Re: ide-cs stack_dump

[Jani Monoses <jani@iv.ro>]

On Tue, 5 Aug 2003 01:56:01 +0200 (MET DST)
Bartlomiej Zolnierkiewicz <B.Zolnierkiewicz@elka.pw.edu.pl> wrote:

> 
> On Mon, 4 Aug 2003, Jani Monoses wrote:
> 
> I think I know how this happens:
> 
> register_disk()->blkdev_get()->do_open(), then disk->fops->open()
> (idedisk_open() for ide-disk) calls check_disk_change().
> check_disk_change() calls disk->fops->media_changed().
> idedisk_media_changes() returns drive->removable, so instead of
> returning from check_disk_change() block_device is invalidated
> and bdev->bd_invalidated is set to 1.  Later in do_open(),
> bdev->bd_invalidated flag is checked and since it is 1
> rescan_partitions() is triggered.  Thus partitions are checked and
> added twice: in do_open()->rescan_partitions() and in register_disk().
> 
> [ The same applies to ide-floppy driver and probably some other
> drivers. ]
> 
> Ufff... I hope it is a correct description (I don't have hardware to
> reproduce the problem).

You're probably right in the description. One ugly way I solved this for
2.6.0-test1 was setting bd_invalidated to 0 
 
> Easy way is to fix is to add drive->attach flag, set it in
> idedisk_attach() and check+clear in idedisk_media_changed(),
> but I don't like this solution (patch below, Jani, can you test it?).

I can test is by tomorrow, it's another box that has the PCCard slot.

Jani