linuxsh-dev

PCMCIA broken on 2.6.21-rc

[Manuel L. <ma...@ro...>]

Hi Paul,

PCMCIA is (again) broken, this time on 2.6.21-rc1.
After doing a p3_ioremap() of AREA5/6, accesses to it
will hang (a simple CIS dump for example simply hangs
on the first readb()). Kernel is not dead (sysrq-t from usb kbd works).

X breaks too, btw. It just sits there and consumes 80% CPU.

Reverting "lazy dcache writeback optimizations" fixes both issues on
my SH7760.

Thanks,

-- 
 ml.

Re: PCMCIA broken on 2.6.21-rc

[Paul M. <le...@li...>]

Hi Manuel,

On Wed, Feb 21, 2007 at 01:38:56PM +0100, Manuel Lauss wrote:
> Reverting "lazy dcache writeback optimizations" fixes both issues on
> my SH7760.
> 
Can you try to run Saito-san's test case and see if it ends up breaking
for you as well? And if so, whether it works for you after reverting the
changes?

Re: PCMCIA broken on 2.6.21-rc

[Manuel L. <ma...@ro...>]

Hi Paul,

On Mon, Feb 26, 2007 at 12:24:15PM +0900, Paul Mundt wrote:
> On Wed, Feb 21, 2007 at 01:38:56PM +0100, Manuel Lauss wrote:
> > Reverting "lazy dcache writeback optimizations" fixes both issues on
> > my SH7760.
> > 
> Can you try to run Saito-san's test case and see if it ends up breaking
> for you as well? And if so, whether it works for you after reverting the
> changes?

With and without the patch reverted, the program does not report any
failures (I let it run 15 minutes both times).

Thanks!

-- 
 ml.

Re: PCMCIA broken on 2.6.21-rc

[Paul M. <le...@li...>]

On Mon, Feb 26, 2007 at 10:39:11AM +0100, Manuel Lauss wrote:
> On Mon, Feb 26, 2007 at 12:24:15PM +0900, Paul Mundt wrote:
> > On Wed, Feb 21, 2007 at 01:38:56PM +0100, Manuel Lauss wrote:
> > > Reverting "lazy dcache writeback optimizations" fixes both issues on
> > > my SH7760.
> > > 
> > Can you try to run Saito-san's test case and see if it ends up breaking
> > for you as well? And if so, whether it works for you after reverting the
> > changes?
> 
> With and without the patch reverted, the program does not report any
> failures (I let it run 15 minutes both times).
> 
Figures. I suppose this too will need to be reverted then until it's had
more testing on the older parts.. I'll queue up a reversion patch and
push it out with some other updates, so at least -rc2 should work for you.

Re: PCMCIA broken on 2.6.21-rc

[Paul M. <le...@li...>]

On Wed, Feb 21, 2007 at 01:38:56PM +0100, Manuel Lauss wrote:
> PCMCIA is (again) broken, this time on 2.6.21-rc1.
> After doing a p3_ioremap() of AREA5/6, accesses to it
> will hang (a simple CIS dump for example simply hangs
> on the first readb()). Kernel is not dead (sysrq-t from usb kbd works).
> 
> X breaks too, btw. It just sits there and consumes 80% CPU.
> 
> Reverting "lazy dcache writeback optimizations" fixes both issues on
> my SH7760.
> 
Can you verify whether this works?

diff --git a/arch/sh/mm/tlb-flush.c b/arch/sh/mm/tlb-flush.c
index d2f7b4a..927414f 100644
--- a/arch/sh/mm/tlb-flush.c
+++ b/arch/sh/mm/tlb-flush.c
@@ -160,7 +160,7 @@ void update_mmu_cache(struct vm_area_struct *vma,
 		unsigned long phys = pte_val(pte) & PTE_PHYS_MASK;
 		int dirty = test_and_clear_bit(PG_dcache_dirty, &page->flags);
 
-		if (dirty)
+		if (dirty || mapping_mapped(mapping))
 			__flush_wback_region((void *)P1SEGADDR(phys),
 					     PAGE_SIZE);
 	}

Re: PCMCIA broken on 2.6.21-rc

[Manuel L. <ma...@ro...>]

Hello Paul,

> > PCMCIA is (again) broken, this time on 2.6.21-rc1.
> > After doing a p3_ioremap() of AREA5/6, accesses to it
> > will hang (a simple CIS dump for example simply hangs
> > on the first readb()). Kernel is not dead (sysrq-t from usb kbd works).
> > 
> > X breaks too, btw. It just sits there and consumes 80% CPU.
> > 
> > Reverting "lazy dcache writeback optimizations" fixes both issues on
> > my SH7760.
> > 
> Can you verify whether this works?
> 
> diff --git a/arch/sh/mm/tlb-flush.c b/arch/sh/mm/tlb-flush.c
> index d2f7b4a..927414f 100644
> --- a/arch/sh/mm/tlb-flush.c
> +++ b/arch/sh/mm/tlb-flush.c
> @@ -160,7 +160,7 @@ void update_mmu_cache(struct vm_area_struct *vma,
>  		unsigned long phys = pte_val(pte) & PTE_PHYS_MASK;
>  		int dirty = test_and_clear_bit(PG_dcache_dirty, &page->flags);
>  
> -		if (dirty)
> +		if (dirty || mapping_mapped(mapping))
>  			__flush_wback_region((void *)P1SEGADDR(phys),
>  					     PAGE_SIZE);
>  	}

Unfortunately not, same symptoms as described above.

-- 
 ml.