kernel page size explanation

kernel page size explanation

[Gaspar Bakos]

Hi,

Sorry for this nursery-school question.

Could someone briefly explain me :
1. what is the kernel page size (any _useful_ pointer on the web is fine),
2. how can one tune it (for 2.6.*)?
3. what kind of effect does it have on system performance, if it is
tuneable, and if it worth changing this at all?

I am a bit confused; at one place i see someone saying that the kernel
page size is 4kb for i386.
At another place I see a statement:
"I tried all four possible page sizes on Itanium (4k, 8k, 16k and 64k)"

How can i figure out the page size of the kernel i am currently using?

Cheers
Gaspar

Re: kernel page size explanation

[Jesper Juhl]

On 7/22/05, Gaspar Bakos <gbakos@cfa.harvard.edu> wrote:
> Hi,
> 
> Sorry for this nursery-school question.
> 
> Could someone briefly explain me :
> 1. what is the kernel page size (any _useful_ pointer on the web is fine),

Depends on arch. Take a look at PAGE_SIZE and PAGE_SHIFT - look in
include/asm-*/page.h
Here's a nice web interface for browsing the source and quickly
finding the info you need :)  : http://lxr.linux.no/ident?i=PAGE_SIZE

> 2. how can one tune it (for 2.6.*)?

For some archs the page size can be set at compile-time with
CONFIG_PAGE_SIZE_4KB, CONFIG_PAGE_SIZE_8KB etc - mips is an example of
such an arch (also take a look at CONFIG_HUGETLB_PAGE and friends).

> 3. what kind of effect does it have on system performance, if it is
> tuneable, and if it worth changing this at all?
> 
Depends on your workload.

> I am a bit confused; at one place i see someone saying that the kernel
> page size is 4kb for i386.
> At another place I see a statement:
> "I tried all four possible page sizes on Itanium (4k, 8k, 16k and 64k)"
> 
That makes perfect sense - i386 uses a 4K page size, ia64 is one of
the archs that support different page sizes (via
CONFIG_IA64_PAGE_SIZE_* ).
some i386 machines can also use 4MB pages IIRC, but I don't think
Linux lets you configure that for i386, but I'm not entirely sure.

> How can i figure out the page size of the kernel i am currently using?
> 
You can
 A) look in the .config file for your current kernel (if your arch
supports different page sizes at all).
 B) You can use the  getpagesize(2) syscall at runtime. getpagesize()
returns the nr of bytes in a page - man getpagesize - I'm not sure
that's universally supported though.
 C) You can look at /proc/cpuinfo or /proc/meminfo , IIRC some archs
report page size there - not quite sure, can't remember...


-- 
Jesper Juhl <jesper.juhl@gmail.com>
Don't top-post  http://www.catb.org/~esr/jargon/html/T/top-post.html
Plain text mails only, please      http://www.expita.com/nomime.html

Re: kernel page size explanation

[Gaspar Bakos]

Hi, Jesper,

RE:
> > 2. how can one tune it (for 2.6.*)?
>
> For some archs the page size can be set at compile-time with
> CONFIG_PAGE_SIZE_4KB, CONFIG_PAGE_SIZE_8KB etc - mips is an example of
> such an arch (also take a look at CONFIG_HUGETLB_PAGE and friends).

OK, now i figured it out. On AMD opteron 64 bit processors the only
option is the default: 4Kb. For itanium and mips, indeed, there are
multiple options. My CPUs are opteron 64 bit ones; that is why i was
not able to find this feature (of tuning it).

> > How can i figure out the page size of the kernel i am currently using?
> >
> You can
>  A) look in the .config file for your current kernel (if your arch
> supports different page sizes at all).

Not there.

>  C) You can look at /proc/cpuinfo or /proc/meminfo , IIRC some archs
> report page size there - not quite sure, can't remember...

Probably not in meminfo, but possibly in cpuinfo:

cat /proc/cpuinfo
...
TLB size        : 1024 4K pages
...

cat /proc/meminfo
MemTotal:      4010956 kB
MemFree:       3848060 kB
Buffers:             0 kB
Cached:          45696 kB
SwapCached:          0 kB
Active:          90432 kB
Inactive:         4820 kB
HighTotal:           0 kB
HighFree:            0 kB
LowTotal:      4010956 kB
LowFree:       3848060 kB
SwapTotal:     7823576 kB
SwapFree:      7823352 kB
Dirty:              20 kB
Writeback:           0 kB
Mapped:          78668 kB
Slab:            46216 kB
CommitLimit:   9829052 kB
Committed_AS:   142036 kB
PageTables:       3252 kB
VmallocTotal: 34359738367 kB
VmallocUsed:      5724 kB
VmallocChunk: 34359732179 kB
HugePages_Total:     0
HugePages_Free:      0
Hugepagesize:     2048 kB

Cheers
Gaspar

Re: kernel page size explanation

[Nix]

On 22 Jul 2005, Jesper Juhl suggested tentatively:
> You can
>  A) look in the .config file for your current kernel (if your arch
> supports different page sizes at all).
>  B) You can use the  getpagesize(2) syscall at runtime. getpagesize()
> returns the nr of bytes in a page - man getpagesize - I'm not sure
> that's universally supported though.
>  C) You can look at /proc/cpuinfo or /proc/meminfo , IIRC some archs
> report page size there - not quite sure, can't remember...

D) getconf PAGE_SIZE should work, although what it does on arches
   with variable page sizes isn't clear to me.

-- 
`But of course, GR is the very best relativity for the masses.'
 --- Wayne Throop

Re: kernel page size explanation

[VASM]

i had one question 
does the linux kernel support only one default page size even if the
processor on which it is working supports multiple ?

On 7/25/05, Nix <nix@esperi.org.uk> wrote:
> On 22 Jul 2005, Jesper Juhl suggested tentatively:
> > You can
> >  A) look in the .config file for your current kernel (if your arch
> > supports different page sizes at all).
> >  B) You can use the  getpagesize(2) syscall at runtime. getpagesize()
> > returns the nr of bytes in a page - man getpagesize - I'm not sure
> > that's universally supported though.
> >  C) You can look at /proc/cpuinfo or /proc/meminfo , IIRC some archs
> > report page size there - not quite sure, can't remember...
> 
> D) getconf PAGE_SIZE should work, although what it does on arches
>   with variable page sizes isn't clear to me.
> 
> --
> `But of course, GR is the very best relativity for the masses.'
>  --- Wayne Throop
> -
> To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
> the body of a message to majordomo@vger.kernel.org
> More majordomo info at  http://vger.kernel.org/majordomo-info.html
> Please read the FAQ at  http://www.tux.org/lkml/
>

Re: kernel page size explanation

[Fawad Lateef]

On 7/25/05, VASM <vasm85@gmail.com> wrote:
> i had one question
> does the linux kernel support only one default page size even if the
> processor on which it is working supports multiple ?
> 

The PAGE_SIZE depends on the architecture and it do supports different
page_sizes depending on the architecture (AFAIK for almost all
supported architectures)

-- 
Fawad Lateef

Re: kernel page size explanation

[Nix]

On Mon, 25 Jul 2005, VASM wrote:
> i had one question 
> does the linux kernel support only one default page size even if the
> processor on which it is working supports multiple ?

No. Some architectures have compile-time support for multiple different
page sizes (e.g. Itanium, SPARC64); many have support for a
(non-swappable) `large pages) system, and a filesystem backed by huge
pages. (Often, the kernel is stored in huge pages, to keep the number
of page table entries wasted by the nonswappable kernel to a minimum.)

What is *not* presently supported is using multiple page sizes to
back userspace processes; that size is currently fixed at compile-time,
even on architectures supporting multiple variably-sized pages.

-- 
`But of course, GR is the very best relativity for the masses.'
 --- Wayne Throop

Re: kernel page size explanation

[VASM]

On 7/25/05, Nix <nix@esperi.org.uk> wrote:
> On Mon, 25 Jul 2005, VASM wrote:
> > i had one question
> > does the linux kernel support only one default page size even if the
> > processor on which it is working supports multiple ?
> 
> No. Some architectures have compile-time support for multiple different
> page sizes (e.g. Itanium, SPARC64); many have support for a
> (non-swappable) `large pages) system, and a filesystem backed by huge
> pages. (Often, the kernel is stored in huge pages, to keep the number
> of page table entries wasted by the nonswappable kernel to a minimum.)
> 
> What is *not* presently supported is using multiple page sizes to
> back userspace processes; that size is currently fixed at compile-time,
> even on architectures supporting multiple variably-sized pages.
> 
are there any specific reasons for not using large page size for
userspace processes

> --
> `But of course, GR is the very best relativity for the masses.'
>  --- Wayne Throop
>

Re: kernel page size explanation

[Valdis.Kletnieks]

[-- Attachment #1: Type: text/plain, Size: 277 bytes --]

On Mon, 25 Jul 2005 19:12:47 +0530, VASM said:
> are there any specific reasons for not using large page size for
> userspace processes

Assume you can use 4K or 4M page sizes.  Compute the total memory usage
for a system that has 50 processes running, each 1556K in size....


[-- Attachment #2: Type: application/pgp-signature, Size: 226 bytes --]

Re: kernel page size explanation

[Robert Hancock]

Gaspar Bakos wrote:
> Hi,
> 
> Sorry for this nursery-school question.
> 
> Could someone briefly explain me :
> 1. what is the kernel page size (any _useful_ pointer on the web is fine),
> 2. how can one tune it (for 2.6.*)?
> 3. what kind of effect does it have on system performance, if it is
> tuneable, and if it worth changing this at all?

This is dependent on the hardware, not really the OS. On x86 the 
normally used page size is 4KB. 4MB pages are also supported but are 
usually used only for special purposes (ex: hugetlbfs).

As you mentioned some other architectures like Itanium support different 
page sizes.

-- 
Robert Hancock      Saskatoon, SK, Canada
To email, remove "nospam" from hancockr@nospamshaw.ca
Home Page: http://www.roberthancock.com/