recent udev upgrade failure on alpha

recent udev upgrade failure on alpha

[Bob Tracy]

Well...  Yes, "sid" is cantankerous and known to occasionally break
working systems, but this is the first time the thermonuclear software
upgrade option has been used effectively on my poor Alpha :-(.

Specifically, I've got about 50 installed but unconfigured packages and
an almost unbootable (single-user mode only) machine because udev 166-1
absolutely requires inotify support, and it doesn't seems to be present
in spite of "CONFIG_INOTIFY_USER=y" being set in all the kernels I've
got installed on my system.

Running "dpkg --configure udev" gives me the following:

udevd[pid]: inotify_init failed: Function not implemented
udevd[pid]: error initializing inotify

and the post-installation script fails.

So...  What's missing or unimplemented on Alpha?  Prior versions of
"udev" worked just fine.

As far as recovery options, I'm open to suggestions including selective
downgrading of packages, but "dependency hell" may make that option
unworkable from a practical standpoint.  I'd rather go forward.  Current
system state: running in single-user mode with a static /dev, and I *do*
have network connectivity for transferring stuff in/out as needed.

-- 
------------------------------------------------------------------------
Bob Tracy          |  "Every normal man must be tempted at times to spit
rct@frus.com       |   upon his hands, hoist the black flag, and begin
                   |   slitting throats."	-- H.L. Mencken
------------------------------------------------------------------------

Re: recent udev upgrade failure on alpha

[Bob Tracy]

On Thu, Feb 24, 2011 at 01:02:03AM -0600, Bob Tracy wrote:
> Running "dpkg --configure udev" gives me the following:
> 
> udevd[pid]: inotify_init failed: Function not implemented
> udevd[pid]: error initializing inotify
> 
> and the post-installation script fails.
> 
> So...  What's missing or unimplemented on Alpha?  Prior versions of
> "udev" worked just fine.

Found it.  inotify_init1() is a stub function on Alpha.  A heated
discussion *somewhere* produced a udev patch that *may* work:

--- a/udev/udev-watch.c
+++ b/udev/udev-watch.c
@@ -38,8 +38,10 @@ static int inotify_fd = -1;
  */
 int udev_watch_init(struct udev *udev)
 {
-       inotify_fd = inotify_init1(IN_CLOEXEC);
-       if (inotify_fd < 0)
+       inotify_fd = inotify_init();
+       if (inotify_fd >= 0)
+               util_set_fd_cloexec(inotify_fd);
+       else
                err(udev, "inotify_init failed: %m\n");
        return inotify_fd;
 }

(Formatting of the above is probably bogus due to conversion from a
formatted html presentation, but the gist of the fix should be apparent).

Time to retrieve the source package and rebuild it...

-- 
------------------------------------------------------------------------
Bob Tracy          |  "Every normal man must be tempted at times to spit
rct@frus.com       |   upon his hands, hoist the black flag, and begin
                   |   slitting throats."	-- H.L. Mencken
------------------------------------------------------------------------

RE: recent udev upgrade failure on alpha

[Uwe Schindler]

Hi Bob,

I downgraded udev on this unstable machine by downloading and archiving the
old source package (which is still on the servers for "stable"). This works
for now.

My bigger problem is that recent kernels (even the stable one no longer work
for me). On mounting my root  xfs filesystem it produces a message about a
wrong relocation format / some other linking problem when loading the XFS
.ko (the exact message is not known, its somewhere in the screen buffer of
the already rebooted with older kernel machine, I can only remember what the
problem was - next time I try I will take a pen and write down the message
on console). This happens with the last stable kernel version, did not test
2.7.37-1 until now. 
The last kernel that works for me is 2.6.31-1-alpha-generic. Does anybody
knows what the problem loading the XFS module is?

Unstable is else still running fine!

Uwe

-----
Uwe Schindler
H.-H.-Meier-Allee 63, D-28213 Bremen
http://www.thetaphi.de
eMail: uwe@thetaphi.de


> -----Original Message-----
> From: Bob Tracy [mailto:rct@gherkin.frus.com]
> Sent: Thursday, February 24, 2011 8:52 AM
> To: debian-alpha@lists.debian.org
> Cc: linux-alpha@vger.kernel.org; mattst88@gmail.com; mcree@orcon.net.nz
> Subject: Re: recent udev upgrade failure on alpha
> 
> On Thu, Feb 24, 2011 at 01:02:03AM -0600, Bob Tracy wrote:
> > Running "dpkg --configure udev" gives me the following:
> >
> > udevd[pid]: inotify_init failed: Function not implemented
> > udevd[pid]: error initializing inotify
> >
> > and the post-installation script fails.
> >
> > So...  What's missing or unimplemented on Alpha?  Prior versions of
> > "udev" worked just fine.
> 
> Found it.  inotify_init1() is a stub function on Alpha.  A heated
discussion
> *somewhere* produced a udev patch that *may* work:
> 
> --- a/udev/udev-watch.c
> +++ b/udev/udev-watch.c
> @@ -38,8 +38,10 @@ static int inotify_fd = -1;
>   */
>  int udev_watch_init(struct udev *udev)
>  {
> -       inotify_fd = inotify_init1(IN_CLOEXEC);
> -       if (inotify_fd < 0)
> +       inotify_fd = inotify_init();
> +       if (inotify_fd >= 0)
> +               util_set_fd_cloexec(inotify_fd);
> +       else
>                 err(udev, "inotify_init failed: %m\n");
>         return inotify_fd;
>  }
> 
> (Formatting of the above is probably bogus due to conversion from a
> formatted html presentation, but the gist of the fix should be apparent).
> 
> Time to retrieve the source package and rebuild it...
> 
> --
> ------------------------------------------------------------------------
> Bob Tracy          |  "Every normal man must be tempted at times to spit
> rct@frus.com       |   upon his hands, hoist the black flag, and begin
>                    |   slitting throats."	-- H.L. Mencken
> ------------------------------------------------------------------------
> 
> 
> --
> To UNSUBSCRIBE, email to debian-alpha-REQUEST@lists.debian.org
> with a subject of "unsubscribe". Trouble? Contact
listmaster@lists.debian.org
> Archive: http://lists.debian.org/20110224075149.GA28779@gherkin.frus.com

Re: recent udev upgrade failure on alpha

[Matt Turner]

On Thu, Feb 24, 2011 at 10:09 AM, Uwe Schindler <uwe@thetaphi.de> wrote:
> My bigger problem is that recent kernels (even the stable one no longer work
> for me). On mounting my root  xfs filesystem it produces a message about a
> wrong relocation format / some other linking problem when loading the XFS
> .ko (the exact message is not known, its somewhere in the screen buffer of
> the already rebooted with older kernel machine, I can only remember what the
> problem was - next time I try I will take a pen and write down the message
> on console). This happens with the last stable kernel version, did not test
> 2.7.37-1 until now.
> The last kernel that works for me is 2.6.31-1-alpha-generic. Does anybody
> knows what the problem loading the XFS module is?
>
> Unstable is else still running fine!

This is the first I've heard of XFS not working (though, I've never
used it myself). Have you reported this anywhere?

Please mail linux-alpha@vger.kernel.org and the appropriate XFS
mailing list to try to get this fixed.

Matt
--
To unsubscribe from this list: send the line "unsubscribe linux-alpha" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html

RE: recent udev upgrade failure on alpha

[Uwe Schindler]

From: Matt Turner [mailto:mattst88@gmail.com]
> On Thu, Feb 24, 2011 at 10:09 AM, Uwe Schindler <uwe@thetaphi.de>
> wrote:
> > My bigger problem is that recent kernels (even the stable one no 
> > longer work for me). On mounting my root  xfs filesystem it produces 
> > a message about a wrong relocation format / some other linking 
> > problem when loading the XFS .ko (the exact message is not known, 
> > its somewhere in the screen buffer of the already rebooted with 
> > older kernel machine, I can only remember what the problem was - 
> > next time I try I will take a pen and write down the message on 
> > console). This happens with the last stable kernel version, did not 
> > test
> > 2.7.37-1 until now.
> > The last kernel that works for me is 2.6.31-1-alpha-generic. Does 
> > anybody knows what the problem loading the XFS module is?
> >
> > Unstable is else still running fine!
> 
> This is the first I've heard of XFS not working (though, I've never 
> used it myself). Have you reported this anywhere?
> 
> Please mail linux-alpha@vger.kernel.org and the appropriate XFS 
> mailing list to try to get this fixed.

This mail thread is already going to these mailing lists
(linux-alpha@vger.kernel.org).

I will try 2.6.27-1-alpha-generic ASAP and report if it is maybe solved,
else I will post the exact error message XFS kernel module is presenting (at
least it's not a problem of my installation, because 2.6.31 and previous
works fine with XFS since years on an 128 Gig disk). Is the UDEV problem
gone with 2.6.37 kernel, too? I think it's 2.6.33 that fixed it (from your
previous mail on this thread). 

Finally I still have to tell that Debian Unstable is still running very well
with this AlphaStation 500 with EV56! So I never understood why there was no
Debian release for Alpha anymore :(

Uwe

RE: recent udev upgrade failure on alpha

[Uwe Schindler]

Hi,

thanks for response!

> -----Original Message-----
> From: Witold Baryluk [mailto:baryluk@smp.if.uj.edu.pl]
> Sent: Wednesday, May 18, 2011 4:56 PM
> To: Uwe Schindler
> Subject: Re: recent udev upgrade failure on alpha
> 
> On 05-18 14:15, Uwe Schindler wrote:
> > Hi,
> >
> > I had some time today to try again an upgrade of my AlphaStation 500
> > to "linux-image-2.6.38-2-alpha-generic". I got again an error on
> > loading the xfs.ko module from initrd  (I typed it manually from the
> > boot screen to this mail *g*):
> >
> > Module xfs: Relocation (type4) overflow vs _xfs_buf_find
> > FATAL: Error inserting xfs
(/lib/modules/2.6.38-2-alpha-generic/.../xfs.ko):
> > Invalid module format
> >
> > All other modules before (like SCSI QLA1040 and SATA SiI3114, Tulip
> > Ethernet,... load successfully). The last kernel that works is (as
> > written
> > below): 2.6.31-1-alpha-generic, all after 2.6.32 fail with the same
> > error message (with some minor differences).
> >
> > I had to go back to the working kernel and of course prevent upgrading
> > udev...
> 
> Strange. I was using xfs on 2.6.32 without problem (but not as root). Can
try
> on 2.6.38.

2.6.32 is also broken here (at least in the debian alpha-generic kernel).

> > Has anybody any idea what's wrong with the XFS module? My root file
> > system on a SATA disk (not /boot) is XFS. Maybe its again a binutils
bug.
> 
> Looks to be, but why only this module?

That's the big question. It's not a corrupted initrd, because i regenerated
it quite often, reinstalled kernels,... It is broken since .32 for this
machine.

Here some more infos about the machine (with working kernel):

[    0.000000] Initializing cgroup subsys cpuset
[    0.000000] Initializing cgroup subsys cpu
[    0.000000] Linux version 2.6.31-1-alpha-generic (Debian 2.6.31-2)
(ben@decadent.org.uk) (gcc version 4.3.4 (Debian 4.3.4-6) ) #1 Mon Nov 16
05:56:06 UTC 2009
[    0.000000] Booting GENERIC on Alcor variation Alcor using machine vector
Alcor from SRM
[    0.000000] Major Options: MAGIC_SYSRQ
[    0.000000] Command line: ro  root=/dev/disk/by-label/root
...

$  cat /proc/cpuinfo
cpu                     : Alpha
cpu model               : EV56
cpu variation           : 7
cpu revision            : 0
cpu serial number       :
system type             : Alcor
system variation        : Alcor
system revision         : 0
system serial number    :
cycle frequency [Hz]    : 500000000
timer frequency [Hz]    : 1024.00
page size [bytes]       : 8192
phys. address bits      : 40
max. addr. space #      : 127
BogoMIPS                : 990.32
kernel unaligned acc    : 0 (pc=0,va=0)
user unaligned acc      : 0 (pc=0,va=0)
platform string         : Digital AlphaStation 500/500
cpus detected           : 1
L1 Icache               : 8K, 1-way, 32b line
L1 Dcache               : 8K, 1-way, 32b line
L2 cache                : 96K, 3-way, 64b line
L3 cache                : 8192K, 1-way, 64b line

$ lspci
00:06.0 Ethernet controller: Digital Equipment Corporation DECchip 21040
[Tulip] (rev 26)
00:07.0 USB Controller: NEC Corporation USB (rev 43)
00:07.1 USB Controller: NEC Corporation USB (rev 43)
00:07.2 USB Controller: NEC Corporation USB 2.0 (rev 04)
00:08.0 RAID bus controller: Silicon Image, Inc. SiI 3114
[SATALink/SATARaid] Serial ATA Controller (rev 02)
00:09.0 SCSI storage controller: QLogic Corp. ISP1020 Fast-wide SCSI (rev
02)
00:0a.0 EISA bridge: Intel Corporation 82375EB/SB PCI to EISA Bridge (rev
15)
00:0b.0 Ethernet controller: Digital Equipment Corporation DECchip 21140
[FasterNet] (rev 20)
00:0c.0 VGA compatible controller: Digital Equipment Corporation PBXGB
[TGA2] (rev 22)

$ lsscsi
[0:0:0:0]    disk    IOMEGA   ZIP 100          J.03  /dev/sda   <-- this
uses special alpha partition table  with ext2 as /boot (contains aboot and
kernel)
[0:0:6:0]    cd/dvd  TEAC     CD-R56S          1.0P  /dev/sr0
[1:0:0:0]    disk    ATA      ST3120827AS      3.42  /dev/sdb  <-- on this
disk is xfs filesystem

Uwe

Re: recent udev upgrade failure on alpha

[Bob Tracy]

On Wed, May 18, 2011 at 05:28:42PM +0200, Uwe Schindler wrote:
> > > I had some time today to try again an upgrade of my AlphaStation 500
> > > to "linux-image-2.6.38-2-alpha-generic". I got again an error on
> > > loading the xfs.ko module from initrd  (I typed it manually from the
> > > boot screen to this mail *g*):
> > >
> > > Module xfs: Relocation (type4) overflow vs _xfs_buf_find
> > > FATAL: Error inserting xfs
> (/lib/modules/2.6.38-2-alpha-generic/.../xfs.ko):
> > > Invalid module format

Another data point...

My PWS 433au with a 2.6.38 kernel built from the standard kernel.org
source has a "xfs.ko" that will load correctly using "modprobe".  I
haven't tried the 2.6.38-2-alpha-generic kernel.

--Bob

RE: recent udev upgrade failure on alpha

[Uwe Schindler]

Hi,

today I had some time to get my machine out of the XFS relocation failure on
booting, it now boots with EXT4 as root file system (was hard word, see
below). I have now linux-image-2.6.38-2-alpha-generic running (installed
from debian-ports) and system runs fine for now.

I tried then to upgrade the package "udev" (was on hold, to prevent me from
making system unbootable), with the hope that the inotify is working with
this debian kernel. Unfortunately that's not the case, UDEV complains about
missing inotify.

So is the patch below already included in some kernel images available? Or
does somebody have a recent udev version as .deb packages?

> On Thu, Feb 24, 2011 at 01:02:03AM -0600, Bob Tracy wrote:
> > Running "dpkg --configure udev" gives me the following:
> >
> > udevd[pid]: inotify_init failed: Function not implemented
> > udevd[pid]: error initializing inotify
> >
> > and the post-installation script fails.
> >
> > So...  What's missing or unimplemented on Alpha?  Prior versions of
> > "udev" worked just fine.
> 
> Found it.  inotify_init1() is a stub function on Alpha.  A heated
discussion
> *somewhere* produced a udev patch that *may* work:
> 
> --- a/udev/udev-watch.c
> +++ b/udev/udev-watch.c
> @@ -38,8 +38,10 @@ static int inotify_fd = -1;
>   */
>  int udev_watch_init(struct udev *udev)
>  {
> -       inotify_fd = inotify_init1(IN_CLOEXEC);
> -       if (inotify_fd < 0)
> +       inotify_fd = inotify_init();
> +       if (inotify_fd >= 0)
> +               util_set_fd_cloexec(inotify_fd);
> +       else
>                 err(udev, "inotify_init failed: %m\n");
>         return inotify_fd;
>  }

Uwe

P.S.: It was hard work to get my root file system converted, as the CD-ROM
drive is broken and I had no success setting up netboot. Finally this is
what I did to get rid of XFS:
- In aboot type in the kernel boot line by hand, but remove the root=/....
parameter. Tis makes initrd fail on mounting the root file system (this is
what I wanted to achieve). It falls to a shell, yipee
- Unfortunately the shell has no tools at all, so I created /mnt and /mnt2,
mounted my root filesystem to /mnt using busybox, also a USB2.0 harddisk as
backup to second /mnt2 point
- tarred the whole root file system to USB's /mnt2 (using busybox tar)
- copied /mnt/lib and /mnt/sbin to / (just to have it available after
unmounting everything again)
- unmount mounted root
- produce symlink in rootfs's /etc/mtab to /proc/mounts (to make ext2-tools
happy)
- recreate filesystem as EXT4 using the copied tools
- mount disk again to /mnt
- untar from USB HDD to /mnt
- flush and unmounts everything from busybox shell
- reboot and it was working with ext4 :-)
(so its possible without a boot disk, only with the default initramfs to fix
and create your root filesystem - phantastic)

Re: recent udev upgrade failure on alpha

[Bob Tracy]

On Fri, Jul 15, 2011 at 04:04:00PM +0200, Uwe Schindler wrote:
> (...)
> I tried then to upgrade the package "udev" (was on hold, to prevent me from
> making system unbootable), with the hope that the inotify is working with
> this debian kernel. Unfortunately that's not the case, UDEV complains about
> missing inotify.
> 
> So is the patch below already included in some kernel images available? Or
> does somebody have a recent udev version as .deb packages?

I still have (and am using) my modified "udev" package
(udev_166-1_alpha.deb: includes the patch you quoted in your original
message).  I can send it your way if you wish.  In truth, the only thing
you need is "/sbin/udevd" from that package to be operational, and I can
send you just that piece if you prefer.  Just let me know.

The underlying problem isn't anything missing in the kernel (at least
when built from the mainline kernel.org source): inotify_init1() is
definitely there.  The problem is the userspace call to inotify_init1()
in the current Alpha version of libc6.1 is a stub function.  I elected
to rebuild the "udev" package with a valid workaround rather than risk
modifying and having to debug "libc" :-).

-- 
------------------------------------------------------------------------
Bob Tracy          |  "Every normal man must be tempted at times to spit
rct@frus.com       |   upon his hands, hoist the black flag, and begin
                   |   slitting throats."	-- H.L. Mencken
------------------------------------------------------------------------

RE: recent udev upgrade failure on alpha

[Uwe Schindler]

Hi Bob,

> > (...)
> > I tried then to upgrade the package "udev" (was on hold, to prevent me
> > from making system unbootable), with the hope that the inotify is
> > working with this debian kernel. Unfortunately that's not the case,
> > UDEV complains about missing inotify.
> >
> > So is the patch below already included in some kernel images
> > available? Or does somebody have a recent udev version as .deb
> packages?
> 
> I still have (and am using) my modified "udev" package
> (udev_166-1_alpha.deb: includes the patch you quoted in your original
> message).  I can send it your way if you wish.  In truth, the only thing
you
> need is "/sbin/udevd" from that package to be operational, and I can send
> you just that piece if you prefer.  Just let me know.

A patch for udev would be the best (do you have it available as a patch),
then I can build a custom udev whenever an update occurs. Do you have it
somewhere on the web/dropbox whatever? But /sbin/udev is also a good choice
for now. Maybe send both as attachment to my private mail. THANKS!

From http://bugs.debian.org/cgi-bin/bugreport.cgi?bug=614099 it seems that
this is fixed in glibc 2.13-1, unfortunately on debian-ports is only
2.11.2-13. So we have to wait until buildds are working :( Maybe I try to
compile the libc package (hours of waiting included...)

> The underlying problem isn't anything missing in the kernel (at least when
> built from the mainline kernel.org source): inotify_init1() is definitely
there.
> The problem is the userspace call to inotify_init1() in the current Alpha
> version of libc6.1 is a stub function.  I elected to rebuild the "udev"
package
> with a valid workaround rather than risk modifying and having to debug
"libc"
> :-).

I was confused about the whole discussion, thanks for making it clear.

Uwe

RE: recent udev upgrade failure on alpha

[Uwe Schindler]

Today, my alpha installed Debian libc6.1-2.13-11, I will try to revert udev
to the original and remove the package hold here.

I will tell if it works!

Uwe

-----
Uwe Schindler
H.-H.-Meier-Allee 63, D-28213 Bremen
http://www.thetaphi.de
eMail: uwe@thetaphi.de


> -----Original Message-----
> From: Uwe Schindler [mailto:uwe@thetaphi.de]
> Sent: Friday, July 15, 2011 5:47 PM
> To: 'Bob Tracy'
> Cc: debian-alpha@lists.debian.org; linux-alpha@vger.kernel.org;
> mattst88@gmail.com; mcree@orcon.net.nz
> Subject: RE: recent udev upgrade failure on alpha
> 
> Hi Bob,
> 
> > > (...)
> > > I tried then to upgrade the package "udev" (was on hold, to prevent
> > > me from making system unbootable), with the hope that the inotify is
> > > working with this debian kernel. Unfortunately that's not the case,
> > > UDEV complains about missing inotify.
> > >
> > > So is the patch below already included in some kernel images
> > > available? Or does somebody have a recent udev version as .deb
> > packages?
> >
> > I still have (and am using) my modified "udev" package
> > (udev_166-1_alpha.deb: includes the patch you quoted in your original
> > message).  I can send it your way if you wish.  In truth, the only
> > thing
> you
> > need is "/sbin/udevd" from that package to be operational, and I can
> > send you just that piece if you prefer.  Just let me know.
> 
> A patch for udev would be the best (do you have it available as a patch),
then I
> can build a custom udev whenever an update occurs. Do you have it
> somewhere on the web/dropbox whatever? But /sbin/udev is also a good
> choice for now. Maybe send both as attachment to my private mail. THANKS!
> 
> >From http://bugs.debian.org/cgi-bin/bugreport.cgi?bug=614099 it seems
> >that
> this is fixed in glibc 2.13-1, unfortunately on debian-ports is only
2.11.2-13. So
> we have to wait until buildds are working :( Maybe I try to compile the
libc
> package (hours of waiting included...)
> 
> > The underlying problem isn't anything missing in the kernel (at least
> > when built from the mainline kernel.org source): inotify_init1() is
> > definitely
> there.
> > The problem is the userspace call to inotify_init1() in the current
> > Alpha version of libc6.1 is a stub function.  I elected to rebuild the
"udev"
> package
> > with a valid workaround rather than risk modifying and having to debug
> "libc"
> > :-).
> 
> I was confused about the whole discussion, thanks for making it clear.
> 
> Uwe
> 
> 
> --
> To UNSUBSCRIBE, email to debian-alpha-REQUEST@lists.debian.org
> with a subject of "unsubscribe". Trouble? Contact
listmaster@lists.debian.org
> Archive:
> http://lists.debian.org/00cd01cc4306$6269fe00$273dfa00$@thetaphi.de

RE: recent udev upgrade failure on alpha

[Uwe Schindler]

Success! With the latest libc update, udev works as it should. I reverted
back to the original udevd binary from the deb package by "apt-get install
--reinstall udev" and it started fine. Also regened initrd works with
l-i-2.6.38-2-alpha-generic after rebooting.

Uwe

-----
Uwe Schindler
H.-H.-Meier-Allee 63, D-28213 Bremen
http://www.thetaphi.de
eMail: uwe@thetaphi.de


> -----Original Message-----
> From: Uwe Schindler [mailto:uwe@thetaphi.de]
> Sent: Monday, August 15, 2011 7:09 PM
> To: 'Bob Tracy'
> Cc: debian-alpha@lists.debian.org; linux-alpha@vger.kernel.org;
> mattst88@gmail.com; mcree@orcon.net.nz
> Subject: RE: recent udev upgrade failure on alpha
> 
> Today, my alpha installed Debian libc6.1-2.13-11, I will try to revert
udev to the
> original and remove the package hold here.
> 
> I will tell if it works!
> 
> Uwe
> 
> -----
> Uwe Schindler
> H.-H.-Meier-Allee 63, D-28213 Bremen
> http://www.thetaphi.de
> eMail: uwe@thetaphi.de
> 
> 
> > -----Original Message-----
> > From: Uwe Schindler [mailto:uwe@thetaphi.de]
> > Sent: Friday, July 15, 2011 5:47 PM
> > To: 'Bob Tracy'
> > Cc: debian-alpha@lists.debian.org; linux-alpha@vger.kernel.org;
> > mattst88@gmail.com; mcree@orcon.net.nz
> > Subject: RE: recent udev upgrade failure on alpha
> >
> > Hi Bob,
> >
> > > > (...)
> > > > I tried then to upgrade the package "udev" (was on hold, to
> > > > prevent me from making system unbootable), with the hope that the
> > > > inotify is working with this debian kernel. Unfortunately that's
> > > > not the case, UDEV complains about missing inotify.
> > > >
> > > > So is the patch below already included in some kernel images
> > > > available? Or does somebody have a recent udev version as .deb
> > > packages?
> > >
> > > I still have (and am using) my modified "udev" package
> > > (udev_166-1_alpha.deb: includes the patch you quoted in your
> > > original message).  I can send it your way if you wish.  In truth,
> > > the only thing
> > you
> > > need is "/sbin/udevd" from that package to be operational, and I can
> > > send you just that piece if you prefer.  Just let me know.
> >
> > A patch for udev would be the best (do you have it available as a
> > patch),
> then I
> > can build a custom udev whenever an update occurs. Do you have it
> > somewhere on the web/dropbox whatever? But /sbin/udev is also a good
> > choice for now. Maybe send both as attachment to my private mail.
THANKS!
> >
> > >From http://bugs.debian.org/cgi-bin/bugreport.cgi?bug=614099 it seems
> > >that
> > this is fixed in glibc 2.13-1, unfortunately on debian-ports is only
> 2.11.2-13. So
> > we have to wait until buildds are working :( Maybe I try to compile
> > the
> libc
> > package (hours of waiting included...)
> >
> > > The underlying problem isn't anything missing in the kernel (at
> > > least when built from the mainline kernel.org source):
> > > inotify_init1() is definitely
> > there.
> > > The problem is the userspace call to inotify_init1() in the current
> > > Alpha version of libc6.1 is a stub function.  I elected to rebuild
> > > the
> "udev"
> > package
> > > with a valid workaround rather than risk modifying and having to
> > > debug
> > "libc"
> > > :-).
> >
> > I was confused about the whole discussion, thanks for making it clear.
> >
> > Uwe
> >
> >
> > --
> > To UNSUBSCRIBE, email to debian-alpha-REQUEST@lists.debian.org
> > with a subject of "unsubscribe". Trouble? Contact
> listmaster@lists.debian.org
> > Archive:
> > http://lists.debian.org/00cd01cc4306$6269fe00$273dfa00$@thetaphi.de
> 
> 
> --
> To UNSUBSCRIBE, email to debian-alpha-REQUEST@lists.debian.org
> with a subject of "unsubscribe". Trouble? Contact
listmaster@lists.debian.org
> Archive:
> http://lists.debian.org/007701cc5b6e$092c7360$1b855a20$@thetaphi.de

Re: recent udev upgrade failure on alpha

[Michael Cree]

On 24/02/11 20:02, Bob Tracy wrote:
> Well...  Yes, "sid" is cantankerous and known to occasionally break
> working systems, but this is the first time the thermonuclear software
> upgrade option has been used effectively on my poor Alpha :-(.
>
> Specifically, I've got about 50 installed but unconfigured packages and
> an almost unbootable (single-user mode only) machine because udev 166-1
> absolutely requires inotify support,

Oops, sorry, I should've sent a message to the list to warn people of this!

As you have identified the problem is that the syscall inotify_init1() 
is not implemented in the Debian supplied kernel.  It did not make it 
into the kernel until version 2.6.33. I did submit a bug report and a 
patch to update the kernel to the Debian bugzilla but I was too late and 
the Squeeze freeze beat me to it and it didn't get included.

So unless you are prepared to:

1) Download a new kernel and compile and install it yourself,

2) run "make headers_install" in the kernel and copy the file 
usr/include/asm/unistd.h header from the kernel sources to the system 
/usr/include/asm/unistd.h,

3) and rebuild libc6.1 from source (which is a time consuming build) and 
install it

don't upgrade to Sid!!!  I think a new kernel may have appeared in Sid 
in the last day or two so the first two steps above may no longer be 
needed, but the third step is needed until a new libc6.1 package comes 
through.

And if you have upgraded to Sid in the last week (or so) and hit the 
unconfigured udev problem don't switch off or reboot your machine until 
you have done all the above steps else you may find your Alpha unbootable.

And the other problem I have hit is that the upgraded apt and apt-utils 
packages are broken and it leaves one unable to upgrade their system! So 
make sure you have copies of the deb files for apt and apt-utils for the 
latest version that made it into Squeeze on hand.  I think I have placed 
a hold on apt and apt-utils at version 0.8.10.3.   I hadn't made a bug 
report at the time I encountered this problem as there were other bug 
reports already made that looked a little similar.  But the apt that I 
got a couple of days ago was still borked.

Cheers
Michael.

Re: recent udev upgrade failure on alpha

[Bob Tracy]

On Thu, Feb 24, 2011 at 10:54:27PM +1300, Michael Cree wrote:
> Oops, sorry, I should've sent a message to the list to warn people of this!

Not your job to protect us from ourselves :-).  Besides...  What better
way to see who's still actively (in some sense) involved in the Alpha
community, eh?

> As you have identified the problem is that the syscall inotify_init1() 
> is not implemented in the Debian supplied kernel.  It did not make it 
> into the kernel until version 2.6.33. I did submit a bug report and a 
> patch to update the kernel to the Debian bugzilla but I was too late and 
> the Squeeze freeze beat me to it and it didn't get included.

Actually, I've been running "sid" for a long time...  The risk of a
broken system has always been there, but needless to say I've been
lucky.  Besides, that's the only way to get current packages on the
Alpha from Debian these days.  That's a roundabout way of saying I'm
running late enough kernels (both Debian-supplied and built from
kernel.org source) that inotify_init1() was definitely present in the
running kernel when I did the upgrade.  As you point out later in your
reply, libc6.1 hasn't caught up on Alpha: I noticed the stubbed-out
function.

As I type this, I'm back up and running with a patched udev_166-1.  It
was easier and faster to patch/build "udev" than do a libc6.1 build.
When the smoke cleared, I simply copied in the new version of
"/sbin/udevd" and was finally able to configure the installed package
(as well as the other 49-or-so packages).  That's more excitement than I
wanted this evening: local time is now 0424 and I've been up since 0700
yesterday :-(.

-- 
------------------------------------------------------------------------
Bob Tracy          |  "Every normal man must be tempted at times to spit
rct@frus.com       |   upon his hands, hoist the black flag, and begin
                   |   slitting throats."	-- H.L. Mencken
------------------------------------------------------------------------