Is this -HEAD? Did you define MALLOC_PRODUCTION?
On 18 September 2012 07:11, Johann Hugo <jh...@meraka.csir.co.za> wrote:
> On Wednesday 05 September 2012 01:20:10 Adrian Chadd wrote:
>> On 28 August 2012 21:00, Adrian Chadd <adr...@freebsd.org> wrote:
>> > The default (most things as modules) kernel build is still ~ 3.5mb.
>> > That's rather large for a board with 16MB of RAM.
>> > Additionally, I can't load the ath device after the system has come up
>> > single user - the driver fails to allocate descriptors/buffer entries.
>> > I'll do some digging about it in a couple of weeks. I'd like some
>> > pointers on how to dump the current state of memory so I can see
>> > exactly what's going on.
>> > For reference, 'sysctl vm.vmtotal' shows the total free RAM is around
>> > 1 to 1.5MB. I'm guessing some are allocated to buffers but I'd like to
>> > see what else is actually holding onto memory..
>> It turns out memory fragmentation due to a huge (for embedded)
>> userland are to blame.
>> I'll see if I can get ray@ to help me port phkmalloc to -HEAD again
>> and try to tweak the default allocator parameters to not pre-allocate
>> large pools for each process.
>> Seeing login, sh, etc have an RSS of 1MB is kind of scary.
>> I'll do some more digging with procstat (when I fit it into the
>> current image) and report back.
>> There's also the problem of a 3.5MB "stripped down nothing but what's
>> needed to boot and read the rootfs from SPI flash + FFS" kernel size
>> problem. That seems slightly(!) too big..
> Hi Adrian
> Did you manage make any progress with this ?
> I'm also seeing the same symptoms on our ARM boards.
> PID USERNAME THR PRI NICE SIZE RES STATE TIME WCPU
> 45 root 1 40 0 10936K 10044K RUN 0:00
> 6.46% top
> 39 root 1 8 0 10416K 10080K wait 0:00
> 0.24% sh
> email@example.com mailing list
> To unsubscribe, send any mail to "freebsd-wireless-unsubscr...@freebsd.org"
firstname.lastname@example.org mailing list
To unsubscribe, send any mail to "freebsd-wireless-unsubscr...@freebsd.org"