Leo Weppelman wrote:
> I think you've just found a race-condition in the scsi code... I think
> what happens is this:
>     scsipi_execute_xs()
>        scsipi_command_direct() = SUCCESSFULLY_QUEUED
>                ..... Some interrupts happen along the way causing the
>                scsi-command to have finished in the mean time. NOSLEEP
>                is set, so no wakeup/return is done, the xs-structure
>                is freed instead (argh!).
>     -> now the resuming sections of scsipi_execute_xs() use an already
>        freed xs....

This is very possible.  Looking at my debug output, it's likely that there
are 2 calls going on simultaneously.  Now, I think that one has SCSI_NOSLEEP
set but the other doesn't.

> One way that might work to get around this is the addition of a FREEWHENDONE
> flag. So the code in scsipi_execute_xs() looks like:
>         ....
>        switch (scsipi_command_direct(xs)) {
>        case SUCCESSFULLY_QUEUED:
>                 if ((xs->flags & (SCSI_NOSLEEP | SCSI_POLL)) == SCSI_NOSLEEP) {
>                      xs->flags |= SCSI_FREEWHENDONE;
>                      return (EJUSTRETURN);
>                 }
>        .....
> 
> The tail of scsipi_done() will look like:
> 
>        .....
>        if (xs->flags & SCSI_FREEWHENDONE)
>               scsipi_free_xs(xs, SCSI_NOSLEEP);
>        if (bp)
>        .....

Erm, I'm not sure about this.  I think [*] that scsipi_done() is called 
from inside scsipi_command_direct(xs)), so the flag is getting clobbered
in scsipi_free_xs().  Thus, setting it after the 'case SUCCESSFULLY_QUEUED:'
won't have any effect.  I'll check this evening though.

Thanks,

J

[*] I haven't got all my notes with me at work.

PS.  For (most) people who have missed the start of the thread, I'm trying
to get a Cabletron EA412 SCSI<->ethernet box (if_se.c) to work with an
Atari TT.

-- 
    1024/55A5BC19        0F 3F 62 56 18 10 8B 84  43 8F F4 94 93 37 76 AA

S.E.P.