Re: 2.5.12-dj1: IDE trouble - RZ1000 still won't finish booting

Re: 2.5.12-dj1: IDE trouble - RZ1000 still won't finish booting

[Martin Dalecki]

Uz.ytkownik Mark Orr napisa?:
> I posted a log about this about a week ago regarding
> Linux 2.5.10 not completing boot on an older system with a
> RZ1000 IDE interface.   (and that RZ1000 by itself, w/ CMD640
> and generic PCI IDE disabled in make config, wouldnt even
> compile  -- it still doesnt with 2.5.12-dj1)
> 
> 
> It still crashes at the same place - when the root
> filesystem is remounted with R/W.   The messages look different,
> i.e. the "unexpected interrupt <x> <y>" is gone, but it looks
> like it's still complaining about the same thing:
> 
> hda: task_mulout_intr: error=0x04  { DriveStatusError }
> 
> then several of these:
> 
>    { task_mulout_intr }
> hda:  ide_set_handler: handler not null  old=<some hex> new=<some other hex>
> bug: kernel timer added twice
> 
> finally:
> 
> end_request: I/O error, dev 03:00, sector 456628
> end_buffer_io_sync: I/O error
> hda: ata_irq_request: hwgroup was not busy!?
> Unable to hand kernel NULL pointer dereference...
> 
> ...and the usual dumpage.    (which isnt logged, naturally)

There where no reaction becouse problems with the multi sector
IO code path in the driver are well known... and unfortunately still
not resolved becouse there are expected changes in the whole
handling of them. However to confirm this you could for the time
beeing please just try to displabe multi sector writes
on this driver by editing the following code in ide-disk.c

#ifdef CONFIG_IDEDISK_MULTI_MODE
		id->multsect = ((id->max_multsect/2) > 1) ? id->max_multsect : 0;
		id->multsect_valid = id->multsect ? 1 : 0;
		drive->mult_req = id->multsect_valid ? id->max_multsect : INITIAL_MULT_COUNT;
		if (drive->mult_req)
			drive->special_cmd |= ATA_SPECIAL_MMODE;
#else
		/* original, pre IDE-NFG, per request of AC */
		drive->mult_req = INITIAL_MULT_COUNT;
		if (drive->mult_req > id->max_multsect)
			drive->mult_req = id->max_multsect;
		if (drive->mult_req || ((id->multsect_valid & 1) && id->multsect))
			drive->special_cmd |= ATA_SPECIAL_MMODE;
#endif

to always set mult_req to 0.

Peerhaps you could of course check bfore hand first
if CONFIG_IDEDISK_MULTI_MODE is set and
INITIAL_MULT_COUNT allways equal to 0.

Thank you in advance.

2.5.12-dj1: IDE trouble - RZ1000 still won't finish booting

[Mark Orr]

I posted a log about this about a week ago regarding
Linux 2.5.10 not completing boot on an older system with a
RZ1000 IDE interface.   (and that RZ1000 by itself, w/ CMD640
and generic PCI IDE disabled in make config, wouldnt even
compile  -- it still doesnt with 2.5.12-dj1)


It still crashes at the same place - when the root
filesystem is remounted with R/W.   The messages look different,
i.e. the "unexpected interrupt <x> <y>" is gone, but it looks
like it's still complaining about the same thing:

hda: task_mulout_intr: error=0x04  { DriveStatusError }

then several of these:

   { task_mulout_intr }
hda:  ide_set_handler: handler not null  old=<some hex> new=<some other hex>
bug: kernel timer added twice

finally:

end_request: I/O error, dev 03:00, sector 456628
end_buffer_io_sync: I/O error
hda: ata_irq_request: hwgroup was not busy!?
Unable to hand kernel NULL pointer dereference...

...and the usual dumpage.    (which isnt logged, naturally)

--
Mark Orr
markorr@intersurf.com

Re: 2.5.12-dj1: IDE trouble - RZ1000 still won't finish booting

[Dave Jones]

On Thu, May 02, 2002 at 03:44:25PM -0500, Mark Orr wrote:

 > hda: task_mulout_intr: error=0x04  { DriveStatusError }
 > 
 > then several of these:
 > 
 >    { task_mulout_intr }
 > hda:  ide_set_handler: handler not null  old=<some hex> new=<some other hex>
 > bug: kernel timer added twice

Yup, I saw the same problem with my dual P5 a few nights ago, shortly
before it ate the filesystem.  It's a known problem, and several people
claim they have a handle on it, but no-one seems in a hurry to fix it.

-- 
| Dave Jones.        http://www.codemonkey.org.uk
| SuSE Labs