bug 193400: sfbuf changes on mips32 may have exposed some vm/pmap issues on mips32?

[Adrian Chadd]

On 6 September 2014 15:28, Alan Cox <alc at rice.edu> wrote:
> On 09/06/2014 17:24, Adrian Chadd wrote:
>> Hi!
>>
>> This bug:
>>
>> https://bugs.freebsd.org/bugzilla/show_bug.cgi?id=193400
>>
>> is what I reported to alc a couple weeks ago as breaking on my 128mb
>> RAM mipsbe board.
>>
>> It turns out that the difference between the MIPS32 sfbuf code and the
>> new stuff gleb added was the refcounting side of things. With the MIPS
>> code, there was no refcounting at all - each sfbuf allocation would
>> allocate a new sfbuf. Now, I don't know how correct that is (my
>> reading of the code is "not correct!") but it wasn't panicing things.
>>
>> So I think maybe gleb's patch to sfbuf exposed a MIPS32 pmap/VM bug?
>>
>> Alan - how would we go about figuring out what could be the issue here?
>
> I just sent you an email a few minutes ago describing how to restore the
> old behavior.
>

Yup. I've also implemented the local sf_buf map and unmap routines; it
seems to fix things.


-a