Re: [PATCH net 3/3] net/mlx5: DR, Fix out-of-bounds write when sync to HW fails
From: Yevgeny Kliteynik
Date: Sun Nov 15 2026 - 19:36:46 EST
On 05-Nov-26 17:02, Yevgeny Kliteynik wrote:
On 27-Oct-26 10:56, Paolo Abeni wrote:
On 10/23/26 12:52, netdev-bot+sashiko@xxxxxxxxxx wrote:
- [Medium] After a single SYNC_STEERING failure the pool latches into a
permanent error state while allocation stays open, so device SW ICM…
This one looks legit.
/P
Agree.
Actually, there should be two aspects to this fix.
First part is as you commented - stop serving new chunks once the pool
is in error state. Second part is allow for some transient sync failures.
I'll issue v2 that handles both: a failed sync no longer disables the pool
right away. hot_chunks_arr has spare room, the hot chunks stay queued, and the
following frees keep retrying the sync. Only if the sync keeps failing until
the spare room is used up does the pool go into an error state.
After some more digging, there is no point in complicating this fix.
I don't expect to see "transient" sync failures, so increasing memory
footprint just to prepare for sync failures is not a good tradeoff.
The v2 will contain just the suggested fix - once the poll is in error
state, the chunks will not be allocated/freed.
-- YK