CAMBRIA and more than one atheros card
Adrian Chadd
adrian at freebsd.org
Mon Jul 14 19:55:20 UTC 2014
Right. What's the output of hw.ath ?
-a
On 14 July 2014 12:49, John Hay <jhay at meraka.org.za> wrote:
> On Mon, Jul 14, 2014 at 12:23:47PM -0700, Adrian Chadd wrote:
>> Ah, you're going to need more than 64 bounce pages. That's just not
>> enough for all the mbufs the 11n support requires.
>>
>> It likely didn't show up in previous incarnations of this because
>> pre-11n support had a much smaller pool of buffers in use.
>>
>> For doing 11n, there's hm, 256 TX and 256 RX buffers for each NIC? Or
>> is it 512 and 512? It's something large like that. So assuming 512,
>> that's 1024 * 4096 bytes each, so 4mbyte per NIC needed of bounce
>> buffers.
>>
>> You could tweak ATH_TXBUF and ATH_RXBUF down to something like 128 TX
>> and 128 RX, but you can't go much lower than that per NIC as an A-MPDU
>> aggregate requires up to 64 buffers to transmit and/or receive and if
>> you're not careful you'll actually overrun the RX FIFO and/or starve
>> the TX code from having mbufs available to transmit with.
>
> I don't mind up the bounce buffers, if that will help. I'll see how
> high I can go before it explode. :-) I guess one cannot easily use
> the RAM above 64M for ram disks and code and then not bouncing at
> all. :-/
>
> I just had an ok boot, ie. all 3 aths came up without an error. So
> just after a login, the bounce stats looked like this:
>
> tst-11-arm:~ # sysctl hw.busdma
> hw.busdma.total_bpages: 64
> hw.busdma.zone0.total_bpages: 64
> hw.busdma.zone0.free_bpages: 57
> hw.busdma.zone0.reserved_bpages: 0
> hw.busdma.zone0.active_bpages: 7
> hw.busdma.zone0.total_bounced: 525
> hw.busdma.zone0.total_deferred: 0
> hw.busdma.zone0.lowaddr: 0x4000fff
> hw.busdma.zone0.alignment: 4096
>
> The usage seems to slowly climb though. After 30 minutes it looks like
> this:
>
> tst-11-arm:~ # sysctl hw.busdma
> hw.busdma.total_bpages: 64
> hw.busdma.zone0.total_bpages: 64
> hw.busdma.zone0.free_bpages: 43
> hw.busdma.zone0.reserved_bpages: 0
> hw.busdma.zone0.active_bpages: 21
> hw.busdma.zone0.total_bounced: 4713
> hw.busdma.zone0.total_deferred: 0
> hw.busdma.zone0.lowaddr: 0x4000fff
> hw.busdma.zone0.alignment: 4096
>
> The board is pretty idle, with no routing daemon or traffic except for
> the occasional broadcast and multicasts that arrive.
>
> John
>
>>
>>
>> -a
>>
>>
>> On 14 July 2014 12:14, John Hay <jhay at meraka.org.za> wrote:
>> > On Mon, Jul 14, 2014 at 09:09:46PM +0200, John Hay wrote:
>> >> On Mon, Jul 14, 2014 at 12:06:34PM -0700, Adrian Chadd wrote:
>> >> > .. why's it need bounce buffers?
>> >>
>> >> I found this:
>> >> arm/xscale/ixp425/ixp425_pci.c: /* NB: PCI dma window is 64M so anything above must be bounced */
>> >
>> > I found a sysctl to show some bounce info. :-)
>> >
>> > On the boot that ath1 and ath2 failed during ifconfig and ath0 started
>> > to complain after a while:
>> >
>> > #####################
>> > tst-11-arm:~ # sysctl hw.busdma
>> > hw.busdma.total_bpages: 64
>> > hw.busdma.zone0.total_bpages: 64
>> > hw.busdma.zone0.free_bpages: 0
>> > hw.busdma.zone0.reserved_bpages: 0
>> > hw.busdma.zone0.active_bpages: 64
>> > hw.busdma.zone0.total_bounced: 1191
>> > hw.busdma.zone0.total_deferred: 0
>> > hw.busdma.zone0.lowaddr: 0x4000fff
>> > hw.busdma.zone0.alignment: 4096
>> > #####################
>> >
>> > I then thought to down wlan0 to see if the page count changed, but
>> > do not know. :-)
>> >
>> > #####################
>> > tst-11-arm:~ # ifconfig wlan0 down
>> > dev.ath.0.debug: 0 -> 8
>> > /sbin/ifconfig.bin wlan0 down
>> > ath0: ath_stop_locked: invalid 0 if_flags 0x8802
>> > Sleeping thread (tid 100019, pid 0) owns a non-sleepable lock
>> > panic: sleeping thread
>> > Uptime: 32m46s
>> > Automatic reboot in 15 seconds - press a key on the console to abort
>> > Rebooting...
>> > #####################
>> >
>> > John
>> >
>> >>
>> >> John
>> >> >
>> >> >
>> >> > -a
>> >> >
>> >> >
>> >> > On 14 July 2014 11:42, John Hay <jhay at meraka.org.za> wrote:
>> >> > > On Mon, Jul 14, 2014 at 09:48:56AM -0700, Adrian Chadd wrote:
>> >> > >> Hm, it could be some bus related stupidity. It's allocating 2k or 4k
>> >> > >> mbufs for the receive path because wifi frames are bigger than
>> >> > >> ethernet frames. If you're not seeing failures in 4k mbuf allocations
>> >> > >> then I'm not sure what it could be.
>> >> > >>
>> >> > >> I'll see about firing it up locally and checking.
>> >> > >
>> >> > > I'm hunting a bit more and it looks like the fail is in the bounce pages.
>> >> > > It looks like the calls look like this:
>> >> > >
>> >> > > ath_legacy_rxbuf_init()
>> >> > > bus_dmamap_load_mbuf_sg()
>> >> > > _bus_dmamap_load_buffer()
>> >> > > _bus_dmamap_reserve_pages()
>> >> > > reserve_bounce_pages()
>> >> > >
>> >> > > I have added a printf at the start of alloc_bounce_pages() and it seems
>> >> > > that it is only called (twice) when ath0 is probed. Is bus_dma_tag_t dmat,
>> >> > > the first argument to alloc_bounce_pages(), common for the whole pci bus?
>> >> > >
>> >> > > The start of the boot, with my printf looks like this:
>> >> > >
>> >> > > ####################
>> >> > > FreeBSD ARM (Gateworks Cambria) boot2 v0.4
>> >> > > -
>> >> > > Default: /boot/kernel/kernel
>> >> > > boot:
>> >> > > Copyright (c) 1992-2014 The FreeBSD Project.
>> >> > > Copyright (c) 1979, 1980, 1983, 1986, 1988, 1989, 1991, 1992, 1993, 1994
>> >> > > The Regents of the University of California. All rights reserved.
>> >> > > FreeBSD is a registered trademark of The FreeBSD Foundation.
>> >> > > FreeBSD 11.0-CURRENT #9 r268502M: Mon Jul 14 15:30:15 SAST 2014
>> >> > > jhay at dolphin.meraka.csir.co.za:/usr/obj/arm.armeb/snaps/arm/11-tst/src/sys/S
>> >> > > MALL-CAMBRIA arm
>> >> > > gcc version 4.2.1 20070831 patched [FreeBSD]
>> >> > > CPU: IXP435 rev 1 (ARMv5TE) (XScale core)
>> >> > > Big-endian DC enabled IC enabled WB enabled LABT branch prediction enabled
>> >> > > 32KB/32B 32-way instruction cache
>> >> > > 32KB/32B 32-way write-back-locking data cache
>> >> > > real memory = 134213632 (127 MB)
>> >> > > avail memory = 124440576 (118 MB)
>> >> > > random device not loaded; using insecure entropy
>> >> > > wlan: mac acl policy registered
>> >> > > random: <Software, Yarrow> initialized
>> >> > > ixp0: <Intel IXP4XX>
>> >> > > ixp0: 37e7f<RCOMP,USB,HASH,AES,DES,HDLC,AAL,ETH0,ETH1,PCI>
>> >> > > pcib0: <IXP4XX PCI Bus> on ixp0
>> >> > > pci0: <PCI bus> on pcib0
>> >> > > ath0: <Atheros 9220> irq 28 at device 1.0 on pci0
>> >> > > alloc_bounce_pages: numpages 63
>> >> > > [ath] enabling AN_TOP2_FIXUP
>> >> > > alloc_bounce_pages: numpages 1
>> >> > > ath0: AR9220 mac 128.2 RF5133 phy 13.0
>> >> > > ath0: 2GHz radio: 0x0000; 5GHz radio: 0x00c0
>> >> > > ath1: <Atheros 9220> irq 27 at device 2.0 on pci0
>> >> > > [ath] enabling AN_TOP2_FIXUP
>> >> > > ath1: AR9220 mac 128.2 RF5133 phy 13.0
>> >> > > ath1: 2GHz radio: 0x0000; 5GHz radio: 0x00c0
>> >> > > ath2: <Atheros 5413> irq 26 at device 3.0 on pci0
>> >> > > ath2: AR5413 mac 10.5 RF5413 phy 6.1
>> >> > > ath2: 2GHz radio: 0x0000; 5GHz radio: 0x0063
>> >> > > ixpclk0: <IXP4XX Timer> on ixp0
>> >> > > ####################
>> >> > >
>> >> > > Interesting, with this boot, with the new kernel, the aths did not fail.
>> >> > > Could the printf have changed something or was it just coincidence? With
>> >> > > a reboot ath1 and ath2 failed again during configuration and a little
>> >> > > while later ath0 started to complain:
>> >> > >
>> >> > > ath0: ath_rx_proc: no mbuf!
>> >> > >
>> >> > > John
>> >> > >
>> >> > >>
>> >> > >>
>> >> > >>
>> >> > >>
>> >> > >> -a
>> >> > >>
>> >> > >> On 14 July 2014 01:07, John Hay <jhay at meraka.org.za> wrote:
>> >> > >> > On Thu, Jul 10, 2014 at 06:02:27AM +0200, John Hay wrote:
>> >> > >> >> On Wed, Jul 09, 2014 at 06:46:12AM +0200, John Hay wrote:
>> >> > >> >> > Hi Guys,
>> >> > >> >> >
>> >> > >> >> > The problem is back / still there. I initially saw the problem during
>> >> > >> >> > boot, with the interface configs in rc.conf, but because it is so
>> >> > >> >> > mixed with the rest, I took it out and put it in the script, then
>> >> > >> >> > after multi-user boot was finished, I did a login and ran the script,
>> >> > >> >> > with the output I showed in the initial post.
>> >> > >> >> >
>> >> > >> >> > So I put the interface configs back into rc.conf and I'm seeing the
>> >> > >> >> > same problem, here is a cut during boot:
>> >> > >> >> >
>> >> > >> >> > ###############
>> >> > >> >> > Starting file system checks:
>> >> > >> >> > /dev/ad0s1a: FILE SYSTEM CLEAN; SKIPPING CHECKS
>> >> > >> >> > /dev/ad0s1a: clean, 93385 free (33 frags, 11669 blocks, 0.0% fragmentation)
>> >> > >> >> > Mounting local file systems:.
>> >> > >> >> > Writing entropy file:.
>> >> > >> >> > Setting hostname: tst-cambria-11.
>> >> > >> >> > wlan0: Ethernet address: 00:21:a4:35:70:42
>> >> > >> >> > wlan1: Ethernet address: 00:21:a4:35:6c:96
>> >> > >> >> > ath1: unable to start recv logic
>> >> > >> >> > wlan2: Ethernet address: 00:21:a4:32:38:c2
>> >> > >> >> > ath2: unable to start recv logic
>> >> > >> >> > ###############
>> >> > >> >> >
>> >> > >> >> > Looking at the vmstat -z output the 256 Bucket fail is much higher than
>> >> > >> >> > if I let it boot to multiuser and then configured the interfaces. It
>> >> > >> >> > was in the 6000, while now it is much higher:
>> >> > >> >> >
>> >> > >> >> > Without wlan configs in rc.conf, but configured afterwards:
>> >> > >> >> > 16 Bucket: 64, 0, 15, 300, 3139, 16, 0
>> >> > >> >> > 256 Bucket: 1024, 0, 31, 1, 592,6062, 0
>> >> > >> >> > vmem btag: 28, 0, 4496, 256, 4496, 32, 0
>> >> > >> >> >
>> >> > >> >> > With wlan configs in rc.conf:
>> >> > >> >> > 16 Bucket: 64, 0, 16, 299, 8611, 16, 0
>> >> > >> >> > 256 Bucket: 1024, 0, 26, 6, 773,16928, 0
>> >> > >> >> > vmem btag: 28, 0, 4405, 59, 4405, 30, 0
>> >> > >> >> >
>> >> > >> >> > Both of them boot from a ro compact-flash with md /etc and /var, but
>> >> > >> >> > they are small 4.3M and 2.1M. I have hacked in the use of tmpfs, but
>> >> > >> >> > that did not make a difference.
>> >> > >> >> >
>> >> > >> >
>> >> > >> > Friday I power cycled the board and it came up without an error. All 3
>> >> > >> > atheros cards configured in rc.conf. So I left it on for the weekend
>> >> > >> > and by this morning there was still no errors, so I rebooted it and
>> >> > >> > again saw the "ath1: unable to start recv logic" message for ath1 and
>> >> > >> > ath2. I power cycled it, but still get the error, so Friday was just
>> >> > >> > lucky with some timing thing, maybe if the card receive while it is
>> >> > >> > still configuring? Also if I leave the board booted in this state,
>> >> > >> > ath0 also start to give problems:
>> >> > >> >
>> >> > >> > #################
>> >> > >> > ath0: ath_rx_proc: no mbuf!
>> >> > >> > ath0: ath_rx_proc: no mbuf!
>> >> > >> > ...
>> >> > >> > ath0: device timeout
>> >> > >> > ath0: ath_reset: unable to start recv logic
>> >> > >> > ...
>> >> > >> > #################
>> >> > >> >
>> >> > >> > In anycase it seems that memory allocation problem. How do I figure out
>> >> > >> > where it is? "netstat -m" does not seem to point to an error. The mbuf
>> >> > >> > info in vmstat -z also look ok. It is only "256 Bucket" in vmstat -z
>> >> > >> > that point to an alloc failure. How do I figure out where that is and
>> >> > >> > how do I fix it or work around it?
>> >> > >> >
>> >> > >> > ###################
>> >> > >> > tst-11-arm:~ # netstat -m
>> >> > >> > 128/382/510 mbufs in use (current/cache/total)
>> >> > >> > 127/129/256/7654 mbuf clusters in use (current/cache/total/max)
>> >> > >> > 127/126 mbuf+clusters out of packet secondary zone in use (current/cache)
>> >> > >> > 0/0/0/3827 4k (page size) jumbo clusters in use (current/cache/total/max)
>> >> > >> > 0/0/0/1134 9k jumbo clusters in use (current/cache/total/max)
>> >> > >> > 0/0/0/637 16k jumbo clusters in use (current/cache/total/max)
>> >> > >> > 286K/353K/639K bytes allocated to network (current/cache/total)
>> >> > >> > 0/0/0 requests for mbufs denied (mbufs/clusters/mbuf+clusters)
>> >> > >> > 0/0/0 requests for mbufs delayed (mbufs/clusters/mbuf+clusters)
>> >> > >> > 0/0/0 requests for jumbo clusters delayed (4k/9k/16k)
>> >> > >> > 0/0/0 requests for jumbo clusters denied (4k/9k/16k)
>> >> > >> > 0/3/1488 sfbufs in use (current/peak/max)
>> >> > >> > 0 requests for sfbufs denied
>> >> > >> > 0 requests for sfbufs delayed
>> >> > >> > 0 requests for I/O initiated by sendfile
>> >> > >> > tst-11-arm:~ # vmstat -z | grep mbuf
>> >> > >> > mbuf_packet: 256, 48990, 127, 126, 2294, 0, 0
>> >> > >> > mbuf: 256, 48990, 1, 256, 920, 0, 0
>> >> > >> > mbuf_cluster: 2048, 7654, 253, 3, 253, 0, 0
>> >> > >> > mbuf_jumbo_page: 4096, 3827, 0, 0, 0, 0, 0
>> >> > >> > mbuf_jumbo_9k: 9216, 1134, 0, 0, 0, 0, 0
>> >> > >> > mbuf_jumbo_16k: 16384, 637, 0, 0, 0, 0, 0
>> >> > >> > mbuf_ext_refcnt: 4, 0, 0, 504, 1, 0, 0
>> >> > >> > tst-11-arm:~ # vmstat -z | grep Bucket
>> >> > >> > 4 Bucket: 16, 0, 7, 497, 1565, 0, 0
>> >> > >> > 6 Bucket: 24, 0, 0, 0, 0, 0, 0
>> >> > >> > 8 Bucket: 32, 0, 3, 375, 83, 0, 0
>> >> > >> > 12 Bucket: 48, 0, 2, 334, 5, 0, 0
>> >> > >> > 16 Bucket: 64, 0, 14, 301, 43687, 16, 0
>> >> > >> > 32 Bucket: 128, 0, 7, 148, 196, 0, 0
>> >> > >> > 64 Bucket: 256, 0, 19, 56, 77, 0, 0
>> >> > >> > 128 Bucket: 512, 0, 16, 16, 48, 0, 0
>> >> > >> > 256 Bucket: 1024, 0, 29, 3, 1521,86738, 0
>> >> > >> > tst-11-arm:~ #
>> >> > >> > ###################
>> >> > >> >
>> >> > >> > Regards
>> >> > >> >
>> >> > >> > John
>> >> > >> > --
>> >> > >> > John Hay -- jhay at meraka.csir.co.za / jhay at meraka.org.za / jhay at FreeBSD.org
>> >> _______________________________________________
>> >> freebsd-wireless at freebsd.org mailing list
>> >> http://lists.freebsd.org/mailman/listinfo/freebsd-wireless
>> >> To unsubscribe, send any mail to "freebsd-wireless-unsubscribe at freebsd.org"
More information about the freebsd-wireless
mailing list