[Linux-ia64] about linux/ia64 mm

[Linux-ia64] about linux/ia64 mm

[唐 宇]

hi all,
   i was a little confused by linux/ia64 mm; as i know, linux requires 
physical memory to be mapped contiguously to kernel space( don't care 
highmem); and PAGE_OFFSET is  defined to 0xe000000000000000 in linux/ia64; 
that's region 7, its page size is fixed to 64M(2.4.12). how could 
__get_free_pages() allocate such a big _page_? it's absolutely impossible.
   so i think i made a big mistake somewhere, would someone help me out? 
thx :-)

rgds, tang
   

_________________________________________________________________
您可以在 MSN Hotmail 站点 http://www.hotmail.com/cn 免费收发电子邮件

Re: [Linux-ia64] about linux/ia64 mm

[David Mosberger]

>>>>> On Thu, 29 Nov 2001 23:19:53 +0800, "=?gb2312?B?zMYg0+4=?=" <ytang5@hotmail.com> said:

  Tang> hi all, i was a little confused by linux/ia64 mm; as i know,
  Tang> linux requires physical memory to be mapped contiguously to
  Tang> kernel space( don't care highmem); and PAGE_OFFSET is defined
  Tang> to 0xe000000000000000 in linux/ia64; that's region 7, its page
  Tang> size is fixed to 64M(2.4.12). how could __get_free_pages()
  Tang> allocate such a big _page_? it's absolutely impossible.  so i
  Tang> think i made a big mistake somewhere, would someone help me
  Tang> out?  thx :-)

The page allocator works at the page frame level, which can be
configured to have a size of 4, 8, 16, or 64KB.  The large TLB
mappings are used merely to access the underlying memory (think of it
as a physical address space with a non-zero starting address).

	--david

Re: [Linux-ia64] about linux/ia64 mm

[唐 宇]

thx a lot first;  
and as _pa(x) is defined to ((x)-PAGE_OFFSET), so i think, kernel space 
from PAGE_OFFSET to (PAGE_OFFSET + Size of Memroy) should be mapped 
directly to 0 -Size of Memory; so what happens if codes access address 
returned by functions such as get_free_pages()(it should range in 
[PAGE_OFFSET,PAGE_OFFSET+...])? how could it be translated by MMU?
and what you mean, there are both 64M-page and configurable 
PAGE_SIZE(4,8k,etc) for Region 7? or i just missed something?

rgds

>From: David Mosberger <davidm@hpl.hp.com>
>Reply-To: davidm@hpl.hp.com
>To: "唐 宇" <ytang5@hotmail.com>
>CC: linux-ia64@linuxia64.org
>Subject: Re: [Linux-ia64] about linux/ia64 mm
>Date: Thu, 29 Nov 2001 21:43:18 -0800
>
> >>>>> On Thu, 29 Nov 2001 23:19:53 +0800, "=?gb2312?B?zMYg0+4=?=" 
<ytang5@hotmail.com> said:
>
>   Tang> hi all, i was a little confused by linux/ia64 mm; as i know,
>   Tang> linux requires physical memory to be mapped contiguously to
>   Tang> kernel space( don't care highmem); and PAGE_OFFSET is defined
>   Tang> to 0xe000000000000000 in linux/ia64; that's region 7, its page
>   Tang> size is fixed to 64M(2.4.12). how could __get_free_pages()
>   Tang> allocate such a big _page_? it's absolutely impossible.  so i
>   Tang> think i made a big mistake somewhere, would someone help me
>   Tang> out?  thx :-)
>
>The page allocator works at the page frame level, which can be
>configured to have a size of 4, 8, 16, or 64KB.  The large TLB
>mappings are used merely to access the underlying memory (think of it
>as a physical address space with a non-zero starting address).
>
>	--david


_________________________________________________________________
您可以在 MSN Hotmail 站点 http://www.hotmail.com/cn 免费收发电子邮件

Re: [Linux-ia64] about linux/ia64 mm

[唐 宇]

ok, i found my answer;
for region 7,bit rr.ve is 0, so the VHPT walker is disabled for these two 
regions; fault handler alt_[d/i]tlb_miss implement region-specific 
translation algorithms(more simple and effecient) for it;

thx again;

>From: David Mosberger <davidm@hpl.hp.com>
>Reply-To: davidm@hpl.hp.com
>To: "唐 宇" <ytang5@hotmail.com>
>CC: linux-ia64@linuxia64.org
>Subject: Re: [Linux-ia64] about linux/ia64 mm
>Date: Thu, 29 Nov 2001 21:43:18 -0800
>
> >>>>> On Thu, 29 Nov 2001 23:19:53 +0800, "=?gb2312?B?zMYg0+4=?=" 
<ytang5@hotmail.com> said:
>
>   Tang> hi all, i was a little confused by linux/ia64 mm; as i know,
>   Tang> linux requires physical memory to be mapped contiguously to
>   Tang> kernel space( don't care highmem); and PAGE_OFFSET is defined
>   Tang> to 0xe000000000000000 in linux/ia64; that's region 7, its page
>   Tang> size is fixed to 64M(2.4.12). how could __get_free_pages()
>   Tang> allocate such a big _page_? it's absolutely impossible.  so i
>   Tang> think i made a big mistake somewhere, would someone help me
>   Tang> out?  thx :-)
>
>The page allocator works at the page frame level, which can be
>configured to have a size of 4, 8, 16, or 64KB.  The large TLB
>mappings are used merely to access the underlying memory (think of it
>as a physical address space with a non-zero starting address).
>
>	--david


_________________________________________________________________
您可以在 MSN Hotmail 站点 http://www.hotmail.com/cn 免费收发电子邮件