Re: [patch] corrupt memstat_kvm_malloc(3) output and dtrace

On May 12, 2009, at 10:28 AM, John Baldwin wrote:
> On Wednesday 06 May 2009 1:00:09 am Ben Kelly wrote:
>> On May 5, 2009, at 10:18 AM, Ben Kelly wrote:
>>> While debugging a problem recently with Alexander Leidinger we
>>> noticed that crashinfo(8) was producing corrupt vmstat -m output.
>>> After doing some digging it appears that memstat_kvm_malloc(3) might
>>> have been broken by this commit:
>>>
>>> http://svn.freebsd.org/viewvc/base?view=revision&revision=179222
>>>
>>> The problem is that memstat_kvm_malloc(3) assumes that
>>> malloc_type_internal starts with an array of malloc_types_stats
>>> structures.  This assumption is no longer true, though, as
>>> mti_probes was inserted at the start of the structure.
>>>
>>> It appears that this (untested) patch might fix the problem:
>>>
>>> http://www.wanderview.com/svn/public/misc/zfs/vmstat_kvm_malloc.diff
>>>
>>> I'm not very familiar with dtrace, though.  Does anyone know if this
>>> would cause problems?
>>
>> Just FYI, I was able to recompile and test this patch tonight.  It
>> seems to have fixed vmstat -M $core -m output on my machine.
>
> I think that it would be better to update memstat_kvm_malloc(3)  
> though.  I'm
> also curious why libmemstat isn't just using malloc_types_internal  
> directly.

I've updated the patch so that  memstat_kvm_malloc(3) uses the  
malloc_type_internal definition to find the start of the statistics  
array.  This also appears to fix the problem on my server.

Thanks.

- Ben