Large memory system
Hi,
I've jumped the gun a little bit and recommended a Professor buy 4GB of RAM on a Xeon machine to run Linux on and he did. After he got it I read the large memory howto which states that the max memory size for Linux 2.2.x is 2GB physical/2GB virtual. The memory size seems to limited by the 32bit nature of the x86 architecture. The Xeon seems to have a 36bit memory addressing mode. Can Linux be easily expanded to use the 36bit addressing?
Thanks for any info on the subject.
- Daniel (Who needs to read more before recommending computers.)
--
Daniel Blakeley (N2YEN) Cornell Center for Materials Research
[email protected] E20 Clark Hall
--
To unsubscribe, send a message with 'unsubscribe linux-mm my@address' in the body to [email protected]. For more info on Linux MM,
see: http://humbolt.geo.uu.nl/Linux-MM/
Quoted 9 lines · Daniel Blakeley
Hi,
I've jumped the gun a little bit and recommended a Professor buy 4GB of RAM on a Xeon machine to run Linux on and he did. After he got it I read the large memory howto which states that the max memory size for Linux 2.2.x is 2GB physical/2GB virtual. The memory size seems to limited by the 32bit nature of the x86 architecture. The Xeon seems to have a 36bit memory addressing mode. Can Linux be easily expanded to use the 36bit addressing?
Easily isn't a good way of putting it, unless you're talking about doing something like mmap on /dev/mem, in which case you could make the user/kernel virtual spilt weigh heavy on the user side and do memory allocation yourself. If you're talking about doing it transparently, you're best bet is to do something like davem's suggested high mem approach, and only use non-kernel mapped memory for user pages... if you want to be able to support the page cache in high memory, things get messy.
-ben
--
To unsubscribe, send a message with 'unsubscribe linux-mm my@address' in the body to [email protected]. For more info on Linux MM,
see: http://humbolt.geo.uu.nl/Linux-MM/
Hi,
Quoted 8 lines · Benjamin C.R. LaHaise
Easily isn't a good way of putting it, unless you're talking about doing something like mmap on
/dev/mem, in which case you could make the user/kernel virtual spilt weigh heavy on the user side and do memory allocation yourself. If you're talking about doing it transparently, you're best bet is to do something like davem's suggested high mem approach, and only use non-kernel mapped memory for user pages... if you want to be able to support the page cache in high memory, things get messy.
No it doesn't! The only tricky thing is IO, but we need to have bounce buffers to high memory anyway for swapping. The page cache uses "struct page" addresses in preference to actual page data pointers almost everywhere anyway, and whenever we are doing something like read(2) or write(2) functions, we just need a single per-CPU virtual pte in the vmalloc region to temporarily map the page into memory while we copy to user space (and remember that we do this from the context of the user process anyway, so we don't have to remap the user page even if it is in high memory).
--Stephen
--
To unsubscribe, send a message with 'unsubscribe linux-mm my@address' in the body to [email protected]. For more info on Linux MM,
see: http://humbolt.geo.uu.nl/Linux-MM/
Quoted 1 line
Quoted 1 line
Quoted 1 line
Quoted 1 line
Quoted 1 line
"ST" == Stephen C Tweedie <[email protected]> writes:
ST> Hi,
ST> On Sat, 30 Jan 1999 12:00:53 -0500 (EST), "Benjamin C.R. LaHaise"
Quoted 8 lines · ST> <[email protected]> said:
Quoted 8 lines
Easily isn't a good way of putting it, unless you're talking about doing something like mmap on
/dev/mem, in which case you could make the user/kernel virtual spilt weigh heavy on the user side and do memory allocation yourself. If you're talking about doing it transparently, you're best bet is to do something like davem's suggested high mem approach, and only use non-kernel mapped memory for user pages... if you want to be able to support the page cache in high memory, things get messy.
ST> No it doesn't! The only tricky thing is IO, but we need to have bounce ST> buffers to high memory anyway for swapping. The page cache uses "struct ST> page" addresses in preference to actual page data pointers almost ST> everywhere anyway, and whenever we are doing something like read(2) or ST> write(2) functions, we just need a single per-CPU virtual pte in the ST> vmalloc region to temporarily map the page into memory while we copy to ST> user space (and remember that we do this from the context of the user ST> process anyway, so we don't have to remap the user page even if it is in ST> high memory).
Cool. We now have an idea that sounds possible.
The only remaining question is how much of a performance hit would changing the contents of a pte around all of the time be?
Every single page read/write syscall, as well as copying down to I/O bounce buffers sounds common enough that we probably would see a performance hit.
The other thing that happens is we start breaking assumptions about fixed limits based on architecture size. Things like the swap entry may need to be expanded.
Eric
--
To unsubscribe, send a message with 'unsubscribe linux-mm my@address' in the body to [email protected]. For more info on Linux MM,
see: http://humbolt.geo.uu.nl/Linux-MM/
Hi,
Quoted 4 lines · On 08 Feb 1999 09:31:11 -0600, [email protected] (Eric W. Biederman) said:
Cool. We now have an idea that sounds possible.
The only remaining question is how much of a performance hit would changing the contents of a pte around all of the time be?
Very little: there's the cost of the page invalidate (a couple of cycles), plus the cost of the CPU refilling that tlb from the page tables. It's completely lost in the noise compared to the cost of transferring an entire page of data to/from user space.
Quoted 3 lines
Every single page read/write syscall, as well as copying down to I/O bounce buffers sounds common enough that we probably would see a performance hit.
I doubt that it would be measurable.
Quoted 3 lines
The other thing that happens is we start breaking assumptions about fixed limits based on architecture size. Things like the swap entry may need to be expanded.
The swap entry can probably stay completely independent; most people with 8G of ram are going to be trying hard never to hit swap anyway. :) Besides, we already have support for 16G of swap as things stand.
--Stephen
--
To unsubscribe, send a message with 'unsubscribe linux-mm my@address' in the body to [email protected]. For more info on Linux MM,
see: http://humbolt.geo.uu.nl/Linux-MM/
Quoted 7 lines · Daniel Blakeley
I've jumped the gun a little bit and recommended a Professor buy 4GB of RAM on a Xeon machine to run Linux on and he did. After he got it I read the large memory howto which states that the max memory size for Linux 2.2.x is 2GB physical/2GB virtual. The memory size seems to limited by the 32bit nature of the x86 architecture. The Xeon seems to have a 36bit memory addressing mode. Can Linux be easily expanded to use the 36bit addressing?
Just today there was a patch on linux-kernel with a patch that allows you to use the top 2 GB as a RAM disk or something like that.
You can use that for swap and to mmap() stuff on. I think this could be quite useful for large simulations and stuff like that.
36-bit addressing is a bit difficult at the moment, but undoubtedly someone will code up something like that for Linux 2.3 (maybe the prof could let some (under)graduate student do this as a major project?).
succes,
Rik -- If a Microsoft product fails, who do you sue?
+-------------------------------------------------------------------+ | Linux Memory Management site: http://humbolt.geo.uu.nl/Linux-MM/ | | Nederlandse Linux documentatie: http://www.nl.linux.org/ | +-------------------------------------------------------------------+
--
To unsubscribe, send a message with 'unsubscribe linux-mm my@address' in the body to [email protected]. For more info on Linux MM,
see: http://humbolt.geo.uu.nl/Linux-MM/
Hi,
Quoted 7 lines · Daniel Blakeley
I've jumped the gun a little bit and recommended a Professor buy 4GB of RAM on a Xeon machine to run Linux on and he did. After he got it I read the large memory howto which states that the max memory size for Linux 2.2.x is 2GB physical/2GB virtual. The memory size seems to limited by the 32bit nature of the x86 architecture. The Xeon seems to have a 36bit memory addressing mode. Can Linux be easily expanded to use the 36bit addressing?
It's not exactly trivial, but it can (and will) be done. For now, you can only use 4G on a 64-bit architecture (Alpha or Sparc64), but basically we know how to address it on Intel too, transparently to the user.
--Stephen
--
To unsubscribe, send a message with 'unsubscribe linux-mm my@address' in the body to [email protected]. For more info on Linux MM,
see: http://humbolt.geo.uu.nl/Linux-MM/