On Monday 24 August 2009, Shaohua Li wrote:
> Add an implementation how to detect wakeup event for PCI. PCI device can
> invoke PME, platform or PCIe native approach can collect the event and
> report to OS. OS should identify exactly which device invokes PME as
> several devices can share PME pin.
>
> In platform approach (ACPI in this case), some BIOS give exact device which
> invokes PME but others doesn't.
I don't think the BIOS can reliably say which device signals PME# in the PCI
(non-PCIe) case, unless the PME# is routed separately from each device to the
chipset. Also, two or more devices may signal PME# at the same time.
So, in this case we generally need to scan the entire hierarchy each time
we get a PME#.
> In PCIe native approach, if PME source device is a pcie endpoint, the device
> is the exact PME source. If the device is root port or pcie-to-pci bridge,
> we need scan the hierarchy under the device.
Why do we have to scan if the source is a root port itself?
> To identify PME source, the patch does:
> 1. if the source is a pci device, the device is the only source for PME
That need not be true in the non-PCIe case IMO.
> 2. if the source is a bridge, scan the hierarchy under the bridge. Several
> devices under the bridge could be the sources.
I think we need a function that will scan the hierarchy below a bridge
(that may be the root bridge in the non-PCIe case or a PCIe-to-PCI
bridge in the PCIe case) and if a device has PME_Status set, it will
(a) clear it and (b) call pm_request_resume() for the device.
> Signed-off-by: Shaohua Li <shaohua.li@xxxxxxxxx>
> ---
> drivers/pci/pci-driver.c | 77 +++++++++++++++++++++++++++++++++++++++++++++++
> drivers/pci/pci.h | 2 +
> 2 files changed, 79 insertions(+)
>
> Index: linux/drivers/pci/pci-driver.c
> ===================================================================
> --- linux.orig/drivers/pci/pci-driver.c 2009-08-24 10:50:12.000000000 +0800
> +++ linux/drivers/pci/pci-driver.c 2009-08-24 15:42:49.000000000 +0800
> @@ -17,6 +17,7 @@
> #include <linux/slab.h>
> #include <linux/sched.h>
> #include <linux/cpu.h>
> +#include <linux/pm_runtime.h>
> #include "pci.h"
>
> /*
> @@ -570,6 +571,82 @@ static void pci_pm_complete(struct devic
> drv->pm->complete(dev);
> }
>
> +/*
> + * Called when dev is suspected to invoke a wakeup event
> + * */
Please add a full kerneldoc comment.
> +static bool pci_pm_check_wakeup_event(struct pci_dev *pdev)
> +{
> + int pme_pos = pdev->pm_cap;
> + u16 pmcsr;
> + bool spurious = false;
> +
> + if (pme_pos == 0) {
if (!pme_pos) would be better IMO. Also the braces are not necessary and I'd
move the comment somewhere else (that function is going to be called for
devices that don't have the PM capability other than the USB controllers).
> + /*
> + * Some USB devices haven't PME, but have specific registers to
> + * control wakeup
> + */
> + return false;
> + }
> +
> + /* clear PME status and disable PME to avoid interrupt flood */
> + pci_read_config_word(pdev, pme_pos + PCI_PM_CTRL, &pmcsr);
> + if (!(pmcsr & PCI_PM_CTRL_PME_STATUS))
> + return false;
> + /* I see spurious PME here, just ignore it for now */
> + if (!(pmcsr & PCI_PM_CTRL_PME_ENABLE))
> + spurious = true;
> + else
> + pmcsr &= ~PCI_PM_CTRL_PME_ENABLE;
> + pmcsr |= PCI_PM_CTRL_PME_STATUS;
> + pci_write_config_word(pdev, pme_pos + PCI_PM_CTRL, pmcsr);
> +
> + if (spurious)
> + return false;
> + return true;
Fold these three lines into one:
+ return !spurious;
Thus, I wouldn't call that variable 'spurious'. I'd rather call it 'ret' and
make it so that we can just return its value at the end.
So, the above can be rewritten as
+ /* Check if PME status is set. */
+ pci_read_config_word(pdev, pme_pos + PCI_PM_CTRL, &pmcsr);
+ if (!(pmcsr & PCI_PM_CTRL_PME_STATUS))
+ return false;
+ /* Clear PME status and disable PME to avoid interrupt flood. */
+ pmcsr |= PCI_PM_CTRL_PME_STATUS;
+ if (pmcsr & PCI_PM_CTRL_PME_ENABLE)
+ pmcsr &= ~PCI_PM_CTRL_PME_ENABLE;
+ else
+ ret = false;
+ pci_write_config_word(pdev, pme_pos + PCI_PM_CTRL, pmcsr);
+ return ret;
> +}
> +
> +/* @target is suspected to invoke a wakeup event. Return true if it's true */
Please add full kerneldoc comment.
> +bool pci_pm_handle_wakeup_event(struct pci_dev *target)
> +{
> + bool ret;
> + struct pci_dev *tmp = NULL;
> + int domain_nr, bus_start, bus_end;
> +
> + /*
> + * @target could be a bridge or a device.
> + * if target is device, trust the device invokes PME. If target is a
> + * bridge, scan devices under the bridge and only trust device invokes
> + * PME which we can detect
> + **/
> + ret = pci_pm_check_wakeup_event(target);
> + if (!target->subordinate || (target->is_pcie &&
> + target->pcie_type != PCI_EXP_TYPE_ROOT_PORT &&
> + target->pcie_type != PCI_EXP_TYPE_PCI_BRIDGE)) {
> + /* always trust the device invokes PME even we can't detect */
> + pm_request_resume(&target->dev);
I don't think we can return here. There may be other devices signalling
PME.
Well, perhaps move the pm_request_resume(&target->dev) to the function above
along with the conditional? In that case you could call the function
pci_pm_wakeup() which would be nicer IMO.
Also I'm not sure about the conditional. Namely, I don't think we'll need to
call this function in the PCIe case except for PCI-to-PCIe bridges, because
the root port will give us the requester ID of the device the interrupt was for
if that's an endpoint. So, perhaps we can assume we scan over non-PCIe
devices here?
> + return true;
> + }
> +
> + if (ret)
> + pm_request_resume(&target->dev);
Do we want to do that for bridges?
> +
> + /* scan devices under the bridge */
> + domain_nr = pci_domain_nr(target->bus);
> + bus_start = target->subordinate->secondary;
> + bus_end = target->subordinate->subordinate;
> + while ((tmp = pci_get_device(PCI_ANY_ID, PCI_ANY_ID, tmp)) != NULL) {
> + if (pci_domain_nr(tmp->bus) == domain_nr &&
> + tmp->bus->number >= bus_start &&
> + tmp->bus->number <= bus_end) {
> + if (pci_pm_check_wakeup_event(tmp)) {
> + ret = true;
> + pm_request_resume(&tmp->dev);
> + }
> + }
> + }
> + return ret;
> +}
> +
> #ifdef CONFIG_SUSPEND
>
> static int pci_pm_suspend(struct device *dev)
> Index: linux/drivers/pci/pci.h
> ===================================================================
> --- linux.orig/drivers/pci/pci.h 2009-08-24 11:04:22.000000000 +0800
> +++ linux/drivers/pci/pci.h 2009-08-24 11:05:24.000000000 +0800
> @@ -83,6 +83,8 @@ static inline void pci_vpd_release(struc
> dev->vpd->ops->release(dev);
> }
>
> +extern bool pci_pm_handle_wakeup_event(struct pci_dev *target);
> +
> /* PCI /proc functions */
> #ifdef CONFIG_PROC_FS
> extern int pci_proc_attach_device(struct pci_dev *dev);
Thanks,
Rafael
_______________________________________________
linux-pm mailing list
linux-pm@xxxxxxxxxxxxxxxxxxxxxxxxxx
https://lists.linux-foundation.org/mailman/listinfo/linux-pm
