DragonFlyBSD

On Thu, Sep 22, 2011 at 02:24:53AM +0000, Justin C. Sherrill (via DragonFly issue tracker) wrote:

New submission from Justin C. Sherrill <justin@shiningsilence.com>:

Here's things that should be cleared before the 2.12 release.

Our exactly-6-months due date is the 26th of this month, I think. The next
quarterly release of pkgsrc is scheduled for the 25th. Building a full set of
packages for the next release will take ~2 weeks, which pushes us to mid-October.

Bug 1998: kqueue issues with nginx. http://bugs.dragonflybsd.org/issue1998

(This has a workaround - build with something other than kqueue)

Bug 2134: sdiff causes stack overflow. http://bugs.dragonflybsd.org/issue2134

Updates of file, openssl, and openssh (done by Peter on 9/20)

Please add as necessary.

Two new ones, almost certainly kqueue-related:

Issue 1893: Firefox consumes 100% CPU while polling.
http://bugs.dragonflybsd.org/issue1893

Issue 2028: hylafax uses 100% cpu in a select() loop
http://bugs.dragonflybsd.org/issue2028

Also: Any Summer of Code 2011 projects - dependent on students and mentors. The
PUFFS port requires some functions on x86_64 to be added before it can be
integrated, I think.

On Thu, Sep 22, 2011 at 1:39 PM, Francois Tigeot <ftigeot@wolfpond.org> wrote:

Issue 2028: hylafax uses 100% cpu in a select() loop
http://bugs.dragonflybsd.org/issue2028

I would suspect this is a user land software bug. Could you help me
to see whether:
/var/spool/hylafax/FIFO.ttyd0
is regular file or not (it looks like a regular file from the output of fstat)?

I think select() for read on a regular file will return immediately.

Best Regards,
sephe

On Fri, Sep 23, 2011 at 07:11:44PM +0800, Sepherosa Ziehau wrote:

On Thu, Sep 22, 2011 at 1:39 PM, Francois Tigeot <ftigeot@wolfpond.org> wrote:

Issue 2028: hylafax uses 100% cpu in a select() loop
http://bugs.dragonflybsd.org/issue2028

I would suspect this is a user land software bug.

I doubt it; FreeBSD was previously installed on this machine and Hylafax had
no such problem then.

Could you help me
to see whether:
/var/spool/hylafax/FIFO.ttyd0
is regular file or not (it looks like a regular file from the output of fstat)?

It's not:

1. ls ls /var/spool/hylafax/FIFO.ttyd0
   prw------ 1 uucp dialer 0B Dec 26 2010 /var/spool/hylafax/FIFO.ttyd0

1. file /var/spool/hylafax/FIFO.ttyd0
   /var/spool/hylafax/FIFO.ttyd0: fifo (named pipe)

Bug 1998 fixed in 3bcb6e5e80cb05cecaddd866d4b15b428334b9aa.

On Fri, Sep 23, 2011 at 7:30 PM, Francois Tigeot <ftigeot@wolfpond.org> wrote:

On Fri, Sep 23, 2011 at 07:11:44PM +0800, Sepherosa Ziehau wrote:

On Thu, Sep 22, 2011 at 1:39 PM, Francois Tigeot <ftigeot@wolfpond.org> wrote:

Issue 2028: hylafax uses 100% cpu in a select() loop
http://bugs.dragonflybsd.org/issue2028

I would suspect this is a user land software bug.

I doubt it; FreeBSD was previously installed on this machine and Hylafax had
no such problem then.

Could you help me
to see whether:
/var/spool/hylafax/FIFO.ttyd0
is regular file or not (it looks like a regular file from the output of fstat)?

It's not:

1. ls ls /var/spool/hylafax/FIFO.ttyd0
   prw------  1 uucp  dialer     0B Dec 26  2010 /var/spool/hylafax/FIFO.ttyd0

1. file /var/spool/hylafax/FIFO.ttyd0
   /var/spool/hylafax/FIFO.ttyd0: fifo (named pipe)

Yeah, that's it. I think is fixed, please test the latest master:
010f86dc1259be3b29303bfc336532138acd077d

Best Regards,
sephe

On Thu, Sep 22, 2011 at 1:39 PM, Francois Tigeot <ftigeot@wolfpond.org> wrote:

Issue 1893: Firefox consumes 100% CPU while polling.
http://bugs.dragonflybsd.org/issue1893

Could you grab a simple kdump trace for this one? And fstat of the xulrunner?

I didn't see anything obviously wrong here (firefox 3.6.12)

Best Regards,
sephe

On Sat, Sep 24, 2011 at 10:33:23PM +0800, Sepherosa Ziehau wrote:

On Thu, Sep 22, 2011 at 1:39 PM, Francois Tigeot <ftigeot@wolfpond.org> wrote:

Issue 2028: hylafax uses 100% cpu in a select() loop
http://bugs.dragonflybsd.org/issue2028

Yeah, that's it. I think is fixed, please test the latest master:
010f86dc1259be3b29303bfc336532138acd077d

Fix confirmed for 2028, thanks !

On Sat, Sep 24, 2011 at 10:51:09PM +0800, Sepherosa Ziehau wrote:

On Thu, Sep 22, 2011 at 1:39 PM, Francois Tigeot <ftigeot@wolfpond.org> wrote:

Issue 1893: Firefox consumes 100% CPU while polling.
http://bugs.dragonflybsd.org/issue1893

Could you grab a simple kdump trace for this one? And fstat of the xulrunner?

I didn't see anything obviously wrong here (firefox 3.6.12)

I've updated issue 1893 with a kdump extract; the fstat output is in this
file:
http://leaf.dragonflybsd.org/~ftigeot/fstat_seamonkey.txt

I'm using seamonkey 2.3.1, it should be equivalent to firefox 4+

Bugs 2028 and 1893 fixed.

Remaining:
Bug 2134: sdiff causes stack overflow
Summer of Code imports, if any
pkgsrc-2011q3 release, now scheduled for Oct. 2.

On Mon, Sep 26, 2011 at 01:08:33AM +0000, Justin C. Sherrill (via DragonFly issue tracker) wrote:

Justin C. Sherrill <justin@shiningsilence.com> added the comment:

Bugs 2028 and 1893 fixed.

Remaining:
Bug 2134: sdiff causes stack overflow
Summer of Code imports, if any
pkgsrc-2011q3 release, now scheduled for Oct. 2.

Issue 2111: Command 'hammer snaprm' removes PFS symlink

There's a patch attached to this bug report and possibly another one
was sent to submit@

Bugs 2111 and 2134 fixed.

Remaining fixes to MFC:

3b264617432fc557fb7711129856cf8a1f236012
Fix spin lock held across contigfree/kfree in bounce code

315422416abed4ef71e97ac2d2a35d23e4a4827a (may be optional for 2.12)
Fix bug in lwkt_token_swap()

4005878e6418361bb0e5bb547ef3ade8e8998133
Resolve NULL ptr dereference in fill_kinfo_proc.

This is fixed in dcf32f81f3c5059fb6d7b6d0d4cfb496d52c8d9d.

4005878e6418361bb0e5bb547ef3ade8e8998133
Resolve NULL ptr dereference in fill_kinfo_proc.

Commit 316672b1896b289b16733682730638b488d1ca27 contains some fixes to the
VM_object path, which should resolve a recent panic marino@ saw.

Remaining fixes:

3b264617432fc557fb7711129856cf8a1f236012
Fix spin lock held across contigfree/kfree in bounce code

Mysterious VM bug - Crashes in sh(1). Duplicate by `bmake clean && bmake >
/dev/null` in devel/gettext-lib, over and over. Easiest to reproduce in x86_64
in a virtual machine.