[Linux-ia64] do_mmap_fake question

[Linux-ia64] do_mmap_fake question

[Manfred Spraul]

do_mmap_fake seems to be buggy:

You use simple memset/memcpy to initialize user memory.
AFAICS this can oops on SMP: one thread calls do_mmap_fake, and a second
thread calls sys_munmap() for that memory range.
memset/memcpy will fail, and you don't have an exception handler
installed.

Btw, why do you use llseek?
Is there a hidden problem with read(,,,&off)?

- llseek(,,off);
- r = file->f_op->read(file,(char*)addr, len, &file->f_pos);
+ r = file->f_op->read(file,(char*)addr, len, &off);

--
	Manfred

Re: [Linux-ia64] do_mmap_fake question

[Don Dugger]

Manfred-

Thanks for looking over the code but I don't think there is a
problem.  `do_mmap_fake' is only called from `sys32_mmap' which
grabs the semaphore `current->mm->mmap_sem' (it also grabs the
kernel lock but we won't go into that).  Any other thread trying
to unmap memmory will block until `do_mmap_fake' completes.

The call to `llseek' is a hold over from my application programming
days, I'll change that in the next release.

Tnx.

On Sun, Jul 09, 2000 at 10:51:44AM +0200, Manfred Spraul wrote:
> do_mmap_fake seems to be buggy:
> 
> You use simple memset/memcpy to initialize user memory.
> AFAICS this can oops on SMP: one thread calls do_mmap_fake, and a second
> thread calls sys_munmap() for that memory range.
> memset/memcpy will fail, and you don't have an exception handler
> installed.
> 
> Btw, why do you use llseek?
> Is there a hidden problem with read(,,,&off)?
> 
> - llseek(,,off);
> - r = file->f_op->read(file,(char*)addr, len, &file->f_pos);
> + r = file->f_op->read(file,(char*)addr, len, &off);
> 
> --
> 	Manfred
> 
> 
> _______________________________________________
> Linux-IA64 mailing list
> Linux-IA64@linuxia64.org
> http://lists.linuxia64.org/lists/listinfo/linux-ia64

-- 
Don Dugger
"Censeo Toto nos in Kansa esse decisse." - D. Gale
n0ano@valinux.com
Ph: 303/938-9838