On 3/27/07, Martyn Welch <martyn.welch@xxxxxxxxxxxxxx> wrote:
> -----Original Message-----
> From: kernelnewbies-bounce@xxxxxxxxxxxx
> [mailto:kernelnewbies-bounce@xxxxxxxxxxxx ] On Behalf Of Pharaoh .
> Sent: 27 March 2007 08:58 AM
> To: kernelnewbies@xxxxxxxxxxxx
> Subject: vmalloc - page tables
>
> Hi
>
> I am confused about the way vmalloc works. AFAIK, only 128 MB
> is available for vmalloc on x86 and this memory is only
> virtually contiguous and not physically when allocated by a
> process. When this memory is allocted then the processe's
> page table does not contain the mappings, instead the
> mappings are only stored in the init_mm->pgd i.e. the master
> page table and when the vmalloced memory is allocated then
> page fault occurs and the current processes page table is
> modified according to the master page table. Is this correct?
> The entire thing I am talking about?
>
I believe the page fault occurs when a memory access is attempted. But
other than this, that's how I understand it.
> If yes then, why it is done this way? I mean why only the
> master page table is modified and then we have to wait for a
> page fault to occur and then sync up the process's page
> table, isnt this an overhead? For normal memory allocation
> this is not applicable, right?
>
I remember correctly (I'm in the process of learning this myself) it
reduces the overall overhead. There's a chance that a process will
malloc a reasonable amount of memory that it will not use straight away,
in fact it may never use some of it. By waiting until the page is
actually being used this overhead can be broken down and pontentially
saved.
Is that right?
Martyn
----
Martyn Welch
Principal Software Engineer
Radstone Digital Processing
Part of GE Fanuc Embedded Systems
Phone: +44 (0) 1327 322748
email: martyn.welch@xxxxxxxxxxxxxx
web: www.radstone.com
________________________________________________________________________
This e-mail has been scanned for all viruses by Star.The service is powered by MessageLabs.
________________________________________________________________________
Sandeep,
can you elucidate it a little more? It might be little naive to ask about this but can you tell how this entire thing is? I meant entire thing about master page table and per process page tables.
Martin,
For describing the area allocated using vmalloc kernel has a small struct vm_struct which is much smaller than a full fledged struct which describes VMA, here they save memory.
There's a chance that a process will
malloc a reasonable amount of memory that it will not use straight away,
in fact it may never use some of it. By waiting until the page is
actually being used this overhead can be broken down and pontentially
saved.
I am not getting it how this is applicable only to vmalloced memory and not kmalloced....Do you get what I am trying to say here? I dont think this is the reason here, please CMIIW.
-pharaoh.
