Re: broken dpt_i2o in 2.6.23 (was: ext2 check page: bad entry in directory) (fwd)

[James Bottomley <James.Bottomley@HansenPartnership.com>]

On Wed, 2007-12-12 at 11:16 -0800, Andrew Morton wrote:
> On Wed, 12 Dec 2007 14:43:42 +0100 Anders Henke <anders.henke@1und1.de> wrote:
> 
> > Am 12.12.2007 schrieb Miquel van Smoorenburg:
> > > On Wed, 2007-12-12 at 03:38 -0800, Andrew Morton wrote:
> > > > On Wed, 12 Dec 2007 11:58:41 +0100 Anders Henke <anders.henke@1und1.de> wrote:
> > > > 
> > > > > Hi,
> > > > > 
> > > > > I'd like to let you now that my boxes are running a 32-bit kernel, so
> > > > > the 64-bit-uncleanliness shouldn't apply to my boxes; however,
> > > > > 
> > > > > http://www.miquels.cistron.nl/linux/dpt_i2o-64bit-2.6.23.patch
> > > > > 
> > > > > fixed the issue on my testbox.
> > > > > 
> > > > > I took a clean 2.6.23, applied patch, recompiled the kernel, reboot: works.
> > > > 
> > > > What a huge patch :(
> > > > 
> > > > We already reverted the offening patch so I assume that 2.6.24-rc5 is
> > > > working for you?
> > > > 
> > > > I guess we need to look at restoring "dpt_i2o: convert to SCSI hotplug
> > > > model" and then absorbing what Miquel has done there.
> > > 
> > > This was just a patch I had lying around, if it worked it would confirm
> > > my suspicion, which it has.
> > > 
> > > The minimal patch which is suitable for 2.6.23-stable and 2.6.24 would
> > > be the attached one-liner. The "dpt_i2o: convert to SCSI hotplug model"
> > > patch could be restored then.
> > > 
> > > (if the list eats the attachment, it's also available here:
> > > http://www.miquels.cistron.nl/linux/linux-2.6.23+24-dpt_i2o-dma64.patch 
> > > )
> > > 
> > > Anders, does this one-liner patch work for you ?
> > 
> > Got it - and it works!
> > 
> > I took a clean 2.6.23, applied the patch, recompiled the kernel and
> > rebooted my testbox: came up with the fresh-compiled kernel 
> > (verified by "uname -a").
> > 
> 
> That looks appropriate for 2.6.23.x:
> 
> --- linux-2.6.23.9.orig/drivers/scsi/dpt_i2o.c	2007-11-26 18:51:43.000000000 +0100
> +++ linux-2.6.23.9/drivers/scsi/dpt_i2o.c	2007-12-12 13:21:05.000000000 +0100
> @@ -905,8 +905,7 @@
>  	}
>  
>  	pci_set_master(pDev);
> -	if (pci_set_dma_mask(pDev, DMA_64BIT_MASK) &&
> -	    pci_set_dma_mask(pDev, DMA_32BIT_MASK))
> +	if (pci_set_dma_mask(pDev, DMA_32BIT_MASK))
>  		return -EINVAL;

Yes, this has to be in ... the mptr filling the scatterlist on the
current driver is only a u32 and so will silently truncate.

>  	base_addr0_phys = pci_resource_start(pDev,0);
> 
> 
> However it is a bit mystifying that
> 55d9fcf57ba5ec427544fca7abc335cf3da78160 would cause a dma mask problem
> (isn't it?)
> 
> The scsi people might want to restore
> 55d9fcf57ba5ec427544fca7abc335cf3da78160 and then apply Miquel's patch on
> top for 2.6.24, or do it for 2.6.25?

I think it's a bit late in the game for 2.6.24, so I'm happy to leave
the hotplug reverted.  We'll try adding back hotplug plus this for
2.6.25 I think.

James