Skip to contentSkip to conversation

perf/amd/ibs: Assorted fixes

PATCH Applied Ravi Bangoria 5 participants 20 messages Feb 16 – Mar 10

Subsystems (2)
lore
Share & tools
lore
#0 lore raw

v2 cover perf/amd/ibs: Assorted fixes

Address several issues in the AMD IBS perf code: interrupt accounting for discarded samples, a Zen5-specific quirk, PhyAddrVal handling, NMI-safety with perf_allow_kernel(), and a race between event add and NMI.

Patches are prepared on tip/perf/core (7db06e329af3)

v1: https://lore.kernel.org/r/[email protected]
v1->v2:
- Split fixes series from future enhancements
- Add interrupt throttling for zero-rip erratum #1197
- Avoid calling perf_allow_kernel() in NMI context (new patch)

Ravi Bangoria (5):
perf/amd/ibs: Account interrupt for discarded samples perf/amd/ibs: Limit ldlat->l3missonly dependency to Zen5 perf/amd/ibs: Preserve PhyAddrVal bit when clearing PhyAddr MSR perf/amd/ibs: Avoid calling perf_allow_kernel() from the IBS NMI
handler
perf/amd/ibs: Avoid race between event add and NMI

 arch/x86/events/amd/ibs.c          | 28 ++++++++++++++++++++++------
 arch/x86/events/perf_event_flags.h |  1 +
 2 files changed, 23 insertions(+), 6 deletions(-)
Signature · 1 line
2.43.0
#1 lore raw

PATCH v2 1/5 perf/amd/ibs: Account interrupt for discarded samples

Add interrupt throttling accounting for below cases:

o IBS Op PMU: A software filter (in addition to the hardware filter) drops samples whose load latency is below the user-specified threshold.

o IBS Fetch PMU: Samples discarded due to the zero-RIP erratum (#1197).

Although these samples are discarded, the NMI cost is still incurred, so they should be counted for interrupt throttling.

---
 arch/x86/events/amd/ibs.c | 8 ++++++--
 1 file changed, 6 insertions(+), 2 deletions(-)
arch/x86/events/amd/ibs.c +6 -2
index aca89f23d2e0..705ef43325be 100644
@@ -1293,8 +1293,10 @@ static int perf_ibs_handle_irq(struct perf_ibs *perf_ibs, struct pt_regs *iregs)
 		 * within [128, 2048] range.
 		 */
 		if (!op_data3.ld_op || !op_data3.dc_miss ||
-		    op_data3.dc_miss_lat <= (event->attr.config1 & 0xFFF))
+		    op_data3.dc_miss_lat <= (event->attr.config1 & 0xFFF)) {
+			throttle = perf_event_account_interrupt(event);
 			goto out;
+		}
 	}
 
 	/*
@@ -1326,8 +1328,10 @@ static int perf_ibs_handle_irq(struct perf_ibs *perf_ibs, struct pt_regs *iregs)
 		regs.flags &= ~PERF_EFLAGS_EXACT;
 	} else {
 		/* Workaround for erratum #1197 */
-		if (perf_ibs->fetch_ignore_if_zero_rip && !(ibs_data.regs[1]))
+		if (perf_ibs->fetch_ignore_if_zero_rip && !(ibs_data.regs[1])) {
+			throttle = perf_event_account_interrupt(event);
 			goto out;
+		}
 
 		set_linear_ip(&regs, ibs_data.regs[1]);
 		regs.flags |= PERF_EFLAGS_EXACT;
Signature · 1 line
2.43.0
#2 lore raw

Re: [PATCH v2 1/5] perf/amd/ibs: Account interrupt for discarded samples

Hi Ravi,

Quoted 11 lines · Ravi Bangoria

Add interrupt throttling accounting for below cases:

o IBS Op PMU: A software filter (in addition to the hardware filter) drops samples whose load latency is below the user-specified threshold.

o IBS Fetch PMU: Samples discarded due to the zero-RIP erratum (#1197).

Although these samples are discarded, the NMI cost is still incurred, so they should be counted for interrupt throttling.

Maybe worth tagging

Fixes 26db2e0c51fe83e1dd852c1321407835b481806e ("perf/x86/amd/ibs: Work around erratum #1197")
Fixes d20610c19b4a22bc69085b7eb7a02741d51de30e ("perf/amd/ibs: Add support for OP Load Latency Filtering")

?

Cheers,
Michael

Quoted 38 lines
---
 arch/x86/events/amd/ibs.c | 8 ++++++--
 1 file changed, 6 insertions(+), 2 deletions(-)
arch/x86/events/amd/ibs.c +6 -2
index aca89f23d2e0..705ef43325be 100644
@@ -1293,8 +1293,10 @@ static int perf_ibs_handle_irq(struct perf_ibs *perf_ibs, struct pt_regs *iregs)
 		 * within [128, 2048] range.
 		 */
 		if (!op_data3.ld_op || !op_data3.dc_miss ||
-		    op_data3.dc_miss_lat <= (event->attr.config1 & 0xFFF))
+		    op_data3.dc_miss_lat <= (event->attr.config1 & 0xFFF)) {
+			throttle = perf_event_account_interrupt(event);
 			goto out;
+		}
 	}
 
 	/*
@@ -1326,8 +1328,10 @@ static int perf_ibs_handle_irq(struct perf_ibs *perf_ibs, struct pt_regs *iregs)
 		regs.flags &= ~PERF_EFLAGS_EXACT;
 	} else {
 		/* Workaround for erratum #1197 */
-		if (perf_ibs->fetch_ignore_if_zero_rip && !(ibs_data.regs[1]))
+		if (perf_ibs->fetch_ignore_if_zero_rip && !(ibs_data.regs[1])) {
+			throttle = perf_event_account_interrupt(event);
 			goto out;
+		}
 
 		set_linear_ip(&regs, ibs_data.regs[1]);
 		regs.flags |= PERF_EFLAGS_EXACT;
Signature · 1 line
2.43.0
#3 lore raw

Re: [PATCH v2 1/5] perf/amd/ibs: Account interrupt for discarded samples

Hi Michael,

Quoted 8 lines
Quoted 8 lines

Add interrupt throttling accounting for below cases:

o IBS Op PMU: A software filter (in addition to the hardware filter) drops samples whose load latency is below the user-specified threshold.

o IBS Fetch PMU: Samples discarded due to the zero-RIP erratum (#1197).

Quoted 8 lines
Quoted 3 lines

Although these samples are discarded, the NMI cost is still incurred, so they should be counted for interrupt throttling.

Maybe worth tagging

Fixes 26db2e0c51fe83e1dd852c1321407835b481806e ("perf/x86/amd/ibs: Work around erratum #1197")
Fixes d20610c19b4a22bc69085b7eb7a02741d51de30e ("perf/amd/ibs: Add support for OP Load Latency Filtering")

Yes, makes sense.

Ingo / Peter, Please let me know if you want me to respin.

Thanks,
Ravi

maintainer
#4 lore raw

Re: [PATCH v2 1/5] perf/amd/ibs: Account interrupt for discarded samples

Quoted 14 lines · Ravi Bangoria

Hi Michael,

Quoted 12 lines
Quoted 11 lines

Add interrupt throttling accounting for below cases:

o IBS Op PMU: A software filter (in addition to the hardware filter) drops samples whose load latency is below the user-specified threshold.

o IBS Fetch PMU: Samples discarded due to the zero-RIP erratum (#1197).

Although these samples are discarded, the NMI cost is still incurred, so they should be counted for interrupt throttling.

Quoted 8 lines
Quoted 4 lines

Maybe worth tagging

Fixes 26db2e0c51fe83e1dd852c1321407835b481806e ("perf/x86/amd/ibs: Work around erratum #1197")
Fixes d20610c19b4a22bc69085b7eb7a02741d51de30e ("perf/amd/ibs: Add support for OP Load Latency Filtering")

Yes, makes sense.

Ingo / Peter, Please let me know if you want me to respin.

I've added them. Thanks!

merge bot Applied
#5 lore raw

[tip: perf/core] perf/amd/ibs: Account interrupt for discarded samples

In reply to Ravi Bangoria
The following commit has been merged into the perf/core branch of tip:

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

Commit-ID:     01336b5559785a136de1cac49705f63a70a755bc
Gitweb:        https://git.kernel.org/tip/01336b5559785a136de1cac49705f63a70a755bc
Author:        Ravi Bangoria <[email protected]>
AuthorDate:    Mon, 16 Feb 2026 04:22:12 
Committer:     Peter Zijlstra <[email protected]>
CommitterDate: Fri, 27 Feb 2026 16:40:22 +01:00

perf/amd/ibs: Account interrupt for discarded samples

Add interrupt throttling accounting for below cases:

o IBS Op PMU: A software filter (in addition to the hardware filter) drops samples whose load latency is below the user-specified threshold.

o IBS Fetch PMU: Samples discarded due to the zero-RIP erratum (#1197).

Although these samples are discarded, the NMI cost is still incurred, so they should be counted for interrupt throttling.

Fixes 26db2e0c51fe83e1dd852c1321407835b481806e ("perf/x86/amd/ibs: Work around erratum #1197")
Fixes d20610c19b4a22bc69085b7eb7a02741d51de30e ("perf/amd/ibs: Add support for OP Load Latency Filtering")
---
 arch/x86/events/amd/ibs.c | 8 ++++++--
 1 file changed, 6 insertions(+), 2 deletions(-)
arch/x86/events/amd/ibs.c +6 -2
index aca89f2..705ef43 100644
@@ -1293,8 +1293,10 @@ fail:
 		 * within [128, 2048] range.
 		 */
 		if (!op_data3.ld_op || !op_data3.dc_miss ||
-		    op_data3.dc_miss_lat <= (event->attr.config1 & 0xFFF))
+		    op_data3.dc_miss_lat <= (event->attr.config1 & 0xFFF)) {
+			throttle = perf_event_account_interrupt(event);
 			goto out;
+		}
 	}
 
 	/*
@@ -1326,8 +1328,10 @@ fail:
 		regs.flags &= ~PERF_EFLAGS_EXACT;
 	} else {
 		/* Workaround for erratum #1197 */
-		if (perf_ibs->fetch_ignore_if_zero_rip && !(ibs_data.regs[1]))
+		if (perf_ibs->fetch_ignore_if_zero_rip && !(ibs_data.regs[1])) {
+			throttle = perf_event_account_interrupt(event);
 			goto out;
+		}
 
 		set_linear_ip(&regs, ibs_data.regs[1]);
 		regs.flags |= PERF_EFLAGS_EXACT;
#6 lore raw

PATCH v2 2/5 perf/amd/ibs: Limit ldlat->l3missonly dependency to Zen5

In reply to Ravi Bangoria

The ldlat dependency on l3missonly is specific to Zen 5; newer generations are not affected. This quirk is documented as an erratum in the following Revision Guide.

Erratum: 1606 IBS (Instruction Based Sampling) OP Load Latency Filtering
May Capture Unwanted Samples When L3Miss Filtering is Disabled

Revision Guide for AMD Family 1Ah Models 00h-0Fh Processors, Pub. 58251 Rev. 1.30 July 2025
https://bugzilla.kernel.org/attachment.cgi?id=309193

---
 arch/x86/events/amd/ibs.c | 5 ++++-
 1 file changed, 4 insertions(+), 1 deletion(-)
arch/x86/events/amd/ibs.c +4 -1
index 705ef43325be..e0b64cb13bf9 100644
@@ -356,7 +356,10 @@ static int perf_ibs_init(struct perf_event *event)
 		ldlat >>= 7;
 
 		config |= (ldlat - 1) << 59;
-		config |= IBS_OP_L3MISSONLY | IBS_OP_LDLAT_EN;
+
+		config |= IBS_OP_LDLAT_EN;
+		if (cpu_feature_enabled(X86_FEATURE_ZEN5))
+			config |= IBS_OP_L3MISSONLY;
 	}
 
 	/*
Signature · 1 line
2.43.0
merge bot Applied
#7 lore raw

[tip: perf/core] perf/amd/ibs: Limit ldlat->l3missonly dependency to Zen5

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

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

Commit-ID:     898138efc99096c3ee836fea439ba6da3cfafa4d
Gitweb:        https://git.kernel.org/tip/898138efc99096c3ee836fea439ba6da3cfafa4d
Author:        Ravi Bangoria <[email protected]>
AuthorDate:    Mon, 16 Feb 2026 04:22:13 
Committer:     Peter Zijlstra <[email protected]>
CommitterDate: Fri, 27 Feb 2026 16:40:22 +01:00

perf/amd/ibs: Limit ldlat->l3missonly dependency to Zen5

The ldlat dependency on l3missonly is specific to Zen 5; newer generations are not affected. This quirk is documented as an erratum in the following Revision Guide.

Erratum: 1606 IBS (Instruction Based Sampling) OP Load Latency Filtering
May Capture Unwanted Samples When L3Miss Filtering is Disabled

Revision Guide for AMD Family 1Ah Models 00h-0Fh Processors, Pub. 58251 Rev. 1.30 July 2025
https://bugzilla.kernel.org/attachment.cgi?id=309193

---
 arch/x86/events/amd/ibs.c | 5 ++++-
 1 file changed, 4 insertions(+), 1 deletion(-)
arch/x86/events/amd/ibs.c +4 -1
index 705ef43..e0b64cb 100644
@@ -356,7 +356,10 @@ static int perf_ibs_init(struct perf_event *event)
 		ldlat >>= 7;
 
 		config |= (ldlat - 1) << 59;
-		config |= IBS_OP_L3MISSONLY | IBS_OP_LDLAT_EN;
+
+		config |= IBS_OP_LDLAT_EN;
+		if (cpu_feature_enabled(X86_FEATURE_ZEN5))
+			config |= IBS_OP_L3MISSONLY;
 	}
 
 	/*
#8 lore raw

PATCH v2 3/5 perf/amd/ibs: Preserve PhyAddrVal bit when clearing PhyAddr MSR

In reply to Ravi Bangoria

Commit 50a53b60e141 ("perf/amd/ibs: Prevent leaking sensitive data to userspace") zeroed the physical address and also cleared the PhyAddrVal flag before copying the value into a perf sample to avoid exposing physical addresses to unprivileged users.

Clearing PhyAddrVal, however, has an unintended side-effect: several other IBS fields are considered valid only when this bit is set. As a result, those otherwise correct fields are discarded, reducing IBS functionality.

Continue to zero the physical address, but keep the PhyAddrVal bit intact so the related fields remain usable while still preventing any address leak.

Fixes 50a53b60e141 ("perf/amd/ibs: Prevent leaking sensitive data to userspace")
---
 arch/x86/events/amd/ibs.c | 2 --
 1 file changed, 2 deletions(-)
arch/x86/events/amd/ibs.c +0 -2
index e0b64cb13bf9..05b7c9f2ec33 100644
@@ -1217,12 +1217,10 @@ static void perf_ibs_phyaddr_clear(struct perf_ibs *perf_ibs,
 				   struct perf_ibs_data *ibs_data)
 {
 	if (perf_ibs == &perf_ibs_op) {
-		ibs_data->regs[ibs_op_msr_idx(MSR_AMD64_IBSOPDATA3)] &= ~(1ULL << 18);
 		ibs_data->regs[ibs_op_msr_idx(MSR_AMD64_IBSDCPHYSAD)] = 0;
 		return;
 	}
 
-	ibs_data->regs[ibs_fetch_msr_idx(MSR_AMD64_IBSFETCHCTL)] &= ~(1ULL << 52);
 	ibs_data->regs[ibs_fetch_msr_idx(MSR_AMD64_IBSFETCHPHYSAD)] = 0;
 }
 
Signature · 1 line
2.43.0
merge bot Applied
#9 lore raw

[tip: perf/core] perf/amd/ibs: Preserve PhyAddrVal bit when clearing PhyAddr MSR

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

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

Commit-ID:     723a290326e015b07931eabc603d3735999377be
Gitweb:        https://git.kernel.org/tip/723a290326e015b07931eabc603d3735999377be
Author:        Ravi Bangoria <[email protected]>
AuthorDate:    Mon, 16 Feb 2026 04:22:14 
Committer:     Peter Zijlstra <[email protected]>
CommitterDate: Fri, 27 Feb 2026 16:40:23 +01:00

perf/amd/ibs: Preserve PhyAddrVal bit when clearing PhyAddr MSR

Commit 50a53b60e141 ("perf/amd/ibs: Prevent leaking sensitive data to userspace") zeroed the physical address and also cleared the PhyAddrVal flag before copying the value into a perf sample to avoid exposing physical addresses to unprivileged users.

Clearing PhyAddrVal, however, has an unintended side-effect: several other IBS fields are considered valid only when this bit is set. As a result, those otherwise correct fields are discarded, reducing IBS functionality.

Continue to zero the physical address, but keep the PhyAddrVal bit intact so the related fields remain usable while still preventing any address leak.

Fixes 50a53b60e141 ("perf/amd/ibs: Prevent leaking sensitive data to userspace")
---
 arch/x86/events/amd/ibs.c | 2 --
 1 file changed, 2 deletions(-)
arch/x86/events/amd/ibs.c +0 -2
index e0b64cb..05b7c9f 100644
@@ -1217,12 +1217,10 @@ static void perf_ibs_phyaddr_clear(struct perf_ibs *perf_ibs,
 				   struct perf_ibs_data *ibs_data)
 {
 	if (perf_ibs == &perf_ibs_op) {
-		ibs_data->regs[ibs_op_msr_idx(MSR_AMD64_IBSOPDATA3)] &= ~(1ULL << 18);
 		ibs_data->regs[ibs_op_msr_idx(MSR_AMD64_IBSDCPHYSAD)] = 0;
 		return;
 	}
 
-	ibs_data->regs[ibs_fetch_msr_idx(MSR_AMD64_IBSFETCHCTL)] &= ~(1ULL << 52);
 	ibs_data->regs[ibs_fetch_msr_idx(MSR_AMD64_IBSFETCHPHYSAD)] = 0;
 }
 
#10 lore raw

PATCH v2 4/5 perf/amd/ibs: Avoid calling perf_allow_kernel() from the IBS NMI handler

In reply to Ravi Bangoria

Calling perf_allow_kernel() from the NMI context is unsafe and could be fatal. Capture the permission at event-initialization time by storing it in event->hw.flags, and have the NMI handler rely on that cached flag instead of making the call directly.

Fixes 50a53b60e141d ("perf/amd/ibs: Prevent leaking sensitive data to userspace")
---
 arch/x86/events/amd/ibs.c          | 5 ++++-
 arch/x86/events/perf_event_flags.h | 1 +
 2 files changed, 5 insertions(+), 1 deletion(-)
arch/x86/events/amd/ibs.c +4 -1
index 05b7c9f2ec33..004226b52ac7 100644
@@ -313,6 +313,9 @@ static int perf_ibs_init(struct perf_event *event)
 	if (ret)
 		return ret;
 
+	if (perf_allow_kernel())
+		hwc->flags |= PERF_X86_EVENT_UNPRIVILEGED;
+
 	if (hwc->sample_period) {
 		if (config & perf_ibs->cnt_mask)
 			/* raw max_cnt may not be set */
@@ -1349,7 +1352,7 @@ static int perf_ibs_handle_irq(struct perf_ibs *perf_ibs, struct pt_regs *iregs)
 	 * unprivileged users.
 	 */
 	if ((event->attr.sample_type & PERF_SAMPLE_RAW) &&
-	    perf_allow_kernel()) {
+	    (hwc->flags & PERF_X86_EVENT_UNPRIVILEGED)) {
 		perf_ibs_phyaddr_clear(perf_ibs, &ibs_data);
 	}
 
arch/x86/events/perf_event_flags.h +1 -0
index 70078334e4a3..47f84ee8f540 100644
@@ -23,3 +23,4 @@ PERF_ARCH(PEBS_LAT_HYBRID,	0x0020000) /* ld and st lat for hybrid */
 PERF_ARCH(NEEDS_BRANCH_STACK,	0x0040000) /* require branch stack setup */
 PERF_ARCH(BRANCH_COUNTERS,	0x0080000) /* logs the counters in the extra space of each branch */
 PERF_ARCH(ACR,			0x0100000) /* Auto counter reload */
+PERF_ARCH(UNPRIVILEGED,		0x0200000) /* Unprivileged event (wrt perf_allow_kernel()) */
Signature · 1 line
2.43.0
merge bot Applied
#11 lore raw

[tip: perf/core] perf/amd/ibs: Avoid calling perf_allow_kernel() from the IBS NMI handler

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

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

Commit-ID:     b0a09142622a994c4f4088c3f61db5da87cfc711
Gitweb:        https://git.kernel.org/tip/b0a09142622a994c4f4088c3f61db5da87cfc711
Author:        Ravi Bangoria <[email protected]>
AuthorDate:    Mon, 16 Feb 2026 04:22:15 
Committer:     Peter Zijlstra <[email protected]>
CommitterDate: Fri, 27 Feb 2026 16:40:23 +01:00

perf/amd/ibs: Avoid calling perf_allow_kernel() from the IBS NMI handler

Calling perf_allow_kernel() from the NMI context is unsafe and could be fatal. Capture the permission at event-initialization time by storing it in event->hw.flags, and have the NMI handler rely on that cached flag instead of making the call directly.

Fixes 50a53b60e141d ("perf/amd/ibs: Prevent leaking sensitive data to userspace")
---
 arch/x86/events/amd/ibs.c          | 5 ++++-
 arch/x86/events/perf_event_flags.h | 1 +
 2 files changed, 5 insertions(+), 1 deletion(-)
arch/x86/events/amd/ibs.c +4 -1
index 05b7c9f..004226b 100644
@@ -313,6 +313,9 @@ static int perf_ibs_init(struct perf_event *event)
 	if (ret)
 		return ret;
 
+	if (perf_allow_kernel())
+		hwc->flags |= PERF_X86_EVENT_UNPRIVILEGED;
+
 	if (hwc->sample_period) {
 		if (config & perf_ibs->cnt_mask)
 			/* raw max_cnt may not be set */
@@ -1349,7 +1352,7 @@ fail:
 	 * unprivileged users.
 	 */
 	if ((event->attr.sample_type & PERF_SAMPLE_RAW) &&
-	    perf_allow_kernel()) {
+	    (hwc->flags & PERF_X86_EVENT_UNPRIVILEGED)) {
 		perf_ibs_phyaddr_clear(perf_ibs, &ibs_data);
 	}
 
arch/x86/events/perf_event_flags.h +1 -0
index 7007833..47f84ee 100644
@@ -23,3 +23,4 @@ PERF_ARCH(PEBS_LAT_HYBRID,	0x0020000) /* ld and st lat for hybrid */
 PERF_ARCH(NEEDS_BRANCH_STACK,	0x0040000) /* require branch stack setup */
 PERF_ARCH(BRANCH_COUNTERS,	0x0080000) /* logs the counters in the extra space of each branch */
 PERF_ARCH(ACR,			0x0100000) /* Auto counter reload */
+PERF_ARCH(UNPRIVILEGED,		0x0200000) /* Unprivileged event (wrt perf_allow_kernel()) */
Tested-by on v2 4/5
#12 lore raw

Re: [PATCH v2 4/5] perf/amd/ibs: Avoid calling perf_allow_kernel() from the IBS NMI handler

In reply to Ravi Bangoria
Quoted 5 lines · Ravi Bangoria

Calling perf_allow_kernel() from the NMI context is unsafe and could be fatal. Capture the permission at event-initialization time by storing it in event->hw.flags, and have the NMI handler rely on that cached flag instead of making the call directly.

Hi Ravi.

I have tested this fix on a machine (below), which without this fix can be easily crashed by recording ibs_op/l3missonly=1/ event.

With this fix, it survived 4900+ iterations without any crash.

CPU details
===========

BIOS CPU family:      107
CPU family:            26
Model:                  2
Stepping:               1

Thanks for fixing.

Cheers,
Michael

Quoted 3 lines
Fixes 50a53b60e141d ("perf/amd/ibs: Prevent leaking sensitive data to userspace")
Quoted 42 lines
---
 arch/x86/events/amd/ibs.c          | 5 ++++-
 arch/x86/events/perf_event_flags.h | 1 +
 2 files changed, 5 insertions(+), 1 deletion(-)
arch/x86/events/amd/ibs.c +4 -1
index 05b7c9f2ec33..004226b52ac7 100644
@@ -313,6 +313,9 @@ static int perf_ibs_init(struct perf_event *event)
 	if (ret)
 		return ret;
 
+	if (perf_allow_kernel())
+		hwc->flags |= PERF_X86_EVENT_UNPRIVILEGED;
+
 	if (hwc->sample_period) {
 		if (config & perf_ibs->cnt_mask)
 			/* raw max_cnt may not be set */
@@ -1349,7 +1352,7 @@ static int perf_ibs_handle_irq(struct perf_ibs *perf_ibs, struct pt_regs *iregs)
 	 * unprivileged users.
 	 */
 	if ((event->attr.sample_type & PERF_SAMPLE_RAW) &&
-	    perf_allow_kernel()) {
+	    (hwc->flags & PERF_X86_EVENT_UNPRIVILEGED)) {
 		perf_ibs_phyaddr_clear(perf_ibs, &ibs_data);
 	}
 
arch/x86/events/perf_event_flags.h +1 -0
index 70078334e4a3..47f84ee8f540 100644
@@ -23,3 +23,4 @@ PERF_ARCH(PEBS_LAT_HYBRID,	0x0020000) /* ld and st lat for hybrid */
 PERF_ARCH(NEEDS_BRANCH_STACK,	0x0040000) /* require branch stack setup */
 PERF_ARCH(BRANCH_COUNTERS,	0x0080000) /* logs the counters in the extra space of each branch */
 PERF_ARCH(ACR,			0x0100000) /* Auto counter reload */
+PERF_ARCH(UNPRIVILEGED,		0x0200000) /* Unprivileged event (wrt perf_allow_kernel()) */
Signature · 1 line
2.43.0
#13 lore raw

Re: [PATCH v2 4/5] perf/amd/ibs: Avoid calling perf_allow_kernel() from the IBS NMI handler

In reply to Ravi Bangoria

Hello Ravi.

By evaluating the privileges at the event-init moment and using cached value later, couldn't it lead to the cached value being obsolete at some point? E.g. a setuid program later dropping the privileges but still being able to read physical addresses? I don't think it can be a solid attack-surface, but still may be worth of a thought... ?

Michael

Quoted 50 lines · Ravi Bangoria

Calling perf_allow_kernel() from the NMI context is unsafe and could be fatal. Capture the permission at event-initialization time by storing it in event->hw.flags, and have the NMI handler rely on that cached flag instead of making the call directly.

Fixes 50a53b60e141d ("perf/amd/ibs: Prevent leaking sensitive data to userspace")
---
 arch/x86/events/amd/ibs.c          | 5 ++++-
 arch/x86/events/perf_event_flags.h | 1 +
 2 files changed, 5 insertions(+), 1 deletion(-)
arch/x86/events/amd/ibs.c +4 -1
index 05b7c9f2ec33..004226b52ac7 100644
@@ -313,6 +313,9 @@ static int perf_ibs_init(struct perf_event *event)
 	if (ret)
 		return ret;
 
+	if (perf_allow_kernel())
+		hwc->flags |= PERF_X86_EVENT_UNPRIVILEGED;
+
 	if (hwc->sample_period) {
 		if (config & perf_ibs->cnt_mask)
 			/* raw max_cnt may not be set */
@@ -1349,7 +1352,7 @@ static int perf_ibs_handle_irq(struct perf_ibs *perf_ibs, struct pt_regs *iregs)
 	 * unprivileged users.
 	 */
 	if ((event->attr.sample_type & PERF_SAMPLE_RAW) &&
-	    perf_allow_kernel()) {
+	    (hwc->flags & PERF_X86_EVENT_UNPRIVILEGED)) {
 		perf_ibs_phyaddr_clear(perf_ibs, &ibs_data);
 	}
 
arch/x86/events/perf_event_flags.h +1 -0
index 70078334e4a3..47f84ee8f540 100644
@@ -23,3 +23,4 @@ PERF_ARCH(PEBS_LAT_HYBRID,	0x0020000) /* ld and st lat for hybrid */
 PERF_ARCH(NEEDS_BRANCH_STACK,	0x0040000) /* require branch stack setup */
 PERF_ARCH(BRANCH_COUNTERS,	0x0080000) /* logs the counters in the extra space of each branch */
 PERF_ARCH(ACR,			0x0100000) /* Auto counter reload */
+PERF_ARCH(UNPRIVILEGED,		0x0200000) /* Unprivileged event (wrt perf_allow_kernel()) */
Signature · 1 line
2.43.0
#14 lore raw

Re: [PATCH v2 4/5] perf/amd/ibs: Avoid calling perf_allow_kernel() from the IBS NMI handler

Hi Michael,

Quoted 4 lines

By evaluating the privileges at the event-init moment and using cached value later, couldn't it lead to the cached value being obsolete at some point? E.g. a setuid program later dropping the privileges but still being able to read physical addresses?

Yes, but wouldn't the same concern apply to all other call sites of perf_allow_kernel() ?

Thanks,
Ravi

#15 lore raw

Re: [PATCH v2 4/5] perf/amd/ibs: Avoid calling perf_allow_kernel() from the IBS NMI handler

Quoted 9 lines · Ravi Bangoria

Hi Michael,

Quoted 4 lines

By evaluating the privileges at the event-init moment and using cached value later, couldn't it lead to the cached value being obsolete at some point? E.g. a setuid program later dropping the privileges but still being able to read physical addresses?

Yes, but wouldn't the same concern apply to all other call sites of perf_allow_kernel() ?

Well, I don't say this case is worse than the others, just raising a more generic question whether this design is fairly acceptable or would deserve e.g. privilege adjusting at the time when they are dropped or something... Nothing more, just a note... :)

Michael

Quoted 5 lines

Thanks,
Ravi

#16 lore raw

Re: [PATCH v2 4/5] perf/amd/ibs: Avoid calling perf_allow_kernel() from the IBS NMI handler

Hi Michael,

Quoted 5 lines
Quoted 5 lines
Quoted 4 lines

By evaluating the privileges at the event-init moment and using cached value later, couldn't it lead to the cached value being obsolete at some point? E.g. a setuid program later dropping the privileges but still being able to read physical addresses?

Quoted 7 lines
Quoted 2 lines

Yes, but wouldn't the same concern apply to all other call sites of perf_allow_kernel() ?

Well, I don't say this case is worse than the others, just raising a more generic question whether this design is fairly acceptable or would deserve e.g. privilege adjusting at the time when they are dropped or something... Nothing more, just a note... :)

I understand. However, performing the privilege check at the beginning and then continuing to allow accessing the privileged resource even after privileges have been dropped is a common design pattern, isn't it? For example, I assume something like this is allowed:

  fd = open("/root/only/file");
  setresuid(); /* drop the privilege */
  read(fd);

Thanks,
Ravi

#17 lore raw

PATCH v2 5/5 perf/amd/ibs: Avoid race between event add and NMI

In reply to Ravi Bangoria

Consider the following race:

--------
o OP_CTL contains stale value: OP_CTL[Val]=1, OP_CTL[En]=0
o A new IBS OP event is being added
o [P]: Process context, [N]: NMI context

  [P] perf_ibs_add(event) {
  [P]     if (test_and_set_bit(IBS_ENABLED, pcpu->state))
  [P]         return;
  [P]     /* pcpu->state = IBS_ENABLED */
  [P]
  [P]     pcpu->event = event;
  [P]
  [P]     perf_ibs_start(event) {
  [P]         set_bit(IBS_STARTED, pcpu->state);
  [P]         /* pcpu->state = IBS_ENABLED | IBS_STARTED */
  [P]         clear_bit(IBS_STOPPING, pcpu->state);
  [P]         /* pcpu->state = IBS_ENABLED | IBS_STARTED */

[N] --> NMI due to genuine FETCH event. perf_ibs_handle_irq()
[N] called for OP PMU as well.
[N]
[N] perf_ibs_handle_irq(perf_ibs) {
[N] event = pcpu->event; /* See line 6 */
[N]
[N] if (!test_bit(IBS_STARTED, pcpu->state)) /* false */
[N] return 0;
[N]
[N] if (WARN_ON_ONCE(!event)) /* false */
[N] goto fail;
[N]
[N] if (!(*buf++ & perf_ibs->valid_mask)) /* false due to stale
[N] * IBS_OP_CTL value */
[N] goto fail;
[N]
[N] ...
[N]
[N] perf_ibs_enable_event() // *Accidentally* enable the event.
[N] }
[N]
[N] /*

  [N]  * Repeated NMIs may follow due to accidentally enabled IBS OP
  [N]  * event if the sample period is very low. It could also lead
  [N]  * to pcpu->state corruption if the event gets throttled due
  [N]  * to too frequent NMIs.
  [N]  */
  [P]         perf_ibs_enable_event();
  [P]     }
  [P] }
  --------

We cannot safely clear IBS_{FETCH|OP}_CTL while disabling the event, because the register might be read again later. So, clear the register in the enable path - before we update pcpu->state and enable the event. This guarantees that any NMI that lands in the gap finds Val=0 and bails out cleanly.

---
 arch/x86/events/amd/ibs.c | 8 ++++++++
 1 file changed, 8 insertions(+)
arch/x86/events/amd/ibs.c +8 -0
index 004226b52ac7..32e6456cb5e5 100644
@@ -493,6 +493,14 @@ static void perf_ibs_start(struct perf_event *event, int flags)
 	}
 	config |= period >> 4;
 
+	/*
+	 * Reset the IBS_{FETCH|OP}_CTL MSR before updating pcpu->state.
+	 * Doing so prevents a race condition in which an NMI due to other
+	 * source might accidentally activate the event before we enable
+	 * it ourselves.
+	 */
+	perf_ibs_disable_event(perf_ibs, hwc, 0);
+
 	/*
 	 * Set STARTED before enabling the hardware, such that a subsequent NMI
 	 * must observe it.
Signature · 1 line
2.43.0
merge bot Applied
#18 lore raw

[tip: perf/core] perf/amd/ibs: Avoid race between event add and NMI

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

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

Commit-ID:     1b044ff3c17e9d7fd93ffc0ba541ccdeb992d7f5
Gitweb:        https://git.kernel.org/tip/1b044ff3c17e9d7fd93ffc0ba541ccdeb992d7f5
Author:        Ravi Bangoria <[email protected]>
AuthorDate:    Mon, 16 Feb 2026 04:22:16 
Committer:     Peter Zijlstra <[email protected]>
CommitterDate: Fri, 27 Feb 2026 16:40:23 +01:00

perf/amd/ibs: Avoid race between event add and NMI

Consider the following race:

--------
o OP_CTL contains stale value: OP_CTL[Val]=1, OP_CTL[En]=0
o A new IBS OP event is being added
o [P]: Process context, [N]: NMI context

  [P] perf_ibs_add(event) {
  [P]     if (test_and_set_bit(IBS_ENABLED, pcpu->state))
  [P]         return;
  [P]     /* pcpu->state = IBS_ENABLED */
  [P]
  [P]     pcpu->event = event;
  [P]
  [P]     perf_ibs_start(event) {
  [P]         set_bit(IBS_STARTED, pcpu->state);
  [P]         /* pcpu->state = IBS_ENABLED | IBS_STARTED */
  [P]         clear_bit(IBS_STOPPING, pcpu->state);
  [P]         /* pcpu->state = IBS_ENABLED | IBS_STARTED */

[N] --> NMI due to genuine FETCH event. perf_ibs_handle_irq()
[N] called for OP PMU as well.
[N]
[N] perf_ibs_handle_irq(perf_ibs) {
[N] event = pcpu->event; /* See line 6 */
[N]
[N] if (!test_bit(IBS_STARTED, pcpu->state)) /* false */
[N] return 0;
[N]
[N] if (WARN_ON_ONCE(!event)) /* false */
[N] goto fail;
[N]
[N] if (!(*buf++ & perf_ibs->valid_mask)) /* false due to stale
[N] * IBS_OP_CTL value */
[N] goto fail;
[N]
[N] ...
[N]
[N] perf_ibs_enable_event() // *Accidentally* enable the event.
[N] }
[N]
[N] /*

  [N]  * Repeated NMIs may follow due to accidentally enabled IBS OP
  [N]  * event if the sample period is very low. It could also lead
  [N]  * to pcpu->state corruption if the event gets throttled due
  [N]  * to too frequent NMIs.
  [N]  */
  [P]         perf_ibs_enable_event();
  [P]     }
  [P] }
  --------

We cannot safely clear IBS_{FETCH|OP}_CTL while disabling the event, because the register might be read again later. So, clear the register in the enable path - before we update pcpu->state and enable the event. This guarantees that any NMI that lands in the gap finds Val=0 and bails out cleanly.

---
 arch/x86/events/amd/ibs.c | 8 ++++++++
 1 file changed, 8 insertions(+)
arch/x86/events/amd/ibs.c +8 -0
index 004226b..32e6456 100644
@@ -494,6 +494,14 @@ static void perf_ibs_start(struct perf_event *event, int flags)
 	config |= period >> 4;
 
 	/*
+	 * Reset the IBS_{FETCH|OP}_CTL MSR before updating pcpu->state.
+	 * Doing so prevents a race condition in which an NMI due to other
+	 * source might accidentally activate the event before we enable
+	 * it ourselves.
+	 */
+	perf_ibs_disable_event(perf_ibs, hwc, 0);
+
+	/*
 	 * Set STARTED before enabling the hardware, such that a subsequent NMI
 	 * must observe it.
 	 */
maintainer Acked-by on the v2 series
#19 lore raw
In reply to Ravi Bangoria

Hello Ravi,

Quoted 13 lines · Ravi Bangoria

Address several issues in the AMD IBS perf code: interrupt accounting for discarded samples, a Zen5-specific quirk, PhyAddrVal handling, NMI-safety with perf_allow_kernel(), and a race between event add and NMI.

Patches are prepared on tip/perf/core (7db06e329af3)

v1: https://lore.kernel.org/r/[email protected]
v1->v2:
- Split fixes series from future enhancements
- Add interrupt throttling for zero-rip erratum #1197
- Avoid calling perf_allow_kernel() in NMI context (new patch)

Quoted 7 lines

Ravi Bangoria (5):
perf/amd/ibs: Account interrupt for discarded samples perf/amd/ibs: Limit ldlat->l3missonly dependency to Zen5 perf/amd/ibs: Preserve PhyAddrVal bit when clearing PhyAddr MSR perf/amd/ibs: Avoid calling perf_allow_kernel() from the IBS NMI
handler
perf/amd/ibs: Avoid race between event add and NMI

Thanks,
Namhyung

Quoted 8 lines
 arch/x86/events/amd/ibs.c          | 28 ++++++++++++++++++++++------
 arch/x86/events/perf_event_flags.h |  1 +
 2 files changed, 23 insertions(+), 6 deletions(-)
Signature · 1 line
2.43.0