[Bug 219857] panic in scsi_cd code

bugzilla-noreply at freebsd.org bugzilla-noreply at freebsd.org
Tue Dec 17 20:29:59 UTC 2019


https://bugs.freebsd.org/bugzilla/show_bug.cgi?id=219857

--- Comment #33 from commit-hook at freebsd.org ---
A commit references this bug:

Author: ken
Date: Tue Dec 17 20:29:47 UTC 2019
New revision: 355862
URL: https://svnweb.freebsd.org/changeset/base/355862

Log:
  MFC r355299:

    ------------------------------------------------------------------------
    r355299 | ken | 2019-12-02 14:57:39 -0500 (Mon, 02 Dec 2019) | 52 lines

    Fix a hang introduced in r351599.

    My changes in 351599 (kindly committed by avg) made the cd(4) media check
    asynchronous to avoid a sleep while holding a mutex.

    There was a difficult to reproduce bug with those changes that caused a
    hang on boot on some single processor machines/VMs.  Leandro Lupori
    managed to reproduce the bug, diagnose it, and supplied a patch!  Here is
    his analysis, from the PR:

    ======
    I was able to reproduce the problem described in comment#14.

    Actually, I wasn't trying to reproduce it, I just started seeing it a few
    weeks ago, in CURRENT.

    I can reproduce it consistently, by using QEMU to run a PowerPC64 VM with a
    single core/thread (-smp 1).

    It happens only when there is no media in the emulated CD-ROM, a device
    that QEMU adds by default, unless -nodefaults is specified in command line.

    I've debugged it and this is what I've found:

    1- After the CD probe is successful, GEOM will try to open the device,
    which will end up calling cdcheckmedia(), that sets CD state to
    CD_STATE_MEDIA_PREVENT.
    2- Next, scsi_prevent() is executed and succeeds, the CD_FLAG_DISC_LOCKED
    flag is set and CD state moves to CD_STATE_MEDIA_SIZE.
    3- Next, scsi_read_capacity() is executed and fails, state is set to
    CD_STATE_MEDIA_ALLOW, cdmediaprobedone() is called and wakes up
    cdcheckmedia().
    4- Then, when cdstart() is invoked to process CD_STATE_MEDIA_ALLOW, it
    first checks if CD_FLAG_DISC_LOCKED is set, and if so skips directly to
    CD_STATE_MEDIA_SIZE state. This will repeat the steps of bullet 3, entering
    an infinite MEDIA_SIZE command loop.

    When there is a least another core/thread, the GEOM thread that performed
    the initial cdopen() will get scheduled again, closing the CD device, that
    will call cdprevent(PR_ALLOW) that clears the CD_FLAG_DISC_LOCKED flag and
    breaks the loop.

    So, apparently, the problem is CD_STATE_MEDIA_ALLOW being skipped when
    CD_FLAG_DISC_LOCKED is set. If I understand correctly, in this case, the
    state should be advanced to CD_STATE_MEDIA size only when the current state
    is CD_STATE_MEDIA_PREVENT.
    =====
    ------------------------------------------------------------------------

  PR:           kern/219857
  Submitted by: Leandro Lupori <leandro.lupori at gmail.com>

Changes:
_U  stable/12/
  stable/12/sys/cam/scsi/scsi_cd.c

-- 
You are receiving this mail because:
You are on the CC list for the bug.


More information about the freebsd-scsi mailing list