Re: mongo_copy: cp: cannot stat `/mnt/testfs/testdir0-0-0/f92': Input/output error

[Carl-Daniel Hailfinger <c-d.hailfinger.kernel.2004@gmx.net>]

Hans Reiser schrieb:
> Christophe Saout wrote:
> 
>> Am Mittwoch, den 04.08.2004, 16:57 -0700 schrieb Hans Reiser:
>>
>>  
>>>>> V4 likes stack.  If someone can convince me that kmalloc is as
>>>>> efficient as stack, I'd like to hear it, because I know I lack
>>>>> expertise on the topic.
>>>>
>>>> The disassembled fast path of kmalloc (with a size of the object known
>>>> at compile time) fits on one page. Most of the time it will just return
>>>> a pointer from a pool of cached and cache-hot objects.
>>>
>>> whereas using stack is 0 instructions or?
>>
>> More or less. But who's counting? ;-)
>
> 0 is less than a page, by a lot, if done a lot, and we do.

Yes, but... (there's always a but) certain hardware drivers also use a lot
of stack. If you manage to get another (hardware) interrupt in a reiser4
path with >4k stack already used, it's very likely your machine will
explode even without 4k stacks.
And before you ask, some of these problems may be triggered intentionally.
IMHO no user should be able to trigger a kernel crash (maybe except when
writing to kernel memory).
Besides that, some time ago a discussion on linux-kernel about available
stack size without the 4KSTACKS option had very interesting results
(that's where the information in the above paragraphs came from).

Regards,
Carl-Daniel
-- 
http://www.hailfinger.org/