Hello all,

I just bisected an issue introduced recently for me where I get screen
corruption before even getting a TTY, and it came down to this commit.

I've got a Navi10 card, and after this commit all that is displayed is a
blank screen (with some obvious corruption artifacts), before I even get
to a tty.

If you'd rather a GitLab issue, just let me know in a reply and I can
file one, but I thought it would be easier to ping those involved since
I already bisected the issue.

I've attached a log of a boot/shutdown sequence from this commit on
amd-staging-drm-next to aid with debugging.

Let me know if you need help reproducing, or fixing. I haven't had my
hands in the display/dc code at all yet, so it would take me some time
to ramp up but I'd be more than willing to help if necessary.

Thanks in advance,
Matt

On 10/14/20 7:20 AM, Kazlauskas, Nicholas wrote:
> On 2020-10-14 3:04 a.m., Yifan Zhang wrote:
>> Change-Id: I831a5ade8b87c23d21a63d08cc4d338468769e2b
>> Signed-off-by: Yifan Zhang <yifan1.zhang@amd.com>
>> ---
>>   .../gpu/drm/amd/display/amdgpu_dm/amdgpu_dm.c | 61 +++++++++++++++++++
>>   1 file changed, 61 insertions(+)
>>
>> diff --git a/drivers/gpu/drm/amd/display/amdgpu_dm/amdgpu_dm.c
>> b/drivers/gpu/drm/amd/display/amdgpu_dm/amdgpu_dm.c
>> index 3cf4e08931bb..aaff8800c7a0 100644
>> --- a/drivers/gpu/drm/amd/display/amdgpu_dm/amdgpu_dm.c
>> +++ b/drivers/gpu/drm/amd/display/amdgpu_dm/amdgpu_dm.c
>> @@ -887,12 +887,67 @@ static void
>> amdgpu_check_debugfs_connector_property_change(struct amdgpu_device
>>       }
>>   }
>>   +static void mmhub_read_system_context(struct amdgpu_device *adev,
>> struct dc_phy_addr_space_config *pa_config)
>> +{
>> +    uint64_t pt_base;
>> +    uint32_t logical_addr_low;
>> +    uint32_t logical_addr_high;
>> +    uint32_t agp_base, agp_bot, agp_top;
>> +    PHYSICAL_ADDRESS_LOC page_table_start, page_table_end,
>> page_table_base;
>> +
>> +    logical_addr_low  = min(adev->gmc.fb_start, adev->gmc.agp_start)
>> >> 18;
>> +    pt_base = amdgpu_gmc_pd_addr(adev->gart.bo);
>> +
>> +    if (adev->apu_flags & AMD_APU_IS_RAVEN2)
>> +        /*
>> +         * Raven2 has a HW issue that it is unable to use the vram which
>> +         * is out of MC_VM_SYSTEM_APERTURE_HIGH_ADDR. So here is the
>> +         * workaround that increase system aperture high address (add 1)
>> +         * to get rid of the VM fault and hardware hang.
>> +         */
>> +        logical_addr_high = max((adev->gmc.fb_end >> 18) + 0x1,
>> adev->gmc.agp_end >> 18);
>> +    else
>> +        logical_addr_high = max(adev->gmc.fb_end, adev->gmc.agp_end)
>> >> 18;
>> +
>> +    agp_base = 0;
>> +    agp_bot = adev->gmc.agp_start >> 24;
>> +    agp_top = adev->gmc.agp_end >> 24;
>> +
>> +
>> +    page_table_start.high_part = (u32)(adev->gmc.gart_start >> 44) &
>> 0xF;
>> +    page_table_start.low_part = (u32)(adev->gmc.gart_start >> 12);
>> +    page_table_end.high_part = (u32)(adev->gmc.gart_end >> 44) & 0xF;
>> +    page_table_end.low_part = (u32)(adev->gmc.gart_end >> 12);
>> +    page_table_base.high_part = upper_32_bits(pt_base) & 0xF;
>> +    page_table_base.low_part = lower_32_bits(pt_base);
>> +
>> +    pa_config->system_aperture.start_addr =
>> (uint64_t)logical_addr_low << 18;
>> +    pa_config->system_aperture.end_addr = (uint64_t)logical_addr_high
>> << 18;
>> +
>> +    pa_config->system_aperture.agp_base = (uint64_t)agp_base << 24 ;
>> +    pa_config->system_aperture.agp_bot = (uint64_t)agp_bot << 24;
>> +    pa_config->system_aperture.agp_top = (uint64_t)agp_top << 24;
>> +
>> +    pa_config->system_aperture.fb_base = adev->gmc.fb_start;
>> +    pa_config->system_aperture.fb_offset = adev->gmc.aper_base;
>> +    pa_config->system_aperture.fb_top = adev->gmc.fb_end;
>> +
>> +    pa_config->gart_config.page_table_start_addr =
>> page_table_start.quad_part << 12;
>> +    pa_config->gart_config.page_table_end_addr =
>> page_table_end.quad_part << 12;
>> +    pa_config->gart_config.page_table_base_addr =
>> page_table_base.quad_part;
>> +
>> +    pa_config->is_hvm_enabled = 0;
>> +
>> +}
>> +
>> +
>>   static int amdgpu_dm_init(struct amdgpu_device *adev)
>>   {
>>       struct dc_init_data init_data;
>>   #ifdef CONFIG_DRM_AMD_DC_HDCP
>>       struct dc_callback_init init_params;
>>   #endif
>> +    struct dc_phy_addr_space_config pa_config;
>>       int r;
>>         adev->dm.ddev = adev_to_drm(adev);
>> @@ -1040,6 +1095,12 @@ static int amdgpu_dm_init(struct amdgpu_device
>> *adev)
>>           goto error;
>>       }
>>   +    mmhub_read_system_context(adev, &pa_config);
>> +
>> +    // Call the DC init_memory func
>> +    dc_setup_system_context(adev->dm.dc, &pa_config);
>> +
>> +
> 
> The dc_setup_system_context should come directly after dc_hardware_init().
> 
> With that fixed this series is
> 
> Reviewed-by: Nicholas Kazlauskas <nicholas.kazlauskas@amd.com>
> 
> There's the vmid module as well that could be created after if needed
> but for s/g suport alone that's not necessary.
> 
> Regards,
> Nicholas Kazlauskas
> 
>>       DRM_DEBUG_DRIVER("KMS initialized.\n");
>>         return 0;
>>
> 
> _______________________________________________
> amd-gfx mailing list
> amd-gfx@lists.freedesktop.org
> https://lists.freedesktop.org/mailman/listinfo/amd-gfx