New testing kernel. 2.6.10-1.727_FC3

[David Zeuthen]

On Thu, 2005-01-06 at 12:24 -0500, Alan Cox wrote:
> On Thu, Jan 06, 2005 at 10:59:01AM -0500, David Zeuthen wrote:
> > > Boot single user, remove haldaemon, reboot. Let me know if that helps (if I've
> > > asked you specifically this already ignore me)
> > > 
> > Alan - any news of when ide-cd is fixed so these issues will be a thing 
> > of the past?
> 

Thanks for the update,

> General
> 	- the cd at boot crash is fixed
> 	- end of media is fixed
> 	- the bogus messages in the log should be fixed
> 	- vanishing CD with no serial number is fixed
> 	- request queue corruptor is fixed (as of yesterday)
> 
> Burning
> 	- request filtering is in all our 2.6.9 trees
> 	- there have been some fixes to that
> 	- We do not yet have tools for adjusting the filter list
> 		[Daniel at gentoo and others were playing with an fs for it]
> 	- the burn v haldaemon stuff has to be userspace I think
> 

I'm really surprised that hald can disrupt burning - when hald polls 
the drive it uses O_EXCL and so does cdrecord (if it doesn't it's a
packaging bug).

> Unsolved
> 	- corner case race that can hang on DMA timeout/speed change
> 	- the great dell v hal drive probing (probably HAL should blacklist)

As of version 0.4.2, HAL does blacklist such devices. It's being tracked
here

 https://bugzilla.redhat.com/bugzilla/show_bug.cgi?id=138148

This bug report discusses why this happens - see comment 14 and 15. It
appears
that one can poll if you keep a file descriptor open, e.g.

 fd = open ("/dev/cdrom", O_RDONLY)
 <poll on fd>
 sleep (2)
 <poll on fd>
 sleep (2)
 <poll on fd>
 ...

does not cause the slowdown on the hard disk on the same channel.
However
if you do

 fd = open ("/dev/cdrom", O_RDONLY)
 <poll on fd>
 close (fd)
 sleep (2)
 fd = open ("/dev/cdrom", O_RDONLY)
 <poll on fd>
 close (fd)
 sleep (2)
 ...

we indeed get the slowdown. Hence, the kernel does something when the
last
closer closes the device - dunno what to do here from userspace short of
just not poll.

> 	- some other drive v hal cases I don't understand even what is at
> 	  fault yet.
> 
> Pending
> 	- CD-ROM drive hotplug for thinkpad and other laptops
> 
> And probably some I only have on paper. Its been a pita because its been a lot
> of little bugs not one big one

There's also this one here

 https://bugzilla.redhat.com/bugzilla/show_bug.cgi?id=144036

which I completely fail to understand. Long term it looks like hald
should be
using MMC commands to detect media if available - any idea of where I
can find
documentation/code for this?

Thanks,
David