Re: Possible regression: MSI vector leakage since 2.6.18-rc5ish (Unable to repeatedly allocate/free MSI interrupt)

[Auke Kok <auke-jan.h.kok@intel.com>]

Eric W. Biederman wrote:
> Auke Kok <auke-jan.h.kok@intel.com> writes:
> 
>> I highly doubt it - I've seen the problem even on this weeks git on
>> x86_64. Moreover, I'm at home for the weekend and testing resources are limited
>> :). I'll see what I can do
> 
> Thanks.  There may be more to it than what I suspect, but I could not
> reproduce it on x86_64.
> 
> Now I may have missed something as I optimized my tested based on the fact
> that close and open are triggered when you up and down a network interface.
> so I didn't do a complete rmmod, (since my network driver wasn't modular).
> 
> Since you have seen this on x86_64 I will look deeper.  

gah, strike that.

my only x86_64 system here survived the test with latest git tree.

my 386 system here has no msi devices and I can't reinstall my x86_64 system 
since it's headless, so I can't test anything until monday. I'll give it a full 
test again and see which 2.6.20rc kernels did fail, most likely a much older 
tree (I suspect).

Auke