2002-01-10 18:11:10

by Fernando Jimenez

[permalink] [raw]
Subject: array size limit in module?

Hi

I am trying to code a simple kernel module and I have found a problem I don't
quite understand.

Here is the offending part of code:

char *sectors_array = NULL;
........
secs_size=131072;
sectors_array = kmalloc(secs_size*sizeof(char), GFP_KERNEL);
for(i=0; i<secs_size; i++) {
sectors_array[i]=0;
}

This bit of code, as it is, works fine. However, if I increment secs_size by
one, ie, I do 'secs_size=131073;' instead of 131072, I get the following:

Unable to handle kernel NULL pointer dereference at virtual address 00000000
Jan 10 18:14:47 localhost kernel: printing eip:
Jan 10 18:14:47 localhost kernel: c4829475
Jan 10 18:14:47 localhost kernel: *pde = 00000000
Jan 10 18:14:47 localhost kernel: Oops: 0002
Jan 10 18:14:47 localhost kernel: CPU: 0
Jan 10 18:14:47 localhost kernel: EIP: 0010:[<c4829475>] Not tainted
Jan 10 18:14:47 localhost kernel: EFLAGS: 00000297
Jan 10 18:14:47 localhost kernel: eax: 00000000 ebx: 00000000 ecx:
00000001 edx: c1467f64
Jan 10 18:14:47 localhost kernel: esi: 000000fe edi: 00000000 ebp:
c3567f28 esp: c3567f20
Jan 10 18:14:47 localhost kernel: ds: 0018 es: 0018 ss: 0018
Jan 10 18:14:47 localhost kernel: Process insmod (pid: 979,
stackpage=c3567000)
Jan 10 18:14:47 localhost kernel: Stack: c4829000 00000000 0806fc58 c011a1e5
00000000 c2f70000 00000d30 c3c0a000
Jan 10 18:14:47 localhost kernel: 00000060 ffffffea 00000007 c0a423cc
00000060 c02caac0 c4829060 00000f90
Jan 10 18:14:47 localhost kernel: 00000000 00000000 00000000 00000000
00000000 00000000 00000000 00000000
Jan 10 18:14:47 localhost kernel: Call Trace: [<c011a1e5>] [<c4829060>]
[<c01075bb>]
Jan 10 18:14:47 localhost kernel:
Jan 10 18:14:47 localhost kernel: Code: c6 04 03 00 83 ec 08 53 68 ce 9c 82
c4 e8 a9 f8 8e fb 83 c4

I'm using kernel version 2.4.17 under RH7.2. I originally thought It could be
related to lack of memory. 'free' returns this:

total used free shared buffers cached
Mem: 61672 54828 6844 0 1972 41060
-/+ buffers/cache: 11796 49876
Swap: 321292 0 321292

but there is plenty of swap space in there. I also tried increasing the RAM
to 128Mb but that didn't help either

Any help will be very much appreciated, and sorry if I'm making an obvious
mistake. I'm quite new to kernel programming. :)

FJ


2002-01-10 18:18:59

by Brian Gerst

[permalink] [raw]
Subject: Re: array size limit in module?

Fernando Jimenez wrote:
>
> Hi
>
> I am trying to code a simple kernel module and I have found a problem I don't
> quite understand.
>
> Here is the offending part of code:
>
> char *sectors_array = NULL;
> ........
> secs_size=131072;
> sectors_array = kmalloc(secs_size*sizeof(char), GFP_KERNEL);
> for(i=0; i<secs_size; i++) {
> sectors_array[i]=0;
> }
>
> This bit of code, as it is, works fine. However, if I increment secs_size by
> one, ie, I do 'secs_size=131073;' instead of 131072, I get the following:

Use vmalloc for allocations that large, unless you must have the memory
physically contiguous. 128k is the largest amount of memory you can
allocate with kmalloc.

--

Brian Gerst

2002-01-10 18:45:39

by Benjamin LaHaise

[permalink] [raw]
Subject: Re: array size limit in module?

On Thu, Jan 10, 2002 at 01:18:35PM -0500, Brian Gerst wrote:
> Use vmalloc for allocations that large, unless you must have the memory
> physically contiguous. 128k is the largest amount of memory you can
> allocate with kmalloc.

Even before he uses vmalloc, he should take a programming 101 course that
explains the importance of error checking. kmalloc returned NULL, the
code didn't handle the case. Switching to vmalloc will only make the code
usually work, but still susceptible to crashing.

-ben

Subject: Re: array size limit in module?

> Here is the offending part of code:
>
> char *sectors_array = NULL;
> ........
> secs_size=131072;
> sectors_array = kmalloc(secs_size*sizeof(char), GFP_KERNEL);

<===== missing check: if (!sectors_array) ....

> for(i=0; i<secs_size; i++) {
> sectors_array[i]=0;

You appear to be missing something that checks for
(even transient) out of memory conditions.

kmalloc() has an internal sensible limit to
allocations of 128Mb (see mm/slab.c, cache_sizes
array). It BUG()s if >128Mb is asked for.
You can get more with __get_free_pages()
and/or vmalloc().

In any case, kmalloc has to allocate contiguous
pages, whilst there may be 4 pages free, there may not be
4 contiguous pages free. This aside, kmalloc()
may /still/ fail.

However, if you are reading sectors probably
wise to group them by page and allocate
each page separately.

--
Alex Bligh