| [Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index] |
I just found a paper "How NOT to write kernel drivers". In
section 2.2, some ogfs code is used as a bad example:
--
Most tutorials warn about using kmalloc for al-locating big areas of
memory, and it seems a lot of people also notice these limits in
practice:
static __inline void *
osi_malloc(unsigned int size)
{
if (size > 2*PAGE_SIZE)
return __vmalloc(size, GFP_NOFS|__GFP_HIGHMEM, PAGE_KERNEL);
return kmalloc(size, GFP_NOFS);
}
[due to cut and paste problems with acroread, I inserted the code
from the ogfs cvs]
from OpenGFS 0.99.2 There are several problems with such an approach.
The first one is performance: vmalloc() ed memory requires more TLBs,
which are a rare resource on most CPUs. A more serious problem arises
when the osi_free() routine gets involved: it calls vfree() for the
vmalloc()ed memory, and it s illegal to call vfree() from inter-rupt
context. This abstraction makes it okay to call the osi_free()
function from interrupt context sometimes, but not for big
allocations. A third issue that most tutorials don t mention is that
while you can use vmalloc() to allo-cate larger chunks of memory, the
total amount you can allocate is rather limited, in the or-der of 64
Megabytes on modern machines (de-pending on what PCI hardware is
present).
--
And indeed this code ise still in the sources
(include/osi_linux_2_4.h). I'm not deep enough into writing
kernel code to figure out what the correct fix is. Does the
author suggest to always use kmalloc instead of vmalloc?
Ciao
Dominik ^_^ ^_^
-------------------------------------------------------
This sf.net email is sponsored by:ThinkGeek
Welcome to geek heaven.
http://thinkgeek.com/sf
_______________________________________________
Opengfs-devel mailing list
Opengfs-devel@lists.sourceforge.net
https://lists.sourceforge.net/lists/listinfo/opengfs-devel