Bug 496586

The delay between two attempts to access the floppy disk drive is exactly 2 secs, so there might happen some interference between a system timer and the floppy device.

Disk access cycle starts upon logging into GNOME. However, it does not stop upon terminating the GNOME session but it does so after switching to run level 1.

Does running the command 'hal-disable-polling' make it stop?

No, it doesn't, and the command 'hal-disable-polling --device /dev/fd0' returns:

  "Polling is already disabled on the given drive."

While the initial bug report report had been posted for the current F12 development tree, the system has been reverted to F11 in the meantime and kernel 2.6.30-0.58.rc2.git1.fc12.x86_64 installed alongside with kernel-2.6.29.1-104.fc11.x86_64.
In addition to the floppy disk issue, it is now not even possible to open a GNOME terminal. A corresponding error message appears in .xsession-errors reporting a failure when trying to connect to /tmp/ibus-$USER/ibus-unix-0 (No such file or directory).

- Issue persists for kernel 2.6.30-0.66.rc3.fc12.x86_64.
- Adding 'noapic' to the kernel boot options does not improve things either.
- File /proc/interrupts exhibits a related entry:

     "6:         39   IO-APIC-edge      floppy"

  after a short uptime which is absent for kernel 2.6.29.1-103.fc11.x86_64.
  So, somehow spurious interrupts seem to be triggered.

Same issue for kernel 2.6.30-0.72.rc3.git3.fc12.i686.PAE on a different Intel P4 system. The physical presence of a floppy disk drice is not required to trigger the /dev/fd0 error message. Again, a new GNOME terminal window will disappear instantaneously. The "rawhide" kernel 2.6.29.1-111.fc11.i686.PAE behaves correctly in these respects.

No improvement for kernel-2.6.30-0.81.rc5.git1.fc12.x86_64.

No problem after installing the latest Ubuntu 9.10 snapshot for AMD64. Here, 'uname -r' returns 2.6.30-5-generic.

This bug appears to have been reported against 'rawhide' during the Fedora 11 development cycle.
Changing version to '11'.

More information and reason for this action is here:
http://fedoraproject.org/wiki/BugZappers/HouseKeeping

At least with 2.6.30-0.97.rc8.fc12.x86_64 from the current rawhide the picture is more complicated than the above comments would suggest.

1. Booting to level 3, with or without media in a floppy drive does NOT light up a floppy activity LED.  /dev/fd0 is there and floppy is mountable from a command line without any issues.

2. That LED lights up only during a startup of a desktop session.  Once there it  stays on and drive without media is making pretty unhealthy clicking noises. Logging out and taking a machine to level 3 does not extinguish this light but going all the way to level 1 eventually it does. One may see a stream of repeating messages "end_request: I/O error, dev fd0, sector 0", possibly interspersed with "VFS: busy inodes on changed media or resized disk fd0",
and one that starts it does not stop but this does not always show up and I cannot tell what triggers it.

3. While in a desktop session attempts to open a "Computer" window fail and result in the following in logs:

gvfsd-computer[2897]: segfault at 0 ip 00007fb5874a3e0a sp 00007fffe0de5b48 error 4 in libc-2.10.1.so[7fb587425000+164000]
gvfsd-computer[2899]: segfault at 0 ip 00007f73368bee0a sp 00007fff58117288 error 4 in libc-2.10.1.so[7f7336840000+164000]
gvfsd-computer[2902]: segfault at 0 ip 00007f47dfda1e0a sp 00007fff2a65f1d8 error 4 in libc-2.10.1.so[7f47dfd23000+164000]
gvfsd-computer[2904]: segfault at 0 ip 00007fbc00018e0a sp 00007fffa83e8cd8 error 4 in libc-2.10.1.so[7fbbfff9a000+164000]

4. Even with a floppy drive "busy" a floppy remains mountable and running
'gnome-mount' in terminal window results in the following:

$ gnome-mount -t -v -d /dev/fd0
gnome-mount 0.8
** (gnome-mount:2860): DEBUG: Will attempts methods on drive object
** (gnome-mount:2860): DEBUG: Mounting /org/freedesktop/Hal/devices/platform_floppy_0_storage_platform_floppy
GConf Error: Bad key or directory name: "/system/storage/default_options/(null)/fstype_override": `(' is an invalid character in key/directory names
** (gnome-mount:2860): DEBUG: Mounting /org/freedesktop/Hal/devices/platform_floppy_0_storage_platform_floppy with mount_point='', fstype='', num_options=1
** (gnome-mount:2860): DEBUG:   option='uid=400'
Mounted /dev/fd0 at "/media/disk"

That action drops a floppy icon on a desktop and a floppy is accesible and,
albeit really slow, browsable (with its LED on all the time).  OTOH attempts to use floppy menu to unmount that bring such alerts:

Unable to unmount disk
/sbin/umount.hal: /media/disk is not recognized by hal

OTOH 'gnome-umount' gives this:

$ gnome-umount -t -v -d /dev/fd0
gnome-mount 0.8
** (gnome-umount:2962): DEBUG: Will attempts methods on drive object
** (gnome-umount:2962): DEBUG: Unmounting /org/freedesktop/Hal/devices/platform_floppy_0_storage_platform_floppy
** (gnome-umount:2962): DEBUG: Setting up 750ms timer for Flushing Cache dialog
** (gnome-umount:2962): DEBUG: in unmount_done : user_data = 0x0
Unmounted /dev/fd0

and the icon disappears.

5. Rebooting to 2.6.29.4-167.fc11.x86_64 immediately cuts all this nonsense but this apparently because /dev/fd0 is not there and a floppy is simply non-accessible.  Radical but not that useful.

This bug appears to have been reported against 'rawhide' during the Fedora 11 development cycle.
Changing version to '11'.

More information and reason for this action is here:
http://fedoraproject.org/wiki/BugZappers/HouseKeeping

As noted in comment #11 this bug applies to the current _rawhide_.  It is possible, which I have no way to check in this moment, that with F11 a floppy would be not recognized at all.  That would replace there this bug with another one and arguably even worse.

I can confirm this bug on kernel 2.6.29.4-167.fc11.i586 on a system just upgraded from F10 to F11. The behaviour is identical. Something is trying to access fd0 every second second, filling the syslog with "end_request: I/O error, dev fd0, sector 0".

Running hal-disable-polling has no effect.

# hal-disable-polling --device /dev/fd0
Polling is already disabled on the given drive.

The problem starts after starting a gnome session.

Since I want to be CC'ed on this I'd might as well  give some further data:
- not seen on kernel-2.6.29.2-126.fc11.i586, kernel-2.6.29.3-155.fc11.i586 and kernel-2.6.29.4-167.fc11.i586;
- first seen on kernel-2.6.30-0.97.rc8.fc12.i586 (but do note that this package was installed with almost a gigabyte worth of other packages - somehow the release of F11 triggered a major upgrade to F12 yesterday, on a machine that is tracking Rawhide);
- on a ThinkPad T41 without any floppy whatsoever;
- the link to the gnome session, made in comment #14, does _look_ plausible (haven't investigated yet);
- trivial workaround is to blacklist the floppy module.

AFAICS its caused by the devkit-disks polling of /dev/fd0 which behaves differently with 29 and 30 kernels.
devkit-disks --inhibit-polling /dev/fd0 
as the current console user works for me to stop it.

devkit-disks goes like 
....
open("/dev/fd0", O_RDONLY)              = -1 ENXIO (No such device or address)
poll([{fd=3, events=POLLIN}], 1, 1974)  = 0 (Timeout)
...
each second

(In reply to comment #15)
> - the link to the gnome session, made in comment #14, does _look_ plausible
Trivial test: this doesn't occur in runlevel 3. (Only a few instances of this message are seen somewhere early in the boot sequence.)

(In reply to comment #16)
> each second
For what it's worth, I got these messages about every twelve seconds: slow machine?

I started seeing this after upgrading from f11 to f12 rawhide today
with kernel-2.6.30-0.97.fc8.fc12.i586.

The fdo polling errors start soon after logging into gnome.

Booting into f11 kernel and everything is fine again.

Actually the difference between 29 and 30 is that with 29 the floppy module doesn't get autoloaded and with 30 it does.
http://git.kernel.org/?p=linux/kernel/git/torvalds/linux-2.6.git;a=commit;h=83f9ef463bcb4ba7b4fee1d6212fac7d277010d3

loading the floopy module under 29 leads to devkit-disks doing the same polling loop. 
This should probably be moved to DeviceKit-disks

*** Bug 505417 has been marked as a duplicate of this bug. ***

Issue now also appears on F11 after upgrading to kernel 2.6.29.6-209.rc1.fc11. Unloading module "floppy" allows to stop accesses to /dev/fd0 successfully.

Actual attempts to fix this one are tracked in bug 489083  so perhaps this one should be marked a dup.

*** This bug has been marked as a duplicate of bug 489083 ***