Re: [PATCH 2/8] perf/x86/amd/ibs: Add workaround for erratum #1,197

From
Namhyung Kim <namhyung@kernel.org>
Date
2021-08-19 22:29:26
Also in
linux-perf-users
Reply to
Kim Phillips, 2021-08-17
Hello,

On Tue, Aug 17, 2021 at 3:11 PM Kim Phillips [off-list ref] wrote:
quoted hunk ↗ jump to hunk
Erratum #1197 "IBS (Instruction Based Sampling) Register State May be
Incorrect After Restore From CC6" is published in a document available
at the link tag below:

"Revision Guide for AMD Family 19h Models 00h-0Fh Processors"
56683 Rev. 1.04 July 2021

Implement the erratum's suggested workaround and ignore IBS samples
if MSRC001_1031 == 0.

Link: https://bugzilla.kernel.org/show_bug.cgi?id=206537
Signed-off-by: Kim Phillips <redacted>
Cc: Alexander Shishkin <alexander.shishkin@linux.intel.com>
Cc: Andrew Murray <redacted>
Cc: Arnaldo Carvalho de Melo <acme@kernel.org>
Cc: Borislav Petkov <bp@alien8.de>
Cc: Boris Ostrovsky <boris.ostrovsky@oracle.com>
Cc: "H. Peter Anvin" <hpa@zytor.com>
Cc: Ian Rogers <irogers@google.com>
Cc: Ingo Molnar <mingo@redhat.com>
Cc: Jiri Olsa <redacted>
Cc: Joao Martins <redacted>
Cc: Konrad Rzeszutek Wilk <redacted>
Cc: Mark Rutland <mark.rutland@arm.com>
Cc: Michael Petlan <redacted>
Cc: Namhyung Kim <namhyung@kernel.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Robert Richter <redacted>
Cc: Stephane Eranian <redacted>
Cc: Thomas Gleixner <redacted>
Cc: <redacted>
Cc: <redacted>
Cc: x86@kernel.org
---
 arch/x86/events/amd/ibs.c | 8 ++++++++
 1 file changed, 8 insertions(+)
diff --git a/arch/x86/events/amd/ibs.c b/arch/x86/events/amd/ibs.c
index 8c25fbd5142e..222c890527a2 100644
--- a/arch/x86/events/amd/ibs.c
+++ b/arch/x86/events/amd/ibs.c
@@ -90,6 +90,7 @@ struct perf_ibs {
        unsigned long                   offset_mask[1];
        int                             offset_max;
        unsigned int                    fetch_count_reset_broken : 1;
+       unsigned int                    fetch_ignore_if_zero_rip : 1;
        struct cpu_perf_ibs __percpu    *pcpu;

        struct attribute                **format_attrs;
@@ -673,6 +674,10 @@ static int perf_ibs_handle_irq(struct perf_ibs *perf_ibs, struct pt_regs *iregs)
        if (check_rip && (ibs_data.regs[2] & IBS_RIP_INVALID)) {
                regs.flags &= ~PERF_EFLAGS_EXACT;
        } else {
+               /* Workaround for erratum #1,197 */
+               if (perf_ibs->fetch_ignore_if_zero_rip && !(ibs_data.regs[1]))
+                       goto out;
Can we just use the iregs.ip instead of dropping the sample?
Users might care about the accurate number of samples..

Thanks,
Namhyung

quoted hunk ↗ jump to hunk
+
                set_linear_ip(&regs, ibs_data.regs[1]);
                regs.flags |= PERF_EFLAGS_EXACT;
        }
@@ -770,6 +775,9 @@ static __init void perf_event_ibs_init(void)
        if (boot_cpu_data.x86 >= 0x16 && boot_cpu_data.x86 <= 0x18)
                perf_ibs_fetch.fetch_count_reset_broken = 1;

+       if (boot_cpu_data.x86 == 0x19 && boot_cpu_data.x86_model < 0x10)
+               perf_ibs_fetch.fetch_ignore_if_zero_rip = 1;
+
        perf_ibs_pmu_init(&perf_ibs_fetch, "ibs_fetch");

        if (ibs_caps & IBS_CAPS_OPCNT) {
--
2.31.1
Thread (27 messages) 27 messages, 3 authors, last reply 2021-08-26
Keyboard shortcuts
hback out one level
jnext message in thread / next series-diff position
kprevious message in thread / previous series-diff position
ldrill in
Escclose help / fold thread tree
?toggle this help