Re: kern/58043: kernel crash in assert_sleepable() in -current, dk(4) driver?

[triaxx%NetBSD.org@localhost]

The following reply was made to PR kern/58043; it has been noted by GNATS.

From: triaxx%NetBSD.org@localhost
To: Paul Goyette <paul%whooppee.com@localhost>, gnats-bugs%netbsd.org@localhost
Cc: kern-bug-people%netbsd.org@localhost, gnats-admin%netbsd.org@localhost, netbsd-bugs%netbsd.org@localhost
Subject: Re: kern/58043: kernel crash in assert_sleepable() in -current, dk(4)
 driver?
Date: Wed, 27 Mar 2024 14:01:19 +0100

 I have two kernels on this system: netbsd-GENERIC and 
 netbsd-GOLIATH-noamdgpu.
 
 Surprisingly, GOLIATH-noamdgpu crashes when GENERIC doesn't. The diff 
 between both configurations is:
 
 1c1
 < ### START CONFIG FILE "/usr/src/sys/arch/amd64/conf/GENERIC"
 ---
  > ### START CONFIG FILE 
 "/home/triaxx/NetBSD/src/sys/arch/amd64/conf/GOLIATH-noamdgpu"
 118,119c118,119
 < #options      DEBUG           # expensive debugging checks/support
 < #options      LOCKDEBUG       # expensive locking checks/support
 ---
  > options       DEBUG           # expensive debugging checks/support
  > options       LOCKDEBUG       # expensive locking checks/support
 130,131c130,131
 < #options      KGDB            # remote debugger
 < #options      KGDB_DEVNAME="\"com\"",KGDB_DEVADDR=0x3f8,KGDB_DEVRATE=9600
 ---
  > options       KGDB            # remote debugger
  > options       KGDB_DEVNAME="\"com\"",KGDB_DEVADDR=0x3f8,KGDB_DEVRATE=9600
 255c255
 < #options      ACPIVERBOSE     # verbose ACPI configuration messages
 ---
  > options       ACPIVERBOSE     # verbose ACPI configuration messages
 461,462c461,462
 < i915drmkms*   at pci? dev ? function ?
 < intelfb*      at intelfbbus?
 ---
  > #i915drmkms*  at pci? dev ? function ?
  > #intelfb*     at intelfbbus?
 464,465c464,465
 < radeon*       at pci? dev ? function ?
 < radeondrmkmsfb* at radeonfbbus?
 ---
  > #radeon*      at pci? dev ? function ?
  > #radeondrmkmsfb* at radeonfbbus?
 470,471c470,471
 < nouveau*      at pci? dev ? function ?
 < nouveaufb*    at nouveaufbbus?
 ---
  > #nouveau*     at pci? dev ? function ?
  > #nouveaufb*   at nouveaufbbus?
 1245c1245
 < ### END CONFIG FILE "/usr/src/sys/arch/amd64/conf/GENERIC"
 ---
  > ### END CONFIG FILE 
 "/home/triaxx/NetBSD/src/sys/arch/amd64/conf/GOLIATH-noamdgpu"
 
 > I have a system with:
 > 
 >  Â Â Â  $ dmesg -t | grep ccd
 >  Â Â Â  dk7 at wd1: "ccd0p0", 488397088 blocks at 40, type: ccd
 >  Â Â Â  dk8 at wd2: "ccd0p1", 488397088 blocks at 40, type: ccd
 >  Â Â Â  ccd0: Interleaving 2 components (63 block interleave)
 >  Â Â Â  ccd0: /dev/dk7 (488397042 blocks)
 >  Â Â Â  ccd0: /dev/dk8 (488397042 blocks)
 >  Â Â Â  ccd0: total 976794084 blocks
 >  Â Â Â  ccd0: GPT GUID: 546b7b1d-bf71-46a2-9788-226ebaf7ae2d
 >  Â Â Â  dk12 at ccd0: "ccd0", 976794008 blocks at 40, type: ffs
 > 
 > The patch fixes the issue with the kernel that crashes on this system.
 > 
 >> I've already converted my config to raid0 instead of ccd, so I am sorry
 >> that I am unable to test the patch.  :-(
 >>
 >> On Mon, 18 Mar 2024, J. Hannken-Illjes wrote:
 >>
 >>> The following reply was made to PR kern/58043; it has been noted by 
 >>> GNATS.
 >>>
 >>> From: "J. Hannken-Illjes" <hannken%mailbox.org@localhost>
 >>> To: gnats-bugs%netbsd.org@localhost
 >>> Cc:
 >>> Subject: Re: kern/58043: kernel crash in assert_sleepable() in 
 >>> -current, dk(4)
 >>> driver?
 >>> Date: Mon, 18 Mar 2024 10:52:09 +0100
 >>>
 >>> Paul,
 >>>
 >>> please try the attached patch -- if it prevents assert_sleepable() to 
 >>> fire
 >>> it is the call to CCD_GETBUF() aka. pool_cache_get(ccd_cache, PR_WAITOK)
 >>> from softint context that has to be fixed.
 >>>
 >>> -- 
 >>> J. Hannken-Illjes - hannken%mailbox.org@localhost <mailto:hannken%mailbox.org@localhost>
 >>>
 >>> ccd_defer
 >>>
 >>> Always defer requests to the helper thread so ccdstart() doesn't
 >>> get called from softint anymore.
 >>>
 >>> diff -r 8b8d2498ffd9 -r 2ec4e85f1120 sys/dev/ccd.c
 >>> --- sys/dev/ccd.c
 >>> +++ sys/dev/ccd.c
 >>> @@ -777,17 +777,10 @@ ccdstrategy(struct buf *bp)
 >>> Â Â Â Â Â Â Â Â  return;
 >>> Â Â Â Â  }
 >>>
 >>> -Â Â Â  /* Defer to thread if system is low on memory. */
 >>> +Â Â Â  /* Always defer to thread. */
 >>> Â Â Â Â  bufq_put(cs->sc_bufq, bp);
 >>> -Â Â Â  if (__predict_false(ccdbackoff(cs))) {
 >>> -Â Â Â Â Â Â Â  mutex_exit(cs->sc_iolock);
 >>> -#ifdef DEBUG
 >>> -Â Â Â Â Â Â Â Â  if (ccddebug & CCDB_FOLLOW)
 >>> -Â Â Â Â Â Â Â Â Â Â Â Â  printf("ccdstrategy: holding off on I/O\n");
 >>> -#endif
 >>> -Â Â Â Â Â Â Â  return;
 >>> -Â Â Â  }
 >>> -Â Â Â  ccdstart(cs);
 >>> +Â Â Â  cv_broadcast(&cs->sc_push);
 >>> +Â Â Â  mutex_exit(cs->sc_iolock);
 >>> Â }
 >>>
 >>> Â static void
 >