Re: [PATCH 3/3] perf/core: Don't zero perf_sample_data in do_perf_sw_event()

From: Usama Arif

Date: Fri Nov 20 2026 - 15:05:27 EST




On 20/11/2026 01:12, Peter Zijlstra wrote:
> On Fri, Nov 20, 2026 at 10:11:51AM +0100, Peter Zijlstra wrote:
>> On Thu, Nov 19, 2026 at 01:10:46PM -0800, Usama Arif wrote:
>>
>>> kernel/events/core.c | 2 +-
>>> 1 file changed, 1 insertion(+), 1 deletion(-)
>>>
>>> diff --git a/kernel/events/core.c b/kernel/events/core.c
>>> index 57ea7e3c01a025..fa10ad7cfd99af 100644
>>> --- a/kernel/events/core.c
>>> +++ b/kernel/events/core.c
>>> @@ -11212,7 +11212,7 @@ static void do_perf_sw_event(enum perf_type_id type, u32 event_id,
>>> u64 nr, struct pt_regs *regs, u64 addr)
>>> {
>>> struct swevent_htable *swhash = this_cpu_ptr(&swevent_htable);
>>> - struct perf_sample_data data;
>>> + struct perf_sample_data data __uninitialized;
>>> struct perf_event *event;
>>> struct hlist_head *head;
>>>
>>
>> I like this one, didn't know about the __uninitialized thing. But I
>> think we wants to be more consistent with this.
>>
>> A little git-grep seems to suggest we ought to do the same elsewhere.
>> Shall I fold the little something below in this patch?
>>
>
> And this should be a separate patch again I think.

Agreed, but IBS is not safe yet. For PERF_SAMPLE_WEIGHT_STRUCT,
perf_ibs_parse_ld_st_data() writes only var1_dw and var2_w but sets
PERF_SAMPLE_WEIGHT_TYPE, so perf_output_sample() copies var3_w from
whatever is on the stack. INIT_STACK_ALL_ZERO hides that today, and
INIT_STACK_NONE kernels have leaked those 2 bytes since 6b2ae4952ef8.
With __uninitialized in perf_ibs_handle_irq() every kernel would.

We need the below:

commit e662f8539ca00096db3ee4d570efe97767427fbe
Author: Usama Arif <usama.arif@xxxxxxxxx>
Date: Fri Nov 20 09:04:13 2026 -0800

perf/x86/amd/ibs: Don't leak stack contents through PERF_SAMPLE_WEIGHT_STRUCT

For a load that missed the data cache, perf_ibs_parse_ld_st_data()
reports the miss latency as the sample weight and marks
PERF_SAMPLE_WEIGHT_TYPE as done, so perf_prepare_sample() leaves the
field alone. With PERF_SAMPLE_WEIGHT_STRUCT it writes only var1_dw and
var2_w, but perf_output_sample() copies all 8 bytes of the union into
the ring buffer. var3_w therefore carries whatever was left in the
on-stack struct perf_sample_data in perf_ibs_handle_irq().

CONFIG_INIT_STACK_ALL_ZERO hides this by clearing the struct on entry.
Without it, every such sample hands 2 bytes of kernel stack to user
space. "perf mem record" asks for PERF_SAMPLE_WEIGHT_STRUCT on x86, and
since v6.14 swfilt lets an unprivileged user sample their own tasks with
IBS.

Write var3_w as well.

Fixes: 6b2ae4952ef8 ("perf/x86/amd: Support PERF_SAMPLE_{WEIGHT|WEIGHT_STRUCT}")
Cc: stable@xxxxxxxxxxxxxxx
Signed-off-by: Usama Arif <usama.arif@xxxxxxxxx>

diff --git a/arch/x86/events/amd/ibs.c b/arch/x86/events/amd/ibs.c
index 66e67e27224f18..9aa47751b37f6c 100644
--- a/arch/x86/events/amd/ibs.c
+++ b/arch/x86/events/amd/ibs.c
@@ -1376,6 +1376,7 @@ static void perf_ibs_parse_ld_st_data(__u64 sample_type,
if (sample_type & PERF_SAMPLE_WEIGHT_STRUCT) {
data->weight.var1_dw = op_data3.dc_miss_lat;
data->weight.var2_w = op_data.tag_to_ret_ctr;
+ data->weight.var3_w = 0;
} else if (sample_type & PERF_SAMPLE_WEIGHT) {
data->weight.full = op_data3.dc_miss_lat;
}


The other sites look fine to me, including arch-PEBS: the drain skips
records whose first fragment has no basic group, so var3_w is always
written there.



>
> ---
> diff --git a/arch/x86/events/amd/core.c b/arch/x86/events/amd/core.c
> index 49b6b8fce566..db6c87552594 100644
> --- a/arch/x86/events/amd/core.c
> +++ b/arch/x86/events/amd/core.c
> @@ -958,7 +958,7 @@ static int amd_pmu_v2_handle_irq(struct pt_regs *regs)
> struct cpu_hw_events *cpuc = this_cpu_ptr(&cpu_hw_events);
> static atomic64_t status_warned = ATOMIC64_INIT(0);
> u64 reserved, status, mask, new_bits, prev_bits;
> - struct perf_sample_data data;
> + struct perf_sample_data data __uninitialized;
> struct hw_perf_event *hwc;
> struct perf_event *event;
> int handled = 0, idx;
> diff --git a/arch/x86/events/amd/ibs.c b/arch/x86/events/amd/ibs.c
> index 96c96369480a..cf0eafd67430 100644
> --- a/arch/x86/events/amd/ibs.c
> +++ b/arch/x86/events/amd/ibs.c
> @@ -1483,7 +1483,7 @@ static int perf_ibs_handle_irq(struct perf_ibs *perf_ibs, struct pt_regs *iregs)
> struct cpu_perf_ibs *pcpu = this_cpu_ptr(perf_ibs->pcpu);
> struct perf_event *event = pcpu->event;
> struct hw_perf_event *hwc;
> - struct perf_sample_data data;
> + struct perf_sample_data data __uninitialized;
> struct perf_raw_record raw;
> struct pt_regs regs;
> struct perf_ibs_data ibs_data;
> diff --git a/arch/x86/events/core.c b/arch/x86/events/core.c
> index 4e344fb1ae3e..5f8bfc65b256 100644
> --- a/arch/x86/events/core.c
> +++ b/arch/x86/events/core.c
> @@ -2157,7 +2157,7 @@ void x86_pmu_update_perf_regs(struct perf_event *event,
>
> int x86_pmu_handle_irq(struct pt_regs *regs)
> {
> - struct perf_sample_data data;
> + struct perf_sample_data data __uninitialized;
> struct cpu_hw_events *cpuc;
> struct perf_event *event;
> int idx, handled = 0;
> diff --git a/arch/x86/events/intel/core.c b/arch/x86/events/intel/core.c
> index 0a34d674df59..5d84c18096a9 100644
> --- a/arch/x86/events/intel/core.c
> +++ b/arch/x86/events/intel/core.c
> @@ -3819,7 +3819,7 @@ static int x86_pmu_handle_guest_pebs(struct pt_regs *regs,
>
> static int handle_pmi_common(struct pt_regs *regs, u64 status)
> {
> - struct perf_sample_data data;
> + struct perf_sample_data data __uninitialized;
> struct cpu_hw_events *cpuc = this_cpu_ptr(&cpu_hw_events);
> int bit;
> int handled = 0;
> diff --git a/arch/x86/events/intel/ds.c b/arch/x86/events/intel/ds.c
> index 8444670cee4a..73f6fd66856f 100644
> --- a/arch/x86/events/intel/ds.c
> +++ b/arch/x86/events/intel/ds.c
> @@ -1184,7 +1184,7 @@ int intel_pmu_drain_bts_buffer(void)
> struct bts_record *at, *base, *top;
> struct perf_output_handle handle;
> struct perf_event_header header;
> - struct perf_sample_data data;
> + struct perf_sample_data data __uninitialized;
> unsigned long skip = 0;
> struct pt_regs regs;
>
> @@ -1265,7 +1265,7 @@ int intel_pmu_drain_bts_buffer(void)
> void intel_pmu_drain_pebs_buffer(void)
> {
> struct cpu_hw_events *cpuc = this_cpu_ptr(&cpu_hw_events);
> - struct perf_sample_data data;
> + struct perf_sample_data data __uninitialized;
>
> WARN_ON_ONCE(cpuc->enabled);
>
> diff --git a/arch/x86/events/intel/knc.c b/arch/x86/events/intel/knc.c
> index e887adc108ac..aa38b0ac3c7b 100644
> --- a/arch/x86/events/intel/knc.c
> +++ b/arch/x86/events/intel/knc.c
> @@ -213,7 +213,7 @@ static inline void knc_pmu_ack_status(u64 ack)
>
> static int knc_pmu_handle_irq(struct pt_regs *regs)
> {
> - struct perf_sample_data data;
> + struct perf_sample_data data __uninitialized;
> struct cpu_hw_events *cpuc;
> int handled = 0;
> int bit, loops;
> diff --git a/arch/x86/events/intel/p4.c b/arch/x86/events/intel/p4.c
> index 5368dc31787c..4ac429e55935 100644
> --- a/arch/x86/events/intel/p4.c
> +++ b/arch/x86/events/intel/p4.c
> @@ -1033,7 +1033,7 @@ static int p4_pmu_set_period(struct perf_event *event)
>
> static int p4_pmu_handle_irq(struct pt_regs *regs)
> {
> - struct perf_sample_data data;
> + struct perf_sample_data data __uninitialized;
> struct cpu_hw_events *cpuc;
> struct perf_event *event;
> struct hw_perf_event *hwc;
> diff --git a/arch/x86/events/zhaoxin/core.c b/arch/x86/events/zhaoxin/core.c
> index e506f677db57..fc98a7fdd3bc 100644
> --- a/arch/x86/events/zhaoxin/core.c
> +++ b/arch/x86/events/zhaoxin/core.c
> @@ -355,7 +355,7 @@ static void zhaoxin_pmu_enable_event(struct perf_event *event)
> */
> static int zhaoxin_pmu_handle_irq(struct pt_regs *regs)
> {
> - struct perf_sample_data data;
> + struct perf_sample_data data __uninitialized;
> struct cpu_hw_events *cpuc;
> int handled = 0;
> u64 status;
> diff --git a/arch/x86/include/asm/stop_machine.h b/arch/x86/include/asm/stop_machine.h
> deleted file mode 100644
> index e69de29bb2d1..000000000000