@@ -1513,7 +1543,7 @@ void nhmex_rbox_alter_er(struct intel_uncore_box *box, struct perf_event *event)
1513 struct hw_perf_event_extra *reg1 = &hwc->extra_reg; 1543 struct hw_perf_event_extra *reg1 = &hwc->extra_reg;
1514 int port; 1544 int port;
1515 1545
1516 /* adjust the main event selector */ 1546 /* adjust the main event selector and extra register index */
1517 if (reg1->idx % 2) { 1547 if (reg1->idx % 2) {
1518 reg1->idx--; 1548 reg1->idx--;
1519 hwc->config -= 1 << NHMEX_R_PMON_CTL_EV_SEL_SHIFT; 1549 hwc->config -= 1 << NHMEX_R_PMON_CTL_EV_SEL_SHIFT;
@@ -1522,29 +1552,17 @@ void nhmex_rbox_alter_er(struct intel_uncore_box *box, struct perf_event *event)
1522 hwc->config += 1 << NHMEX_R_PMON_CTL_EV_SEL_SHIFT; 1552 hwc->config += 1 << NHMEX_R_PMON_CTL_EV_SEL_SHIFT;
1523 } 1553 }
1524 1554
1525 /* adjust address or config of extra register */ 1555 /* adjust extra register config */
1526 port = reg1->idx / 6 + box->pmu->pmu_idx * 4; 1556 port = reg1->idx / 6 + box->pmu->pmu_idx * 4;
1527 switch (reg1->idx % 6) { 1557 switch (reg1->idx % 6) {
1528 case 0:
1529 reg1->reg = NHMEX_R_MSR_PORTN_IPERF_CFG0(port);
1530 break;
1531 case 1:
1532 reg1->reg = NHMEX_R_MSR_PORTN_IPERF_CFG1(port);
1533 break;
1534 case 2: 1558 case 2:
1535 /* the 8~15 bits to the 0~7 bits */ 1559 /* shift the 8~15 bits to the 0~7 bits */
1536 reg1->config >>= 8; 1560 reg1->config >>= 8;
1537 break; 1561 break;
1538 case 3: 1562 case 3:
1539 /* the 0~7 bits to the 8~15 bits */ 1563 /* shift the 0~7 bits to the 8~15 bits */
1540 reg1->config <<= 8; 1564 reg1->config <<= 8;
1541 break; 1565 break;
1542 case 4:
1543 reg1->reg = NHMEX_R_MSR_PORTN_XBR_SET1_MM_CFG(port);
1544 break;
1545 case 5:
1546 reg1->reg = NHMEX_R_MSR_PORTN_XBR_SET2_MM_CFG(port);
1547 break;
1548 }; 1566 };
1549} 1567}
1550 1568
@@ -1671,7 +1689,7 @@ static int nhmex_rbox_hw_config(struct intel_uncore_box *box, struct perf_event
1671 struct hw_perf_event *hwc = &event->hw; 1689 struct hw_perf_event *hwc = &event->hw;
1672 struct hw_perf_event_extra *reg1 = &event->hw.extra_reg; 1690 struct hw_perf_event_extra *reg1 = &event->hw.extra_reg;
1673 struct hw_perf_event_extra *reg2 = &event->hw.branch_reg; 1691 struct hw_perf_event_extra *reg2 = &event->hw.branch_reg;
1674 int port, idx; 1692 int idx;
1675 1693
1676 idx = (event->hw.config & NHMEX_R_PMON_CTL_EV_SEL_MASK) >> 1694 idx = (event->hw.config & NHMEX_R_PMON_CTL_EV_SEL_MASK) >>
1677 NHMEX_R_PMON_CTL_EV_SEL_SHIFT; 1695 NHMEX_R_PMON_CTL_EV_SEL_SHIFT;
@@ -1681,27 +1699,11 @@ static int nhmex_rbox_hw_config(struct intel_uncore_box *box, struct perf_event
1681 reg1->idx = idx; 1699 reg1->idx = idx;
1682 reg1->config = event->attr.config1; 1700 reg1->config = event->attr.config1;
1683 1701
1684 port = idx / 6 + box->pmu->pmu_idx * 4; 1702 switch (idx % 6) {
1685 idx %= 6;
1686 switch (idx) {
1687 case 0:
1688 reg1->reg = NHMEX_R_MSR_PORTN_IPERF_CFG0(port);
1689 break;
1690 case 1:
1691 reg1->reg = NHMEX_R_MSR_PORTN_IPERF_CFG1(port);
1692 break;
1693 case 2:
1694 case 3:
1695 reg1->reg = NHMEX_R_MSR_PORTN_QLX_CFG(port);
1696 break;
1697 case 4: 1703 case 4:
1698 case 5: 1704 case 5:
1699 if (idx == 4)
1700 reg1->reg = NHMEX_R_MSR_PORTN_XBR_SET1_MM_CFG(port);
1701 else
1702 reg1->reg = NHMEX_R_MSR_PORTN_XBR_SET2_MM_CFG(port);
1703 reg2->config = event->attr.config2;
1704 hwc->config |= event->attr.config & (~0ULL << 32); 1705 hwc->config |= event->attr.config & (~0ULL << 32);
1706 reg2->config = event->attr.config2;
1705 break; 1707 break;
1706 }; 1708 };
1707 return 0; 1709 return 0;
@@ -1727,28 +1729,34 @@ static void nhmex_rbox_msr_enable_event(struct intel_uncore_box *box, struct per
1727 struct hw_perf_event *hwc = &event->hw; 1729 struct hw_perf_event *hwc = &event->hw;
1728 struct hw_perf_event_extra *reg1 = &hwc->extra_reg; 1730 struct hw_perf_event_extra *reg1 = &hwc->extra_reg;
1729 struct hw_perf_event_extra *reg2 = &hwc->branch_reg; 1731 struct hw_perf_event_extra *reg2 = &hwc->branch_reg;
1730 int idx, er_idx; 1732 int idx, port;
1731 1733
1732 idx = reg1->idx % 6; 1734 idx = reg1->idx;
1733 er_idx = idx; 1735 port = idx / 6 + box->pmu->pmu_idx * 4;
1734 if (er_idx > 2)
1735 er_idx--;
1736 er_idx += (reg1->idx / 6) * 5;
1737 1736
1738 switch (idx) { 1737 switch (idx % 6) {
1739 case 0: 1738 case 0:
1739 wrmsrl(NHMEX_R_MSR_PORTN_IPERF_CFG0(port), reg1->config);
1740 break;
1740 case 1: 1741 case 1:
1741 wrmsrl(reg1->reg, reg1->config); 1742 wrmsrl(NHMEX_R_MSR_PORTN_IPERF_CFG1(port), reg1->config);
1742 break; 1743 break;
1743 case 2: 1744 case 2:
1744 case 3: 1745 case 3:
1745 wrmsrl(reg1->reg, nhmex_rbox_shared_reg_config(box, er_idx)); 1746 wrmsrl(NHMEX_R_MSR_PORTN_QLX_CFG(port),
1747 nhmex_rbox_shared_reg_config(box, 2 + (idx / 6) * 5));
1746 break; 1748 break;
1747 case 4: 1749 case 4:
1750 wrmsrl(NHMEX_R_MSR_PORTN_XBR_SET1_MM_CFG(port),
1751 hwc->config >> 32);
1752 wrmsrl(NHMEX_R_MSR_PORTN_XBR_SET1_MATCH(port), reg1->config);
1753 wrmsrl(NHMEX_R_MSR_PORTN_XBR_SET1_MASK(port), reg2->config);
1754 break;
1748 case 5: 1755 case 5:
1749 wrmsrl(reg1->reg, reg1->config); 1756 wrmsrl(NHMEX_R_MSR_PORTN_XBR_SET2_MM_CFG(port),
1750 wrmsrl(reg1->reg + 1, hwc->config >> 32); 1757 hwc->config >> 32);
1751 wrmsrl(reg1->reg + 2, reg2->config); 1758 wrmsrl(NHMEX_R_MSR_PORTN_XBR_SET2_MATCH(port), reg1->config);
1759 wrmsrl(NHMEX_R_MSR_PORTN_XBR_SET2_MASK(port), reg2->config);
1752 break; 1760 break;
1753 }; 1761 };
1754 1762
@@ -1756,8 +1764,8 @@ static void nhmex_rbox_msr_enable_event(struct intel_uncore_box *box, struct per
1756 (hwc->config & NHMEX_R_PMON_CTL_EV_SEL_MASK)); 1764 (hwc->config & NHMEX_R_PMON_CTL_EV_SEL_MASK));
1757} 1765}
1758 1766
1759DEFINE_UNCORE_FORMAT_ATTR(xbr_match, xbr_match, "config:32-63"); 1767DEFINE_UNCORE_FORMAT_ATTR(xbr_mm_cfg, xbr_mm_cfg, "config:32-63");
1760DEFINE_UNCORE_FORMAT_ATTR(xbr_mm_cfg, xbr_mm_cfg, "config1:0-63"); 1768DEFINE_UNCORE_FORMAT_ATTR(xbr_match, xbr_match, "config1:0-63");
1761DEFINE_UNCORE_FORMAT_ATTR(xbr_mask, xbr_mask, "config2:0-63"); 1769DEFINE_UNCORE_FORMAT_ATTR(xbr_mask, xbr_mask, "config2:0-63");
1762DEFINE_UNCORE_FORMAT_ATTR(qlx_cfg, qlx_cfg, "config1:0-15"); 1770DEFINE_UNCORE_FORMAT_ATTR(qlx_cfg, qlx_cfg, "config1:0-15");
1763DEFINE_UNCORE_FORMAT_ATTR(iperf_cfg, iperf_cfg, "config1:0-31"); 1771DEFINE_UNCORE_FORMAT_ATTR(iperf_cfg, iperf_cfg, "config1:0-31");
@@ -2303,6 +2311,7 @@ int uncore_pmu_event_init(struct perf_event *event)
2303 event->hw.idx = -1; 2311 event->hw.idx = -1;
2304 event->hw.last_tag = ~0ULL; 2312 event->hw.last_tag = ~0ULL;
2305 event->hw.extra_reg.idx = EXTRA_REG_NONE; 2313 event->hw.extra_reg.idx = EXTRA_REG_NONE;
2314 event->hw.branch_reg.idx = EXTRA_REG_NONE;
2306 2315
2307 if (event->attr.config == UNCORE_FIXED_EVENT) { 2316 if (event->attr.config == UNCORE_FIXED_EVENT) {
2308 /* no fixed counter */ 2317 /* no fixed counter */
@@ -2373,7 +2382,7 @@ static void __init uncore_type_exit(struct intel_uncore_type *type)
2373 type->attr_groups[1] = NULL; 2382 type->attr_groups[1] = NULL;
2374} 2383}
2375 2384
2376static void uncore_types_exit(struct intel_uncore_type **types) 2385static void __init uncore_types_exit(struct intel_uncore_type **types)
2377{ 2386{
2378 int i; 2387 int i;
2379 for (i = 0; types[i]; i++) 2388 for (i = 0; types[i]; i++)
@@ -2814,7 +2823,13 @@ static int __init uncore_cpu_init(void)
2814 snbep_uncore_cbox.num_boxes = max_cores; 2823 snbep_uncore_cbox.num_boxes = max_cores;
2815 msr_uncores = snbep_msr_uncores; 2824 msr_uncores = snbep_msr_uncores;
2816 break; 2825 break;
2817 case 46: 2826 case 46: /* Nehalem-EX */
2827 uncore_nhmex = true;
2828 case 47: /* Westmere-EX aka. Xeon E7 */
2829 if (!uncore_nhmex)
2830 nhmex_uncore_mbox.event_descs = wsmex_uncore_mbox_events;
2831 if (nhmex_uncore_cbox.num_boxes > max_cores)
2832 nhmex_uncore_cbox.num_boxes = max_cores;
2818 msr_uncores = nhmex_msr_uncores; 2833 msr_uncores = nhmex_msr_uncores;
2819 break; 2834 break;
2820 default: 2835 default:
diff --git a/arch/x86/kernel/cpu/perf_event_intel_uncore.h b/arch/x86/kernel/cpu/perf_event_intel_uncore.h
index f3851892e077..5b81c1856aac 100644
--- a/arch/x86/kernel/cpu/perf_event_intel_uncore.h
+++ b/arch/x86/kernel/cpu/perf_event_intel_uncore.h
@@ -5,7 +5,7 @@
5#include "perf_event.h" 5#include "perf_event.h"
6 6
7#define UNCORE_PMU_NAME_LEN 32 7#define UNCORE_PMU_NAME_LEN 32
8#define UNCORE_PMU_HRTIMER_INTERVAL (60 * NSEC_PER_SEC) 8#define UNCORE_PMU_HRTIMER_INTERVAL (60LL * NSEC_PER_SEC)
9 9
10#define UNCORE_FIXED_EVENT 0xff 10#define UNCORE_FIXED_EVENT 0xff
11#define UNCORE_PMC_IDX_MAX_GENERIC 8 11#define UNCORE_PMC_IDX_MAX_GENERIC 8
@@ -230,6 +230,7 @@
230#define NHMEX_S1_MSR_MASK 0xe5a 230#define NHMEX_S1_MSR_MASK 0xe5a
231 231
232#define NHMEX_S_PMON_MM_CFG_EN (0x1ULL << 63) 232#define NHMEX_S_PMON_MM_CFG_EN (0x1ULL << 63)
233#define NHMEX_S_EVENT_TO_R_PROG_EV 0
233 234
234/* NHM-EX Mbox */ 235/* NHM-EX Mbox */
235#define NHMEX_M0_MSR_GLOBAL_CTL 0xca0 236#define NHMEX_M0_MSR_GLOBAL_CTL 0xca0
@@ -275,18 +276,12 @@
275 NHMEX_M_PMON_CTL_INC_SEL_MASK | \ 276 NHMEX_M_PMON_CTL_INC_SEL_MASK | \
276 NHMEX_M_PMON_CTL_SET_FLAG_SEL_MASK) 277 NHMEX_M_PMON_CTL_SET_FLAG_SEL_MASK)
277 278
278 279#define NHMEX_M_PMON_ZDP_CTL_FVC_MASK (((1 << 11) - 1) | (1 << 23))
279#define NHMEX_M_PMON_ZDP_CTL_FVC_FVID_MASK 0x1f
280#define NHMEX_M_PMON_ZDP_CTL_FVC_BCMD_MASK (0x7 << 5)
281#define NHMEX_M_PMON_ZDP_CTL_FVC_RSP_MASK (0x7 << 8)
282#define NHMEX_M_PMON_ZDP_CTL_FVC_PBOX_INIT_ERR (1 << 23)
283#define NHMEX_M_PMON_ZDP_CTL_FVC_MASK \
284 (NHMEX_M_PMON_ZDP_CTL_FVC_FVID_MASK | \
285 NHMEX_M_PMON_ZDP_CTL_FVC_BCMD_MASK | \
286 NHMEX_M_PMON_ZDP_CTL_FVC_RSP_MASK | \
287 NHMEX_M_PMON_ZDP_CTL_FVC_PBOX_INIT_ERR)
288#define NHMEX_M_PMON_ZDP_CTL_FVC_EVENT_MASK(n) (0x7 << (11 + 3 * (n))) 280#define NHMEX_M_PMON_ZDP_CTL_FVC_EVENT_MASK(n) (0x7 << (11 + 3 * (n)))
289 281
282#define WSMEX_M_PMON_ZDP_CTL_FVC_MASK (((1 << 12) - 1) | (1 << 24))
283#define WSMEX_M_PMON_ZDP_CTL_FVC_EVENT_MASK(n) (0x7 << (12 + 3 * (n)))
284
290/* 285/*
291 * use the 9~13 bits to select event If the 7th bit is not set, 286 * use the 9~13 bits to select event If the 7th bit is not set,
292 * otherwise use the 19~21 bits to select event. 287 * otherwise use the 19~21 bits to select event.
@@ -368,6 +363,7 @@ struct intel_uncore_type {
368 unsigned num_shared_regs:8; 363 unsigned num_shared_regs:8;
369 unsigned single_fixed:1; 364 unsigned single_fixed:1;
370 unsigned pair_ctr_ctl:1; 365 unsigned pair_ctr_ctl:1;
366 unsigned *msr_offsets;
371 struct event_constraint unconstrainted; 367 struct event_constraint unconstrainted;
372 struct event_constraint *constraints; 368 struct event_constraint *constraints;
373 struct intel_uncore_pmu *pmus; 369 struct intel_uncore_pmu *pmus;
@@ -485,29 +481,31 @@ unsigned uncore_pci_perf_ctr(struct intel_uncore_box *box, int idx)
485 return idx * 8 + box->pmu->type->perf_ctr; 481 return idx * 8 + box->pmu->type->perf_ctr;
486} 482}
487 483
488static inline 484static inline unsigned uncore_msr_box_offset(struct intel_uncore_box *box)
489unsigned uncore_msr_box_ctl(struct intel_uncore_box *box) 485{
486 struct intel_uncore_pmu *pmu = box->pmu;
487 return pmu->type->msr_offsets ?
488 pmu->type->msr_offsets[pmu->pmu_idx] :
489 pmu->type->msr_offset * pmu->pmu_idx;
490}
491
492static inline unsigned uncore_msr_box_ctl(struct intel_uncore_box *box)
490{ 493{
491 if (!box->pmu->type->box_ctl) 494 if (!box->pmu->type->box_ctl)
492 return 0; 495 return 0;
493 return box->pmu->type->box_ctl + 496 return box->pmu->type->box_ctl + uncore_msr_box_offset(box);
494 box->pmu->type->msr_offset * box->pmu->pmu_idx;
495} 497}
496 498
497static inline 499static inline unsigned uncore_msr_fixed_ctl(struct intel_uncore_box *box)
498unsigned uncore_msr_fixed_ctl(struct intel_uncore_box *box)
499{ 500{
500 if (!box->pmu->type->fixed_ctl) 501 if (!box->pmu->type->fixed_ctl)
501 return 0; 502 return 0;
502 return box->pmu->type->fixed_ctl + 503 return box->pmu->type->fixed_ctl + uncore_msr_box_offset(box);
503 box->pmu->type->msr_offset * box->pmu->pmu_idx;
504} 504}
505 505
506static inline 506static inline unsigned uncore_msr_fixed_ctr(struct intel_uncore_box *box)
507unsigned uncore_msr_fixed_ctr(struct intel_uncore_box *box)
508{ 507{
509 return box->pmu->type->fixed_ctr + 508 return box->pmu->type->fixed_ctr + uncore_msr_box_offset(box);
510 box->pmu->type->msr_offset * box->pmu->pmu_idx;
511} 509}
512 510
513static inline 511static inline
@@ -515,7 +513,7 @@ unsigned uncore_msr_event_ctl(struct intel_uncore_box *box, int idx)
515{ 513{
516 return box->pmu->type->event_ctl + 514 return box->pmu->type->event_ctl +
517 (box->pmu->type->pair_ctr_ctl ? 2 * idx : idx) + 515 (box->pmu->type->pair_ctr_ctl ? 2 * idx : idx) +
518 box->pmu->type->msr_offset * box->pmu->pmu_idx; 516 uncore_msr_box_offset(box);
519} 517}
520 518
521static inline 519static inline
@@ -523,7 +521,7 @@ unsigned uncore_msr_perf_ctr(struct intel_uncore_box *box, int idx)
523{ 521{
524 return box->pmu->type->perf_ctr + 522 return box->pmu->type->perf_ctr +
525 (box->pmu->type->pair_ctr_ctl ? 2 * idx : idx) + 523 (box->pmu->type->pair_ctr_ctl ? 2 * idx : idx) +
526 box->pmu->type->msr_offset * box->pmu->pmu_idx; 524 uncore_msr_box_offset(box);
527} 525}
528 526
529static inline 527static inline