[PATCH 3/3] drm/xe/guc: Cancel ongoing H2G requests when stopping CT

[Michal Wajdeczko]

On 09.07.2025 21:07, Cavitt, Jonathan wrote:
> 
>>
>> -----Original Message-----
> From: Intel-xe <intel-xe-bounces at lists.freedesktop.org> On Behalf Of Michal Wajdeczko
> Sent: Wednesday, July 9, 2025 10:41 AM
> To: intel-xe at lists.freedesktop.org
> Cc: Wajdeczko, Michal <Michal.Wajdeczko at intel.com>; Brost, Matthew <matthew.brost at intel.com>
> Subject: [PATCH 3/3] drm/xe/guc: Cancel ongoing H2G requests when stopping CT
>>
>> Once we have started a GT reset sequence, which includes stopping
>> GuC CTB communication, we should also cancel all onging H2G send-
>> recv requests, as either GuC is already dead, or due to imminent
>> reset GuC will not be able to reply, or due to internal cleanup
>> we will loose pending fences. With this we will report dedicated
>> -ECANCELED error instead of misleading -ETIME.
>>
>> Signed-off-by: Michal Wajdeczko <michal.wajdeczko at intel.com>
>> Cc: Matthew Brost <matthew.brost at intel.com>
> 
> s/onging/ongoing
> s/loose/lose
> 
> And also, are we certain that reporting -ECANCELED here won't result
> in some test failure due to expecting -ETIME in this failure case?  

I don't think there are any tests that would rely on the specific error
code to be seen during reset, and even if there is one, it should be
modified to expect -ECANCELED since this is what we already use to
report broken attempts to send H2G when we are in STOPPED state

> I'd
> not expect so, but I'd like to be certain before proceeding.

and as expected, CI didn't catch anything ;)

> Otherwise, though:
> Reviewed-by: Jonathan Cavitt <jonathan.cavitt at intel.com>
> -Jonathan Cavitt

thanks

> 
>> ---
>>  drivers/gpu/drm/xe/xe_guc_ct.c | 24 ++++++++++++++++++++++++
>>  1 file changed, 24 insertions(+)
>>
>> diff --git a/drivers/gpu/drm/xe/xe_guc_ct.c b/drivers/gpu/drm/xe/xe_guc_ct.c
>> index 17e5870baf33..b6acccfcd351 100644
>> --- a/drivers/gpu/drm/xe/xe_guc_ct.c
>> +++ b/drivers/gpu/drm/xe/xe_guc_ct.c
>> @@ -85,6 +85,7 @@ struct g2h_fence {
>>  	u16 error;
>>  	u16 hint;
>>  	u16 reason;
>> +	bool cancel;
>>  	bool retry;
>>  	bool fail;
>>  	bool done;
>> @@ -103,6 +104,13 @@ static void g2h_fence_init(struct g2h_fence *g2h_fence, u32 *response_buffer)
>>  	g2h_fence->seqno = ~0x0;
>>  }
>>  
>> +static void g2h_fence_cancel(struct g2h_fence *g2h_fence)
>> +{
>> +	g2h_fence->cancel = true;
>> +	g2h_fence->fail = true;
>> +	g2h_fence->done = true;
>> +}
>> +
>>  static bool g2h_fence_needs_alloc(struct g2h_fence *g2h_fence)
>>  {
>>  	return g2h_fence->seqno == ~0x0;
>> @@ -388,6 +396,8 @@ static void guc_ct_change_state(struct xe_guc_ct *ct,
>>  				enum xe_guc_ct_state state)
>>  {
>>  	struct xe_gt *gt = ct_to_gt(ct);
>> +	struct g2h_fence *g2h_fence;
>> +	unsigned long idx;
>>  
>>  	mutex_lock(&ct->lock);		/* Serialise dequeue_one_g2h() */
>>  	spin_lock_irq(&ct->fast_lock);	/* Serialise CT fast-path */
>> @@ -406,6 +416,14 @@ static void guc_ct_change_state(struct xe_guc_ct *ct,
>>  
>>  	spin_unlock_irq(&ct->fast_lock);
>>  
>> +	/* cancel all in-flight send-recv requests */
>> +	xa_for_each(&ct->fence_lookup, idx, g2h_fence)
>> +		g2h_fence_cancel(g2h_fence);
>> +
>> +	/* make sure guc_ct_send_recv() will see g2h_fence changes */
>> +	smp_mb();
>> +	wake_up_all(&ct->g2h_fence_wq);
>> +
>>  	/*
>>  	 * Lockdep doesn't like this under the fast lock and he destroy only
>>  	 * needs to be serialized with the send path which ct lock provides.
>> @@ -1098,6 +1116,11 @@ static int guc_ct_send_recv(struct xe_guc_ct *ct, const u32 *action, u32 len,
>>  		goto retry;
>>  	}
>>  	if (g2h_fence.fail) {
>> +		if (g2h_fence.cancel) {
>> +			xe_gt_dbg(gt, "H2G request %#x canceled!\n", action[0]);
>> +			ret = -ECANCELED;
>> +			goto unlock;
>> +		}
>>  		xe_gt_err(gt, "H2G request %#x failed: error %#x hint %#x\n",
>>  			  action[0], g2h_fence.error, g2h_fence.hint);
>>  		ret = -EIO;
>> @@ -1106,6 +1129,7 @@ static int guc_ct_send_recv(struct xe_guc_ct *ct, const u32 *action, u32 len,
>>  	if (ret > 0)
>>  		ret = response_buffer ? g2h_fence.response_len : g2h_fence.response_data;
>>  
>> +unlock:
>>  	mutex_unlock(&ct->lock);
>>  
>>  	return ret;
>> -- 
>> 2.47.1
>>
>>