Thread (64 messages) 64 messages, 5 authors, 2010-05-28
STALE5877d

[PATCH 2/8] ARM: Implement read/write for ownership in theARMv6 DMA cache ops

From: George G. Davis <hidden>
Date: 2010-05-15 01:26:10

Hi,

On Fri, May 14, 2010 at 05:42:31PM +0100, Catalin Marinas wrote:
On Fri, 2010-05-14 at 17:29 +0100, Catalin Marinas wrote:
quoted
What I think I missed in my patch (and didn't show up with a sata drive)
is the "sync" operations. The dma_sync_single_for_device(FROMDEVICE)
corrupts the existing buffer.

I'll post an updated patch shortly.
Actually I think the hunk that I posted earlier should be enough:
Um, apologies for the noise but I didn't understand if Ronen said
this resolved the e1000 problem or not.

Thanks!

--
Regards,
George
quoted hunk ↗ jump to hunk
diff --git a/arch/arm/mm/cache-v6.S b/arch/arm/mm/cache-v6.S
index a4a6840..aa3ac32 100644
--- a/arch/arm/mm/cache-v6.S
+++ b/arch/arm/mm/cache-v6.S
@@ -216,7 +216,8 @@ v6_dma_inv_range:
 #endif
 1:
 #ifdef CONFIG_SMP
-	str	r0, [r0]			@ write for ownership
+	ldr	r2, [r0]			@ read for ownership
+	str	r2, [r0]			@ write for ownership
 #endif
 #ifdef HARVARD_CACHE
 	mcr	p15, 0, r0, c7, c6, 1		@ invalidate D line

The dma_sync_single_for_cpu() eventually calls dmac_unmap_area() which
on ARM11MPCore became a no-op. There is no problem since this function
call is always preceded by a call to either dma_map_single() or
dma_sync_single_to_device(), both of them calling dma_map_area() which
does the cache invalidation.

With the patch above, v6_dma_inv_range() is no longer destructive so the
transferred data to the SDRAM is preserved. We could even re-instate the
v6_dma_unmap_area() function but for performance reasons I would leave
it as a no-op.

Russell, are you ok with such change? Since you already applied the
original patch as #6111, I will send an additional patch with this fix
(I'm sending it directly to the patch system as I won't have access to
my work PC this weekend. Feel free to apply or reject).

Thanks.

-- 
Catalin
Keyboard shortcuts
hback out one level
jnext message in thread
kprevious message in thread
ldrill in
Escclose help / fold thread tree
?toggle this help