[PATCH 1/2] Revert "drm/amdkfd: fix boot failure when iommu is disabled in Picasso."

[PATCH 1/2] Revert "drm/amdkfd: fix boot failure when iommu is disabled in Picasso."

[James Zhu]

To revert wrong merge during rebase.
This reverts commit afd18180c07026f94a80ff024acef5f4159084a4.
---
 drivers/gpu/drm/amd/amdkfd/kfd_device.c | 3 ---
 1 file changed, 3 deletions(-)

diff --git a/drivers/gpu/drm/amd/amdkfd/kfd_device.c b/drivers/gpu/drm/amd/amdkfd/kfd_device.c
index 7677ced16a27..be26c4016ade 100644
--- a/drivers/gpu/drm/amd/amdkfd/kfd_device.c
+++ b/drivers/gpu/drm/amd/amdkfd/kfd_device.c
@@ -1031,9 +1031,6 @@ bool kgd2kfd_device_init(struct kfd_dev *kfd,
 
 	svm_migrate_init(kfd->adev);
 
-	if(kgd2kfd_resume_iommu(kfd))
-		goto device_iommu_error;
-
 	if (kfd_resume(kfd))
 		goto kfd_resume_error;
 
-- 
2.25.1

[PATCH 2/2] drm/amdkfd: fix boot failure when iommu is disabled in Picasso.

[James Zhu]

From: Yifan Zhang <yifan1.zhang@amd.com>

When IOMMU disabled in sbios and kfd in iommuv2 path, iommuv2
init will fail. But this failure should not block amdgpu driver init.

Reported-by: youling <youling257@gmail.com>
Tested-by: youling <youling257@gmail.com>
Signed-off-by: Yifan Zhang <yifan1.zhang@amd.com>
Reviewed-by: James Zhu <James.Zhu@amd.com>
---
 drivers/gpu/drm/amd/amdgpu/amdgpu_device.c | 4 ----
 drivers/gpu/drm/amd/amdkfd/kfd_device.c    | 3 +++
 2 files changed, 3 insertions(+), 4 deletions(-)

diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_device.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_device.c
index e56bc925afcf..f77823ce7ae8 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_device.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_device.c
@@ -2398,10 +2398,6 @@ static int amdgpu_device_ip_init(struct amdgpu_device *adev)
 	if (!adev->gmc.xgmi.pending_reset)
 		amdgpu_amdkfd_device_init(adev);
 
-	r = amdgpu_amdkfd_resume_iommu(adev);
-	if (r)
-		goto init_failed;
-
 	amdgpu_fru_get_product_info(adev);
 
 init_failed:
diff --git a/drivers/gpu/drm/amd/amdkfd/kfd_device.c b/drivers/gpu/drm/amd/amdkfd/kfd_device.c
index be26c4016ade..7677ced16a27 100644
--- a/drivers/gpu/drm/amd/amdkfd/kfd_device.c
+++ b/drivers/gpu/drm/amd/amdkfd/kfd_device.c
@@ -1031,6 +1031,9 @@ bool kgd2kfd_device_init(struct kfd_dev *kfd,
 
 	svm_migrate_init(kfd->adev);
 
+	if(kgd2kfd_resume_iommu(kfd))
+		goto device_iommu_error;
+
 	if (kfd_resume(kfd))
 		goto kfd_resume_error;
 
-- 
2.25.1

Re: [PATCH 1/2] Revert "drm/amdkfd: fix boot failure when iommu is disabled in Picasso."

[Alex Deucher]

On Tue, Nov 2, 2021 at 3:23 PM James Zhu <James.Zhu@amd.com> wrote:
>
> To revert wrong merge during rebase.
> This reverts commit afd18180c07026f94a80ff024acef5f4159084a4.

Missing your signed-off-by.  Might be better to just merge these two
patches into one unless they are individually applicable to different
kernels (e.g., 5.15 vs drm-next).  Also these will probably fix the
following:
Bug: https://bugzilla.kernel.org/show_bug.cgi?id=214859
Bug: https://gitlab.freedesktop.org/drm/amd/-/issues/1770

Alex

> ---
>  drivers/gpu/drm/amd/amdkfd/kfd_device.c | 3 ---
>  1 file changed, 3 deletions(-)
>
> diff --git a/drivers/gpu/drm/amd/amdkfd/kfd_device.c b/drivers/gpu/drm/amd/amdkfd/kfd_device.c
> index 7677ced16a27..be26c4016ade 100644
> --- a/drivers/gpu/drm/amd/amdkfd/kfd_device.c
> +++ b/drivers/gpu/drm/amd/amdkfd/kfd_device.c
> @@ -1031,9 +1031,6 @@ bool kgd2kfd_device_init(struct kfd_dev *kfd,
>
>         svm_migrate_init(kfd->adev);
>
> -       if(kgd2kfd_resume_iommu(kfd))
> -               goto device_iommu_error;
> -
>         if (kfd_resume(kfd))
>                 goto kfd_resume_error;
>
> --
> 2.25.1
>

Re: [PATCH 1/2] Revert "drm/amdkfd: fix boot failure when iommu is disabled in Picasso."

[Zhu, James]

[-- Attachment #1: Type: text/plain, Size: 2515 bytes --]

[AMD Official Use Only]


Hi Alex

Sure, Let me check it can be merged into one patch.


Thanks & Best Regards!


James Zhu

________________________________
From: Alex Deucher <alexdeucher@gmail.com>
Sent: Tuesday, November 2, 2021 4:34 PM
To: Zhu, James <James.Zhu@amd.com>
Cc: amd-gfx list <amd-gfx@lists.freedesktop.org>; Deucher, Alexander <Alexander.Deucher@amd.com>; Zhang, Yifan <Yifan1.Zhang@amd.com>; Zhu, James <James.Zhu@amd.com>; Ken Moffat <zarniwhoop@ntlworld.com>
Subject: Re: [PATCH 1/2] Revert "drm/amdkfd: fix boot failure when iommu is disabled in Picasso."

On Tue, Nov 2, 2021 at 3:23 PM James Zhu <James.Zhu@amd.com> wrote:
>
> To revert wrong merge during rebase.
> This reverts commit afd18180c07026f94a80ff024acef5f4159084a4.

Missing your signed-off-by.  Might be better to just merge these two
patches into one unless they are individually applicable to different
kernels (e.g., 5.15 vs drm-next).  Also these will probably fix the
following:
Bug: https://nam11.safelinks.protection.outlook.com/?url=https%3A%2F%2Fbugzilla.kernel.org%2Fshow_bug.cgi%3Fid%3D214859&amp;data=04%7C01%7CJames.Zhu%40amd.com%7Cd89a2e71a6204428008908d99e40368f%7C3dd8961fe4884e608e11a82d994e183d%7C0%7C0%7C637714820908943450%7CUnknown%7CTWFpbGZsb3d8eyJWIjoiMC4wLjAwMDAiLCJQIjoiV2luMzIiLCJBTiI6Ik1haWwiLCJXVCI6Mn0%3D%7C1000&amp;sdata=wOmavKInLZrl3Qge2N83kgMh2dATo4aUH%2BW6QrPvL30%3D&amp;reserved=0
Bug: https://nam11.safelinks.protection.outlook.com/?url=https%3A%2F%2Fgitlab.freedesktop.org%2Fdrm%2Famd%2F-%2Fissues%2F1770&amp;data=04%7C01%7CJames.Zhu%40amd.com%7Cd89a2e71a6204428008908d99e40368f%7C3dd8961fe4884e608e11a82d994e183d%7C0%7C0%7C637714820908943450%7CUnknown%7CTWFpbGZsb3d8eyJWIjoiMC4wLjAwMDAiLCJQIjoiV2luMzIiLCJBTiI6Ik1haWwiLCJXVCI6Mn0%3D%7C1000&amp;sdata=k3GJBSFKw8FJhGs9Urpm8PmFPzN7m0LFl5p7Wv1FfGY%3D&amp;reserved=0

Alex

> ---
>  drivers/gpu/drm/amd/amdkfd/kfd_device.c | 3 ---
>  1 file changed, 3 deletions(-)
>
> diff --git a/drivers/gpu/drm/amd/amdkfd/kfd_device.c b/drivers/gpu/drm/amd/amdkfd/kfd_device.c
> index 7677ced16a27..be26c4016ade 100644
> --- a/drivers/gpu/drm/amd/amdkfd/kfd_device.c
> +++ b/drivers/gpu/drm/amd/amdkfd/kfd_device.c
> @@ -1031,9 +1031,6 @@ bool kgd2kfd_device_init(struct kfd_dev *kfd,
>
>         svm_migrate_init(kfd->adev);
>
> -       if(kgd2kfd_resume_iommu(kfd))
> -               goto device_iommu_error;
> -
>         if (kfd_resume(kfd))
>                 goto kfd_resume_error;
>
> --
> 2.25.1
>

[-- Attachment #2: Type: text/html, Size: 5646 bytes --]

Re: [PATCH 1/2] Revert "drm/amdkfd: fix boot failure when iommu is disabled in Picasso."

[Ken Moffat]

On Tue, Nov 02, 2021 at 09:00:04PM +0000, Zhu, James wrote:
> [AMD Official Use Only]
> 
> 
> Hi Alex
> 
> Sure, Let me check it can be merged into one patch.
> 
> 
> Thanks & Best Regards!
> 
> 
> James Zhu
> 

Hi James,

I'm trying to apply these to 5.15.0 (using the dry-run option of
traditional patch, kernel source is from the tarball).

the first one fails:

Reversed (or previously applied) patch detected!  Assume -R? [n]

it will revert:
Hunk #1 succeeded at 924 with fuzz 2 (offset -107 lines).

In anything connected to the kernel, fuzz on a patch worries me.
Should I apply this with patch -p1 -R ?

For the second patch:

checking file drivers/gpu/drm/amd/amdgpu/amdgpu_device.c
Hunk #1 succeeded at 2432 (offset 34 lines).
checking file drivers/gpu/drm/amd/amdkfd/kfd_device.c
Hunk #1 succeeded at 924 with fuzz 2 (offset -107 lines).

Again, note the fuzz in kfd_device.c.

Using 'git apply --check' instead of 'patch' both patches report the
same failure for kfd_device.c:

error: patch failed: drivers/gpu/drm/amd/amdkfd/kfd_device.c:1031
error: drivers/gpu/drm/amd/amdkfd/kfd_device.c: patch does not apply

ĸen
-- 
Vetinari smiled. "Can you keep a secret, Mister Lipwig?"
"Oh, yes, sir. I've kept lots."
"Capital. And the point is, so can I. You do not need to know.”

Re: [PATCH 2/2] drm/amdkfd: fix boot failure when iommu is disabled in Picasso.

[Lazar, Lijo]

On 11/3/2021 12:53 AM, James Zhu wrote:
> From: Yifan Zhang <yifan1.zhang@amd.com>
> 
> When IOMMU disabled in sbios and kfd in iommuv2 path, iommuv2
> init will fail. But this failure should not block amdgpu driver init.
> 
> Reported-by: youling <youling257@gmail.com>
> Tested-by: youling <youling257@gmail.com>
> Signed-off-by: Yifan Zhang <yifan1.zhang@amd.com>
> Reviewed-by: James Zhu <James.Zhu@amd.com>
> ---
>   drivers/gpu/drm/amd/amdgpu/amdgpu_device.c | 4 ----
>   drivers/gpu/drm/amd/amdkfd/kfd_device.c    | 3 +++
>   2 files changed, 3 insertions(+), 4 deletions(-)
> 
> diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_device.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_device.c
> index e56bc925afcf..f77823ce7ae8 100644
> --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_device.c
> +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_device.c
> @@ -2398,10 +2398,6 @@ static int amdgpu_device_ip_init(struct amdgpu_device *adev)
>   	if (!adev->gmc.xgmi.pending_reset)
>   		amdgpu_amdkfd_device_init(adev);
>   
> -	r = amdgpu_amdkfd_resume_iommu(adev);
> -	if (r)
> -		goto init_failed;
> -
>   	amdgpu_fru_get_product_info(adev);
>   
>   init_failed:
> diff --git a/drivers/gpu/drm/amd/amdkfd/kfd_device.c b/drivers/gpu/drm/amd/amdkfd/kfd_device.c
> index be26c4016ade..7677ced16a27 100644
> --- a/drivers/gpu/drm/amd/amdkfd/kfd_device.c
> +++ b/drivers/gpu/drm/amd/amdkfd/kfd_device.c
> @@ -1031,6 +1031,9 @@ bool kgd2kfd_device_init(struct kfd_dev *kfd,
>   
>   	svm_migrate_init(kfd->adev);
>   
> +	if(kgd2kfd_resume_iommu(kfd))
> +		goto device_iommu_error;
> +

This also brings a duplicate iommu resume in the reset path -
https://elixir.bootlin.com/linux/v5.15/source/drivers/gpu/drm/amd/amdgpu/amdgpu_device.c#L4612

Thanks,
Lijo
>   	if (kfd_resume(kfd))
>   		goto kfd_resume_error;
>   
> 

Re: [PATCH 2/2] drm/amdkfd: fix boot failure when iommu is disabled in Picasso.

[James Zhu]

On 2021-11-02 11:53 p.m., Lazar, Lijo wrote:
>
>
> On 11/3/2021 12:53 AM, James Zhu wrote:
>> From: Yifan Zhang <yifan1.zhang@amd.com>
>>
>> When IOMMU disabled in sbios and kfd in iommuv2 path, iommuv2
>> init will fail. But this failure should not block amdgpu driver init.
>>
>> Reported-by: youling <youling257@gmail.com>
>> Tested-by: youling <youling257@gmail.com>
>> Signed-off-by: Yifan Zhang <yifan1.zhang@amd.com>
>> Reviewed-by: James Zhu <James.Zhu@amd.com>
>> ---
>>   drivers/gpu/drm/amd/amdgpu/amdgpu_device.c | 4 ----
>>   drivers/gpu/drm/amd/amdkfd/kfd_device.c    | 3 +++
>>   2 files changed, 3 insertions(+), 4 deletions(-)
>>
>> diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_device.c 
>> b/drivers/gpu/drm/amd/amdgpu/amdgpu_device.c
>> index e56bc925afcf..f77823ce7ae8 100644
>> --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_device.c
>> +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_device.c
>> @@ -2398,10 +2398,6 @@ static int amdgpu_device_ip_init(struct 
>> amdgpu_device *adev)
>>       if (!adev->gmc.xgmi.pending_reset)
>>           amdgpu_amdkfd_device_init(adev);
>>   -    r = amdgpu_amdkfd_resume_iommu(adev);
>> -    if (r)
>> -        goto init_failed;
>> -
>>       amdgpu_fru_get_product_info(adev);
>>     init_failed:
>> diff --git a/drivers/gpu/drm/amd/amdkfd/kfd_device.c 
>> b/drivers/gpu/drm/amd/amdkfd/kfd_device.c
>> index be26c4016ade..7677ced16a27 100644
>> --- a/drivers/gpu/drm/amd/amdkfd/kfd_device.c
>> +++ b/drivers/gpu/drm/amd/amdkfd/kfd_device.c
>> @@ -1031,6 +1031,9 @@ bool kgd2kfd_device_init(struct kfd_dev *kfd,
>>         svm_migrate_init(kfd->adev);
>>   +    if(kgd2kfd_resume_iommu(kfd))
>> +        goto device_iommu_error;
>> +
>
> This also brings a duplicate iommu resume in the reset path -
> https://elixir.bootlin.com/linux/v5.15/source/drivers/gpu/drm/amd/amdgpu/amdgpu_device.c#L4612 
>

[JZ] Hi Lijo, please help clarify this duplicate case.

Thanks!

James

>
> Thanks,
> Lijo
>>       if (kfd_resume(kfd))
>>           goto kfd_resume_error;
>>

Re: [PATCH 2/2] drm/amdkfd: fix boot failure when iommu is disabled in Picasso.

[Lazar, Lijo]

[-- Attachment #1: Type: text/plain, Size: 263 bytes --]

[Public]

It's a conditional case for some kind of early reset. Haven't checked details, on a quick glance it appeared to call iommu init again.

https://elixir.bootlin.com/linux/v5.15/source/drivers/gpu/drm/amd/amdgpu/amdgpu_device.c#L5160

Thanks,
Lijo

[-- Attachment #2: Type: text/html, Size: 971 bytes --]