Investigating the issue Dirk reported [1], and after the RFC sent[2],
here is now the series with my conclusions. Patch 1 fixes a real issue
IMHO and is the minimal solution for backporting. Patch 2 adds some
documentation. Patch 3 reduces potential race windows by letting the
timer only run when we really want it. Patch 4 is a simplification
because locking should be as simple as possible, right?
I could trigger the code path meanwhile and did not encounter a
regression. But I could not reproduce the initial report from Dirk, so I
can't say this is a 100% fix. Yet, I think, we want these patches
nonetheless. Dirk metioned that his system could have been shutting
down. If so, patch 1 is probably the solution, but we don't know for
sure.
[1] https://lore.kernel.org/r/[email protected]
[2] https://lore.kernel.org/r/[email protected]
Wolfram Sang (4):
serial: sh-sci: protect invalidating RXDMA on shutdown
serial: sh-sci: describe locking requirements for invalidating RXDMA
serial: sh-sci: let timeout timer only run when DMA is scheduled
serial: sh-sci: simplify locking when re-issuing RXDMA fails
drivers/tty/serial/sh-sci.c | 17 ++++++++++++-----
1 file changed, 12 insertions(+), 5 deletions(-)
--
2.43.0