Re: lwp resource limit

[Matt Thomas <matt%3am-software.com@localhost>]

On Jun 8, 2012, at 1:04 PM, Christos Zoulas wrote:

> In article <A4F96B90-D2E7-4F98-9B44-A040DED326F6%3am-software.com@localhost>,
> Matt Thomas  <matt%3am-software.com@localhost> wrote:
>> 
>> On Jun 8, 2012, at 9:19 AM, Christos Zoulas wrote:
>> 
>>> On Jun 8,  9:00am, matt%3am-software.com@localhost (Matt Thomas) wrote:
>>> -- Subject: Re: lwp resource limit
>>> 
>>> | 
>>> | On Jun 8, 2012, at 5:24 AM, Christos Zoulas wrote:
>>> | 
>>> | >> +      if (l->l_flag & LW_RESCOUNT)
>>> | >> 
>>> | >> I don't see the need for this, why not check p_nlwp == 1?
>>> | >> 
>>> | >> if this is the first lwp for the proc p_nlwp should be 0 so check
>>> | >> maxproc otherwise maxlwp.
>>> | > 
>>> | > You are forgetting compat_linux.
>>> | 
>>> | No I'm not.  I'm not using l_lid but the # of lwps in the proc.
>>> 
>>> I made the change, but now this begs the question to remove the enforce
>>> argument because I don't like the assymmetry. Should I do that too? And
>>> ignore kthreads in using a different criterion?
>> 
>> LW_SYSTEM/LW_INTR threads shouldn't be counted
> 
> New diff in http://www.netbsd.org/~christos/maxlwp.diff

Better.

Rather than have all the #ifdef __HAVE_CPU_MAXLWP how about doing

#ifndef __HAVE_CPU_MAXLWP
static inline int
cpu_maxlwp(void)
{
        return maxlwp;
}
#endif

in an appropriate header file?