user space vs kernel space address space

[Miles MH Chen]

Hi Pritam,

Every page frame is controlled by kernel, but kernel does not have to map
all pages in all time.
On a 32-bit bit, 3:1 split Linux kernel, the kernel has at most 1GB VIRTUAL
address space. (user space has 3GB).
If there are 2GB of physical DRAM on the system, our kernel obviously
cannot map 2GB memory into 1GB virtual address space.

To solve this problem, the kernel map part of DRAM to normal area (1-to-1
mapping). A page from normal zone can be accessed
by kernel by using the existing mapping.
Other memory is called high memory. When kernel wants to access the high
memory pages,
it maps them to kernel address space (by vmalloc, fixed map or permanent
map).

(A 64-bit CPU can make life much easier. We can map a lot of DRAM directly.)

Regards,
MH


On Mon, Dec 30, 2013 at 10:28 PM, Pritam Bankar
<pritambankar1988 at gmail.com>wrote:

>
> Miles thanks a lot for response but I still have some doubts please see my
> comments inline.
>
>
>
> On Sun, Dec 29, 2013 at 7:04 PM, Miles MH Chen <orca.chen at gmail.com>wrote:
>
>> Hi Pritam,
>>
>> 1) Yes, all 512RAM will be direct mapped to kernel address space IF the
>> kernel have a 896MB direct mapping area.
>> Actually you can change the range of kernel direct mapping by the
>> vmalloc=<size> in boot command line.
>> In 32-bit and 3:1 split configuration, kernel direct mapping area +
>> vmalloc area is roughly 1G.
>>
>> 2) User space and kernel space can have different virtual addresses
>> mapping to the same physical frame at the same time.
>>
>> 3) Direct mapped or fixed map or permanent map are describing kernel
>> VIRTUAL address space, you can see the virtual memory
>> layout by 'dmesg'. When user space needs memory, the kernel allocates a
>> free memory frame, and remaps the frame to user space.
>> The frame does not have to have a valid kernel virtual address.
>>
>>
>
> But all physical memory/frames are eventually controlled by kernel , then
> how can there ever exist a frame which doesn't have mapping with kernel or
> kernel virtual address ?
>
>
>
>
>> Regards,
>> MH
>>
>>
>> On Sun, Dec 29, 2013 at 5:42 PM, Pritam Bankar <
>> pritambankar1988 at gmail.com> wrote:
>>
>>> Lets consider 32 bit Linux system with 512 physical RAM. Suppose I have
>>> standard 3:1 address space split. Now what I understand is
>>>
>>> (In general)
>>> 1. In the fourth gigabyte I have kernel space
>>> 2. Out of 1GB for kernel address space only 896MB is used as direct
>>> mapping and other 128 MB is used for Noncontiguous Memory Area Management,
>>> Fixed Mapping and Permanent Mapping.
>>>
>>> PCMIIW
>>>
>>> Following are my doubts :
>>> 1. Since my system has only 512MB RAM, will there be only direct
>>> mappings since 896 is enough to hold 512 RAM?
>>> 2. When user space program do malloc, we get some virtual address from
>>> userspace region (from first 3GB) of process. So will it be like, when I
>>> access some memory from that region, there is some physical frame
>>> associated with it AND same physical frame will also be mapped in kernel
>>> space ?
>>> So what I want  to know, for every physical frame is there a mapping in
>>> userspace as well as kernel space ? (given that some address in user space
>>> of process map to same physical frame)
>>> 3. If we consider example of Linux system with 4GB, now if I do malloc
>>> from user space from which memory region kernel will give memory ? Direct
>>> mapped region or fixed map or permanent map ?
>>>
>>> Thanks and regards,
>>> Pritam Bankar
>>>
>>> _______________________________________________
>>> Kernelnewbies mailing list
>>> Kernelnewbies at kernelnewbies.org
>>> http://lists.kernelnewbies.org/mailman/listinfo/kernelnewbies
>>>
>>>
>>
>
> Thanks and regards,
> Pritam Bankar
>
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: http://lists.kernelnewbies.org/pipermail/kernelnewbies/attachments/20131230/e031c229/attachment-0001.html