Skip to contentSkip to conversation

Remove needless flush_dcache_page call

PATCH In review MinChan Kim 6 participants 13 messages Jan 16, 2009

Subsystems (5)
lore
Share & tools
lore
#0 lore raw

PATCH 1/1 Remove needless flush_dcache_page call

Now, Anyone don't maintain cramfs.
I don't know who is maintain romfs. so I send this patch to linux-mm, lkml, linux-dev.

I am not sure my thought is right.

When readpage is called, page with argument in readpage is just new allocated because kernel can't find that page in page cache.

At this time, any user process can't map the page to their address space. so, I think D-cache aliasing probelm never occur.

It make sense ?

---

arch/arm/mach-integrator/clock.h deleted +0 -0
deleted file mode 100644
index e69de29..0000000
fs/cramfs/inode.c +0 -1
index a07338d..40c8b84 100644
@@ -492,7 +492,6 @@ static int cramfs_readpage(struct file *file, struct page * page)
 		pgdata = kmap(page);
 	memset(pgdata + bytes_filled, 0, PAGE_CACHE_SIZE - bytes_filled);
 	kunmap(page);
-	flush_dcache_page(page);
 	SetPageUptodate(page);
 	unlock_page(page);
 	return 0;
fs/romfs/inode.c +0 -1
index 98a232f..d008262 100644
@@ -454,7 +454,6 @@ romfs_readpage(struct file *file, struct page * page)
 
 	if (!result)
 		SetPageUptodate(page);
-	flush_dcache_page(page);
 
 	unlock_page(page);
 

--
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>

#1 lore raw
Quoted 6 lines · MinChan Kim

Now, Anyone don't maintain cramfs.
I don't know who is maintain romfs. so I send this patch to linux-mm, lkml, linux-dev.

I am not sure my thought is right.

Quoted 7 lines

When readpage is called, page with argument in readpage is just new allocated because kernel can't find that page in page cache.

At this time, any user process can't map the page to their address space. so, I think D-cache aliasing probelm never occur.

It make sense ?

Sorry, no. You have to call fluch_dcache_page() in two situations -- when the kernel is going to read some data that userspace wrote, *and* when userspace is going to read some data that the kernel wrote. From a quick look at the patch, this seems to be the second case. The kernel wrote data to a pagecache page, and userspace should be able to read it.

To understand why this is necessary, consider a processor which is virtually indexed and has a writeback cache. The kernel writes to a page, then a user process reads from the same page through a different address. The cache doesn't find the data the kernel wrote because it has a different virtual index, so userspace reads stale data.

Signature · 4 lines
Matthew Wilcox				Intel Open Source Technology Centre
"Bill, look, we understand that you're interested in selling us this
operating system, but compare it to ours.  We can't possibly take such
a retrograde step."
#2 lore raw
Quoted 21 lines · Matthew Wilcox
Quoted 6 lines · MinChan Kim

Now, Anyone don't maintain cramfs.
I don't know who is maintain romfs. so I send this patch to linux-mm, lkml, linux-dev.

I am not sure my thought is right.

Quoted 7 lines

When readpage is called, page with argument in readpage is just new allocated because kernel can't find that page in page cache.

At this time, any user process can't map the page to their address space. so, I think D-cache aliasing probelm never occur.

It make sense ?

Sorry, no. You have to call fluch_dcache_page() in two situations -- when the kernel is going to read some data that userspace wrote, *and* when userspace is going to read some data that the kernel wrote. From a quick look at the patch, this seems to be the second case. The kernel wrote data to a pagecache page, and userspace should be able to read it.

Quoted 5 lines

To understand why this is necessary, consider a processor which is virtually indexed and has a writeback cache. The kernel writes to a page, then a user process reads from the same page through a different address. The cache doesn't find the data the kernel wrote because it has a different virtual index, so userspace reads stale data.

I see. :)

Thanks for quick reponse and good explaination.
Hmm,.. one more question.

I can't find flush_dcache_page call in mpage_readpage which is generic read function. In case of ext fs, it use mpage_readpage with readpage.

who and where call flush_dcache_page in mpage_readpage call path?

Quoted 6 lines
Signature · 5 lines
-- 
Matthew Wilcox				Intel Open Source Technology Centre
"Bill, look, we understand that you're interested in selling us this
operating system, but compare it to ours.  We can't possibly take such
a retrograde step."
#3 lore raw
Quoted 15 lines · MinChan Kim
Quoted 11 lines · Matthew Wilcox

Sorry, no. You have to call fluch_dcache_page() in two situations -- when the kernel is going to read some data that userspace wrote, *and* when userspace is going to read some data that the kernel wrote. From a quick look at the patch, this seems to be the second case. The kernel wrote data to a pagecache page, and userspace should be able to read it.

To understand why this is necessary, consider a processor which is virtually indexed and has a writeback cache. The kernel writes to a page, then a user process reads from the same page through a different address. The cache doesn't find the data the kernel wrote because it has a different virtual index, so userspace reads stale data.

I see. :)

Quoted 8 lines

Thanks for quick reponse and good explaination.
Hmm,.. one more question.

I can't find flush_dcache_page call in mpage_readpage which is generic read function. In case of ext fs, it use mpage_readpage with readpage.

who and where call flush_dcache_page in mpage_readpage call path?

Most I/O devices will do DMA to the page in question and thus the kernel hasn't written to it and the CPU won't have the data in cache. For the few devices which can't do DMA, it's the responsibility of the device driver to call flush_dcache_page() (or some other flushing primitive). See the gdth driver for an example:

            address = kmap_atomic(sg_page(sl), KM_BIO_SRC_IRQ) + sl->offset;
            memcpy(address, buffer, cpnow);
            flush_dcache_page(sg_page(sl));
            kunmap_atomic(address, KM_BIO_SRC_IRQ);
Signature · 10 lines
Matthew Wilcox				Intel Open Source Technology Centre
"Bill, look, we understand that you're interested in selling us this
operating system, but compare it to ours.  We can't possibly take such
a retrograde step."

--
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>
#4 lore raw
Quoted 23 lines · Matthew Wilcox
Quoted 22 lines · MinChan Kim
Quoted 11 lines · Matthew Wilcox

Sorry, no. You have to call fluch_dcache_page() in two situations -- when the kernel is going to read some data that userspace wrote, *and* when userspace is going to read some data that the kernel wrote. From a quick look at the patch, this seems to be the second case. The kernel wrote data to a pagecache page, and userspace should be able to read it.

To understand why this is necessary, consider a processor which is virtually indexed and has a writeback cache. The kernel writes to a page, then a user process reads from the same page through a different address. The cache doesn't find the data the kernel wrote because it has a different virtual index, so userspace reads stale data.

I see. :)

Thanks for quick reponse and good explaination.
Hmm,.. one more question.

I can't find flush_dcache_page call in mpage_readpage which is generic read function. In case of ext fs, it use mpage_readpage with readpage.

Quoted 6 lines
Quoted 1 line

who and where call flush_dcache_page in mpage_readpage call path?

Most I/O devices will do DMA to the page in question and thus the kernel hasn't written to it and the CPU won't have the data in cache. For the few devices which can't do DMA, it's the responsibility of the device driver to call flush_dcache_page() (or some other flushing primitive).

Hmm.. Now I am confusing.
If devicer driver or with DMA makes sure cache consistency, Why filesystem code have to handle it ?

Quoted 12 lines

See the gdth driver for an example:

            address = kmap_atomic(sg_page(sl), KM_BIO_SRC_IRQ) + sl->offset;
            memcpy(address, buffer, cpnow);
            flush_dcache_page(sg_page(sl));
            kunmap_atomic(address, KM_BIO_SRC_IRQ);
Signature · 4 lines
Matthew Wilcox				Intel Open Source Technology Centre
"Bill, look, we understand that you're interested in selling us this
operating system, but compare it to ours.  We can't possibly take such
a retrograde step."

--
Kinds regards,
MinChan Kim

--
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>

#5 lore raw
Quoted 9 lines · MinChan Kim
Quoted 4 lines · Matthew Wilcox

Most I/O devices will do DMA to the page in question and thus the kernel hasn't written to it and the CPU won't have the data in cache. For the few devices which can't do DMA, it's the responsibility of the device driver to call flush_dcache_page() (or some other flushing primitive).

Hmm.. Now I am confusing.
If devicer driver or with DMA makes sure cache consistency, Why filesystem code have to handle it ?

Because the filesystem is accessing the page directly rathe rthan going to IO.

Basically, whoever reads or writes the page is responsible to avoid user aliases. You see these calls in the VM for anonymous pages, in bounce buffer layers, in filesystems that read or write from pages that are exposed to userspace (ie. metadata generally need not be flushed because it will not be mmapped by userspace).

#6 lore raw
Quoted 11 lines · Nick Piggin
Quoted 9 lines · MinChan Kim
Quoted 4 lines · Matthew Wilcox

Most I/O devices will do DMA to the page in question and thus the kernel hasn't written to it and the CPU won't have the data in cache. For the few devices which can't do DMA, it's the responsibility of the device driver to call flush_dcache_page() (or some other flushing primitive).

Hmm.. Now I am confusing.
If devicer driver or with DMA makes sure cache consistency, Why filesystem code have to handle it ?

Quoted 8 lines

Because the filesystem is accessing the page directly rathe rthan going to IO.

Basically, whoever reads or writes the page is responsible to avoid user aliases. You see these calls in the VM for anonymous pages, in bounce buffer layers, in filesystems that read or write from pages that are exposed to userspace (ie. metadata generally need not be flushed because it will not be mmapped by userspace).

Totally, understand.
Thanks for kind answering to my poor question in patience.

--
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>

#7 lore raw
In reply to MinChan Kim
Quoted 30 lines · MinChan Kim
Quoted 21 lines · Matthew Wilcox
Quoted 6 lines · MinChan Kim

Now, Anyone don't maintain cramfs.
I don't know who is maintain romfs. so I send this patch to linux-mm, lkml, linux-dev.

I am not sure my thought is right.

Quoted 7 lines

When readpage is called, page with argument in readpage is just new allocated because kernel can't find that page in page cache.

At this time, any user process can't map the page to their address space. so, I think D-cache aliasing probelm never occur.

It make sense ?

Sorry, no. You have to call fluch_dcache_page() in two situations -- when the kernel is going to read some data that userspace wrote, *and* when userspace is going to read some data that the kernel wrote. From a quick look at the patch, this seems to be the second case. The kernel wrote data to a pagecache page, and userspace should be able to read it.

Quoted 5 lines

To understand why this is necessary, consider a processor which is virtually indexed and has a writeback cache. The kernel writes to a page, then a user process reads from the same page through a different address. The cache doesn't find the data the kernel wrote because it has a different virtual index, so userspace reads stale data.

I see. :)

Quoted 8 lines

Thanks for quick reponse and good explaination.
Hmm,.. one more question.

I can't find flush_dcache_page call in mpage_readpage which is generic read function. In case of ext fs, it use mpage_readpage with readpage.

who and where call flush_dcache_page in mpage_readpage call path?

I think if the page is populated via IO, then it is responsibility of the IO layers (eg dma API) to ensure caches are consistent. Presumably this would include calling flush_dcache_page if we CPU is being used for the copies (eg. see drivers/block/brd.c).

But there are quite possibly holes around here because not as much testing is done on CPUs with these kinds of caches. Eg. brd probably should be doing a flush_dcache_page in the rw == WRITE direction AFAIKS, so it picks up user aliases here.

#8 lore raw
Quoted 4 lines · Nick Piggin

But there are quite possibly holes around here because not as much testing is done on CPUs with these kinds of caches. Eg. brd probably should be doing a flush_dcache_page in the rw == WRITE direction AFAIKS, so it picks up user aliases here.

Nick, if you wanted me to schlep a parisc machine to LCA for you, you needed to ask me *before* I got as far as Vancouver ;-)

Signature · 4 lines
Matthew Wilcox				Intel Open Source Technology Centre
"Bill, look, we understand that you're interested in selling us this
operating system, but compare it to ours.  We can't possibly take such
a retrograde step."
#9 lore raw
Quoted 8 lines · Matthew Wilcox
Quoted 4 lines · Nick Piggin

But there are quite possibly holes around here because not as much testing is done on CPUs with these kinds of caches. Eg. brd probably should be doing a flush_dcache_page in the rw == WRITE direction AFAIKS, so it picks up user aliases here.

Nick, if you wanted me to schlep a parisc machine to LCA for you, you needed to ask me *before* I got as far as Vancouver ;-)

;)

That actually might not be a bad idea if I had access to one of your crazy systems. Would it really fit on an aeroplane or would you need a shipping container? :)

#10 lore raw
Quoted 3 lines · Nick Piggin
Quoted 2 lines · Matthew Wilcox
Quoted 1 line · Nick Piggin

But there are quite possibly holes around here because not as muc=

h testing

Quoted 1 line
Quoted 1 line
Quoted 1 line

is done on CPUs with these kinds of caches. Eg. brd probably shou=

ld be

Quoted 1 line
Quoted 1 line
Quoted 1 line

doing a flush_dcache_page in the rw =3D=3D WRITE direction AFAIKS=

, so it picks

Quoted 3 lines
Quoted 3 lines
Quoted 1 line

up user aliases here.

=20
Nick, if you wanted me to schlep a parisc machine to LCA for you, y=

ou

Quoted 5 lines
Quoted 1 line

needed to ask me *before* I got as far as Vancouver ;-)

=20
;)
=20
That actually might not be a bad idea if I had access to one of your =

crazy

Quoted 1 line

systems. Would it really fit on an aeroplane or would you need a ship=

ping

Quoted 1 line

container? :)

They can be taken as checked luggage ... however, what Matthew means is that by being in Vancouver he's already in transit to Australia. However, we can give you access to the several remote test rings we have, either Cupertino (send ssh key to Rick Jones <[email protected]> or Thiabut Var=C3=A8ne <[email protected]>)

I thought we also had a wiki page listing all the machines, but I can't find it. They all use the HP Guardian Service Processor console, which gives you remote console, power and all the other good things you need.

James

--
To unsubscribe from this list: send the line "unsubscribe linux-parisc"=
in
the body of a message to [email protected]
More majordomo info at http://vger.kernel.org/majordomo-info.html

#11 lore raw
Quoted 4 lines · James Bottomley
Quoted 3 lines · Nick Piggin
Quoted 2 lines · Matthew Wilcox
Quoted 1 line · Nick Piggin

But there are quite possibly holes around here because not as m=

uch testing

Quoted 1 line
Quoted 1 line
Quoted 1 line
Quoted 1 line

is done on CPUs with these kinds of caches. Eg. brd probably sh=

ould be

Quoted 1 line
Quoted 1 line
Quoted 1 line
Quoted 1 line

doing a flush_dcache_page in the rw =3D=3D WRITE direction AFAI=

KS, so it picks

Quoted 3 lines
Quoted 3 lines
Quoted 3 lines
Quoted 1 line

up user aliases here.

=20
Nick, if you wanted me to schlep a parisc machine to LCA for you,=

you

Quoted 5 lines
Quoted 5 lines
Quoted 1 line

needed to ask me *before* I got as far as Vancouver ;-)

=20
;)
=20
That actually might not be a bad idea if I had access to one of you=

r crazy

Quoted 1 line
Quoted 1 line

systems. Would it really fit on an aeroplane or would you need a sh=

ipping

Quoted 3 lines
Quoted 1 line

container? :)

=20
They can be taken as checked luggage ... however, what Matthew means =

is

Quoted 3 lines

that by being in Vancouver he's already in transit to Australia. However, we can give you access to the several remote test rings we have, either Cupertino (send ssh key to Rick Jones <[email protected]> =

or

Quoted 2 lines

Thiabut Var=E8ne <[email protected]>)
=20

I think Nick already has access to the gsyprf* systems, as I've seen hi= m
logged in there before. ;-)
--
To unsubscribe from this list: send the line "unsubscribe linux-parisc"=
in
the body of a message to [email protected]
More majordomo info at http://vger.kernel.org/majordomo-info.html

#12 lore raw
In reply to James Bottomley

On Fri, Jan 16, 2009 at 09:28:21AM -0500, James Bottomley wrote:
...`

Quoted 5 lines

They can be taken as checked luggage ... however, what Matthew means is that by being in Vancouver he's already in transit to Australia. However, we can give you access to the several remote test rings we have, either Cupertino (send ssh key to Rick Jones <[email protected]> or Thiabut Var??ne <[email protected]>)

Rick just gets me physical access when we have time to mess with the rack.

Kyle is correct that Nick still has a login to Cupertino ring though he might need one of us to update his .ssh/authorized keys:

grundler@lp1000r:~$ finger nick
Login: npiggin                          Name: Nick Piggin
Directory: /home/npiggin                Shell: /bin/bash
Office Phone: +xx x xxxx xxxx
Last login Wed Aug  3 22:35 2005 (PDT) on pts/5 from 203-173-7-11.dyn.iinet.net.au
No mail.
No Plan.

IIRC, i gave him access to a "beefier" ia64 machine at the time (and those are all still there too).

Quoted 2 lines

I thought we also had a wiki page listing all the machines, but I can't find it.

http://www.parisc-linux.org/cluster.html

I've intentionally left it unlinked from the main page.

Quoted 2 lines

They all use the HP Guardian Service Processor console, which gives you remote console, power and all the other good things you need.

And most of the parisc machines are in a sorry state right now. I'll update the web page with current state.

hth,
grant