Re: [PATCH v4] Fixes to DMA state check

On 2021/10/12 14:57, Reimar Döffinger wrote:

On 12 October 2021 04:20:47 CEST, Damien Le Moal [off-list ref] wrote:

quoted

On 10/3/21 22:28, Reimar Döffinger wrote:

quoted

Patch series to add ata_dma_enabled calls instead of incorrectly
checking dev->dma_mode != 0.
Only the first patch is confirmed to have caused real issues
that it indeed fixes, rest based purely on code review.

Changes v4:
- split per file/driver
- added Signed-off-by and Tested-by lines, improved commit messages
Changes v3:
- found and updated more cases in pata_ali, pata_amd and pata_radisys.
Changes v2:
- removed initialization change for SATA. I got confused by the
  ping-pong between libata-eh and libata-core and thought SATA did not
  set up xfermode
- reviewed other cases that used dma_mode in boolean context and those
  seemed to need changing as well, so added them to patch.
  I did not see any places that would set dma_mode to 0 ever, so I
  do think they were all indeed wrong.

This looks all good to me but I do not see any CC Stable tag. Do you
want this backported to stable versions ?

Sorry, I admit I am quite ignorant of these workflow details.
Personally I have no need for a backport as it's only about one piece of legacy HW for me.
I also feel like the untested patches do not belong in a stable backport.
However the first, tested patch might be good to have backported if considering the whole user base and not just my use? Would that be sensible to handle that way? 

Note that I will apply the entire series to for-5.16 so that the patches have
time to go through linux-next for more testing. The first patch will be
backported once we start the 5.16 cycle.

Best regards,
Reimar

-- 
Damien Le Moal
Western Digital Research