Re: [PATCH v6 00/12] Some Panfrost fixes and improvements

[AngeloGioacchino Del Regno <angelogioacchino.delregno@collabora.com>]

Il 15/10/25 02:09, Adrián Larumbe ha scritto:
> This is v6 of https://lore.kernel.org/dri-devel/20251007150216.254250-1-adrian.larumbe@collabora.com/
> 
> This patch series is a collection of minor fixes and improvements I came up
> with while working on driver related stuff.
> 

Whole series is

Reviewed-by: AngeloGioacchino Del Regno <angelogioacchino.delregno@collabora.com>

> Changelog:
>   v6:
>    - Don't optimise page range walk in the event of a double MMU fault
>    - Handle some minots and minor style changes
>    - Rebase patch series onto latest drm-misc-next
>   v5:
>   - Move devfreq record keeping further down job submission function to
>   keep busy count balanced in case of an early bail-out.
>   - In MMU page fault ISR, bail out when sgt for 2MiB page is not assigned,
>   rather than when the page array is populated. Add new commit for this.
>   - Add commit with myself as a new Panfrost maintainer
> 
>   v4:
>    - Rebased older patch series onto latest drm-misc-next
>    - Added patch for renaming JM functions to reflect their actual role
>    - Fixed treatment of error code in perfcnt when enabling sample buffer AS
> 
>   v3:
>    - Minor convenience fixes to patches 3 and 4 in the series
>    - Move unmapping of maped range of BO to the function's error path
>    in case of BO mapping failure, also for putting BO's pages
>    - Split patch 6/8 into two: one makes sure the Job IRQ enablement mask
>    isn't recalculated at every device reset and uses the same expression
>    everywhere in the driver, and another one that breaks the enablement
>    function into two stages.
> 
>   v2:
>    - Removed commit that provided an explicit fence cleanup function
>    - Added commit for removing unused Panfrost device structure member
>    - Refactored how optional job interrupt reenabling during reset is handled
>    - Make the way errors and successful return values are delivered from inside
>     panfrost_mmu_as_get more according to standard.
>    - Simplify unmapping of already mapped area when mapping the pages of a BO
>    - Fixing management of runtime-PM reference counts when failing HW job submission.
> 
> Adrián Larumbe (12):
>    drm/panfrost: Replace DRM driver allocation method with newer one
>    drm/panfrost: Handle inexistent GPU during probe
>    drm/panfrost: Handle job HW submit errors
>    drm/panfrost: Handle error when allocating AS number
>    drm/panfrost: Check sgt to know whether pages are already mapped
>    drm/panfrost: Handle page mapping failure
>    drm/panfrost: Don't rework job IRQ enable mask in the enable path
>    drm/panfrost: Make re-enabling job interrupts at device reset optional
>    drm/panfrost: Add forward declaration and types header
>    drm/panfrost: Remove unused device property
>    drm/panfrost: Rename panfrost_job functions to reflect real role
>    MAINTAINERS: Panfrost: Add Steven Price and Adrian Larumbe
> 
>   MAINTAINERS                                   |   3 +-
>   drivers/gpu/drm/panfrost/panfrost_devfreq.c   |   4 +-
>   drivers/gpu/drm/panfrost/panfrost_device.c    |  68 +++++----
>   drivers/gpu/drm/panfrost/panfrost_device.h    |  13 +-
>   drivers/gpu/drm/panfrost/panfrost_drv.c       |  78 ++++------
>   drivers/gpu/drm/panfrost/panfrost_dump.c      |   8 +-
>   drivers/gpu/drm/panfrost/panfrost_gem.c       |   8 +-
>   .../gpu/drm/panfrost/panfrost_gem_shrinker.c  |   4 +-
>   drivers/gpu/drm/panfrost/panfrost_gpu.c       |  64 ++++----
>   drivers/gpu/drm/panfrost/panfrost_job.c       | 139 +++++++++---------
>   drivers/gpu/drm/panfrost/panfrost_job.h       |  15 +-
>   drivers/gpu/drm/panfrost/panfrost_mmu.c       | 112 +++++++++-----
>   drivers/gpu/drm/panfrost/panfrost_mmu.h       |   3 +-
>   drivers/gpu/drm/panfrost/panfrost_perfcnt.c   |  26 ++--
>   14 files changed, 301 insertions(+), 244 deletions(-)
> 
> 
> base-commit: 7fb19ea1ec6aa85c75905b1fd732d50801e7fb28
> --
> 2.51.0