Re: [LTP] LTP - fcntl23,mincore01,madvise02 - fixes

[Gernot P. <gp...@su...>]

On Thursday 19 May 2005 08:35, pra...@wi... wrote:
> Hello,
> Under most test conditions fcntl23,mincore01,madvise02 were failing on
> ENOMEM test scenarios.

Interesting. I did not experience such problems in recent months.

> madvise02 :
> /*
>  *          TEST(madvise(file,stat.st_size + stat.st_size
> ,MADV_NORMAL));
>  *  prashant yendigeri added the below line because with above line
> ENOMEM was not generated
>  *    under most test conditions.
> */
>           TEST(madvise(file,stat.st_size * pagesize,MADV_NORMAL));

Oh, I already fixed that myself and forgot to send in the patch. ;-)

For ia64 I even use

TEST(madvise(file,stat.st_size + 5 * pagesize,MADV_NORMAL));

> mincore01 :
> /*
> * TC[2].len = global_len*5;
> * prashant yendigeri added the below line because under most test
> scenario,
> * ENOMEM was not generated
> */
>   TC[2].len = global_len*getpagesize();

This won't fix the general problem. You can't be 100% sure that you have any 
unmapped virtual memory following address global_pointer, no matter what 
value is assigned to TC[2].len.
In order to increase the probability you would have to copy the mmap/munmap 
stuff from setup2().

> /*
> * TC[3].addr = global_pointer+2*global_len;
> * prashant yendigeri added the below line because under most TEST
> scenario,
> * ENOMEM was not generated
> */
>   TC[3].addr = global_pointer+global_len * getpagesize();

Here the same applies.

What kernel version and architecture do you use?

> Regards,
> Prashant P Yendigeri

mfg
Gernot