Re: [PATCH V9 10/10] arm64/perf: Implement branch records save on PMU IRQ

From: Anshuman Khandual
Date: Tue May 23 2023 - 23:10:58 EST



On 5/23/23 20:09, James Clark wrote:
>
>
> On 15/03/2023 05:14, Anshuman Khandual wrote:
>> This modifies armv8pmu_branch_read() to concatenate live entries along with
>> task context stored entries and then process the resultant buffer to create
>> perf branch entry array for perf_sample_data. It follows the same principle
>> like task sched out.
>>
>> Cc: Catalin Marinas <catalin.marinas@xxxxxxx>
>> Cc: Will Deacon <will@xxxxxxxxxx>
>> Cc: Mark Rutland <mark.rutland@xxxxxxx>
>> Cc: linux-arm-kernel@xxxxxxxxxxxxxxxxxxx
>> Cc: linux-kernel@xxxxxxxxxxxxxxx
>> Signed-off-by: Anshuman Khandual <anshuman.khandual@xxxxxxx>
>> ---
>
> [...]
>
>> void armv8pmu_branch_read(struct pmu_hw_events *cpuc, struct perf_event *event)
>> {
>> struct brbe_hw_attr *brbe_attr = (struct brbe_hw_attr *)cpuc->percpu_pmu->private;
>> + struct arm64_perf_task_context *task_ctx = event->pmu_ctx->task_ctx_data;
>> + struct brbe_regset live[BRBE_MAX_ENTRIES];
>> + int nr_live, nr_store;
>> u64 brbfcr, brbcr;
>> - int idx, loop1_idx1, loop1_idx2, loop2_idx1, loop2_idx2, count;
>>
>> brbcr = read_sysreg_s(SYS_BRBCR_EL1);
>> brbfcr = read_sysreg_s(SYS_BRBFCR_EL1);
>> @@ -739,36 +743,13 @@ void armv8pmu_branch_read(struct pmu_hw_events *cpuc, struct perf_event *event)
>> write_sysreg_s(brbfcr | BRBFCR_EL1_PAUSED, SYS_BRBFCR_EL1);
>> isb();
>>
>> - /* Determine the indices for each loop */
>> - loop1_idx1 = BRBE_BANK0_IDX_MIN;
>> - if (brbe_attr->brbe_nr <= BRBE_BANK_MAX_ENTRIES) {
>> - loop1_idx2 = brbe_attr->brbe_nr - 1;
>> - loop2_idx1 = BRBE_BANK1_IDX_MIN;
>> - loop2_idx2 = BRBE_BANK0_IDX_MAX;
>> - } else {
>> - loop1_idx2 = BRBE_BANK0_IDX_MAX;
>> - loop2_idx1 = BRBE_BANK1_IDX_MIN;
>> - loop2_idx2 = brbe_attr->brbe_nr - 1;
>> - }
>> -
>> - /* Loop through bank 0 */
>> - select_brbe_bank(BRBE_BANK_IDX_0);
>> - for (idx = 0, count = loop1_idx1; count <= loop1_idx2; idx++, count++) {
>> - if (!capture_branch_entry(cpuc, event, idx))
>> - goto skip_bank_1;
>> - }
>> -
>> - /* Loop through bank 1 */
>> - select_brbe_bank(BRBE_BANK_IDX_1);
>> - for (count = loop2_idx1; count <= loop2_idx2; idx++, count++) {
>> - if (!capture_branch_entry(cpuc, event, idx))
>> - break;
>> - }
>> -
>> -skip_bank_1:
>> - cpuc->branches->branch_stack.nr = idx;
>> - cpuc->branches->branch_stack.hw_idx = -1ULL;
>> + nr_live = capture_brbe_regset(brbe_attr, live);
>> + nr_store = task_ctx->nr_brbe_records;
>> + nr_store = stitch_stored_live_entries(task_ctx->store, live, nr_store,
>> + nr_live, brbe_attr->brbe_nr);
>> + process_branch_entries(cpuc, event, task_ctx->store, nr_store);
>
> Hi Anshuman,
>
> With the following command I get a crash:
>
> perf record --branch-filter any,save_type -a -- ls
>
> [ 101.171822] Unable to handle kernel NULL pointer dereference at
> virtual address 0000000000000600
> ...
> [145380.414654] Call trace:
> [145380.414739] armv8pmu_branch_read+0x7c/0x578
> [145380.414895] armv8pmu_handle_irq+0x104/0x1c0
> [145380.415043] armpmu_dispatch_irq+0x38/0x70
> [145380.415209] __handle_irq_event_percpu+0x124/0x3b8
> [145380.415392] handle_irq_event+0x54/0xc8
> [145380.415567] handle_fasteoi_irq+0x100/0x1e0
> [145380.415718] generic_handle_domain_irq+0x38/0x58
> [145380.415895] gic_handle_irq+0x5c/0x130
> [145380.416025] call_on_irq_stack+0x24/0x58
> [145380.416173] el1_interrupt+0x74/0xc0
> [145380.416321] el1h_64_irq_handler+0x18/0x28
> [145380.416475] el1h_64_irq+0x64/0x68
> [145380.416604] smp_call_function_single+0xe8/0x1f0
> [145380.416745] event_function_call+0xbc/0x1c8
> [145380.416919] _perf_event_enable+0x84/0xa0
> [145380.417069] perf_ioctl+0xe8/0xd68
> [145380.417204] __arm64_sys_ioctl+0x9c/0xe0
> [145380.417353] invoke_syscall+0x4c/0x120
> [145380.417523] el0_svc_common+0xd0/0x120
> [145380.417693] do_el0_svc+0x3c/0xb8
> [145380.417859] el0_svc+0x50/0xc0
> [145380.418004] el0t_64_sync_handler+0x84/0xf0
> [145380.418160] el0t_64_sync+0x190/0x198
>
> When using --branch-filter any,u without -a it seems to be fine so could
> be that task_ctx is null in per-cpu mode, or something to do with the
> userspace only flag?

It's task_ctx is null in per-cpu mode, because armv8pmu_branch_read() used
event->pmu_ctx->task_ctx_data in per-cpu mode where it would not have been
allocated as well. Following change fixes the problem.

diff --git a/drivers/perf/arm_brbe.c b/drivers/perf/arm_brbe.c
index 9e441141a2c3..c8fd581eacf9 100644
--- a/drivers/perf/arm_brbe.c
+++ b/drivers/perf/arm_brbe.c
@@ -744,12 +744,17 @@ void armv8pmu_branch_read(struct pmu_hw_events *cpuc, struct perf_event *event)
isb();

nr_live = capture_brbe_regset(brbe_attr, live);
- nr_store = task_ctx->nr_brbe_records;
- nr_store = stitch_stored_live_entries(task_ctx->store, live, nr_store,
- nr_live, brbe_attr->brbe_nr);
- process_branch_entries(cpuc, event, task_ctx->store, nr_store);
+ if (event->ctx->task) {
+ nr_store = task_ctx->nr_brbe_records;
+ nr_store = stitch_stored_live_entries(task_ctx->store, live, nr_store,
+ nr_live, brbe_attr->brbe_nr);
+ process_branch_entries(cpuc, event, task_ctx->store, nr_store);
+ task_ctx->nr_brbe_records = 0;
+ } else {
+ process_branch_entries(cpuc, event, live, nr_live);
+ }
+
process_branch_aborts(cpuc);
- task_ctx->nr_brbe_records = 0;

/* Unpause the buffer */
write_sysreg_s(brbfcr & ~BRBFCR_EL1_PAUSED, SYS_BRBFCR_EL1);


>
> I'm also wondering if it's possible to collapse some of the last 5
> commits? They seem to mostly modify things in brbe.c which is a new file
> so the history probably isn't important at this point it just makes it a
> bit harder to review.

[PATCH 6/10] enables base perf branch stack sampling on arm64 platform via
BRBE and then subsequent patches represent logical progression up until
save-stitch mechanism is implemented both during normal PMU IRQ and task
switch callbacks.

>
>> process_branch_aborts(cpuc);
>> + task_ctx->nr_brbe_records = 0;
>>
>> /* Unpause the buffer */
>> write_sysreg_s(brbfcr & ~BRBFCR_EL1_PAUSED, SYS_BRBFCR_EL1);