[PATCH 7/7] drm/amdgpu: retry init if it fails due to exclusive mode timeout

[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

 



Thanks Andres, revised in v2.
â?? 
Sincerely Yours,
Pixel








On 23/10/2017, 11:44 PM, "amd-gfx on behalf of Andres Rodriguez" <amd-gfx-bounces at lists.freedesktop.org on behalf of andresx7 at gmail.com> wrote:

>
>
>On 2017-10-23 06:03 AM, Pixel Ding wrote:
>> From: pding <Pixel.Ding at amd.com>
>> 
>> The exclusive mode has real-time limitation in reality, such like being
>> done in 300ms. It's easy observed if running many VF/VMs in single host
>> with heavy CPU workload.
>> 
>> If we find the init fails due to exclusive mode timeout, try it again.
>> 
>> Signed-off-by: pding <Pixel.Ding at amd.com>
>> ---
>>   drivers/gpu/drm/amd/amdgpu/amdgpu_device.c | 10 ++++++++++
>>   drivers/gpu/drm/amd/amdgpu/amdgpu_kms.c    | 15 +++++++++++++--
>>   2 files changed, 23 insertions(+), 2 deletions(-)
>> 
>> diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_device.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_device.c
>> index 3458d46..1935f5a 100644
>> --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_device.c
>> +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_device.c
>> @@ -2306,6 +2306,15 @@ int amdgpu_device_init(struct amdgpu_device *adev,
>>   
>>   	r = amdgpu_init(adev);
>>   	if (r) {
>> +		/* failed in exclusive mode due to timeout */
>> +		if (amdgpu_sriov_vf(adev) &&
>> +		    !amdgpu_sriov_runtime(adev) &&
>> +		    amdgpu_virt_mmio_blocked(adev) &&
>> +		    !amdgpu_virt_wait_reset(adev)) {
>> +			dev_err(adev->dev, "VF exclusive mode timeout\n");
>> +			r = -EAGAIN;
>> +			goto failed;
>> +		}
>>   		dev_err(adev->dev, "amdgpu_init failed\n");
>>   		amdgpu_vf_error_put(adev, AMDGIM_ERROR_VF_AMDGPU_INIT_FAIL, 0, 0);
>>   		amdgpu_fini(adev);
>> @@ -2393,6 +2402,7 @@ int amdgpu_device_init(struct amdgpu_device *adev,
>>   	amdgpu_vf_error_trans_all(adev);
>>   	if (runtime)
>>   		vga_switcheroo_fini_domain_pm_ops(adev->dev);
>> +
>>   	return r;
>>   }
>>   
>> diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_kms.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_kms.c
>> index f2eb7ac..fdc240a 100644
>> --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_kms.c
>> +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_kms.c
>> @@ -86,7 +86,7 @@ void amdgpu_driver_unload_kms(struct drm_device *dev)
>>   int amdgpu_driver_load_kms(struct drm_device *dev, unsigned long flags)
>>   {
>>   	struct amdgpu_device *adev;
>> -	int r, acpi_status;
>> +	int r, acpi_status, retry = 0;
>>   
>>   #ifdef CONFIG_DRM_AMDGPU_SI
>>   	if (!amdgpu_si_support) {
>> @@ -122,6 +122,7 @@ int amdgpu_driver_load_kms(struct drm_device *dev, unsigned long flags)
>>   		}
>>   	}
>>   #endif
>> +retry_init:
>>   
>>   	adev = kzalloc(sizeof(struct amdgpu_device), GFP_KERNEL);
>>   	if (adev == NULL) {
>> @@ -144,7 +145,17 @@ int amdgpu_driver_load_kms(struct drm_device *dev, unsigned long flags)
>>   	 * VRAM allocation
>>   	 */
>>   	r = amdgpu_device_init(adev, dev, dev->pdev, flags);
>> -	if (r) {
>> +	if (++retry != 3 && r == -EAGAIN) {
>
>Minor nitpick here. Might want to rewrite the condition so that it 
>evaluates to false for most values of retry (currently it evaluates to 
>false only for one value of retry).
>
>E.g. if (++retry >= 3 ...)
>
>Or
>
>int retry = 3;
>...
>if (--retry >= 0 ...)
>
>> +		adev->virt.caps &= ~AMDGPU_SRIOV_CAPS_RUNTIME;
>> +		adev->virt.ops = NULL;
>> +		amdgpu_device_fini(adev);
>> +		kfree(adev);
>> +		dev->dev_private = NULL;
>> +		msleep(5000);
>> +		dev_err(&dev->pdev->dev, "retry init %d\n", retry);
>> +		amdgpu_init_log = 0;
>> +		goto retry_init;
>> +	} else if (r) {
>>   		dev_err(&dev->pdev->dev, "Fatal error during GPU init\n");
>>   		goto out;
>>   	}
>> 
>_______________________________________________
>amd-gfx mailing list
>amd-gfx at lists.freedesktop.org
>https://lists.freedesktop.org/mailman/listinfo/amd-gfx


[Index of Archives]     [Linux USB Devel]     [Linux Audio Users]     [Yosemite News]     [Linux Kernel]     [Linux SCSI]

  Powered by Linux