[PATCH] drm/amdgpu: Fix discovery initialization failure during pci rescan

[Ma, Jun]

On 4/1/2024 8:24 PM, Christian König wrote:
> Am 01.04.24 um 12:18 schrieb Ma Jun:
>> Waiting for system ready to fix the discovery initialization
>> failure issue. This failure usually occurs when dGPU is
>> removed and then rescanned via command line.
>> It's caused by following two errors:
>> [1] vram size is 0
>> [2] wrong binary signature
>>
>> Signed-off-by: Ma Jun <Jun.Ma2 at amd.com>
>> ---
>>   drivers/gpu/drm/amd/amdgpu/amdgpu_discovery.c | 15 ++++++++++++---
>>   1 file changed, 12 insertions(+), 3 deletions(-)
>>
>> diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_discovery.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_discovery.c
>> index 07c5fca06178..ac6b2ae6414c 100644
>> --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_discovery.c
>> +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_discovery.c
>> @@ -276,7 +276,12 @@ static int amdgpu_discovery_read_binary_from_mem(struct amdgpu_device *adev,
>>   			msleep(1);
>>   		}
>>   	}
>> -	vram_size = (uint64_t)RREG32(mmRCC_CONFIG_MEMSIZE) << 20;
>> +	for (i = 0; i < 100; i++) {
>> +		vram_size = (uint64_t)RREG32(mmRCC_CONFIG_MEMSIZE) << 20;
>> +		if (vram_size)
>> +			break;
>> +		usleep_range(1000, 1100);
>> +	}
> 
> Nice that we come closer to a solution for this, but that here is really 
> hacky.
> 
> Do we have any idea why mmRCC_CONFIG_MEMSIZE is zero when we re-scan? Is 
> some initialization not completed yet or something like that.
> 
I just found there is a patch related to my issue.
https://lore.kernel.org/all/20240306174446.2042884-1-alexander.deucher@amd.com/

But I think it should not be limited to removable devices or devices with
discovery_version=2 and 4. 
I will submit v2 based on this patch.

Regards,
Ma Jun

> Regards,
> Christian.
> 
>>   
>>   	if (vram_size) {
>>   		uint64_t pos = vram_size - DISCOVERY_TMR_OFFSET;
>> @@ -371,6 +376,7 @@ static int amdgpu_discovery_init(struct amdgpu_device *adev)
>>   {
>>   	struct table_info *info;
>>   	struct binary_header *bhdr;
>> +	int error_count = 0;
>>   	uint16_t offset;
>>   	uint16_t size;
>>   	uint16_t checksum;
>> @@ -380,7 +386,7 @@ static int amdgpu_discovery_init(struct amdgpu_device *adev)
>>   	adev->mman.discovery_bin = kzalloc(adev->mman.discovery_tmr_size, GFP_KERNEL);
>>   	if (!adev->mman.discovery_bin)
>>   		return -ENOMEM;
>> -
>> +retry:
>>   	/* Read from file if it is the preferred option */
>>   	if (amdgpu_discovery == 2) {
>>   		dev_info(adev->dev, "use ip discovery information from file");
>> @@ -401,6 +407,10 @@ static int amdgpu_discovery_init(struct amdgpu_device *adev)
>>   
>>   	/* check the ip discovery binary signature */
>>   	if (!amdgpu_discovery_verify_binary_signature(adev->mman.discovery_bin)) {
>> +		if (error_count++ < 1) {
>> +			msleep(100);
>> +			goto retry;
>> +		}
>>   		dev_err(adev->dev,
>>   			"get invalid ip discovery binary signature\n");
>>   		r = -EINVAL;
>> @@ -515,7 +525,6 @@ static int amdgpu_discovery_init(struct amdgpu_device *adev)
>>   	if (0 && offset) {
>>   		struct mall_info_header *mhdr =
>>   			(struct mall_info_header *)(adev->mman.discovery_bin + offset);
>> -
>>   		if (le32_to_cpu(mhdr->table_id) != MALL_INFO_TABLE_ID) {
>>   			dev_err(adev->dev, "invalid ip discovery mall table id\n");
>>   			r = -EINVAL;
>