Bug#397616: /proc/cpu/alignment should default to warn and fixup on arm
* Riku Voipio <[EMAIL PROTECTED]> [2008-05-04 22:07]: > 2) warnings can cause bad side-effects (imagine a unaligned > error in sysklogd..) > > http://thread.gmane.org/gmane.linux.ports.arm.kernel/34044 I think that's a good argument for not changing the upstream default. So I think we should close this bug report and someone should take this up with upstream (not me). Any volunteers? -- Martin Michlmayr http://www.cyrius.com/ -- To UNSUBSCRIBE, email to [EMAIL PROTECTED] with a subject of "unsubscribe". Trouble? Contact [EMAIL PROTECTED]
Bug#397616: /proc/cpu/alignment should default to warn and fixup on arm
I think this should be pushed via upstream. In the following thread, it seems to be concluded that 1) fixups should be default 2) warnings can cause bad side-effects (imagine a unaligned error in sysklogd..) http://thread.gmane.org/gmane.linux.ports.arm.kernel/34044 On Fri, Apr 25, 2008 at 03:12:08PM +0700, Martin Michlmayr wrote: > I'd like to hear Riku's opinion on this. > > But I'm inclined not to move away from the upstream default, although > I must admit that Sjoerd has good arguments for changing it. > > Maybe you could ask upstream whether it's time to change the default > to a warning? > > > * Sjoerd Simons <[EMAIL PROTECTED]> [2006-11-08 14:02]: > > Package: linux-2.6 > > Severity: normal > > > > Hi, > > > > By default the ARM kernel just ignores unaligned accesses from > > userspace and can just accesses another address then actually > > specified. This can cause strange behaviour by userspace programs. > > While i agree that these programs are somewhat buggy, but doing > > something undefined and not telling anyone doesn't seem like a good > > strategy :) > > > > Some discussion on #debian-arm indicates that it the current default > > made some sense in the old days. Where some programs actually relied > > on the behaviour and the amount of buggy programs was so big that it > > actually caused a flood of warnings. > > > > Times have changed though, no applications in debian should depend on > > this behaviour and turning on warn+fixup doesn't seem cause a flood > > anymore. Most problemeatic should have been fixed by now, as on Sparc > > you'll get a sigbus on unaligned access and iirc Alpha gives a warning > > about it. > > > > For reference, i discovered this issue because powerdns on my arm was > > giving out weird SOA records. So enabling fixup does solves real > > problems or at least shows where they are :) (Yes i've already patched > > pdns and will be sending the patch out after some more testing) > > > > Sjoerd > > > > -- System Information: > > Debian Release: testing/unstable > > APT prefers testing > > APT policy: (500, 'testing'), (101, 'unstable') > > Architecture: arm (armv5tel) > > Shell: /bin/sh linked to /bin/bash > > Kernel: Linux 2.6.18-1-iop32x > > Locale: LANG=C, LC_CTYPE=nl_NL (charmap=ISO-8859-1) > > > > > > -- > > To UNSUBSCRIBE, email to [EMAIL PROTECTED] > > with a subject of "unsubscribe". Trouble? Contact [EMAIL PROTECTED] > > -- > Martin Michlmayr > http://www.cyrius.com/ -- "rm -rf" only sounds scary if you don't have backups -- To UNSUBSCRIBE, email to [EMAIL PROTECTED] with a subject of "unsubscribe". Trouble? Contact [EMAIL PROTECTED]
Bug#397616: /proc/cpu/alignment should default to warn and fixup on arm
I'd like to hear Riku's opinion on this. But I'm inclined not to move away from the upstream default, although I must admit that Sjoerd has good arguments for changing it. Maybe you could ask upstream whether it's time to change the default to a warning? * Sjoerd Simons <[EMAIL PROTECTED]> [2006-11-08 14:02]: > Package: linux-2.6 > Severity: normal > > Hi, > > By default the ARM kernel just ignores unaligned accesses from > userspace and can just accesses another address then actually > specified. This can cause strange behaviour by userspace programs. > While i agree that these programs are somewhat buggy, but doing > something undefined and not telling anyone doesn't seem like a good > strategy :) > > Some discussion on #debian-arm indicates that it the current default > made some sense in the old days. Where some programs actually relied > on the behaviour and the amount of buggy programs was so big that it > actually caused a flood of warnings. > > Times have changed though, no applications in debian should depend on > this behaviour and turning on warn+fixup doesn't seem cause a flood > anymore. Most problemeatic should have been fixed by now, as on Sparc > you'll get a sigbus on unaligned access and iirc Alpha gives a warning > about it. > > For reference, i discovered this issue because powerdns on my arm was > giving out weird SOA records. So enabling fixup does solves real > problems or at least shows where they are :) (Yes i've already patched > pdns and will be sending the patch out after some more testing) > > Sjoerd > > -- System Information: > Debian Release: testing/unstable > APT prefers testing > APT policy: (500, 'testing'), (101, 'unstable') > Architecture: arm (armv5tel) > Shell: /bin/sh linked to /bin/bash > Kernel: Linux 2.6.18-1-iop32x > Locale: LANG=C, LC_CTYPE=nl_NL (charmap=ISO-8859-1) > > > -- > To UNSUBSCRIBE, email to [EMAIL PROTECTED] > with a subject of "unsubscribe". Trouble? Contact [EMAIL PROTECTED] -- Martin Michlmayr http://www.cyrius.com/ -- To UNSUBSCRIBE, email to [EMAIL PROTECTED] with a subject of "unsubscribe". Trouble? Contact [EMAIL PROTECTED]
Bug#397616: /proc/cpu/alignment should default to warn and fixup on arm
* Sjoerd Simons <[EMAIL PROTECTED]> [2006-11-08 14:02]: > By default the ARM kernel just ignores unaligned accesses from > userspace and can just accesses another address then actually > specified. This can cause strange behaviour by userspace programs. ... Nothing has been done with regards to this bug. I suggest that unless you can get upstream to change the behaviour we won't change it either. -- Martin Michlmayr http://www.cyrius.com/ -- To UNSUBSCRIBE, email to [EMAIL PROTECTED] with a subject of "unsubscribe". Trouble? Contact [EMAIL PROTECTED]
Bug#397616: /proc/cpu/alignment should default to warn and fixup on arm
* Sjoerd Simons <[EMAIL PROTECTED]> [2006-11-08 21:07]: > Sure, wookey had some extra comments after i asked for permission to post it > so > i've added those too: > http://beast.luon.net/~sjoerd/arm_unaligned.log http://netwinder.osuosl.org/users/b/brianbr/public_html/alignment.html Ok, thanks for the pointer. I don't think it's a good idea to enable this right now, but we should do it after etch is out. -- Martin Michlmayr http://www.cyrius.com/ -- To UNSUBSCRIBE, email to [EMAIL PROTECTED] with a subject of "unsubscribe". Trouble? Contact [EMAIL PROTECTED]
Bug#397616: /proc/cpu/alignment should default to warn and fixup on arm
On Wed, Nov 08, 2006 at 06:46:38PM +, Martin Michlmayr wrote: > * Sjoerd Simons <[EMAIL PROTECTED]> [2006-11-08 14:02]: > > Some discussion on #debian-arm indicates that it the current default > > Can you make logs of that discussion available? Sure, wookey had some extra comments after i asked for permission to post it so i've added those too: http://beast.luon.net/~sjoerd/arm_unaligned.log Sjoerd -- The amount of time between slipping on the peel and landing on the pavement is precisely 1 bananosecond. -- To UNSUBSCRIBE, email to [EMAIL PROTECTED] with a subject of "unsubscribe". Trouble? Contact [EMAIL PROTECTED]
Bug#397616: /proc/cpu/alignment should default to warn and fixup on arm
* Sjoerd Simons <[EMAIL PROTECTED]> [2006-11-08 14:02]: > Some discussion on #debian-arm indicates that it the current default Can you make logs of that discussion available? -- Martin Michlmayr http://www.cyrius.com/ -- To UNSUBSCRIBE, email to [EMAIL PROTECTED] with a subject of "unsubscribe". Trouble? Contact [EMAIL PROTECTED]
Processed: rename Bug#397616: /proc/cpu/alignment should default to warn and fixup on arm
Processing commands for [EMAIL PROTECTED]: > retitle 397616 [arm] /proc/cpu/alignment should default to warn and fixup on > arm Bug#397616: /proc/cpu/alignment should default to warn and fixup on arm Changed Bug title. > thanks Stopping processing here. Please contact me if you need assistance. Debian bug tracking system administrator (administrator, Debian Bugs database) -- To UNSUBSCRIBE, email to [EMAIL PROTECTED] with a subject of "unsubscribe". Trouble? Contact [EMAIL PROTECTED]
Bug#397616: /proc/cpu/alignment should default to warn and fixup on arm
Package: linux-2.6 Severity: normal Hi, By default the ARM kernel just ignores unaligned accesses from userspace and can just accesses another address then actually specified. This can cause strange behaviour by userspace programs. While i agree that these programs are somewhat buggy, but doing something undefined and not telling anyone doesn't seem like a good strategy :) Some discussion on #debian-arm indicates that it the current default made some sense in the old days. Where some programs actually relied on the behaviour and the amount of buggy programs was so big that it actually caused a flood of warnings. Times have changed though, no applications in debian should depend on this behaviour and turning on warn+fixup doesn't seem cause a flood anymore. Most problemeatic should have been fixed by now, as on Sparc you'll get a sigbus on unaligned access and iirc Alpha gives a warning about it. For reference, i discovered this issue because powerdns on my arm was giving out weird SOA records. So enabling fixup does solves real problems or at least shows where they are :) (Yes i've already patched pdns and will be sending the patch out after some more testing) Sjoerd -- System Information: Debian Release: testing/unstable APT prefers testing APT policy: (500, 'testing'), (101, 'unstable') Architecture: arm (armv5tel) Shell: /bin/sh linked to /bin/bash Kernel: Linux 2.6.18-1-iop32x Locale: LANG=C, LC_CTYPE=nl_NL (charmap=ISO-8859-1) -- To UNSUBSCRIBE, email to [EMAIL PROTECTED] with a subject of "unsubscribe". Trouble? Contact [EMAIL PROTECTED]
