Skip to contentSkip to conversation

sparc32: fix iommu_flush_iotlb end address

PATCH Applied Bob Breuer 4 participants 6 messages Jun 20, 2006 – Jun 26, 2006 SPARC + UltraSPARC (sparc/sparc64)

lore
Share & tools
lore

Bob Breuer

author
#0 lore raw

PATCH 1/1 sparc32: fix iommu_flush_iotlb end address

Fix the calculation of the end address when flushing iotlb entries to ram. This bug has been a cause of esp dma errors, and it affects HyperSPARC systems much worse than SuperSPARC systems.

---

Just in case it's not obvious from the patch as to how the dma was
broken: Each dma mapping sets up iopte entries for the iommu, and the iommu only looks in main memory for the iopte entries. If there is a group of iopte entries that are smaller than a page in size, but straddle a page boundary, the broken code would fail to flush the last page to ram.

Bob

arch/sparc/mm/iommu.c +2 -1
index 77840c8..7215849 100644
@@ -144,8 +144,9 @@ static void iommu_flush_iotlb(iopte_t *i
 	unsigned long start;
 	unsigned long end;
 
-	start = (unsigned long)iopte & PAGE_MASK;
+	start = (unsigned long)iopte;
 	end = PAGE_ALIGN(start + niopte*sizeof(iopte_t));
+	start &= PAGE_MASK;
 	if (viking_mxcc_present) {
 		while(start < end) {
 			viking_mxcc_flush_page(start);
maintainer Applied
#1 lore raw

From: Bob Breuer <[email protected]>
Date: Mon, 19 Jun 2006 23:08:59 -0500

Quoted 5 lines

Fix the calculation of the end address when flushing iotlb entries to ram. This bug has been a cause of esp dma errors, and it affects HyperSPARC systems much worse than SuperSPARC systems.

Good catch, that bug would explain a lot.

Applied, and I'll toss this to -stable too.

#2 lore raw
In reply to Bob Breuer

Hi Bob,

Quoted 3 lines · Bob Breuer

Fix the calculation of the end address when flushing iotlb entries to ram. This bug has been a cause of esp dma errors, and it affects HyperSPARC systems much worse than SuperSPARC systems.

Thanks for this patch! I was going cross-eyed trying to figure this out. I have updated kernel bug 6344 with this info. Can no longer reproduce any esp dma errors.

Thanks,

Signature · 1 line
Martin
#3 lore raw
In reply to Bob Breuer
Quoted 4 lines · Martin Habets

Hi Bob,

On Mon, Jun 19, 2006 at 11:08:59PM -0500, Bob Breuer wrote:

Quoted 8 lines
Quoted 3 lines

Fix the calculation of the end address when flushing iotlb entries to ram. This bug has been a cause of esp dma errors, and it affects HyperSPARC systems much worse than SuperSPARC systems.

Thanks for this patch! I was going cross-eyed trying to figure this out. I have updated kernel bug 6344 with this info. Can no longer reproduce any esp dma errors.

Thanks for this patch. I shall test it in smp configuration. Do you
know the release of the official kernel that supports sparc32 smp mode ? I have tested the 2.6.17.1 kernel without success (ot doesn't build due to a undefined symbol...).

Regards,

JKB

author
#4 lore raw
In reply to Bob Breuer
Quoted 4 lines · BERTRAND Joël

Thanks for this patch. I shall test it in smp configuration. Do you
know the release of the official kernel that supports sparc32 smp mode ? I have tested the 2.6.17.1 kernel without success (ot doesn't build due to a undefined symbol...).

For sparc32 smp without extra patches, 2.6.17.X won't work. You will need at least 2.6.17-git2 or newer.

Bob

#5 lore raw
In reply to Bob Breuer

Hello,

I have build a 2.6.17-git7. It can boot with init=/bin/bash but if I
try to boot with init=2 (debian system), kernel craches when it tries to mount local filesystem (watchdog reset or dead lock).

	Configuration : 448 MB, 2*SM71, CG14, raid1.

	I will try a -git10...

	Regards,

	JKB