Hi Dirk, thanks a ton for this report and RFC! On Tue, Feb 20, 2024 at 07:13:56AM +0100, Dirk Behme wrote: > Analyzing the KASAN report [1] tells us that in > > mmc_request_done+0xcc/0x30c > > what can be resolved to an access to > > mrq->cap_cmd_during_tfr > > in mmc_command_done() called inline from mmc_request_done() "mrq" > becomes invalid. > > In the driver we have two work queues, tmio_mmc_reset_work() and > tmio_mmc_done_work(). Both operate on mrq. > > Synchronize this by extending the spin lock protected sections. As discussed further privately, we both see the problem but want to keep the critical sections minimal. So, after some more investigations, a counter-patch is coming in a minute. Wolfram
Attachment:
signature.asc
Description: PGP signature
