Re: [PATCH 0/2] dmaengine: idxd: fix crashes in the probe error path

[Jingyi Li <jinli@purestorage.com>]

Hi Frank, Vinod,

Thanks Frank for the review and for pointing me to Vinicius's series
"dmaengine: idxd: More error handling fixes" [1]. Patches 1-3 of that
series cover both patches here:

- 1/2 (double free in idxd_free()) is covered by their 2/6, which
  makes the same change to idxd_free().

- 2/2 (destroy of a NULL or already destroyed workqueue) is covered
  by their 1/6 and 3/6 together. 1/6 adds the same NULL check in
  idxd_conf_device_release(), for a probe failure before idxd->wq is
  allocated. 3/6 removes the early destroy_workqueue() calls, so the
  release callback is the only place the workqueue is destroyed.

So please drop this series in favor of theirs. The one caveat is that
1/6 alone only covers the "never allocated" case; without 3/6, a
failure after idxd_setup_internals() still destroys the workqueue
twice. If 1/6 and 3/6 don't both go in (including to stable), 2/2
here would still be needed.

I've replied on their series with test results.

[1] https://lore.kernel.org/dmaengine/20261012-wt-idxd-fixes-v1-0-30a35a445077@intel.com/

Thanks,
Jingyi