Re: why are the watchdog drivers blacklisted?
On Feb 05, Marco d'Itri wrote: > I cannot even find anymore this i810_tco/i8xx_tco module, so in the next > udev upload I will remove all watchdog drivers from the blacklist and > maybe add back the ones reported as broken. Done, with the exception of iTCO_wdt which is still blacklisted as requested here. No complaints. -- ciao, Marco signature.asc Description: Digital signature
Re: why are the watchdog drivers blacklisted?
> Any chances of uploading the new version (after it has stabilized and made > it to testing) to backports.org? Sure, I'm currently just waiting for it to migrate. Michael -- Michael Meskes Michael at Fam-Meskes dot De, Michael at Meskes dot (De|Com|Net|Org) Michael at BorussiaFan dot De, Meskes at (Debian|Postgresql) dot Org ICQ 179140304, AIM/Yahoo/Skype michaelmeskes, Jabber mes...@jabber.org VfL Borussia! Força Barça! Go SF 49ers! Use Debian GNU/Linux, PostgreSQL -- To UNSUBSCRIBE, email to debian-devel-requ...@lists.debian.org with a subject of "unsubscribe". Trouble? Contact listmas...@lists.debian.org Archive: http://lists.debian.org/201002191608.11794.mes...@debian.org
Re: why are the watchdog drivers blacklisted?
On Fri, 19 Feb 2010, Michael Meskes wrote: > > For the record, the watchdog daemon seems not to be able to deal with 1Hz > > cleanly (must have something to do with these zombies it likes to keep > > around for 1s) in my test boxes. It can do 0.5Hz just fine though. > > The latest upload 5.7-4 or an older version? The older versions use sleep() Probably an older version, I don't recall if I tested it in Lenny or Squeeze (but I do know I didn't test with Sid's version). Any chances of uploading the new version (after it has stabilized and made it to testing) to backports.org? -- "One disk to rule them all, One disk to find them. One disk to bring them all and in the darkness grind them. In the Land of Redmond where the shadows lie." -- The Silicon Valley Tarot Henrique Holschuh -- To UNSUBSCRIBE, email to debian-devel-requ...@lists.debian.org with a subject of "unsubscribe". Trouble? Contact listmas...@lists.debian.org Archive: http://lists.debian.org/20100219130912.ga28...@khazad-dum.debian.net
Re: why are the watchdog drivers blacklisted?
> For the record, the watchdog daemon seems not to be able to deal with 1Hz > cleanly (must have something to do with these zombies it likes to keep > around for 1s) in my test boxes. It can do 0.5Hz just fine though. The latest upload 5.7-4 or an older version? The older versions use sleep() which sleep for half the interval which with 1Hz is 0 seconds. I thought I fixed that in -4, if not please give me some details. Michael -- Michael Meskes Michael at Fam-Meskes dot De, Michael at Meskes dot (De|Com|Net|Org) Michael at BorussiaFan dot De, Meskes at (Debian|Postgresql) dot Org ICQ 179140304, AIM/Yahoo/Skype michaelmeskes, Jabber mes...@jabber.org VfL Borussia! Força Barça! Go SF 49ers! Use Debian GNU/Linux, PostgreSQL -- To UNSUBSCRIBE, email to debian-devel-requ...@lists.debian.org with a subject of "unsubscribe". Trouble? Contact listmas...@lists.debian.org Archive: http://lists.debian.org/201002190944.16882.mes...@debian.org
Re: why are the watchdog drivers blacklisted?
On Mon, 08 Feb 2010, Henrique de Moraes Holschuh wrote: > On Tue, 09 Feb 2010, Darren Salt wrote: > > I demand that Guillem Jover may or may not have written... > > > On Mon, 2010-02-08 at 20:15:30 +0100, Michael Meskes wrote: > > >> Please keep in mind the OOM killer will only influence watchdog if it > > >> happens to kill it. If you happen to run out of memory though, you can > > >> tell watchdog to test if enough free mem is available. > > > > > The OOM killer can be disabled for precious processes by writting the > > > string "-17" to ???/proc//oom_adj???. > > > > That sounds to me like a good thing to do by default. > > And while at it, change wd_keepalive and watchdog to default to pat at 1Hz > instead of 0.1Hz. That will reduce a _lot_ the changes of spurious > reboots. For the record, the watchdog daemon seems not to be able to deal with 1Hz cleanly (must have something to do with these zombies it likes to keep around for 1s) in my test boxes. It can do 0.5Hz just fine though. -- "One disk to rule them all, One disk to find them. One disk to bring them all and in the darkness grind them. In the Land of Redmond where the shadows lie." -- The Silicon Valley Tarot Henrique Holschuh -- To UNSUBSCRIBE, email to debian-devel-requ...@lists.debian.org with a subject of "unsubscribe". Trouble? Contact listmas...@lists.debian.org Archive: http://lists.debian.org/20100218184306.gf31...@khazad-dum.debian.net
Re: Re: why are the watchdog drivers blacklisted?
> If you want to test forking ability just enable test-binary test without > giving > it a test-binary or use an empty one. This will make watchdog fork() and react > if not possible. Thinking about this some more, the test for an emtpy test-binary is done *after* the fork, so it should find the forking problem even without a manual test-binary configuration. I do remember testing this feature and it worked. So maybe you had another problem on your system. If you can reproduce this please tell me. Michael -- Michael Meskes Michael at Fam-Meskes dot De, Michael at Meskes dot (De|Com|Net|Org) Michael at BorussiaFan dot De, Meskes at (Debian|Postgresql) dot Org ICQ 179140304, AIM/Yahoo/Skype michaelmeskes, Jabber mes...@jabber.org VfL Borussia! Força Barça! Go SF 49ers! Use Debian GNU/Linux, PostgreSQL -- To UNSUBSCRIBE, email to debian-devel-requ...@lists.debian.org with a subject of "unsubscribe". Trouble? Contact listmas...@lists.debian.org
Re: why are the watchdog drivers blacklisted?
On Tue, 09 Feb 2010, Michael Meskes wrote: > > drivers have different defaults, and as far as I recall, some of them > > default to whatever the BIOS or BMC programmed in the watchdog. > > > > A period of 1s would be a much safer default on the userland side. > > Okay, will change. Thanks. -- "One disk to rule them all, One disk to find them. One disk to bring them all and in the darkness grind them. In the Land of Redmond where the shadows lie." -- The Silicon Valley Tarot Henrique Holschuh -- To UNSUBSCRIBE, email to debian-devel-requ...@lists.debian.org with a subject of "unsubscribe". Trouble? Contact listmas...@lists.debian.org
Re: why are the watchdog drivers blacklisted?
On Tue, Feb 09, 2010 at 12:57:29PM -0200, Henrique de Moraes Holschuh wrote: > The kernel drivers don't always expect pats at 0.016Hz by default. Some That's what it was years ago. > drivers have different defaults, and as far as I recall, some of them > default to whatever the BIOS or BMC programmed in the watchdog. > > A period of 1s would be a much safer default on the userland side. Okay, will change. Michael -- Michael Meskes Michael at Fam-Meskes dot De, Michael at Meskes dot (De|Com|Net|Org) Michael at BorussiaFan dot De, Meskes at (Debian|Postgresql) dot Org ICQ 179140304, AIM/Yahoo/Skype michaelmeskes, Jabber mes...@jabber.org VfL Borussia! Força Barça! Go SF 49ers! Use Debian GNU/Linux, PostgreSQL -- To UNSUBSCRIBE, email to debian-devel-requ...@lists.debian.org with a subject of "unsubscribe". Trouble? Contact listmas...@lists.debian.org
Re: why are the watchdog drivers blacklisted?
On Mon, Feb 08, 2010 at 08:54:23PM +0100, Guillem Jover wrote: > The OOM killer can be disabled for precious processes by writting the > string "-17" to “/proc//oom_adj”. Added in git. Thanks for the hint. Michael -- Michael Meskes Michael at Fam-Meskes dot De, Michael at Meskes dot (De|Com|Net|Org) Michael at BorussiaFan dot De, Meskes at (Debian|Postgresql) dot Org ICQ 179140304, AIM/Yahoo/Skype michaelmeskes, Jabber mes...@jabber.org VfL Borussia! Força Barça! Go SF 49ers! Use Debian GNU/Linux, PostgreSQL -- To UNSUBSCRIBE, email to debian-devel-requ...@lists.debian.org with a subject of "unsubscribe". Trouble? Contact listmas...@lists.debian.org
Re: why are the watchdog drivers blacklisted?
On Tue, Feb 09, 2010 at 12:57:29PM -0200, Henrique de Moraes Holschuh wrote: > On Tue, 09 Feb 2010, Michael Meskes wrote: > > This looks like a workaround for some other problem to me. Patting at 0.1Hz > > should be sufficient if the kernel expects a change at 0.016 Hz. I don't > > have > > any report about a spurious reboot, but if you do have some I'd like to > > learn > > more about it to see where this problem stems from. > The kernel drivers don't always expect pats at 0.016Hz by default. Some > drivers have different defaults, and as far as I recall, some of them > default to whatever the BIOS or BMC programmed in the watchdog. This is definitely the case for embedded drivers, they'll generally inherit the configuration that the hardware has. For many devices a timeout as long as 10s may not even be possible in the hardware. -- To UNSUBSCRIBE, email to debian-devel-requ...@lists.debian.org with a subject of "unsubscribe". Trouble? Contact listmas...@lists.debian.org
Re: why are the watchdog drivers blacklisted?
On Tue, 09 Feb 2010, Michael Meskes wrote: > TOn Mon, Feb 08, 2010 at 11:44:55PM -0200, Henrique de Moraes Holschuh wrote: > > And while at it, change wd_keepalive and watchdog to default to pat at 1Hz > > instead of 0.1Hz. That will reduce a _lot_ the changes of spurious > > reboots. > > This looks like a workaround for some other problem to me. Patting at 0.1Hz > should be sufficient if the kernel expects a change at 0.016 Hz. I don't have > any report about a spurious reboot, but if you do have some I'd like to learn > more about it to see where this problem stems from. The kernel drivers don't always expect pats at 0.016Hz by default. Some drivers have different defaults, and as far as I recall, some of them default to whatever the BIOS or BMC programmed in the watchdog. A period of 1s would be a much safer default on the userland side. -- "One disk to rule them all, One disk to find them. One disk to bring them all and in the darkness grind them. In the Land of Redmond where the shadows lie." -- The Silicon Valley Tarot Henrique Holschuh -- To UNSUBSCRIBE, email to debian-devel-requ...@lists.debian.org with a subject of "unsubscribe". Trouble? Contact listmas...@lists.debian.org
Re: why are the watchdog drivers blacklisted?
TOn Mon, Feb 08, 2010 at 11:44:55PM -0200, Henrique de Moraes Holschuh wrote: > And while at it, change wd_keepalive and watchdog to default to pat at 1Hz > instead of 0.1Hz. That will reduce a _lot_ the changes of spurious > reboots. This looks like a workaround for some other problem to me. Patting at 0.1Hz should be sufficient if the kernel expects a change at 0.016 Hz. I don't have any report about a spurious reboot, but if you do have some I'd like to learn more about it to see where this problem stems from. Michael -- Michael Meskes Michael at Fam-Meskes dot De, Michael at Meskes dot (De|Com|Net|Org) Michael at BorussiaFan dot De, Meskes at (Debian|Postgresql) dot Org ICQ 179140304, AIM/Yahoo/Skype michaelmeskes, Jabber mes...@jabber.org VfL Borussia! Força Barça! Go SF 49ers! Use Debian GNU/Linux, PostgreSQL -- To UNSUBSCRIBE, email to debian-devel-requ...@lists.debian.org with a subject of "unsubscribe". Trouble? Contact listmas...@lists.debian.org
Re: why are the watchdog drivers blacklisted?
On Tue, Feb 09, 2010 at 01:20:31AM +, Darren Salt wrote: > > The OOM killer can be disabled for precious processes by writting the > > string "-17" to “/proc//oom_adj”. > > That sounds to me like a good thing to do by default. Absolutely agreed. As soon as I find the time. Michael -- Michael Meskes Michael at Fam-Meskes dot De, Michael at Meskes dot (De|Com|Net|Org) Michael at BorussiaFan dot De, Meskes at (Debian|Postgresql) dot Org ICQ 179140304, AIM/Yahoo/Skype michaelmeskes, Jabber mes...@jabber.org VfL Borussia! Força Barça! Go SF 49ers! Use Debian GNU/Linux, PostgreSQL -- To UNSUBSCRIBE, email to debian-devel-requ...@lists.debian.org with a subject of "unsubscribe". Trouble? Contact listmas...@lists.debian.org
Re: why are the watchdog drivers blacklisted?
On Tue, 09 Feb 2010, Darren Salt wrote: > I demand that Guillem Jover may or may not have written... > > On Mon, 2010-02-08 at 20:15:30 +0100, Michael Meskes wrote: > >> Please keep in mind the OOM killer will only influence watchdog if it > >> happens to kill it. If you happen to run out of memory though, you can > >> tell watchdog to test if enough free mem is available. > > > The OOM killer can be disabled for precious processes by writting the > > string "-17" to ???/proc//oom_adj???. > > That sounds to me like a good thing to do by default. And while at it, change wd_keepalive and watchdog to default to pat at 1Hz instead of 0.1Hz. That will reduce a _lot_ the changes of spurious reboots. -- "One disk to rule them all, One disk to find them. One disk to bring them all and in the darkness grind them. In the Land of Redmond where the shadows lie." -- The Silicon Valley Tarot Henrique Holschuh -- To UNSUBSCRIBE, email to debian-devel-requ...@lists.debian.org with a subject of "unsubscribe". Trouble? Contact listmas...@lists.debian.org
Re: why are the watchdog drivers blacklisted?
I demand that Guillem Jover may or may not have written... > On Mon, 2010-02-08 at 20:15:30 +0100, Michael Meskes wrote: >> Please keep in mind the OOM killer will only influence watchdog if it >> happens to kill it. If you happen to run out of memory though, you can >> tell watchdog to test if enough free mem is available. > The OOM killer can be disabled for precious processes by writting the > string "-17" to “/proc//oom_adj”. That sounds to me like a good thing to do by default. -- | Darren Salt| linux at youmustbejoking | nr. Ashington, | Doon | using Debian GNU/Linux | or ds,demon,co,uk| Northumberland | Army | + http://www.youmustbejoking.demon.co.uk/ & http://tartarus.org/ds/ Errare umanum est. -- To UNSUBSCRIBE, email to debian-devel-requ...@lists.debian.org with a subject of "unsubscribe". Trouble? Contact listmas...@lists.debian.org
Re: why are the watchdog drivers blacklisted?
Hi! On Mon, 2010-02-08 at 20:15:30 +0100, Michael Meskes wrote: > Please keep in mind the OOM killer will only influence watchdog if it happens > to kill it. If you happen to run out of memory though, you can tell watchdog > to > test if enough free mem is available. The OOM killer can be disabled for precious processes by writting the string "-17" to “/proc//oom_adj”. regards, guillem -- To UNSUBSCRIBE, email to debian-devel-requ...@lists.debian.org with a subject of "unsubscribe". Trouble? Contact listmas...@lists.debian.org
Re: why are the watchdog drivers blacklisted?
On Mon, Feb 08, 2010 at 04:12:10PM +, Mark Brown wrote: > The core problem with watchdog WRT stuff like that that is that it's a > fairly small program and doesn't monitor the state of the rest of Plus it locks itself into main memory and will not suffer from swappiness itself. > userspace so there's error conditions like being deep into swap which > make the actual application unusable but don't stop watchdog soldiering > on. You can configure the watchdog daemon to also test quite some system conditions and react accordingly. This will however be done in user space. Michael -- Michael Meskes Michael at Fam-Meskes dot De, Michael at Meskes dot (De|Com|Net|Org) Michael at BorussiaFan dot De, Meskes at (Debian|Postgresql) dot Org ICQ 179140304, AIM/Yahoo/Skype michaelmeskes, Jabber mes...@jabber.org VfL Borussia! Força Barça! Go SF 49ers! Use Debian GNU/Linux, PostgreSQL -- To UNSUBSCRIBE, email to debian-devel-requ...@lists.debian.org with a subject of "unsubscribe". Trouble? Contact listmas...@lists.debian.org
Re: why are the watchdog drivers blacklisted?
On Mon, Feb 08, 2010 at 04:02:17PM +0100, Marco d'Itri wrote: > I am not aware of any other users of /dev/watchdog. There used to be other programs accessing it, but maybe they all vanished. > I use the default configuration. The system is broken enough that new The default configuration in Debian does nothing, not even accessing /dev/watchdog. So I take it you enabled the device. But then it does exactly just that, triggering the device, which means you will only get a reset if the device can't be triggered anymore because e.g the kernel oopsed. Please keep in mind the OOM killer will only influence watchdog if it happens to kill it. If you happen to run out of memory though, you can tell watchdog to test if enough free mem is available. > TCP connections are established but apparently the running daemons > cannot fork (e.g. I do not get a SMTP banner). > I have no way of reproducing the issue. If you want to test forking ability just enable test-binary test without giving it a test-binary or use an empty one. This will make watchdog fork() and react if not possible. Michael -- Michael Meskes Michael at Fam-Meskes dot De, Michael at Meskes dot (De|Com|Net|Org) Michael at BorussiaFan dot De, Meskes at (Debian|Postgresql) dot Org ICQ 179140304, AIM/Yahoo/Skype michaelmeskes, Jabber mes...@jabber.org VfL Borussia! Força Barça! Go SF 49ers! Use Debian GNU/Linux, PostgreSQL -- To UNSUBSCRIBE, email to debian-devel-requ...@lists.debian.org with a subject of "unsubscribe". Trouble? Contact listmas...@lists.debian.org
Re: why are the watchdog drivers blacklisted?
On Mon, Feb 08, 2010 at 03:51:24PM +0100, Michael Meskes wrote: > On Mon, Feb 08, 2010 at 04:45:53AM +0100, Marco d'Itri wrote: > > (BTW, is there any other watchdog daemon? The watchdog package reliably > > fails to detect when the system is half-killed by OOM.) > How about explaing your problem a little bit better and, if it's really a > failure in watchdog, reporting a bug? What exactly did you do? And how did you > configure watchdog? The core problem with watchdog WRT stuff like that that is that it's a fairly small program and doesn't monitor the state of the rest of userspace so there's error conditions like being deep into swap which make the actual application unusable but don't stop watchdog soldiering on. -- To UNSUBSCRIBE, email to debian-devel-requ...@lists.debian.org with a subject of "unsubscribe". Trouble? Contact listmas...@lists.debian.org
Re: why are the watchdog drivers blacklisted?
On Feb 08, Michael Meskes wrote: > On Mon, Feb 08, 2010 at 04:45:53AM +0100, Marco d'Itri wrote: > > If the defaults for some drivers are wrong then I can't see why they > > should not be fixed, but if default configuration parameters are needed > > then they should be provided by the watchdog package. > If the watchdog package is the only one using these modules that would be > fine. > But watchdog itself just expects a device to be present and only insmods a > module if one is manually configured. Other users might use these modules with > the user space daemon package but they still should get the right > configuration, don't you think? Or what am I missing. I am not aware of any other users of /dev/watchdog. > > (BTW, is there any other watchdog daemon? The watchdog package reliably > > fails to detect when the system is half-killed by OOM.) > How about explaing your problem a little bit better and, if it's really a > failure in watchdog, reporting a bug? What exactly did you do? And how did you > configure watchdog? I use the default configuration. The system is broken enough that new TCP connections are established but apparently the running daemons cannot fork (e.g. I do not get a SMTP banner). I have no way of reproducing the issue. -- ciao, Marco signature.asc Description: Digital signature
Re: why are the watchdog drivers blacklisted?
On Mon, Feb 08, 2010 at 04:45:53AM +0100, Marco d'Itri wrote: > If the defaults for some drivers are wrong then I can't see why they > should not be fixed, but if default configuration parameters are needed > then they should be provided by the watchdog package. If the watchdog package is the only one using these modules that would be fine. But watchdog itself just expects a device to be present and only insmods a module if one is manually configured. Other users might use these modules with the user space daemon package but they still should get the right configuration, don't you think? Or what am I missing. > (BTW, is there any other watchdog daemon? The watchdog package reliably > fails to detect when the system is half-killed by OOM.) How about explaing your problem a little bit better and, if it's really a failure in watchdog, reporting a bug? What exactly did you do? And how did you configure watchdog? Michael -- Michael Meskes Michael at Fam-Meskes dot De, Michael at Meskes dot (De|Com|Net|Org) Michael at BorussiaFan dot De, Meskes at (Debian|Postgresql) dot Org ICQ 179140304, AIM/Yahoo/Skype michaelmeskes, Jabber mes...@jabber.org VfL Borussia! Força Barça! Go SF 49ers! Use Debian GNU/Linux, PostgreSQL -- To UNSUBSCRIBE, email to debian-devel-requ...@lists.debian.org with a subject of "unsubscribe". Trouble? Contact listmas...@lists.debian.org
Re: why are the watchdog drivers blacklisted?
On Feb 07, Henrique de Moraes Holschuh wrote: > I'd rather we had a watchdog mini policy that boils down to: As the udev and module-init-tools maintainer my goal is to support automatically loading all the drivers which their maintainers intended to be automatically loaded and blacklist until they are fixed the ones which are currently too buggy to be automatically loaded. If the defaults for some drivers are wrong then I can't see why they should not be fixed, but if default configuration parameters are needed then they should be provided by the watchdog package. (BTW, is there any other watchdog daemon? The watchdog package reliably fails to detect when the system is half-killed by OOM.) -- ciao, Marco signature.asc Description: Digital signature
Re: why are the watchdog drivers blacklisted?
On Sat, 06 Feb 2010, Marco d'Itri wrote: > On Feb 06, Henrique de Moraes Holschuh wrote: > > It got renamed to wdt_tco, I think, and it will hard-hang a lot of thinkpads > > if it ever triggers, for example: the SMBIOS can't handle it. > OK, I will blacklist this one. I'd rather we had a watchdog mini policy that boils down to: Watchdog drivers have to default to *at least* N seconds timeout (N can't be too large, some watchdogs have hardware/firmware limits). All watchdog-enabled packages have to default to *at most* N/2 seconds timeout (probably N/3 is better, though). The blacklisting of watchdog drivers would be switched to default options to ensure the "N seconds timeout", or alternatively, we would need to fix it on the kernel tree. I prefer the options, because AFAIK some kernel drivers default to "leave the timeout to whatever was set by the BIOS", and it would be tough to get that policy changed upstream. > > Anyway, if for any reason we load a watchdog driver, AND any of the watchdog > > userspace packages by mistake, we can cause data loss. > root can cause data loss by running rm -rf / by mistake as well, so this > is not a great argument. And if that happens because of any default config or package maintainer script, it is a "critical" bug. At least the watchdog issues would warrant at most "grave" bugs (and personally I would place them at severity "important", since spurious reboots are a normal and expected failure mode of a watchdog system). -- "One disk to rule them all, One disk to find them. One disk to bring them all and in the darkness grind them. In the Land of Redmond where the shadows lie." -- The Silicon Valley Tarot Henrique Holschuh -- To UNSUBSCRIBE, email to debian-devel-requ...@lists.debian.org with a subject of "unsubscribe". Trouble? Contact listmas...@lists.debian.org
Re: why are the watchdog drivers blacklisted?
I demand that Marco d'Itri may or may not have written... > On Feb 06, Henrique de Moraes Holschuh wrote: >> It got renamed to wdt_tco, I think, Do you mean iTCO_wdt? If so, then you should know that that's working fine on my EeePC 901. >> and it will hard-hang a lot of thinkpads if it ever triggers, for >> example: the SMBIOS can't handle it. > OK, I will blacklist this one. Via /etc/modprobe.d/*? It looks more like a case for DMI-based blacklisting to me. However, according to the module source, this can be used for computers with broken BIOSes (given CONFIG_ITCO_VENDOR_SUPPORT=y): options iTCO_wdt vendorsupport=901 [snip] -- | Darren Salt| linux at youmustbejoking | nr. Ashington, | Doon | using Debian GNU/Linux | or ds,demon,co,uk| Northumberland | Army | + This comment has been censored. I'd like to, but my mother would never let me hear the end of it. -- To UNSUBSCRIBE, email to debian-devel-requ...@lists.debian.org with a subject of "unsubscribe". Trouble? Contact listmas...@lists.debian.org
Re: why are the watchdog drivers blacklisted?
On Feb 06, Henrique de Moraes Holschuh wrote: > It got renamed to wdt_tco, I think, and it will hard-hang a lot of thinkpads > if it ever triggers, for example: the SMBIOS can't handle it. OK, I will blacklist this one. > Anyway, if for any reason we load a watchdog driver, AND any of the watchdog > userspace packages by mistake, we can cause data loss. root can cause data loss by running rm -rf / by mistake as well, so this is not a great argument. -- ciao, Marco signature.asc Description: Digital signature
Re: why are the watchdog drivers blacklisted?
On Fri, 05 Feb 2010, Marco d'Itri wrote: > On Feb 05, Petter Reinholdtsen wrote: > > do not remember any more details. This was discovered a few years > > ago, and I hope that crazy driver has been fixed in the mean time. > So it looks like this *is* my fault, in my defense I can only say that I > got bad advice from the kernel maintainers... > http://bugs.debian.org/cgi-bin/bugreport.cgi?bug=249600 > > I cannot even find anymore this i810_tco/i8xx_tco module, so in the next > udev upload I will remove all watchdog drivers from the blacklist and > maybe add back the ones reported as broken. It got renamed to wdt_tco, I think, and it will hard-hang a lot of thinkpads if it ever triggers, for example: the SMBIOS can't handle it. Anyway, if for any reason we load a watchdog driver, AND any of the watchdog userspace packages by mistake, we can cause data loss. For example: someone decided that once every 10s was a good default watchdog pat, which is incorrect. If the platform or the kernel driver defaults to 10s or less, bad things happen. I have seen this happen on Lenny. I.e. even if the kernel and platform cooperates, we don't have done our work carefully enough yet to risk these things getting enabled without local setup. -- "One disk to rule them all, One disk to find them. One disk to bring them all and in the darkness grind them. In the Land of Redmond where the shadows lie." -- The Silicon Valley Tarot Henrique Holschuh -- To UNSUBSCRIBE, email to debian-devel-requ...@lists.debian.org with a subject of "unsubscribe". Trouble? Contact listmas...@lists.debian.org
Re: why are the watchdog drivers blacklisted?
On Feb 05, Petter Reinholdtsen wrote: > do not remember any more details. This was discovered a few years > ago, and I hope that crazy driver has been fixed in the mean time. So it looks like this *is* my fault, in my defense I can only say that I got bad advice from the kernel maintainers... http://bugs.debian.org/cgi-bin/bugreport.cgi?bug=249600 I cannot even find anymore this i810_tco/i8xx_tco module, so in the next udev upload I will remove all watchdog drivers from the blacklist and maybe add back the ones reported as broken. -- ciao, Marco signature.asc Description: Digital signature
Re: why are the watchdog drivers blacklisted?
Petter Reinholdtsen wrote: > This caused the installer to stop after 1 minute. Not sure how this could have affected the installer as we don't include any watchdog modules in D-I (at least, not that I'm aware of). Cheers, FJP -- To UNSUBSCRIBE, email to debian-devel-requ...@lists.debian.org with a subject of "unsubscribe". Trouble? Contact listmas...@lists.debian.org
Re: why are the watchdog drivers blacklisted?
On Fri, Feb 05, 2010 at 10:11:25AM +0100, Petter Reinholdtsen wrote: > [Marco d'Itri] > > I maintain the package providing it, but I fear it is the result of > > cargo cult sysadmining. A driver will not engage the watchdog > > anyway until /dev/watchdog is opened. > If I remember correctly, the reason is that the observed behaviour is > that a driver sometimes will engage the watchdog without /dev/watchdog > is opened, triggered by one driver starting without anyone touching > /dev/watchdog. This caused the installer to stop after 1 minute. I > do not remember any more details. This was discovered a few years > ago, and I hope that crazy driver has been fixed in the mean time. That's just a buggy watchdog driver, though (unless the problem was really that the watchdog was already enabled by the hardware prior to startup, but then not loading the driver isn't going to help anything). -- To UNSUBSCRIBE, email to debian-devel-requ...@lists.debian.org with a subject of "unsubscribe". Trouble? Contact listmas...@lists.debian.org
Re: why are the watchdog drivers blacklisted?
[Marco d'Itri] > I maintain the package providing it, but I fear it is the result of > cargo cult sysadmining. A driver will not engage the watchdog > anyway until /dev/watchdog is opened. If I remember correctly, the reason is that the observed behaviour is that a driver sometimes will engage the watchdog without /dev/watchdog is opened, triggered by one driver starting without anyone touching /dev/watchdog. This caused the installer to stop after 1 minute. I do not remember any more details. This was discovered a few years ago, and I hope that crazy driver has been fixed in the mean time. Happy hacking, -- Petter Reinholdtsen -- To UNSUBSCRIBE, email to debian-devel-requ...@lists.debian.org with a subject of "unsubscribe". Trouble? Contact listmas...@lists.debian.org
why are the watchdog drivers blacklisted?
/etc/modprobe.d/blacklist.conf contains this comment, but why? # watchdog drivers should be loaded only if a watchdog daemon is installed I maintain the package providing it, but I fear it is the result of cargo cult sysadmining. A driver will not engage the watchdog anyway until /dev/watchdog is opened. -- ciao, Marco signature.asc Description: Digital signature