[PATCH] perf/x86/intel/lbr: Fix unchecked MSR access error on HSW

[PATCH] perf/x86/intel/lbr: Fix unchecked MSR access error on HSW

[kan.liang]

From: Kan Liang <kan.liang@linux.intel.com>

The fuzzer triggers the below trace.

[ 7763.384369] unchecked MSR access error: WRMSR to 0x689
(tried to write 0x1fffffff8101349e) at rIP: 0xffffffff810704a4
(native_write_msr+0x4/0x20)
[ 7763.397420] Call Trace:
[ 7763.399881]  <TASK>
[ 7763.401994]  intel_pmu_lbr_restore+0x9a/0x1f0
[ 7763.406363]  intel_pmu_lbr_sched_task+0x91/0x1c0
[ 7763.410992]  __perf_event_task_sched_in+0x1cd/0x240

On a machine with the LBR format LBR_FORMAT_EIP_FLAGS2, when the TSX is
disabled, a TSX quirk is required to access LBR from registers.
The lbr_from_signext_quirk_needed() is introduced to determine whether
the TSX quirk should be applied. However, the
lbr_from_signext_quirk_needed() is invoked before the
intel_pmu_lbr_init(), which parses the LBR format information. Without
the correct LBR format information, the TSX quirk never be applied.

Move the lbr_from_signext_quirk_needed() into the intel_pmu_lbr_init().
Checking x86_pmu.lbr_has_tsx in the lbr_from_signext_quirk_needed() is
not required anymore.

Both LBR_FORMAT_EIP_FLAGS2 and LBR_FORMAT_INFO have LBR_TSX flag, but
only the LBR_FORMAT_EIP_FLAGS2 requirs the quirk. Update the comments
accordingly.

Fixes: 1ac7fd8159a8 ("perf/x86/intel/lbr: Support LBR format V7")
Reported-by: Vince Weaver <vincent.weaver@maine.edu>
Signed-off-by: Kan Liang <kan.liang@linux.intel.com>
Cc: stable@vger.kernel.org
---
 arch/x86/events/intel/lbr.c | 19 ++++++++++---------
 1 file changed, 10 insertions(+), 9 deletions(-)

diff --git a/arch/x86/events/intel/lbr.c b/arch/x86/events/intel/lbr.c
index 13179f31fe10..4f70fb6c2c1e 100644
--- a/arch/x86/events/intel/lbr.c
+++ b/arch/x86/events/intel/lbr.c
@@ -278,9 +278,9 @@ enum {
 };
 
 /*
- * For formats with LBR_TSX flags (e.g. LBR_FORMAT_EIP_FLAGS2), bits 61:62 in
- * MSR_LAST_BRANCH_FROM_x are the TSX flags when TSX is supported, but when
- * TSX is not supported they have no consistent behavior:
+ * For format LBR_FORMAT_EIP_FLAGS2, bits 61:62 in MSR_LAST_BRANCH_FROM_x
+ * are the TSX flags when TSX is supported, but when TSX is not supported
+ * they have no consistent behavior:
  *
  *   - For wrmsr(), bits 61:62 are considered part of the sign extension.
  *   - For HW updates (branch captures) bits 61:62 are always OFF and are not
@@ -288,7 +288,7 @@ enum {
  *
  * Therefore, if:
  *
- *   1) LBR has TSX format
+ *   1) LBR format LBR_FORMAT_EIP_FLAGS2
  *   2) CPU has no TSX support enabled
  *
  * ... then any value passed to wrmsr() must be sign extended to 63 bits and any
@@ -300,7 +300,7 @@ static inline bool lbr_from_signext_quirk_needed(void)
 	bool tsx_support = boot_cpu_has(X86_FEATURE_HLE) ||
 			   boot_cpu_has(X86_FEATURE_RTM);
 
-	return !tsx_support && x86_pmu.lbr_has_tsx;
+	return !tsx_support;
 }
 
 static DEFINE_STATIC_KEY_FALSE(lbr_from_quirk_key);
@@ -1609,9 +1609,6 @@ void intel_pmu_lbr_init_hsw(void)
 	x86_pmu.lbr_sel_map  = hsw_lbr_sel_map;
 
 	x86_get_pmu(smp_processor_id())->task_ctx_cache = create_lbr_kmem_cache(size, 0);
-
-	if (lbr_from_signext_quirk_needed())
-		static_branch_enable(&lbr_from_quirk_key);
 }
 
 /* skylake */
@@ -1702,7 +1699,11 @@ void intel_pmu_lbr_init(void)
 	switch (x86_pmu.intel_cap.lbr_format) {
 	case LBR_FORMAT_EIP_FLAGS2:
 		x86_pmu.lbr_has_tsx = 1;
-		fallthrough;
+		x86_pmu.lbr_from_flags = 1;
+		if (lbr_from_signext_quirk_needed())
+			static_branch_enable(&lbr_from_quirk_key);
+		break;
+
 	case LBR_FORMAT_EIP_FLAGS:
 		x86_pmu.lbr_from_flags = 1;
 		break;
-- 
2.35.1

Re: [PATCH] perf/x86/intel/lbr: Fix unchecked MSR access error on HSW

[Peter Zijlstra]

On Thu, Jul 14, 2022 at 11:26:30AM -0700, kan.liang@linux.intel.com wrote:
> From: Kan Liang <kan.liang@linux.intel.com>
> 
> The fuzzer triggers the below trace.
> 
> [ 7763.384369] unchecked MSR access error: WRMSR to 0x689
> (tried to write 0x1fffffff8101349e) at rIP: 0xffffffff810704a4
> (native_write_msr+0x4/0x20)
> [ 7763.397420] Call Trace:
> [ 7763.399881]  <TASK>
> [ 7763.401994]  intel_pmu_lbr_restore+0x9a/0x1f0
> [ 7763.406363]  intel_pmu_lbr_sched_task+0x91/0x1c0
> [ 7763.410992]  __perf_event_task_sched_in+0x1cd/0x240
> 
> On a machine with the LBR format LBR_FORMAT_EIP_FLAGS2, when the TSX is
> disabled, a TSX quirk is required to access LBR from registers.
> The lbr_from_signext_quirk_needed() is introduced to determine whether
> the TSX quirk should be applied. However, the
> lbr_from_signext_quirk_needed() is invoked before the
> intel_pmu_lbr_init(), which parses the LBR format information. Without
> the correct LBR format information, the TSX quirk never be applied.
> 
> Move the lbr_from_signext_quirk_needed() into the intel_pmu_lbr_init().
> Checking x86_pmu.lbr_has_tsx in the lbr_from_signext_quirk_needed() is
> not required anymore.
> 
> Both LBR_FORMAT_EIP_FLAGS2 and LBR_FORMAT_INFO have LBR_TSX flag, but
> only the LBR_FORMAT_EIP_FLAGS2 requirs the quirk. Update the comments
> accordingly.
> 
> Fixes: 1ac7fd8159a8 ("perf/x86/intel/lbr: Support LBR format V7")
> Reported-by: Vince Weaver <vincent.weaver@maine.edu>
> Signed-off-by: Kan Liang <kan.liang@linux.intel.com>

Thanks!

[tip: perf/urgent] perf/x86/intel/lbr: Fix unchecked MSR access error on HSW

[tip-bot2 for Kan Liang]

The following commit has been merged into the perf/urgent branch of tip:

Commit-ID:     b0380e13502adf7dd8be4c47d622c3522aae6c63
Gitweb:        https://git.kernel.org/tip/b0380e13502adf7dd8be4c47d622c3522aae6c63
Author:        Kan Liang <kan.liang@linux.intel.com>
AuthorDate:    Thu, 14 Jul 2022 11:26:30 -07:00
Committer:     Peter Zijlstra <peterz@infradead.org>
CommitterDate: Wed, 20 Jul 2022 19:24:55 +02:00

perf/x86/intel/lbr: Fix unchecked MSR access error on HSW

The fuzzer triggers the below trace.

[ 7763.384369] unchecked MSR access error: WRMSR to 0x689
(tried to write 0x1fffffff8101349e) at rIP: 0xffffffff810704a4
(native_write_msr+0x4/0x20)
[ 7763.397420] Call Trace:
[ 7763.399881]  <TASK>
[ 7763.401994]  intel_pmu_lbr_restore+0x9a/0x1f0
[ 7763.406363]  intel_pmu_lbr_sched_task+0x91/0x1c0
[ 7763.410992]  __perf_event_task_sched_in+0x1cd/0x240

On a machine with the LBR format LBR_FORMAT_EIP_FLAGS2, when the TSX is
disabled, a TSX quirk is required to access LBR from registers.
The lbr_from_signext_quirk_needed() is introduced to determine whether
the TSX quirk should be applied. However, the
lbr_from_signext_quirk_needed() is invoked before the
intel_pmu_lbr_init(), which parses the LBR format information. Without
the correct LBR format information, the TSX quirk never be applied.

Move the lbr_from_signext_quirk_needed() into the intel_pmu_lbr_init().
Checking x86_pmu.lbr_has_tsx in the lbr_from_signext_quirk_needed() is
not required anymore.

Both LBR_FORMAT_EIP_FLAGS2 and LBR_FORMAT_INFO have LBR_TSX flag, but
only the LBR_FORMAT_EIP_FLAGS2 requirs the quirk. Update the comments
accordingly.

Fixes: 1ac7fd8159a8 ("perf/x86/intel/lbr: Support LBR format V7")
Reported-by: Vince Weaver <vincent.weaver@maine.edu>
Signed-off-by: Kan Liang <kan.liang@linux.intel.com>
Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Cc: stable@vger.kernel.org
Link: https://lkml.kernel.org/r/20220714182630.342107-1-kan.liang@linux.intel.com
---
 arch/x86/events/intel/lbr.c | 19 ++++++++++---------
 1 file changed, 10 insertions(+), 9 deletions(-)

diff --git a/arch/x86/events/intel/lbr.c b/arch/x86/events/intel/lbr.c
index 13179f3..4f70fb6 100644
--- a/arch/x86/events/intel/lbr.c
+++ b/arch/x86/events/intel/lbr.c
@@ -278,9 +278,9 @@ enum {
 };
 
 /*
- * For formats with LBR_TSX flags (e.g. LBR_FORMAT_EIP_FLAGS2), bits 61:62 in
- * MSR_LAST_BRANCH_FROM_x are the TSX flags when TSX is supported, but when
- * TSX is not supported they have no consistent behavior:
+ * For format LBR_FORMAT_EIP_FLAGS2, bits 61:62 in MSR_LAST_BRANCH_FROM_x
+ * are the TSX flags when TSX is supported, but when TSX is not supported
+ * they have no consistent behavior:
  *
  *   - For wrmsr(), bits 61:62 are considered part of the sign extension.
  *   - For HW updates (branch captures) bits 61:62 are always OFF and are not
@@ -288,7 +288,7 @@ enum {
  *
  * Therefore, if:
  *
- *   1) LBR has TSX format
+ *   1) LBR format LBR_FORMAT_EIP_FLAGS2
  *   2) CPU has no TSX support enabled
  *
  * ... then any value passed to wrmsr() must be sign extended to 63 bits and any
@@ -300,7 +300,7 @@ static inline bool lbr_from_signext_quirk_needed(void)
 	bool tsx_support = boot_cpu_has(X86_FEATURE_HLE) ||
 			   boot_cpu_has(X86_FEATURE_RTM);
 
-	return !tsx_support && x86_pmu.lbr_has_tsx;
+	return !tsx_support;
 }
 
 static DEFINE_STATIC_KEY_FALSE(lbr_from_quirk_key);
@@ -1609,9 +1609,6 @@ void intel_pmu_lbr_init_hsw(void)
 	x86_pmu.lbr_sel_map  = hsw_lbr_sel_map;
 
 	x86_get_pmu(smp_processor_id())->task_ctx_cache = create_lbr_kmem_cache(size, 0);
-
-	if (lbr_from_signext_quirk_needed())
-		static_branch_enable(&lbr_from_quirk_key);
 }
 
 /* skylake */
@@ -1702,7 +1699,11 @@ void intel_pmu_lbr_init(void)
 	switch (x86_pmu.intel_cap.lbr_format) {
 	case LBR_FORMAT_EIP_FLAGS2:
 		x86_pmu.lbr_has_tsx = 1;
-		fallthrough;
+		x86_pmu.lbr_from_flags = 1;
+		if (lbr_from_signext_quirk_needed())
+			static_branch_enable(&lbr_from_quirk_key);
+		break;
+
 	case LBR_FORMAT_EIP_FLAGS:
 		x86_pmu.lbr_from_flags = 1;
 		break;