Re: Protocol not attached

[Arnaldo Carvalho de Melo <acme@redhat.com>]

Em Thu, Nov 06, 2008 at 04:20:48PM +0100, Gerrit Renker escreveu:
> | > is there any news regarding the CCID/module loading problem on the 64
> | > bit computer with 32 bit kernel?
> | >
> <snip>
> | Hello Gerrit,
> |     yes! I think that we found the commit that inserted the problem (probably).
> | 
> |     This is the list of commits, the first time that the problem
> | occurs is in "f76fd327a8b32d3ad5b51639faf6f54d18be0981". I'm sending
> | to you the tree commits before the BAD that it is working properly.
> | 
> |  - BAD - Gerrit Renker f76fd327a8b32d3ad5b51639faf6f54d18be0981 dccp
> Ah that would explain it. Did you check the logs - if the not-loading
> was due to low timer resolution, then there should be a message like
> "Timer to coarse (xxx usec), need 10usec".
> 
> If that is the case, 
>     1. does the problem disappear when commenting out 
>        'return -ESOCKTNOSUPPORT' in  ccid3_module_init() below or if
>     2. the kernel is compiled with support for high-resolution timers?        
> 
> 
> +       if (tp.tv_sec || tp.tv_nsec > DCCP_TIME_RESOLUTION * NSEC_PER_USEC) {
> +               printk(KERN_ERR "%s: Timer too coarse (%ld usec), need %u-usec"
> +                      " resolution - check your clocksource.\n", __func__,
> +                      tp.tv_nsec/NSEC_PER_USEC, DCCP_TIME_RESOLUTION);
> +               return -ESOCKTNOSUPPORT;
> +       }
> 
> Thanks a lot for following this up, this now looks a much more likely cause.

Makes complete sense, to finish this, Leandro, can you please tell us
what is the value of CONFIG_HZ in your config? 

- Arnaldo