drm/msm: Depopulate platform on probe failure
authorSean Paul <seanpaul@chromium.org>
Mon, 17 Jun 2019 20:12:51 +0000 (16:12 -0400)
committerSean Paul <seanpaul@chromium.org>
Thu, 20 Jun 2019 18:42:05 +0000 (14:42 -0400)
add_display_components() calls of_platform_populate, and we depopluate
on pdev remove, but not when probe fails. So if we get a probe deferral
in one of the components, we won't depopulate the platform. This causes
the core to keep references to devices which should be destroyed, which
causes issues when those same devices try to re-initialize on the next
probe attempt.

I think this is the reason we had issues with the gmu's device-managed
resources on deferral (worked around in commit 94e3a17f33a5).

Reviewed-by: Rob Clark <robdclark@chromium.org>
Signed-off-by: Sean Paul <seanpaul@chromium.org>
Link: https://patchwork.freedesktop.org/patch/msgid/20190617201301.133275-3-sean@poorly.run
drivers/gpu/drm/msm/msm_drv.c

index 4c51063531f10d395b3dc52850f706df71be3e5f..72139ddcede6087ad9ef9d590ee18007b302b8a9 100644 (file)
@@ -1319,16 +1319,24 @@ static int msm_pdev_probe(struct platform_device *pdev)
 
        ret = add_gpu_components(&pdev->dev, &match);
        if (ret)
-               return ret;
+               goto fail;
 
        /* on all devices that I am aware of, iommu's which can map
         * any address the cpu can see are used:
         */
        ret = dma_set_mask_and_coherent(&pdev->dev, ~0);
        if (ret)
-               return ret;
+               goto fail;
+
+       ret = component_master_add_with_match(&pdev->dev, &msm_drm_ops, match);
+       if (ret)
+               goto fail;
 
-       return component_master_add_with_match(&pdev->dev, &msm_drm_ops, match);
+       return 0;
+
+fail:
+       of_platform_depopulate(&pdev->dev);
+       return ret;
 }
 
 static int msm_pdev_remove(struct platform_device *pdev)