[PATCH] coresight: Fix crash when Perf and sysfs modes are used concurrently

[Suzuki K Poulose]

Junhao He,

Please could you test the patch and let us know if it resolves the
problem for you ?

On 06/10/2023 14:14, James Clark wrote:
> Partially revert the change in commit 6148652807ba ("coresight: Enable
> and disable helper devices adjacent to the path") which changed the bare
> call from source_ops(csdev)->enable() to coresight_enable_source() for
> Perf sessions. It was missed that coresight_enable_source() is
> specifically for the sysfs interface, rather than being a generic call.
> This interferes with the sysfs reference counting to cause the following
> crash:
> 
>    $ perf record -e cs_etm/@tmc_etr0/ -C 0 &
>    $ echo 1 > /sys/bus/coresight/devices/tmc_etr0/enable_sink
>    $ echo 1 > /sys/bus/coresight/devices/etm0/enable_source
>    $ echo 0 > /sys/bus/coresight/devices/etm0/enable_source
> 
>    Unable to handle kernel NULL pointer dereference at virtual
>    address 00000000000001d0
>    Internal error: Oops: 0000000096000004 [#1] PREEMPT SMP
>    ...
>    Call trace:
>     etm4_disable+0x54/0x150 [coresight_etm4x]
>     coresight_disable_source+0x6c/0x98 [coresight]
>     coresight_disable+0x74/0x1c0 [coresight]
>     enable_source_store+0x88/0xa0 [coresight]
>     dev_attr_store+0x20/0x40
>     sysfs_kf_write+0x4c/0x68
>     kernfs_fop_write_iter+0x120/0x1b8
>     vfs_write+0x2dc/0x3b0
>     ksys_write+0x70/0x108
>     __arm64_sys_write+0x24/0x38
>     invoke_syscall+0x50/0x128
>     el0_svc_common.constprop.0+0x104/0x130
>     do_el0_svc+0x40/0xb8
>     el0_svc+0x2c/0xb8
>     el0t_64_sync_handler+0xc0/0xc8
>     el0t_64_sync+0x1a4/0x1a8
>    Code: d53cd042 91002000 b9402a81 b8626800 (f940ead5)
>    ---[ end trace 0000000000000000 ]---
> 
> This commit linked below also fixes the issue, but has unlocked updates
> to the mode which could potentially race. So until we come up with a
> more complete solution that takes all locking and interaction between
> both modes into account, just revert back to the old behavior for Perf.
> 
> Reported-by: Junhao He <hejunhao3 at huawei.com>
> Closes: https://lore.kernel.org/linux-arm-kernel/20230921132904.60996-1-hejunhao3@huawei.com/
> Fixes: 6148652807ba ("coresight: Enable and disable helper devices adjacent to the path")
> Signed-off-by: James Clark <james.clark at arm.com>

The patch looks good to me. I will wait for Junhao to test this before
pulling it in.

Suzuki




> ---
>   drivers/hwtracing/coresight/coresight-etm-perf.c | 4 ++--
>   1 file changed, 2 insertions(+), 2 deletions(-)
> 
> diff --git a/drivers/hwtracing/coresight/coresight-etm-perf.c b/drivers/hwtracing/coresight/coresight-etm-perf.c
> index 5ca6278baff4..89e8ed214ea4 100644
> --- a/drivers/hwtracing/coresight/coresight-etm-perf.c
> +++ b/drivers/hwtracing/coresight/coresight-etm-perf.c
> @@ -493,7 +493,7 @@ static void etm_event_start(struct perf_event *event, int flags)
>   		goto fail_end_stop;
>   
>   	/* Finally enable the tracer */
> -	if (coresight_enable_source(csdev, CS_MODE_PERF, event))
> +	if (source_ops(csdev)->enable(csdev, event, CS_MODE_PERF))
>   		goto fail_disable_path;
>   
>   	/*
> @@ -587,7 +587,7 @@ static void etm_event_stop(struct perf_event *event, int mode)
>   		return;
>   
>   	/* stop tracer */
> -	coresight_disable_source(csdev, event);
> +	source_ops(csdev)->disable(csdev, event);
>   
>   	/* tell the core */
>   	event->hw.state = PERF_HES_STOPPED;