Re: [PATCHv3 2/3] drm/amdkfd: Update queue unmap after VM fault with MES

[Felix Kuehling <felix.kuehling@amd.com>]

On 2024-08-16 14:01, Mukul Joshi wrote:
> MEC FW expects MES to unmap all queues when a VM fault is observed
> on a queue and then resumed once the affected process is terminated.
> Use the MES Suspend and Resume APIs to achieve this.
>
> Signed-off-by: Mukul Joshi <mukul.joshi@amd.com>
> Acked-by: Alex Deucher <alexander.deucher@amd.com>

Reviewed-by: Felix Kuehling <felix.kuehling@amd.com>


> ---
> v1->v2:
> - Add MES FW version check.
> - Separate out the kfd_dqm_evict_pasid into another function.
> - Use amdgpu_mes_suspend/amdgpu_mes_resume to suspend/resume queues.
>
> v2->v3:
> - Use down_read_trylock/up_read instead of dqm->is_hws_hang.
> - Increase eviction count if the process is already evicted in
>    kfd_dqm_evict_pasid_mes to make sure the process stays evicted.
>
>   .../drm/amd/amdkfd/kfd_device_queue_manager.c | 87 ++++++++++++++++++-
>   1 file changed, 85 insertions(+), 2 deletions(-)
>
> diff --git a/drivers/gpu/drm/amd/amdkfd/kfd_device_queue_manager.c b/drivers/gpu/drm/amd/amdkfd/kfd_device_queue_manager.c
> index f6e211070299..0ca933d2099c 100644
> --- a/drivers/gpu/drm/amd/amdkfd/kfd_device_queue_manager.c
> +++ b/drivers/gpu/drm/amd/amdkfd/kfd_device_queue_manager.c
> @@ -319,6 +319,46 @@ static int remove_all_queues_mes(struct device_queue_manager *dqm)
>   	return retval;
>   }
>   
> +static int suspend_all_queues_mes(struct device_queue_manager *dqm)
> +{
> +	struct amdgpu_device *adev = (struct amdgpu_device *)dqm->dev->adev;
> +	int r = 0;
> +
> +	if (!down_read_trylock(&adev->reset_domain->sem))
> +		return -EIO;
> +
> +	r = amdgpu_mes_suspend(adev);
> +	up_read(&adev->reset_domain->sem);
> +
> +	if (r) {
> +		dev_err(adev->dev, "failed to suspend gangs from MES\n");
> +		dev_err(adev->dev, "MES might be in unrecoverable state, issue a GPU reset\n");
> +		kfd_hws_hang(dqm);
> +	}
> +
> +	return r;
> +}
> +
> +static int resume_all_queues_mes(struct device_queue_manager *dqm)
> +{
> +	struct amdgpu_device *adev = (struct amdgpu_device *)dqm->dev->adev;
> +	int r = 0;
> +
> +	if (!down_read_trylock(&adev->reset_domain->sem))
> +		return -EIO;
> +
> +	r = amdgpu_mes_resume(adev);
> +	up_read(&adev->reset_domain->sem);
> +
> +	if (r) {
> +		dev_err(adev->dev, "failed to resume gangs from MES\n");
> +		dev_err(adev->dev, "MES might be in unrecoverable state, issue a GPU reset\n");
> +		kfd_hws_hang(dqm);
> +	}
> +
> +	return r;
> +}
> +
>   static void increment_queue_count(struct device_queue_manager *dqm,
>   				  struct qcm_process_device *qpd,
>   				  struct queue *q)
> @@ -2835,6 +2875,44 @@ void device_queue_manager_uninit(struct device_queue_manager *dqm)
>   	kfree(dqm);
>   }
>   
> +static int kfd_dqm_evict_pasid_mes(struct device_queue_manager *dqm,
> +				   struct qcm_process_device *qpd)
> +{
> +	struct device *dev = dqm->dev->adev->dev;
> +	int ret = 0;
> +
> +	/* Check if process is already evicted */
> +	dqm_lock(dqm);
> +	if (qpd->evicted) {
> +		/* Increment the evicted count to make sure the
> +		 * process stays evicted before its terminated.
> +		 */
> +		qpd->evicted++;
> +		dqm_unlock(dqm);
> +		goto out;
> +	}
> +	dqm_unlock(dqm);
> +
> +	ret = suspend_all_queues_mes(dqm);
> +	if (ret) {
> +		dev_err(dev, "Suspending all queues failed");
> +		goto out;
> +	}
> +
> +	ret = dqm->ops.evict_process_queues(dqm, qpd);
> +	if (ret) {
> +		dev_err(dev, "Evicting process queues failed");
> +		goto out;
> +	}
> +
> +	ret = resume_all_queues_mes(dqm);
> +	if (ret)
> +		dev_err(dev, "Resuming all queues failed");
> +
> +out:
> +	return ret;
> +}
> +
>   int kfd_dqm_evict_pasid(struct device_queue_manager *dqm, u32 pasid)
>   {
>   	struct kfd_process_device *pdd;
> @@ -2845,8 +2923,13 @@ int kfd_dqm_evict_pasid(struct device_queue_manager *dqm, u32 pasid)
>   		return -EINVAL;
>   	WARN(debug_evictions, "Evicting pid %d", p->lead_thread->pid);
>   	pdd = kfd_get_process_device_data(dqm->dev, p);
> -	if (pdd)
> -		ret = dqm->ops.evict_process_queues(dqm, &pdd->qpd);
> +	if (pdd) {
> +		if (dqm->dev->kfd->shared_resources.enable_mes)
> +			ret = kfd_dqm_evict_pasid_mes(dqm, &pdd->qpd);
> +		else
> +			ret = dqm->ops.evict_process_queues(dqm, &pdd->qpd);
> +	}
> +
>   	kfd_unref_process(p);
>   
>   	return ret;