Re: [ANNOUNCE] mdb-2.6.27-rc2-ia32-08-07-08.patch

[jmerkey@wolfmountaingroup.com]

> Stefan Richter wrote:
>> jmerkey@wolfmountaingroup.com wrote:
>>> ftp://ftp.wolfmountaingroup.org/pub/mdb/mdb-2.6.27-rc2-ia32-08-07-08.patch
> ...
>>> +typedef struct _RLOCK
>>> +{
>>> +#if defined(CONFIG_SMP)
>>> +    spinlock_t lock;
>>> +#endif
>>> +    unsigned long flags;
>>> +    unsigned long processor;
>>> +    unsigned long count;
>>> +} rlock_t;
> ...
>> The accessors rspin_lock() and rspin_try_lock() peek into spinlock_t and
>> may therefore not be fully portable.  Also, they and rspin_unlock()
>> don't look SMP safe:
> ...
>>> +   if (rlock->lock.raw_lock.slock && rlock->processor == proc)
>
> Correction:  They _are_ not portable.  Look at the first hit in
> http://lxr.linux.no/linux+v2.6.26/+code=raw_spinlock_t .


OK.  One more to fix.  I have to be able to take nested page faults inside
the debugger in the event I am debugging a page fault handler or paging
event and someone's working set routine crashes.

rspin locks are for these types of cases -- so if I fault on the same
processor I took the lock on it just bumps a counter -- yes, it is atomic
and SMP safe to do it this way.  You are correct that its non-portable
but the file this is in is named mdb-IA32 and is not meant as portable to
anything other than intel anyway.

As for coding style and misuse of a raw lock structure, I agree with you
completely, and I will look for a better way to do this without exposing
this structure.

Jeff

> --
> Stefan Richter
> -=====-==--- =--- --===
> http://arcgraph.de/sr/
>