Re: [PATCH 2/2] drm/amdgpu: add fw release for sdma v5_0

["Wang, Kevin(Yang)" <Kevin1.Wang@xxxxxxx>]

[AMD Official Use Only - Internal Distribution Only]

before this patch, the driver has same code logic at "drivers/gpu/drm/amd/amdgpu/sdma_v3_0.c"

related patch: drm/amdgpu: fix sdma3 ucode mem leak

commit 14d83e78c578a6c45163fb399ee760fe0d314bad

Author: Monk Liu <Monk.Liu@xxxxxxx>

Date:   Mon May 30 15:15:32 2016 +0800

    drm/amdgpu: fix sdma3 ucode mem leak

    Signed-off-by: Monk Liu <Monk.Liu@xxxxxxx>

    Reviewed-by: Christian König <christian.koenig@xxxxxxx>

    Signed-off-by: Alex Deucher <alexander.deucher@xxxxxxx>

diff --git a/drivers/gpu/drm/amd/amdgpu/sdma_v3_0.c b/drivers/gpu/drm/amd/amdgpu/sdma_v3_0.c

index 33605d4ac2d9..532ea88da66a 100644

--- a/drivers/gpu/drm/amd/amdgpu/sdma_v3_0.c

+++ b/drivers/gpu/drm/amd/amdgpu/sdma_v3_0.c

@@ -236,6 +236,15 @@ static void sdma_v3_0_init_golden_registers(struct amdgpu_device *adev)

        }

 }

+static void sdma_v3_0_free_microcode(struct amdgpu_device *adev)

+{

+       int i;

+       for (i = 0; i < adev->sdma.num_instances; i++) {

+               release_firmware(adev->sdma.instance[i].fw);

+               adev->sdma.instance[i].fw = NULL;

+       }

+}

+

 /**

  * sdma_v3_0_init_microcode - load ucode images from disk

  *

@@ -1256,6 +1265,7 @@ static int sdma_v3_0_sw_fini(void *handle)

        for (i = 0; i < adev->sdma.num_instances; i++)

                amdgpu_ring_fini(&adev->sdma.instance[i].ring);

+       sdma_v3_0_free_microcode(adev);

        return 0;

 }

Best Regards,

Kevin

---

From: Sheng, Wenhui <Wenhui.Sheng@xxxxxxx>
Sent: Thursday, June 18, 2020 5:21 PM
To: Wang, Kevin(Yang) <Kevin1.Wang@xxxxxxx>; amd-gfx@xxxxxxxxxxxxxxxxxxxxx <amd-gfx@xxxxxxxxxxxxxxxxxxxxx>
Subject: RE: [PATCH 2/2] drm/amdgpu: add fw release for sdma v5_0

 

[AMD Official Use Only - Internal Distribution Only]

[AMD Official Use Only - Internal Distribution Only]

 

Hmmmm….

 

1. There are some places in amdgpu driver using release_firmware like this way
2. This case is a little different, we use it in loop and maybe not all sdma instances’ fw is initialized, just like we did in v5_2

 

Brs

Wenhui

 

From: Wang, Kevin(Yang) <Kevin1.Wang@xxxxxxx>
Sent: Thursday, June 18, 2020 4:48 PM
To: Sheng, Wenhui <Wenhui.Sheng@xxxxxxx>; amd-gfx@xxxxxxxxxxxxxxxxxxxxx
Subject: Re: [PATCH 2/2] drm/amdgpu: add fw release for sdma v5_0

 

[AMD Official Use Only - Internal Distribution Only]

 

afte the API firmware_realease created, the api logic never changed. (the first commit by Linus)

and you can use below command to reference the api usage case in our drm driver.

 

$ grep -nR "release_firmware" -A 5 -B 5 drivers/gpu/drm/ | vim -

 

Best Regards,

Kevin

---

From: Sheng, Wenhui <Wenhui.Sheng@xxxxxxx>
Sent: Thursday, June 18, 2020 4:31 PM
To: Wang, Kevin(Yang) <Kevin1.Wang@xxxxxxx>; amd-gfx@xxxxxxxxxxxxxxxxxxxxx <amd-gfx@xxxxxxxxxxxxxxxxxxxxx>
Subject: RE: [PATCH 2/2] drm/amdgpu: add fw release for sdma v5_0

 

[AMD Official Use Only - Internal Distribution Only]

 

[AMD Official Use Only - Internal Distribution Only]

 

Although we know that release_firmware has null check, but the code is not maintained by ourselves, so I think it’s much more safe to add null  check before invoke release_firmware.

 

 

Brs

Wenhui

From: Wang, Kevin(Yang) <Kevin1.Wang@xxxxxxx>
Sent: Thursday, June 18, 2020 4:25 PM
To: Sheng, Wenhui <Wenhui.Sheng@xxxxxxx>; amd-gfx@xxxxxxxxxxxxxxxxxxxxx
Subject: Re: [PATCH 2/2] drm/amdgpu: add fw release for sdma v5_0

 

[AMD Official Use Only - Internal Distribution Only]

 

 

 

---

From: amd-gfx <amd-gfx-bounces@xxxxxxxxxxxxxxxxxxxxx> on behalf of Wenhui Sheng <Wenhui.Sheng@xxxxxxx>
Sent: Thursday, June 18, 2020 3:53 PM
To: amd-gfx@xxxxxxxxxxxxxxxxxxxxx <amd-gfx@xxxxxxxxxxxxxxxxxxxxx>
Cc: Sheng, Wenhui <Wenhui.Sheng@xxxxxxx>
Subject: [PATCH 2/2] drm/amdgpu: add fw release for sdma v5_0

 

sdma v5_0 fw isn't released when module exit

Signed-off-by: Wenhui Sheng <Wenhui.Sheng@xxxxxxx>
---
 drivers/gpu/drm/amd/amdgpu/sdma_v5_0.c | 6 +++++-
 1 file changed, 5 insertions(+), 1 deletion(-)

diff --git a/drivers/gpu/drm/amd/amdgpu/sdma_v5_0.c b/drivers/gpu/drm/amd/amdgpu/sdma_v5_0.c
index 58d2a80af450..6751ad69ed90 100644
--- a/drivers/gpu/drm/amd/amdgpu/sdma_v5_0.c
+++ b/drivers/gpu/drm/amd/amdgpu/sdma_v5_0.c
@@ -1299,8 +1299,12 @@ static int sdma_v5_0_sw_fini(void *handle)
         struct amdgpu_device *adev = (struct amdgpu_device *)handle;
         int i;
 
-       for (i = 0; i < adev->sdma.num_instances; i++)
+       for (i = 0; i < adev->sdma.num_instances; i++) {
+               if (adev->sdma.instance[i].fw != NULL)
+                       release_firmware(adev->sdma.instance[i].fw);

[kevin]:

the kernel api "release_firmware()" will check argument (is NULL pointer).

i think the patch don't need to double check it.

+
                 amdgpu_ring_fini(&adev->sdma.instance[i].ring);
+       }
 
         return 0;
 }
--
2.17.1

_______________________________________________
amd-gfx mailing list
amd-gfx@xxxxxxxxxxxxxxxxxxxxx
https://nam11.safelinks.protection.outlook.com/?url="">

_______________________________________________
amd-gfx mailing list
amd-gfx@xxxxxxxxxxxxxxxxxxxxx
https://lists.freedesktop.org/mailman/listinfo/amd-gfx