... ahc_lock(); if ((scb =3D ahc_get_scb(ahc)) =3D=3D NULL) { =09 xpt_freeze_simq ... 185,26 @@ ahc_flush_device_writes(struct ahc_softc *ahc) =20 /**************************** Locking Primitives *************************= *****/ /* Lock protecting ...

dragonflybsd-kernel - Peter Avalos - Dec 31 2007 - 20:19

... ahc_lock(); if ((scb =3D ahc_get_scb(ahc)) =3D=3D NULL) { =09 xpt_freeze_simq ... 185,26 @@ ahc_flush_device_writes(struct ahc_softc *ahc) =20 /**************************** Locking Primitives *************************= *****/ /* Lock protecting ...

dragonflybsd-kernel - Peter Avalos - Jan 1 2008 - 22:15

... ahc_lock(); if ((scb =3D ahc_get_scb(ahc)) =3D=3D NULL) { =09 xpt_freeze_simq ... 185,26 @@ ahc_flush_device_writes(struct ahc_softc *ahc) =20 /**************************** Locking Primitives *************************= *****/ /* Lock protecting ...

dragonflybsd-kernel - Peter Avalos - Jan 13 2008 - 20:25

... device drivers with interrupt handlers, such as ahc/ahd, it may be beneficial to use ... the code might be needed since the ahc/ahd interrupt code probably tries to release ... Find the per-controller soft structure for ahc/ahd and throw the lock in ...

dragonflybsd-kernel - Matthew Dillon - Dec 31 2007 - 12:32

... and causing problems with backups. The tape drives are attached via >ahc(4) cards. It also has two spool disks, attached via mpt ... an alternate description of your problem. Seems to be linked to ahc. Any takers? (I am currently busy with NTP release and ...

netbsd-tech-kern - Frank Kardel - May 27 2006 - 03:24

... , Nov 06, 2007 at 12:39:20PM -0800, Steve Kargl wrote: > It seems that some recent change has caused problems > with ahc devices. From dmesg I find > > ahc0: port 0xb800-0xb8ff > mem 0xdeafe000-0xdeafefff irq 16 at ...

freebsd-current - Steve Kargl - Nov 6 2007 - 17:06

On 2007-12-31 03:43, Peter Avalos wrote: > I've attempted to bring in some locking for the ahc and ahd drivers from > FreeBSD. This is my first attempt with locking mechanisms, and I'd > appreciate some input on the attached patch. > > Doing this ...

dragonflybsd-kernel - Erik Wikström - Dec 31 2007 - 06:44

... how to install a serializing > lock w/ the interrupt. Note that some reworking of the code might be > needed since the ahc/ahd interrupt code probably tries to release/rea= quire > and/or do other nasty things when, in fact, you just want ...

dragonflybsd-kernel - Peter Avalos - Dec 31 2007 - 18:21

... serializer to bus_setup_intr()... I suggest having just one serializer per interrupt source (I'm not sure how that works with ahc/ahd)... but its the easiest way to do it. If you do that then the kernel interrupt code will automatically ...

dragonflybsd-kernel - Matthew Dillon - Dec 31 2007 - 21:09

... to install a serializing > > > lock w/ the interrupt. Note that some reworking of the code might be > > > needed since the ahc/ahd interrupt code probably tries to release/reaquire > > > and/or do other nasty things when, in fact, you just ...

dragonflybsd-kernel - Sepherosa Ziehau - Dec 31 2007 - 23:20

... time but it's possible that something got corrupted on-disk during your testing. Softupdates is ultra sensitive to disk corruption. If cam/ahc/ahd were losing track of an IO it would most like panic with 'biodone: bp %p already done!' or ...

dragonflybsd-kernel - Matthew Dillon - Jan 13 2008 - 21:40

... and causing problems with backups. The tape drives are attached via > ahc(4) cards. It also has two spool disks, attached via mpt ... see them in any consistent > fashion either. I believe they're usually on commands where a > tape is being mounted for ...

netbsd-tech-kern - Manuel Bouyer - May 27 2006 - 03:30

... is ultra sensitive to disk corrupti= on. >=20 > If cam/ahc/ahd were losing track of an IO it would most like ... #11 0xc01c0c11 in kern_pwritev (fd=3D3, auio=3D0xe8bffca0, flags=3D0, res= =3D0xe8bffcf0) at file2.h:72 #12 0xc01c1064 in sys_write (uap= ...

dragonflybsd-kernel - Peter Avalos - Jan 13 2008 - 22:20

On Tue, Jan 01, 2008 at 11:20:47AM +0800, Sepherosa Ziehau wrote: > > > > 2. How should the tsleep()s be handled? >=20 > Assum serializer 'ser' is held. >=20 > ... > /* Your code */ > ... > crit_enter(); > tsleep_interlock(your_cond); >

dragonflybsd-kernel - Peter Avalos - Jan 1 2008 - 22:15

:Here's my latest patch. The difference from the last is that I interlocked :the tsleeps and asserted that the serializer is held in the interrupt :handler. It looks pretty good but I have no way to test it. Insofar as the serializer

dragonflybsd-kernel - Matthew Dillon - Jan 3 2008 - 15:06

On Thu, Jan 03, 2008 at 11:06:20AM -0800, Matthew Dillon wrote: >=20 > :Here's my latest patch. The difference from the last is that I interloc= ked > :the tsleeps and asserted that the serializer is held in the interrupt > :handler. >=20 > It

dragonflybsd-kernel - Peter Avalos - Jan 8 2008 - 01:31

On Tue, Jan 08, 2008 at 12:31:19AM -0500, Peter Avalos wrote: > On Thu, Jan 03, 2008 at 11:06:20AM -0800, Matthew Dillon wrote: > >=20 > > :Here's my latest patch. The difference from the last is that I interl= ocked > > :the tsleeps and asserted

dragonflybsd-kernel - Peter Avalos - Jan 8 2008 - 21:26

> The system was totally locked, and I couldn't exit to the debugger. > How is this "readily debuggable"? > > --Peter Were you in a GUI during the crash? When/if my system freezes in a GUI i also cant go to ddb. Have to try to reproduce the

dragonflybsd-kernel - Petr Janda - Jan 8 2008 - 22:52

On Wed, Jan 09, 2008 at 01:52:47PM +1100, Petr Janda wrote: >=20 > > The system was totally locked, and I couldn't exit to the debugger. > > How is this "readily debuggable"? > > > > --Peter >=20 >=20 >=20 > Were you in a GUI during the crash?

dragonflybsd-kernel - Peter Avalos - Jan 8 2008 - 23:21

On Jan 9, 2008 11:21 AM, Peter Avalos wrote: > > On Wed, Jan 09, 2008 at 01:52:47PM +1100, Petr Janda wrote: > > > > > The system was totally locked, and I couldn't exit to the debugger. > > > How is this "readily debuggable"

dragonflybsd-kernel - Sepherosa Ziehau - Jan 8 2008 - 23:29