Re: pci-express hotplug

[Kenji Kaneshige]

Jens Axboe wrote:
> On Tue, Oct 27 2009, Kenji Kaneshige wrote:
> > Jens Axboe wrote:
> > > On Tue, Oct 20 2009, Alex Chiang wrote:
> > > > * Jens Axboe <jens.axboe@xxxxxxxxxx>:
> > > > > On Tue, Oct 13 2009, Alex Chiang wrote:
> > > > > > > > Can you modprobe acpiphp with debug=1? And send the output?
> > > > > > > acpiphp: ACPI Hot Plug PCI Controller Driver version: 0.5
> > > > > > > acpiphp_glue: found PCI-to-PCI bridge at PCI 0000:00:05.0
> > > > > > > acpiphp_glue: found ACPI PCI Hotplug slot 1 at PCI 0000:08:00
> > > > > > > acpiphp: Slot [1] registered
> > > > > > > acpiphp_glue: found PCI-to-PCI bridge at PCI 0000:00:07.0
> > > > > > > acpiphp_glue: found ACPI PCI Hotplug slot 2 at PCI 0000:0b:00
> > > > > > > acpiphp: Slot [2] registered
> > > > > > > acpiphp_glue: found PCI-to-PCI bridge at PCI 0000:80:07.0
> > > > > > > acpiphp_glue: found ACPI PCI Hotplug slot 6 at PCI 0000:84:00
> > > > > > > acpiphp: Slot [6] registered
> > > > > > > acpiphp_glue: found PCI-to-PCI bridge at PCI 0000:80:09.0
> > > > > > > acpiphp_glue: found ACPI PCI Hotplug slot 7 at PCI 0000:87:00
> > > > > > > acpiphp: Slot [7] registered
> > > > > > > acpiphp_glue: Bus 0000:87 has 1 slot
> > > > > > > acpiphp_glue: Bus 0000:84 has 1 slot
> > > > > > > acpiphp_glue: Bus 0000:0b has 1 slot
> > > > > > > acpiphp_glue: Bus 0000:08 has 1 slot
> > > > > > > acpiphp_glue: Total 4 slots
> > > > > > You mentioned in another mail that you echoed 1 into the various
> > > > > > slots' power files.
> > > > > >
> > > > > > Did you do that after modprobing acpiphp with debug=1?
> > > > > >
> > > > > > If so, there should be debug output when you try and turn them
> > > > > > on.
> > > > > It produces:
> > > > >
> > > > > acpiphp: enable_slot - physical_slot = 1
> > > > > acpiphp_glue: acpiphp_enable_slot: Slot status is not ACPI_STA_ALL
> > > > > acpiphp: enable_slot - physical_slot = 2
> > > > > acpiphp_glue: acpiphp_enable_slot: Slot status is not ACPI_STA_ALL
> > > > > acpiphp: enable_slot - physical_slot = 6
> > > > > acpiphp_glue: acpiphp_enable_slot: Slot status is not ACPI_STA_ALL
> > > > > acpiphp: enable_slot - physical_slot = 7
> > > > > acpiphp_glue: acpiphp_enable_slot: Slot status is not ACPI_STA_ALL
> > > > Hm, so for some reason, firmware on your machine is telling us
> > > > that it doesn't think cards are present and/or enabled.
> > > >
> > > > Unfortunately, I don't know why your firmware would be saying
> > > > that. We could add some more debug printks to see what firmware
> > > > thinks about your system... Or we could just wait and see what
> > > > happens after you get your hardware replaced.
> > > New board, the exact same thing happens.
> > >
> > > > > I have a card in one of the slots only this time.
> > > > >
> > > > > > Also, quick dummy check, you are trying to power on populated
> > > > > > slots, right? :)
> > > > > Yes :-)
> > > > >
> > > > > > Can you send the output of lspci -vv? And I like the output of
> > > > > > lspci -vt as well... Both before and after loading acpiphp
> > > > > > please.
> > > > > Send privately.
> > > > No difference in before and after. Odd.
> > > >
> > > > If you want to poke us again after your hardware swap, please do
> > > > so. Sorry for being not so helpful. :-/
> > > Poke :-)
> > >
> > > One more thing I tried was pushing the power button on the slot
> > > manually. With acpiphp, I get the same messages as above. Using pciehp,
> > > I get the same power fault bit interrupt storm. So no difference from
> > > using the sysfs interface or doing it on the box side, doesn't work
> > > either way.
> > I'd like to confirm power fault interrupt storm, just in case.
> > Could you get /proc/interrupts information after power fault
> > problem happens and send it to me?
>
> The box pretty much hangs when I try to power on a slot with pciehp, so
> it's not easy to do... It doesn't hang with acpiphp, but doesn't work
> either (see previous reply to Alex).

Could you try the attached debugging patch? With this patch, power
fault interrupt would be disabled after 100 power fault detected (
I hope so). You can get /proc/interrupts after that.

Thanks,
Kenji Kaneshige


---
 drivers/pci/hotplug/pciehp_hpc.c |    8 ++++++++
 1 file changed, 8 insertions(+)

Index: 20091026/drivers/pci/hotplug/pciehp_hpc.c
===================================================================
--- 20091026.orig/drivers/pci/hotplug/pciehp_hpc.c
+++ 20091026/drivers/pci/hotplug/pciehp_hpc.c
@@ -612,6 +612,7 @@ static irqreturn_t pcie_isr(int irq, voi
 	struct controller *ctrl = (struct controller *)dev_id;
 	struct slot *slot = ctrl->slot;
 	u16 detected, intr_loc;
+	static int nr_power_faults = 0;
 
 	/*
 	 * In order to guarantee that all interrupt events are
@@ -664,6 +665,13 @@ static irqreturn_t pcie_isr(int irq, voi
 	if (intr_loc & PCI_EXP_SLTSTA_PDC)
 		pciehp_handle_presence_change(slot);
 
+	if ((intr_loc & PCI_EXP_SLTSTA_PFD) && (++nr_power_faults > 100)) {
+		u16 reg16;
+		pciehp_readw(ctrl, PCI_EXP_SLTCTL, &reg16);
+		reg16 &= ~PCI_EXP_SLTCTL_PFDE;
+		pciehp_writew(ctrl, PCI_EXP_SLTCTL, reg16);
+	}
+
 	/* Check Power Fault Detected */
 	if ((intr_loc & PCI_EXP_SLTSTA_PFD) && !ctrl->power_fault_detected) {
 		ctrl->power_fault_detected = 1;