Skip to contentSkip to conversation

ARM fixes

GIT PULL Linus Torvalds 2 participants 4 messages Feb 18, 2014 – Feb 19, 2014 1 message not in the archive

Subsystems (2)
Memory management SCSI
lore
Share & tools
lore

A message that is not in the archive (sent off-list or not yet received): [email protected]

Linus Torvalds

author
#1 lore raw
Quoted 5 lines · Russell King

One fix touches code outside of arch/arm, which is related to sorting out the DMA masks correctly. There is a long standing issue with the conversion from PFNs to addresses where people assume that shifting an unsigned long left by PAGE_SHIFT results in a correct address.

You should probably have used PFN_PHYS(), which does this correctly. Your explicit u64 isn't exactly wrong, but phys_addr_t is really the right type for the result.

That said, it's admittedly a disgusting name, and I wonder if we should introduce a nicer-named "pfn_to_phys()" that matches the other "xyz_to_abc()" functions we have (including "pfn_to_virt()")

Looking at it, the Xen people then do this disgusting thing:
"__va(PFN_PHYS(pfn))" which is both ugly and pointless (__va() isn't going to work for a phys_addr_t anyway). And <linux/mm.h> has this gem:

  __va(PFN_PHYS(page_to_pfn(page)));

Ugh. The ugly - it burns. that really should be "pfn_to_virt(page_to_pfn())", I think. Adding a few mailing lists in the hope that some sucker^Whumanitarian person would want to take a look.

Anyway, I pulled your change to scsi_lib.c, since it's certainly no worse than what we used to have, but James and company cc'd too.

Linus

--
To unsubscribe, send a message with 'unsubscribe linux-mm' in the body to [email protected]. For more info on Linux MM,
see: http://www.linux-mm.org/ .
Don't email: <a href=mailto:"[email protected]"> [email protected] </a>

#2 lore raw
Quoted 10 lines · Linus Torvalds
Quoted 5 lines · Russell King

One fix touches code outside of arch/arm, which is related to sorting out the DMA masks correctly. There is a long standing issue with the conversion from PFNs to addresses where people assume that shifting an unsigned long left by PAGE_SHIFT results in a correct address.

You should probably have used PFN_PHYS(), which does this correctly. Your explicit u64 isn't exactly wrong, but phys_addr_t is really the right type for the result.

Almost, but not quite. If we're going to avoid u64, then dma_addr_t woudl be the right type here because we're talking about DMA addresses. We could also switch to keeping this as PFNs - block internally converts it to a PFN anyway:

void blk_queue_bounce_limit(struct request_queue *q, u64 max_addr)

{
        unsigned long b_pfn = max_addr >> PAGE_SHIFT;

...

and that is ultimately assigned to q->limits.bounce_pfn. So, if we arranged for blk_queue_bounce_limit() to take a PFN, and then modified it's two callers, then we don't need to care about converting between phys and pfns.

Maybe blk_queue_bounce_pfn_limit() so we ensure all users get caught?

Quoted 3 lines

That said, it's admittedly a disgusting name, and I wonder if we should introduce a nicer-named "pfn_to_phys()" that matches the other "xyz_to_abc()" functions we have (including "pfn_to_virt()")

We have these on ARM:

arch/arm/include/asm/memory.h:#define __pfn_to_phys(pfn) ((phys_addr_t)(pfn) << PAGE_SHIFT)
arch/arm/include/asm/memory.h:#define __phys_to_pfn(paddr) ((unsigned long)((paddr) >> PAGE_SHIFT))

it probably makes sense to pick those right out, maybe losing the __ prefix on them.

Quoted 6 lines

Looking at it, the Xen people then do this disgusting thing:
"__va(PFN_PHYS(pfn))" which is both ugly and pointless (__va() isn't going to work for a phys_addr_t anyway). And <linux/mm.h> has this gem:

  __va(PFN_PHYS(page_to_pfn(page)));

Wow. Two things spring to mind there... highmem pages, and don't we already have page_address() for that?

Quoted 2 lines

Anyway, I pulled your change to scsi_lib.c, since it's certainly no worse than what we used to have, but James and company cc'd too.

Thanks. I do worry about all the other places which I also found - but the first step is getting concensus on what the macro should be.

Signature · 9 lines
FTTC broadband for 0.8mile line: 5.8Mbps down 500kbps up.  Estimation
in database were 13.1 to 19Mbit for a good line, about 7.5+ for a bad.
Estimate before purchase was "up to 13.2Mbit".

--
To unsubscribe, send a message with 'unsubscribe linux-mm' in
the body to [email protected].  For more info on Linux MM,
see: http://www.linux-mm.org/ .
Don't email: <a href=mailto:"[email protected]"> [email protected] </a>
author
#3 lore raw
Quoted 3 lines · Russell King - ARM Linux

Almost, but not quite. If we're going to avoid u64, then dma_addr_t woudl be the right type here because we're talking about DMA addresses.

Well, phys_addr_t had better be as big as dma_addr_t, because that's what the resource management handles.

Quoted 2 lines

We could also switch to keeping this as PFNs - block internally converts it to a PFN anyway:

Yeah, that definitely sounds like it would be a good idea.

Quoted 6 lines

Maybe blk_queue_bounce_pfn_limit() so we ensure all users get caught?

Quoted 3 lines

That said, it's admittedly a disgusting name, and I wonder if we should introduce a nicer-named "pfn_to_phys()" that matches the other "xyz_to_abc()" functions we have (including "pfn_to_virt()")

Quoted 7 lines

We have these on ARM:

arch/arm/include/asm/memory.h:#define   __pfn_to_phys(pfn)      ((phys_addr_t)(pfn) << PAGE_SHIFT)
arch/arm/include/asm/memory.h:#define   __phys_to_pfn(paddr)    ((unsigned long)((paddr) >> PAGE_SHIFT))

it probably makes sense to pick those right out, maybe losing the __ prefix on them.

Yup.

Quoted 4 lines
Quoted 1 line
  __va(PFN_PHYS(page_to_pfn(page)));

Wow. Two things spring to mind there... highmem pages, and don't we already have page_address() for that?

Well, that code clearly cannot handle highmem anyway, but yes, it really smells like xen should use page_address().

Adding Xen people who I didn't add the last time around.

Linus

--
To unsubscribe, send a message with 'unsubscribe linux-mm' in the body to [email protected]. For more info on Linux MM,
see: http://www.linux-mm.org/ .
Don't email: <a href=mailto:"[email protected]"> [email protected] </a>

author
#4 lore raw
In reply to Russell King - ARM Linux
Quoted 3 lines · Russell King - ARM Linux

Almost, but not quite. If we're going to avoid u64, then dma_addr_t woudl be the right type here because we're talking about DMA addresses.

Well, phys_addr_t had better be as big as dma_addr_t, because that's what the resource management handles.

Quoted 2 lines

We could also switch to keeping this as PFNs - block internally converts it to a PFN anyway:

Yeah, that definitely sounds like it would be a good idea.

Quoted 6 lines

Maybe blk_queue_bounce_pfn_limit() so we ensure all users get caught?

Quoted 3 lines

That said, it's admittedly a disgusting name, and I wonder if we should introduce a nicer-named "pfn_to_phys()" that matches the other "xyz_to_abc()" functions we have (including "pfn_to_virt()")

Quoted 7 lines

We have these on ARM:

arch/arm/include/asm/memory.h:#define   __pfn_to_phys(pfn)      ((phys_addr_t)(pfn) << PAGE_SHIFT)
arch/arm/include/asm/memory.h:#define   __phys_to_pfn(paddr)    ((unsigned long)((paddr) >> PAGE_SHIFT))

it probably makes sense to pick those right out, maybe losing the __ prefix on them.

Yup.

Quoted 4 lines
Quoted 1 line
  __va(PFN_PHYS(page_to_pfn(page)));

Wow. Two things spring to mind there... highmem pages, and don't we already have page_address() for that?

Well, that code clearly cannot handle highmem anyway, but yes, it really smells like xen should use page_address().

Adding Xen people who I didn't add the last time around.

Linus