msk0: interrupt storm

[John Baldwin]

On Tuesday, April 24, 2012 4:07:19 pm YongHyeon PYUN wrote:
> On Mon, Apr 23, 2012 at 10:24:41AM -0400, John Baldwin wrote:
> > On Wednesday, March 07, 2012 3:40:53 pm YongHyeon PYUN wrote:
> > > On Tue, Mar 06, 2012 at 10:36:05AM -0500, John Baldwin wrote:
> > > > On Thursday, March 01, 2012 8:29:55 pm YongHyeon PYUN wrote:
> > > > > On Wed, Feb 29, 2012 at 01:03:29AM +0400, Pavel Gorshkov wrote:
> > > > > > My laptop running 9.0-RELEASE/amd64/GENERIC freezes and
> > > > > > (sometimes) unfreezes intermittently, logging the following:
> > > > > > 
> > > > > > Feb 28 23:07:36 lifebook kernel: interrupt storm detected on 
"irq259:"; 
> > > > throttling interrupt source
> > > > > > 
> > > > > > $ vmstat -i
> > > > > > ...
> > > > > > irq259: mskc0                   11669511       3456
> > > > > > 
> > > > > > 
> > > > > > Looks very similar to this:
> > > > > > http://www.freebsd.org/cgi/query-pr.cgi?pr=164569
> > > > > > 
> > > > > > Any suggestions?
> > > > > 
> > > > > Try disabling MSI and see whether that makes any difference.
> > > > 
> > > > I also get interrupt storms with msk.  They do fix themselves when 
they 
> > > > happen, and I've seen it happen with the machine is idle.  This is on 
my 
> > > > little netbook where msk had several problems initially that have 
since been 
> > > > fixed.
> > > > 
> > > > mskc0: <Marvell Yukon 88E8072 Gigabit Ethernet> port 0x2000-0x20ff mem 
> > > > 0xe0000000-0xe0003fff irq 19 at device 0.0 on pci32
> > > > msk0: <Marvell Technology Group Ltd. Yukon EX Id 0xb5 Rev 0x02> on 
mskc0
> > > > msk0: Ethernet address: 00:24:81:40:e3:ef
> > > > miibus0: <MII bus> on msk0
> > > > e1000phy0: <Marvell 88E1149 Gigabit PHY> PHY 0 on miibus0
> > > > e1000phy0:  none, 10baseT, 10baseT-FDX, 100baseTX, 100baseTX-FDX, 
1000baseT, 
> > > > 1000baseT-master, 1000baseT-FDX, 1000baseT-FDX-master, auto, auto-flow
> > > > 
> > > > mskc0 at pci0:32:0:0:      class=0x020000 card=0x3056103c chip=0x436c11ab 
> > > > rev=0x10 hdr=0x00
> > > >     vendor     = 'Marvell Technology Group Ltd.'
> > > >     device     = '88E8072 PCI-E Gigabit Ethernet Controller'
> > > >     class      = network
> > > >     subclass   = ethernet
> > > > 
> > > 
> > > John, can you let me know the value of B0_Y2_SP_ISRC2 register in
> > > interrupt handler when you see the interrupt storm?
> > 
> > I finally tested this.  I added some KTR traces to dump ISRC2 on each
> > call to msk_intr() and hacked the interrupt thread code to turn KTR 
tracing
> > off when a storm occurred.  The traces look like this:
> > 
> > index  cpu timestamp        trace
> > ------ --- ---------------- ----- 
> >    148   0  111662766108828 msk_intr: B0_Y2_SP_ISRC2 = 0x44000000
> >    147   0  111662765994576 msk_intr: B0_Y2_SP_ISRC2 = 0x44000000
> >    146   0  111662765380260 msk_intr: B0_Y2_SP_ISRC2 = 0x44000000
> >    145   0  111662765257308 msk_intr: B0_Y2_SP_ISRC2 = 0x44000000
> >    144   0  111662765134356 msk_intr: B0_Y2_SP_ISRC2 = 0x44000000
> >    143   0  111662765011560 msk_intr: B0_Y2_SP_ISRC2 = 0x44000000
> >    142   0  111662764888656 msk_intr: B0_Y2_SP_ISRC2 = 0x44000000
> >    141   0  111662764773924 msk_intr: B0_Y2_SP_ISRC2 = 0x44000000
> >    140   0  111662764659360 msk_intr: B0_Y2_SP_ISRC2 = 0x44000000
> >    139   0  111662764528140 msk_intr: B0_Y2_SP_ISRC2 = 0x44000000
> >    138   0  111662764413576 msk_intr: B0_Y2_SP_ISRC2 = 0x44000000
> >    137   0  111662764287852 msk_intr: B0_Y2_SP_ISRC2 = 0x44000000
> > ...
> > 
> > (All traces have the same register value.)  The TSC on this netbook runs
> > at machdep.tsc_freq: 1596035244
> > 
> > (The timestamps above are TSC values.)
> > 
> > Let me know if you'd like me to log more stuff in the driver.  Thanks!
> 
>  wonder why the deivce gets TWSI completion interrupt since the
> driver does not monitor temperature sensor. In addition, the
> interrupt was already disabled so have no idea how this can happen.
> Here, I assume your controller implemented optional temperature
> sensor and it is monitored by H/W.
> Anyway, try attached patch and let me know whether it makes any
> difference.

It does fix the interrupt storms.  I added a debugging printf to fire each 
time msk_intr() sees this bit to see if it storms, etc.  What I see is that 
each time I would previously get a single printf reporting an interrupt storm, 
I now get a single printf reporting that the TWSI_RDY bit was set.

-- 
John Baldwin