linux-kernel - [PATCH] ARM: perf: Use correct unwind info for tracepoint events

lists.openwall.net		lists / announce owl-users owl-dev john-users john-dev passwdqc-users yescrypt popa3d-users / oss-security kernel-hardening musl sabotage tlsify passwords / crypt-dev xvendor / Bugtraq Full-Disclosure linux-kernel linux-netdev linux-ext4 linux-hardening linux-cve-announce PHC
Open Source and information security mailing list archives

Hash Suite: Windows password security audit tool. GUI, reports in PDF.

[<prev] [next>] [day] [month] [year] [list]

Message-ID: <20240227065828.24598-1-haibo.li@mediatek.com>
Date: Tue, 27 Feb 2024 14:58:28 +0800
From: Haibo Li <haibo.li@...iatek.com>
To: <tnovak@...com>
CC: <acme@...nel.org>, <alexander.shishkin@...ux.intel.com>,
	<jolsa@...nel.org>, <linux-arm-kernel@...ts.infradead.org>,
	<linux-kernel@...r.kernel.org>, <linux-perf-users@...r.kernel.org>,
	<linux@...linux.org.uk>, <mark.rutland@....com>, <mingo@...hat.com>,
	<namhyung@...nel.org>, <peterz@...radead.org>, <will@...nel.org>,
	<xiaoming.yu@...iatek.com>, Haibo Li <haibo.li@...iatek.com>
Subject: [PATCH] ARM: perf: Use correct unwind info for tracepoint events

>Unwinding tends to work okay in practice 
>because perf_trace_* functions and their callers 
>usually have similar prologues.A common unwind insn sequence
>is "sp=fp, sp-=X, pop_r4_to_rN", so even if prologues don't match exactly,
>stored LR gets correctly popped off the stack.

It does not work well for all tracepoint events.
perf gets incomplete stacktrace after below commands
if CONFIG_ARM_UNWIND is on:

perf record -e block:block_rq_complete  -g sleep 10
perf script
'''
swapper     0 [000] 50333.927556: block:block_rq_complete:
 254,0 WSM () 376888 + 18 [0]
      804ac458 __traceiter_block_rq_complete+0x3c ([kernel.kallsyms])
      806bdd70 virtblk_done+0xc8 ([kernel.kallsyms])
      10 [unknown] ([unknown])
'''
In experiment,
ext4:ext4_da_write_begin,
block:block_rq_complete,
csd:csd_function_entry
suffer from above problem(maybe more).

As you said,it is caused by mismatched ARM_pc.

After apply similar patch,now perf gets correct stacktrace:
swapper     0 [000]   122.003800: block:block_rq_complete:
 254,0 WSM () 377064 + 16 [0]
 804acdd4 perf_trace_block_rq_completion+0x80 ([kernel.kallsyms])
 804ac2b0 __traceiter_block_rq_complete+0x3c ([kernel.kallsyms])
 804b9388 blk_update_request+0x64 ([kernel.kallsyms])
 804b9ab4 blk_mq_end_request+0x14 ([kernel.kallsyms])
 804c0bac blk_done_softirq+0x58 ([kernel.kallsyms])
 8010135c __softirqentry_text_start+0x138 ([kernel.kallsyms])
 801298f8 __irq_exit_rcu+0x58 ([kernel.kallsyms])
 801299a0 irq_exit+0x8 ([kernel.kallsyms])
 80100ba8 __irq_svc+0x88 ([kernel.kallsyms])
 80165628 arch_local_irq_enable+0x4 ([kernel.kallsyms])
 801659d8 do_idle+0xdc ([kernel.kallsyms])
 80165bf4 cpu_startup_entry+0x28 ([kernel.kallsyms])
 809c8764 kernel_init+0x0 ([kernel.kallsyms])

Acked-by: Haibo Li <haibo.li@...iatek.com>