[PATCH] drm/amd: fix debugfs access for discovery blob

[Mario Limonciello]

On 8/7/2023 10:20, Alex Deucher wrote:
> On Tue, Aug 1, 2023 at 4:15 PM Mario Limonciello
> <mario.limonciello at amd.com> wrote:
>>
>> Accessing the blob for amdgpu discovery from debugfs triggers:
>>
>> [ 1924.487667] kernel BUG at mm/usercopy.c:102!
>>
>> usercopy_abort() explains that it needs to be solved by creating
>> a cache to store the data.
>>
>> Link: https://gitlab.freedesktop.org/drm/amd/-/issues/2748#note_2023519
>> Signed-off-by: Mario Limonciello <mario.limonciello at amd.com>
> 
> Reviewed-by: Alex Deucher <alexander.deucher at amd.com>
> 
Although this avoids the issue; I've concluded it's an inappropriate fix 
and will abandon it.  It turns out to spit out 64k discovery blobs that 
should have been 8k.

6.5-rc and ASDN already picked up a better solution.

db3b5cb64a9c ("drm/amdgpu: Use apt name for FW reserved region")

>> ---
>>   drivers/gpu/drm/amd/amdgpu/amdgpu.h         |  7 +++++--
>>   drivers/gpu/drm/amd/amdgpu/amdgpu_debugfs.c | 17 ++++++++++++++++-
>>   drivers/gpu/drm/amd/amdgpu/amdgpu_drv.c     |  2 ++
>>   3 files changed, 23 insertions(+), 3 deletions(-)
>>
>> diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu.h b/drivers/gpu/drm/amd/amdgpu/amdgpu.h
>> index a3b86b86dc477..66a2251bdeba4 100644
>> --- a/drivers/gpu/drm/amd/amdgpu/amdgpu.h
>> +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu.h
>> @@ -791,8 +791,11 @@ struct amdgpu_device {
>>          bool                            accel_working;
>>          struct notifier_block           acpi_nb;
>>          struct amdgpu_i2c_chan          *i2c_bus[AMDGPU_MAX_I2C_BUS];
>> -       struct debugfs_blob_wrapper     debugfs_vbios_blob;
>> -       struct debugfs_blob_wrapper     debugfs_discovery_blob;
>> +#if defined(CONFIG_DEBUG_FS)
>> +       struct debugfs_blob_wrapper     debugfs_vbios_blob;
>> +       struct debugfs_blob_wrapper     debugfs_discovery_blob;
>> +       struct kmem_cache               *discovery_blob_cache;
>> +#endif
>>          struct mutex                    srbm_mutex;
>>          /* GRBM index mutex. Protects concurrent access to GRBM index */
>>          struct mutex                    grbm_idx_mutex;
>> diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_debugfs.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_debugfs.c
>> index 56e89e76ff179..55ea5be14b188 100644
>> --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_debugfs.c
>> +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_debugfs.c
>> @@ -2180,7 +2180,15 @@ int amdgpu_debugfs_init(struct amdgpu_device *adev)
>>          debugfs_create_blob("amdgpu_vbios", 0444, root,
>>                              &adev->debugfs_vbios_blob);
>>
>> -       adev->debugfs_discovery_blob.data = adev->mman.discovery_bin;
>> +
>> +       adev->discovery_blob_cache = kmem_cache_create_usercopy("amdgpu_discovery",
>> +                                                               adev->mman.discovery_tmr_size,
>> +                                                               0, 0, 0,
>> +                                                               adev->mman.discovery_tmr_size,
>> +                                                               NULL);
>> +       adev->debugfs_discovery_blob.data = kmem_cache_alloc(adev->discovery_blob_cache, GFP_KERNEL);
>> +       memcpy(adev->debugfs_discovery_blob.data, adev->mman.discovery_bin,
>> +              adev->mman.discovery_tmr_size);
>>          adev->debugfs_discovery_blob.size = adev->mman.discovery_tmr_size;
>>          debugfs_create_blob("amdgpu_discovery", 0444, root,
>>                              &adev->debugfs_discovery_blob);
>> @@ -2188,6 +2196,12 @@ int amdgpu_debugfs_init(struct amdgpu_device *adev)
>>          return 0;
>>   }
>>
>> +void amdgpu_debugfs_fini(struct amdgpu_device *adev)
>> +{
>> +       kmem_cache_free(adev->discovery_blob_cache, adev->debugfs_discovery_blob.data);
>> +       kmem_cache_destroy(adev->discovery_blob_cache);
>> +}
>> +
>>   #else
>>   int amdgpu_debugfs_init(struct amdgpu_device *adev)
>>   {
>> @@ -2197,4 +2211,5 @@ int amdgpu_debugfs_regs_init(struct amdgpu_device *adev)
>>   {
>>          return 0;
>>   }
>> +inline void amdgpu_debugfs_fini(struct amdgpu_device *adev) {}
>>   #endif
>> diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_drv.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_drv.c
>> index 0593ef8fe0a63..1a3b30dff5171 100644
>> --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_drv.c
>> +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_drv.c
>> @@ -2276,6 +2276,8 @@ amdgpu_pci_remove(struct pci_dev *pdev)
>>          struct drm_device *dev = pci_get_drvdata(pdev);
>>          struct amdgpu_device *adev = drm_to_adev(dev);
>>
>> +       amdgpu_debugfs_fini(adev);
>> +
>>          amdgpu_xcp_dev_unplug(adev);
>>          drm_dev_unplug(dev);
>>
>> --
>> 2.34.1
>>