[SOLVED] Re: Fwd: Interrupt Overload

Konstantin Belousov kostikbel at gmail.com
Sat Jun 7 20:17:27 UTC 2014


On Sat, Jun 07, 2014 at 09:33:40PM +0200, Dutch Ingraham wrote:
> 
> 
> > Sent: Saturday, June 07, 2014 at 2:41 PM
> > From: "Konstantin Belousov" <kostikbel at gmail.com>
> > To: "Dutch Ingraham" <stoa at gmx.us>
> > Cc: freebsd-hackers at freebsd.org
> > Subject: Re: Fwd: Interrupt Overload
> >
> > On Sat, Jun 07, 2014 at 08:21:08PM +0200, Dutch Ingraham wrote:
> > > 
> > > 
> > > > Sent: Saturday, June 07, 2014 at 1:57 PM
> > > > From: "Konstantin Belousov" <kostikbel at gmail.com>
> > > > To: "Dutch Ingraham" <stoa at gmx.us>
> > > > Cc: freebsd-hackers at freebsd.org
> > > > Subject: Re: Fwd: Interrupt Overload
> > > >
> > > > On Sat, Jun 07, 2014 at 12:48:16PM -0400, Dutch Ingraham wrote:
> > > > > On 06/07/2014 12:04 PM, Vladislav Prodan wrote:
> > > > > > 
> > > > > > 
> > > > > >  
> > > > > >  --- Original message ---
> > > > > >  From: "Dutch Ingraham" <stoa at gmx.us>
> > > > > >  Date: 7 June 2014, 18:33:12
> > > > > >   
> > > > > > 
> > > > > >>
> > > > > >> Thanks for the response.
> > > > > >>
> > > > > >> The output you requested:
> > > > > >>
> > > > > >> kern.eventtimer.choice: HPET1 (440) HPET2 (440) HPET3 (440) HPET4 (440)
> > > > > >> HPET5 (440) HPET6 (440) LAPIC (400) i8254 (100) RTC (0)
> > > > > >>
> > > > > >> kern.eventtimer.choice: HPET (did not specify 1, 2, etc.)
> > > > > >>
> > > > > >> I also changed the type of timer to LAPIC and rebooted; there was no
> > > > > >> appreciable change in the interrupt activity.
> > > > > > 
> > > > > > After reboot what became timer? :)
> > > > > > 
> > > > > > You can change the timer "on the fly", without rebooting the system.
> > > > > > 
> > > > > > If LAPIC does not help, then try other timers.
> > > > > > 
> > > > > > 
> > > > > > --
> > > > > >  Vladislav V. Prodan
> > > > > >  System & Network Administrator
> > > > > >  support.od.ua
> > > > > >  
> > > > > > _______________________________________________
> > > > > > freebsd-hackers at freebsd.org mailing list
> > > > > > http://lists.freebsd.org/mailman/listinfo/freebsd-hackers
> > > > > > To unsubscribe, send any mail to "freebsd-hackers-unsubscribe at freebsd.org"
> > > > > > 
> > > > > 
> > > > > You're right, it is not persistent.  I changed to each different event
> > > > > timer and the only one that made a difference was the i8254; that
> > > > > dropped the cpu load from 30% to 10-12%.  Much better, but still of
> > > > > course not acceptable for a Core II-Duo running at 3.0GHz.  The load
> > > > > averages shown in <top> do also drop proportionally.  Interestingly,
> > > > > though, <systat -vmstat> shows the same interrupt rate - 325K/sec.
> > > > > 
> > > > > What do you make of the fact that when I suspend with <<acpiconf -s 3>
> > > > > and then wake-up, everything is absolutely normal, regardless of event
> > > > > timer type?
> > > > 
> > > > You did not shown _useful_ output of vmstat -i.  Do it when the storm
> > > > occurs.  Also, show the pciconf -lvc output on the machine.
> > > > 
> > > 
> > > Sorry - I was entering that output by hand, so truncated what I thought was not useful.  
> > > In addition, the storm is always occurring, unless I put the machine to sleep and then wake-up.
> > > 
> > > Here is the full vmstat -i:
> > > 
> > > dutch:~:# vmstat -i
> > > interrupt                          total       rate
> > > irq1: atkbd0                          48          0
> > > irq0: attimer0                  12236927       1178
> > > irq8: atrtc0                      146537         14
> > > irq16: uhci0                  3362560857     323946
> > > irq18: atapci0+                    19828          1
> > > irq23: uhci3 ehci1                     2          0
> > > cpu0:timer                        163301         15
> > > irq256: hpet0:t0                 4516011        435
> > > irq257: hpet0:t1                   83960          8
> > > irq264: em0                        31799          3
> > > irq265: hdac0                         95          0
> > > irq266: ahci0:ch0                   8423          0
> > > irq267: ahci0:ch1                  15620          1
> > > cpu1:timer                          1229          0
> > > irq274: vgapci0                    10041          0
> > > Total                         3379794678     325606
> > > dutch:~:#
> > > 
> > > And here is pciconf -lvc:
> > > 
> > > dutch:~:# pciconf -lvc
> > > hostb0 at pci0:0:0:0:	class=0x060000 card=0x04201028 chip=0x2e108086 rev=0x03 hdr=0x00
> > >     vendor     = 'Intel Corporation'
> > >     device     = '4 Series Chipset DRAM Controller'
> > >     class      = bridge
> > >     subclass   = HOST-PCI
> > >     cap 09[e0] = vendor (length 12) Intel cap 6 version 1
> > > pcib1 at pci0:0:1:0:	class=0x060400 card=0x04201028 chip=0x2e118086 rev=0x03 hdr=0x01
> > >     vendor     = 'Intel Corporation'
> > >     device     = '4 Series Chipset PCI Express Root Port'
> > >     class      = bridge
> > >     subclass   = PCI-PCI
> > >     cap 0d[88] = PCI Bridge card=0x04201028
> > >     cap 01[80] = powerspec 3  supports D0 D3  current D0
> > >     cap 05[90] = MSI supports 1 message 
> > >     cap 10[a0] = PCI-Express 2 root port slot max data 128(128) link x0(x16)
> > >                  speed 0.0(5.0) ASPM disabled(L0s)
> > >     ecap 0002[100] = VC 1 max VC0
> > >     ecap 0005[140] = Root Complex Link Declaration 1
> > > vgapci0 at pci0:0:2:0:	class=0x030000 card=0x04201028 chip=0x2e128086 rev=0x03 hdr=0x00
> > >     vendor     = 'Intel Corporation'
> > >     device     = '4 Series Chipset Integrated Graphics Controller'
> > >     class      = display
> > >     subclass   = VGA
> > >     cap 05[90] = MSI supports 1 message enabled with 1 message
> > >     cap 01[d0] = powerspec 2  supports D0 D3  current D0
> > 
> > Try to set the tunable hw.drm.msi to 0 before i915 driver is loaded.
> > I.e. the easiest is to set it at loader prompt.  If you load driver
> > by starting Xorg, then kenv hw.drm.msi=0 would be enough.
> > 
> > Either way, helped it or not, post the vmstat -i output while the Xorg
> > is running.
> > 
> 
> BEAUTIFUL!!  Knocked the uhci interrupts to normal.  Here is the >vmstat -i>
> with X running and hw.drm.msi=0 set:
> 
> dutch at dutch:~ % vmstat -i
> interrupt                          total       rate
> irq1: atkbd0                          48          0
> irq16: uhci0+                        167          1
> irq18: atapci0+                      344          3
> irq23: uhci3 ehci1                     2          0
> irq256: hpet0:t0                   10675        111
> irq257: hpet0:t1                    5897         61
> irq264: em0                          341          3
> irq265: hdac0                         95          0
> irq266: ahci0:ch0                   4274         44
> irq267: ahci0:ch1                    155          1
> Total                              21998        229
> dutch at dutch:~ %
> 
> HPETs are a little high, but overall 1000% better.  I have added
> to my /boot/loader.conf.
Why do you think that HPET is 'a little high' ? It is down from the hz,
which is probably 1024 on your machine, so it is a little low.

Still, I do not believe that the above vmstat -i is from the running
Xorg session.  What interrupt i915 gfx uses, it probably shared on irq16 ?
If so, it cannot be total 167 interrupts for the started X session,
even if dumbed down to X running a single xterm.

> 
> I see where this was discussed back in the 7.2R release notes, but
> I would have never found it.  Many thanks for your help.

There is some unpublished errata for G45/GM45 chipsets where display
interrupts are reported by legacy method even if the MSI is enabled.
I think it is somehow related to the motherboard or BIOS layout since
the situation is very rare.
-------------- next part --------------
A non-text attachment was scrubbed...
Name: not available
Type: application/pgp-signature
Size: 819 bytes
Desc: not available
URL: <http://lists.freebsd.org/pipermail/freebsd-hackers/attachments/20140607/90286ebb/attachment-0001.sig>


More information about the freebsd-hackers mailing list