On Sun, Nov 17, 2024 at 04:16:17PM +1000, David Gwynne wrote:
> On Sat, Nov 16, 2024 at 07:36:37PM -0800, Andrew Hewus Fresh wrote:
> > I finally got around to fixing my alpha, which involved replacing the
> > disk. That meant I have to scp some stuff over to to it and after a bit
> > of time it panics:
> >
> > panic: mtx 0xfffffe000002a628: locking against myself
> > Stopped at db_enter+0x8: lda sp,10(sp)
> > TID PID UID PRFLAGS PFLAGS CPU COMMAND
> > *204699 79774 0 0x14000 0x200 0 softnet0
> > db_enter(0, 7ffffe00e0003f8, 1, 8, 3, 8) at db_enter+0x8
> > panic(?, fffffe000002a628, 1b0, 10, a, 1) at panic+0xe8
> > mtx_enter(?, ?, 1b0, 10, a, 1) at mtx_enter+0xb4
> > ifq_set_oactive(?, ?, 1b0, 10, a, 1) at ifq_set_oactive+0x50
>
> this is from src/sys/net/ifq.c r1.50 where i added a counter for the
> number of times oactive gets set. because there's checks and multiple
> things being tweaked i used the ifq mutex to serialise the updates.
>
> de(4) uses ifq_deq_begin to try and shove an mbuf onto the hardware,
> which takes but doesnt release the ifq mutex until ifq_deq_commit or
> ifq_deq_rollback is called. so while it's holding the mutex is calls
> ifq_set_oactive, which also tries to take the mutex.
>
> i honestly don't understand what de(4) is doing with the hardware and
> packet setup, so i dont feel confident changing the driver to avoid
> this. the least worst alternative i could think of is to provide an
> alternative set_oactive it can call.
>
> the diff below should fix this.
I mean, it "fixed" the locking against myself error :-)
de0 at pci0 dev 11 function 0 "DEC 21040" rev 0x23, DEC 21040 pass 2.3: isa irq
5, address ...
panic: mutex 0xfffffe000002a628 not held in ifq_deq_set_oactive
Stopped at db_enter+0x8: lda sp,10(sp)
TID PID UID PRFLAGS PFLAGS CPU COMMAND
* 0 0 0 0x10000 0x200 0 swapper
db_enter(0, 7ffffe00e0003f8, 1, 8, 3, fffffc0000000008) at db_enter+0x8
panic(?, fffffe000002a628, fffffc0000bc9718, 1, 5, 1) at panic+0xe8
ifq_deq_set_oactive(?, ?, fffffc0000bc9718, 1, 5, 1) at ifq_deq_set_oactive+0x8
8
tulip_txput(?, ?, fffffe000002a000, ?, 5, 1) at tulip_txput+0x5e8
