[Intel-gfx] [PATCH] Manage PIPESTAT pending interrupt values to unblock vblank interrupts
Steven J Newbury
steve at snewbury.org.uk
Sat Nov 8 22:26:47 CET 2008
On Sat, 2008-11-08 at 10:27 -0500, Robert Noland wrote:
> On Fri, 2008-11-07 at 22:00 +0000, Steven J Newbury wrote:
> > On Fri, 2008-11-07 at 21:44 +0000, Steven J Newbury wrote:
> > > On Fri, 2008-11-07 at 20:45 +0000, Steven J Newbury wrote:
> > > > On Fri, 2008-11-07 at 11:11 -0800, Eric Anholt wrote:
> > > > > On Fri, 2008-11-07 at 14:01 +0000, Steven J Newbury wrote:
> > >
> > > > > > I'm on 965GM and I'm having a serious interrupt problem since this patch
> > > > > > went into for-review:
> > > > > >
> > > > > > Nov 7 04:20:22 infinity irq 16: nobody cared (try booting with the
> > > > > > "irqpoll" option)
> > > > > > Nov 7 04:20:22 infinity Pid: 0, comm: swapper Not tainted
> > > > > > 2.6.28-rc3-00236-g1d7eff8 #23
> > > > > > Nov 7 04:20:22 infinity Call Trace:
> > > > > > Nov 7 04:20:22 infinity <IRQ> [<ffffffff80491a25>] ?
> > > > > > i915_driver_irq_handler+0x53/0x186
> > > > > > Nov 7 04:20:22 infinity [<ffffffff80270b55>] __report_bad_irq+0x3d/0x8c
> > > > > > Nov 7 04:20:22 infinity [<ffffffff80270cb7>] note_interrupt+0x113/0x178
> > > > > > Nov 7 04:20:22 infinity [<ffffffff802713db>] handle_fasteoi_irq
> > > > > > +0x99/0xc3
> > > > > > Nov 7 04:20:22 infinity [<ffffffff8020ee5f>] do_IRQ+0x9c/0x11d
> > > > > > Nov 7 04:20:22 infinity [<ffffffff8020c826>] ret_from_intr+0x0/0xa
> > > > > > Nov 7 04:20:22 infinity <EOI> [<ffffffff804572c0>] ?
> > > > > > acpi_idle_enter_simple+0x175/0x1a8
> > > > > > Nov 7 04:20:22 infinity [<ffffffff804572b6>] ? acpi_idle_enter_simple
> > > > > > +0x16b/0x1a8
> > > > > > Nov 7 04:20:22 infinity [<ffffffff8052af56>] ? cpuidle_idle_call
> > > > > > +0xa6/0xe0
> > > > > > Nov 7 04:20:22 infinity [<ffffffff8020b47a>] ? cpu_idle+0x4c/0xb0
> > > > > > Nov 7 04:20:22 infinity [<ffffffff80614551>] ? rest_init+0x75/0x77
> > > > > > Nov 7 04:20:22 infinity handlers:
> > > > > > Nov 7 04:20:22 infinity [<ffffffff804919d2>] (i915_driver_irq_handler
> > > > > > +0x0/0x186)
> > > > > > Nov 7 04:20:22 infinity Disabling IRQ #16
> > > > > >
> > > > > > This happens after a random amount of time in X, athough never very
> > > > > > long. From this point on there are no interrupts generated unless I
> > > > > > switch vts away from X and back again.
> > > I'm wrong here. Switching vts only "fixes" the second problem below.
> > >
> > > > This gets interrupts working
> > > > > > again for a short while.
> > > > >
> > > > > Can you get /proc/dri/0/i915_gem_interrupt from before and just after
> > > > > the problem occurs?
> > > > >
> > > > I'll fire up a for-review kernel and see what it says.
> > >
> > > Before X:
> > >
> > > Interrupt enable: 00000000
> > > Interrupt identity: 00000000
> > > Interrupt mask: fffedfff
> > > Pipe A stat: 00000203
> > > Pipe B stat: 80000206
> > > Interrupts received: 0
> > > Current sequence: 0
> > > Waiter sequence: 0
> > > IRQ sequence: 0
> > >
> > > After X has started:
> > >
> > > Interrupt enable: 00000051
> > > Interrupt identity: 00000002
> > > Interrupt mask: fffedfac
> > > Pipe A stat: 00020204
> > > Pipe B stat: 00000206
> > > Interrupts received: 1327
> > > Current sequence: 1742
> > > Waiter sequence: 0
> > > IRQ sequence: 1738
> > >
> > > Interrupt enable: 00000051
> > > Interrupt identity: 00000002
> > > Interrupt mask: fffedfac
> > > Pipe A stat: 00020204
> > > Pipe B stat: 00000206
> > > Interrupts received: 33424
> > > Current sequence: 43154
> > > Waiter sequence: 0
> > > IRQ sequence: 43132
> > >
> > > Interrupt enable: 00000051
> > > Interrupt identity: 00000002
> > > Interrupt mask: fffedfac
> > > Pipe A stat: 00020204
> > > Pipe B stat: 00020000
> > > Interrupts received: 42250
> > > Current sequence: 58442
> > > Waiter sequence: 0
> > > IRQ sequence: 58434
> > > ____
> > >
> > > After interrupt failure:
> > >
> > > Interrupt enable: 00000051
> > > Interrupt identity: 00000000
> > > Interrupt mask: fffedfac
> > > Pipe A stat: 00020204
> > > Pipe B stat: 00000206
> > > Interrupts received: 200097
> > > Current sequence: 96282
> > > Waiter sequence: 0
> > > IRQ sequence: 96282
> > >
> > > Output of 'cat /proc/interrupts' :
> > > CPU0 CPU1
> > > 0: 309831 301848 IO-APIC-edge timer
> > > 1: 964 1747 IO-APIC-edge i8042
> > > 4: 1 1 IO-APIC-edge
> > > 8: 1 0 IO-APIC-edge rtc0
> > > 9: 0 1 IO-APIC-fasteoi acpi
> > > 12: 11555 16280 IO-APIC-edge i8042
> > > 14: 0 0 IO-APIC-edge ata_piix
> > > 15: 0 0 IO-APIC-edge ata_piix
> > > 16: 99522 100479 IO-APIC-fasteoi i915 at pci:0000:00:02.0
> > > 19: 6 9 IO-APIC-fasteoi yenta, firewire_ohci
> > > 20: 75 63 IO-APIC-fasteoi uhci_hcd:usb1,
> > > uhci_hcd:usb3, ehci_hcd:usb7
> > > 21: 204 216 IO-APIC-fasteoi uhci_hcd:usb2,
> > > uhci_hcd:usb4, HDA Intel
> > > 22: 352 644 IO-APIC-fasteoi uhci_hcd:usb5,
> > > ehci_hcd:usb6
> > > 43: 4898 5996 PCI-MSI-edge ahci
> > > NMI: 0 0 Non-maskable interrupts
> > > LOC: 116278 86951 Local timer interrupts
> > > RES: 27385 27476 Rescheduling interrupts
> > > CAL: 91 32 Function call interrupts
> > > TLB: 32 96 TLB shootdowns
> > > TRM: 0 0 Thermal event interrupts
> > > THR: 0 0 Threshold APIC interrupts
> > > SPU: 0 0 Spurious interrupts
> > > ERR: 0
> > > MIS: 0
> >
> > Curiously, the i915_gem_interrupt count continues to rise despite no
> > more interrupts being recorded in /proc/interrupts. Clearly interrupts
> > are not working, X is very slow, and glxgears reports interrupts are not
> > working correctly.
> >
> > Currently:
> > cat /proc/dri/0/i915_gem_interrupt
> > Interrupt enable: 00000051
> > Interrupt identity: 00000002
> > Interrupt mask: fffedfac
> > Pipe A stat: 00000000
> > Pipe B stat: 00000206
> > Interrupts received: 615479
> > Current sequence: 308340
> > Waiter sequence: 0
> > IRQ sequence: 308338
>
> Unless keithp's most recent patch moving BREADCRUMB_INDEX prevents some
> internal brain damage, messing with IER often seems to be a bad idea, at
> least on 965gm. I've spent most of the week fighting this issue on
> FreeBSD. Last night, I flipped the logic back to setting up IER during
> interrupt handler install and flipping bits in IMR to enable / disable
> irqs and everything is working correctly again. I have made some other
> code changes in the handler, but none of them resolved the issue.
> Inverting the logic got everything working again, for both INTx and MSI.
> I know that it is published that MSI should not be used on the 965gm,
> but I've not seen any issues on my hardware.
>
> robert.
Now this is really weird, if I suspend to RAM and then resume, from that
point everything seems to work fine so far!?! My guess is the
re-installation of the interrupt handler on resume occurs with different
register values compared to the initial setup.
More information about the Intel-gfx
mailing list