Missing critical phys_to_virt in lib/swiotlb.c
PATCH 1/1 Missing critical phys_to_virt in lib/swiotlb.c
From: David Moore <[email protected]>
Adds missing call to phys_to_virt() in the lib/swiotlb.c:swiotlb_sync_sg() function. Without this change, a kernel panic will always occur whenever a SWIOTLB bounce buffer from a scatter-gather list gets synced.
---
This change was originally part of a larger patch by Jan Beulich, which was more extensive and doesn't look destined to make it into 2.6.20:
http://kernel.org/pub/linux/kernel/people/akpm/patches/2.6/2.6.20-rc6/2.6.20-rc6-mm3/broken-out/ia64-swiotlb-bug-fixes.patch
However, considering the severity of this one-liner bug, I would like to request that this simplified patch make it into 2.6.20, despite how close we are to the final cut. It fixes real crashes:
http://lists.opensuse.org/opensuse-bugs/2006-12/msg02943.html
http://qa.mandriva.com/show_bug.cgi?id=28224
http://www.pchdtv.com/forum/viewtopic.php?t=2063&sid=a959a14a4c2db0eebaab7b0df56103ce
lib/swiotlb.c renamed from lib/swiotlb.c.orig +1 -1
@@ -750,7 +750,7 @@ swiotlb_sync_sg(struct device *hwdev, stfor (i =0 ; i < nelems; i++, sg++)if (sg->dma_address !=SG_ENT_PHYS_ADDRESS (sg)) -sync_single (hwdev, (void *) sg->dma_address, +sync_single (hwdev,phys_to_virt (sg->dma_address), sg->dma_length, dir, target); }
David Moore wrote:
[...]
Quoted 3 lines
considering the severity of this one-liner bug, I would like to request that this simplified patch make it into 2.6.20, despite how close we are to the final cut.
So we were too close. Maybe the -stable team likes to have it in 2.6.20.1.
Quoted 4 lines
It fixes real crashes:
http://lists.opensuse.org/opensuse-bugs/2006-12/msg02943.html
http://qa.mandriva.com/show_bug.cgi?id=28224
http://www.pchdtv.com/forum/viewtopic.php?t=2063&sid=a959a14a4c2db0eebaab7b0df56103ce
and FireWire crashes too.
Quoted 11 lines
lib/swiotlb.c renamed from lib/swiotlb.c.orig +1
-1@@ -750,7 +750,7 @@ swiotlb_sync_sg(struct device *hwdev, stfor (i =0 ; i < nelems; i++, sg++)if (sg->dma_address !=SG_ENT_PHYS_ADDRESS (sg)) -sync_single (hwdev, (void *) sg->dma_address, +sync_single (hwdev,phys_to_virt (sg->dma_address), sg->dma_length, dir, target); }
Signature · 3 lines
Stefan Richter -=====-=-=== --=- --=-- http://arcgraph.de/sr/
Andrew Morton
maintainerQuoted 24 lines · On Sun, 04 Feb 2007 13:39:40 -0500 David Moore <[email protected]> wrote:
From: David Moore <[email protected]>
Adds missing call to
phys_to_virt()in the lib/swiotlb.c:swiotlb_sync_sg() function. Without this change, a kernel panic will always occur whenever a SWIOTLB bounce buffer from a scatter-gather list gets synced.Signed-off-by David Moore <[email protected]>---
This change was originally part of a larger patch by Jan Beulich, which was more extensive and doesn't look destined to make it into 2.6.20:
http://kernel.org/pub/linux/kernel/people/akpm/patches/2.6/2.6.20-rc6/2.6.20-rc6-mm3/broken-out/ia64-swiotlb-bug-fixes.patchHowever, considering the severity of this one-liner bug, I would like to request that this simplified patch make it into 2.6.20, despite how close we are to the final cut. It fixes real crashes:
http://lists.opensuse.org/opensuse-bugs/2006-12/msg02943.html
http://qa.mandriva.com/show_bug.cgi?id=28224
http://www.pchdtv.com/forum/viewtopic.php?t=2063&sid=a959a14a4c2db0eebaab7b0df56103ce
lib/swiotlb.c swiotlb_sync_sg(struct device *hwdev, st · line 756
Quoted 8 lines
lib/swiotlb.c renamed from lib/swiotlb.c.orig +1
-1@@ -750,7 +750,7 @@ swiotlb_sync_sg(struct device *hwdev, stfor (i =0 ; i < nelems; i++, sg++)if (sg->dma_address !=SG_ENT_PHYS_ADDRESS (sg)) -sync_single (hwdev, (void *) sg->dma_address, +sync_single (hwdev,phys_to_virt (sg->dma_address), sg->dma_length, dir, target); }
argh. I didn't know that Jan's patches fixed crashes. I thought they were ia64-only things.
Who maintains the swiotlb code?
I shall queue this up, tag it for 2.6.20.1, then I'll fix up Jan's alleged ia64 patches to account for that. I'll still push it through Tony, but the x86_64/ia64 linkage here seems to be a source of problems.
Quoted 28 lines · Andrew Morton
Quoted 25 lines · On Sun, 04 Feb 2007 13:39:40 -0500 David Moore <[email protected]> wrote:
From: David Moore <[email protected]>
Adds missing call to
phys_to_virt()in the lib/swiotlb.c:swiotlb_sync_sg() function. Without this change, a kernel panic will always occur whenever a SWIOTLB bounce buffer from a scatter-gather list gets synced.Signed-off-by David Moore <[email protected]>---
This change was originally part of a larger patch by Jan Beulich, which was more extensive and doesn't look destined to make it into 2.6.20:
http://kernel.org/pub/linux/kernel/people/akpm/patches/2.6/2.6.20-rc6/2.6.20-rc6-mm3/broken-out/ia64-swiotlb-bug-fixes.patchHowever, considering the severity of this one-liner bug, I would like to request that this simplified patch make it into 2.6.20, despite how close we are to the final cut. It fixes real crashes:
http://lists.opensuse.org/opensuse-bugs/2006-12/msg02943.html
http://qa.mandriva.com/show_bug.cgi?id=28224
http://www.pchdtv.com/forum/viewtopic.php?t=2063&sid=a959a14a4c2db0eebaab7b0df56103celib/swiotlb.c renamed from lib/swiotlb.c.orig +0
-0@@ -750,7 +750,7 @@ swiotlb_sync_sg(struct device *hwdev, st
Quoted 10 lines
lib/swiotlb.c swiotlb_sync_sg(struct device *hwdev, st · line 756
Quoted 7 lines
lib/swiotlb.c renamed from lib/swiotlb.c.orig +1
-1for (i =0 ; i < nelems; i++, sg++)if (sg->dma_address !=SG_ENT_PHYS_ADDRESS (sg)) -sync_single (hwdev, (void *) sg->dma_address, +sync_single (hwdev,phys_to_virt (sg->dma_address), sg->dma_length, dir, target); }argh. I didn't know that Jan's patches fixed crashes. I thought they were ia64-only things.
Sounds weird. If this really didn't work much more should be broken (e.g. no cdroms/sound on Intel x86-64 boxes with >4GB)
I'm a little sceptical. Perhaps the TV driver is doing something bogus here?
Also I haven't heard of this problem before at all and I'm sure I would have if sounds/cdroms were broken.
Shouldn't be applied without further analysis.
Quoted 1 line
Who maintains the swiotlb code?
Nobody. But I hacked last on it.
-Andi
Quoted 1 line · Andi Kleen
Quoted 1 line
Quoted 1 line
From: David Moore <[email protected]>
...
Quoted 15 lines
Quoted 15 lines
Quoted 15 lines
It fixes real crashes:
http://lists.opensuse.org/opensuse-bugs/2006-12/msg02943.html
http://qa.mandriva.com/show_bug.cgi?id=28224
http://www.pchdtv.com/forum/viewtopic.php?t=2063&sid=a959a14a4c2db0eebaab7b0df56103celib/swiotlb.c renamed from lib/swiotlb.c.orig +1
-1@@ -750,7 +750,7 @@ swiotlb_sync_sg(struct device *hwdev, stfor (i =0 ; i < nelems; i++, sg++)if (sg->dma_address !=SG_ENT_PHYS_ADDRESS (sg)) -sync_single (hwdev, (void *) sg->dma_address, +sync_single (hwdev,phys_to_virt (sg->dma_address), sg->dma_length, dir, target); }
...
Quoted 2 lines
Sounds weird. If this really didn't work much more should be broken (e.g. no cdroms/sound on Intel x86-64 boxes with >4GB)
Yes, it's weird; e.g. it seems that reports of that on linux1394-user and -devel came in only recently. But I may have missed something. (OTOH, this bug was partially hidden for FireWire users because one of the relevant FireWire drivers was lacking respective _sync_sg calls.)
Quoted 2 lines
I'm a little sceptical. Perhaps the TV driver is doing something bogus here?
sync_single() definitely wants a virtual address there. sg->dma_address is a physical address. The bug and the fix are obvious.
Unfortunately an author of lib/swiotlb.c chose to call many variables holding *virtual* addresses "dma_addr". Note how that file at the same time contains variables like "dma_addr_t dma_handle". A recipe for disaster.
Quoted 4 lines
Also I haven't heard of this problem before at all and I'm sure I would have if sounds/cdroms were broken.
Shouldn't be applied without further analysis.
Maybe some recent changes elsewhere cause more frequent use of swiotlb-driven bounce buffers? That should be verified.
Or maybe people are deploying affected hardware more often now?
Signature · 3 lines
Stefan Richter -=====-=-=== --=- --==- http://arcgraph.de/sr/
Quoted 3 lines · I wrote:
Unfortunately an author of
lib/swiotlb.cchose to call many variables holding *virtual* addresses "dma_addr". Note how that file at the same time contains variables like "dma_addr_tdma_handle".
And there is even one occurrence of a "dma_addr" holding a physical address/ bus address, unlike the other dma_addr's:
intswiotlb_dma_mapping_error(dma_addr_t dma_addr)
{
return (dma_addr == virt_to_phys(io_tlb_overflow_buffer));
}Signature · 3 lines
Stefan Richter -=====-=-=== --=- --==- http://arcgraph.de/sr/
Quoted 1 line
Shouldn't be applied without further analysis.
I don't think further analysis is required before this change be done, the missing conversion is rather obvious when comparing to all other functions in that file.
Whether the observed crashes really origin from the missing bits here is a different question.
Jan