svn commit: r289279 - in head/sys: kern vm

[Jeff Roberson]

On Sun, 1 Nov 2015, NGie Cooper wrote:

>
>> On Nov 1, 2015, at 19:20, Adrian Chadd <adrian.chadd at gmail.com> wrote:
>>
>> hiya jeff,
>>
>> this broke low-memory, no-swap boards (eg MIPS.)
>>
>> On a MIPS board (carambola2) with 32MB of RAM, just scp'ing a kernel
>> into the rootfs on USB hangs the system. After doing some digging, I
>> found this:
>>
>>
>> INTERNAL: Allocating one item from buf free cache(0x83fea7e0)
>> uma_zalloc_arg: Bucketzone returned NULL
>> INTERNAL: Allocating one item from buf free cache(0x83fea7e0)
>> uma_zalloc_arg: Bucketzone returned NULL
>>
>> .. and it was just stuck in a loop trying to allocate them, failing,
>> and trying to allocate them again.
>>
>> I'll see if I can reproduce it with a qemu emulator with sufficiently
>> low RAM so you don't need a MIPS router to reproduce it.
>>
>> It's sufficient to just start the scp; it runs out of RAM within a
>> couple of seconds.
>>
>> Any ideas?
>
> What happens if you change vfs.maxbufspace ?
>
> The reason that I?m noting is that (if I?m reading the code correctly), it?s now allocating 16 clean queues instead of 1 and each is up to vfs.maxbufspace size, which is 256MB per queue based on this line:
>
> + clean_queues = MIN(howmany(maxbufspace, 256*1024*1024), CLEAN_QUEUES);
>
> The 256MB amount seems like it should be a tunable, as well as the CLEAN_QUEUES #define? it?s a bit high for low memory platforms, i.e. platforms with <372MB of scratch space to play around with I?m guessing?

maxbufspace is still a global.  It just makes N queues for every 256MB of 
configured buffer cache size.

We shouldn't be looping forever.  It should sleep and try to re-allocate. 
How many cpus are on the mips board?  We may need to implement something 
to directly flush the buf zone and disable the per-cpu caches when 
allocation fails.

Jeff

>
> Thanks,
> -NGie