[Intel-gfx] [PATCH 1/5] drm/i915: Mark up unlocked update of i915_request.hwsp_seqno

[Mika Kuoppala]

Chris Wilson <chris at chris-wilson.co.uk> writes:

> During i915_request_retire() we decouple the i915_request.hwsp_seqno
> from the intel_timeline so that it may be freed before the request is
> released. However, we need to warn the compiler that the pointer may
> update under its nose.
>
> [  171.438899] BUG: KCSAN: data-race in i915_request_await_dma_fence [i915] / i915_request_retire [i915]
> [  171.438920]
> [  171.438932] write to 0xffff8881e7e28ce0 of 8 bytes by task 148 on cpu 2:
> [  171.439174]  i915_request_retire+0x1ea/0x660 [i915]
> [  171.439408]  retire_requests+0x7a/0xd0 [i915]
> [  171.439640]  engine_retire+0xa1/0xe0 [i915]
> [  171.439657]  process_one_work+0x3b1/0x690
> [  171.439671]  worker_thread+0x80/0x670
> [  171.439685]  kthread+0x19a/0x1e0
> [  171.439701]  ret_from_fork+0x1f/0x30
> [  171.439721]
> [  171.439739] read to 0xffff8881e7e28ce0 of 8 bytes by task 696 on cpu 1:
> [  171.439990]  i915_request_await_dma_fence+0x162/0x520 [i915]
> [  171.440230]  i915_request_await_object+0x2fe/0x470 [i915]
> [  171.440467]  i915_gem_do_execbuffer+0x45dc/0x4c20 [i915]
> [  171.440704]  i915_gem_execbuffer2_ioctl+0x2c3/0x580 [i915]
> [  171.440722]  drm_ioctl_kernel+0xe4/0x120
> [  171.440736]  drm_ioctl+0x297/0x4c7
> [  171.440750]  ksys_ioctl+0x89/0xb0
> [  171.440766]  __x64_sys_ioctl+0x42/0x60
> [  171.440788]  do_syscall_64+0x6e/0x2c0
> [  171.440802]  entry_SYSCALL_64_after_hwframe+0x44/0xa9
>
> Signed-off-by: Chris Wilson <chris at chris-wilson.co.uk>
> Cc: Mika Kuoppala <mika.kuoppala at linux.intel.com>
> ---
>  drivers/gpu/drm/i915/i915_request.h | 7 +++++--
>  1 file changed, 5 insertions(+), 2 deletions(-)
>
> diff --git a/drivers/gpu/drm/i915/i915_request.h b/drivers/gpu/drm/i915/i915_request.h
> index d4bae16b4785..6020d5b2a3df 100644
> --- a/drivers/gpu/drm/i915/i915_request.h
> +++ b/drivers/gpu/drm/i915/i915_request.h
> @@ -396,7 +396,9 @@ static inline bool i915_seqno_passed(u32 seq1, u32 seq2)
>  
>  static inline u32 __hwsp_seqno(const struct i915_request *rq)
>  {
> -	return READ_ONCE(*rq->hwsp_seqno);
> +	const u32 *hwsp = READ_ONCE(rq->hwsp_seqno);
> +
> +	return READ_ONCE(*hwsp);

This is good enough for decouple. But good enough for hardware
might be different thing.

I am paranoid enough to wanting an rmb(), before the final
read once.

and clflush after.

If the hardware can't guarantee coherency in csb, why
would it in the different region in hwsp.

But the patch does the what the commit message says,
Reviewed-by: Mika Kuoppala <mika.kuoppala at linux.intel.com>

>  }
>  
>  /**
> @@ -510,7 +512,8 @@ static inline bool i915_request_completed(const struct i915_request *rq)
>  
>  static inline void i915_request_mark_complete(struct i915_request *rq)
>  {
> -	rq->hwsp_seqno = (u32 *)&rq->fence.seqno; /* decouple from HWSP */
> +	WRITE_ONCE(rq->hwsp_seqno, /* decouple from HWSP */
> +		   (u32 *)&rq->fence.seqno);
>  }
>  
>  static inline bool i915_request_has_waitboost(const struct i915_request *rq)
> -- 
> 2.20.1