[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]
Re: [PATCH 2/4] pcie: update slot power status only is power control is
From: |
Michael S. Tsirkin |
Subject: |
Re: [PATCH 2/4] pcie: update slot power status only is power control is enabled |
Date: |
Fri, 25 Feb 2022 08:48:13 -0500 |
On Fri, Feb 25, 2022 at 02:35:28PM +0100, Igor Mammedov wrote:
> On Fri, 25 Feb 2022 08:08:57 -0500
> "Michael S. Tsirkin" <mst@redhat.com> wrote:
>
> > On Fri, Feb 25, 2022 at 02:02:31PM +0100, Igor Mammedov wrote:
> > > On Fri, 25 Feb 2022 11:12:59 +0100
> > > Gerd Hoffmann <kraxel@redhat.com> wrote:
> > >
> > > > Hi,
> > > >
> > > > > pcie_cap_slot_post_load()
> > > > > -> pcie_cap_update_power()
> > > > > -> pcie_set_power_device()
> > > > > -> pci_set_power()
> > > > > -> pci_update_mappings()
> > > >
> > > > > Fix it by honoring PCI_EXP_SLTCAP_PCP and updating power status
> > > > > only if capability is enabled.
> > > >
> > > > > diff --git a/hw/pci/pcie.c b/hw/pci/pcie.c
> > > > > index d7d73a31e4..2339729a7c 100644
> > > > > --- a/hw/pci/pcie.c
> > > > > +++ b/hw/pci/pcie.c
> > > > > @@ -383,10 +383,9 @@ static void pcie_cap_update_power(PCIDevice
> > > > > *hotplug_dev)
> > > > >
> > > > > if (sltcap & PCI_EXP_SLTCAP_PCP) {
> > > > > power = (sltctl & PCI_EXP_SLTCTL_PCC) ==
> > > > > PCI_EXP_SLTCTL_PWR_ON;
> > > > > + pci_for_each_device(sec_bus, pci_bus_num(sec_bus),
> > > > > + pcie_set_power_device, &power);
> > > > > }
> > > > > -
> > > > > - pci_for_each_device(sec_bus, pci_bus_num(sec_bus),
> > > > > - pcie_set_power_device, &power);
> > > > > }
> > > >
> > > > The change makes sense, although I don't see how that changes qemu
> > > > behavior.
> > >
> > > looks like I need to fix commit message
> > >
> > > >
> > > > 'power' defaults to true, so when SLTCAP_PCP is off it should never
> > > > ever try to power off the devices. And pci_set_power() should figure
> > > > the state didn't change and instantly return without touching the
> > > > device.
> > >
> > >
> > > SLTCAP_PCP is on by default as well, so empty slot ends up with
> > > power disabled PCC state [1]:
> > >
> > > sltctl & SLTCTL_PCC == 0x400
> > >
> > > by the time machine is initialized.
> > >
> > > Then ACPI pcihp callbacks override native hotplug ones
> > > so PCC remains stuck in this state since all power control
> > > is out of picture in case of ACPI based hotplug. Guest OS
> > > doesn't use/or ignore native PCC.
> >
> > So how about when ACPI pcihp overrides native with its callbacks we also
> > set PCC power to on?
>
> with some reworks it should work (i.e. adding an extra knob that will tell
> PCI core not to power off when it should, looks fragile and very hacky).
> It has the same migration implications as this patch, so I'd rather go
> after disabling whole SLTCAP_PCP thing to be correct and keeping PCI
> code free from ACPI hacks.
Hmm I don't get it. I literally mean this:
diff --git a/hw/pci/pcie.c b/hw/pci/pcie.c
index d7d73a31e4..72de72ce7a 100644
--- a/hw/pci/pcie.c
+++ b/hw/pci/pcie.c
@@ -389,6 +389,17 @@ static void pcie_cap_update_power(PCIDevice *hotplug_dev)
pcie_set_power_device, &power);
}
+void pcie_cap_enable_power(PCIDevice *hotplug_dev)
+{
+ uint8_t *exp_cap = hotplug_dev->config + hotplug_dev->exp.exp_cap;
+ uint32_t sltcap = pci_get_long(exp_cap + PCI_EXP_SLTCAP);
+
+ if (sltcap & PCI_EXP_SLTCAP_PCP) {
+ pci_set_word_by_mask(exp_cap + PCI_EXP_SLTCTL,
+ PCI_EXP_SLTCTL_PCC, PCI_EXP_SLTCTL_PWR_ON);
+ }
+}
+
/*
* A PCI Express Hot-Plug Event has occurred, so update slot status register
* and notify OS of the event if necessary.
Then call this from ACPI. How would this have any migration
implications at all? And why do we need a knob not to power off then?
Power will just stay on since there's nothing turning it off.
--
MST
- [PATCH 2/4] pcie: update slot power status only is power control is enabled, (continued)
- [PATCH 2/4] pcie: update slot power status only is power control is enabled, Igor Mammedov, 2022/02/24
- Re: [PATCH 2/4] pcie: update slot power status only is power control is enabled, Michael S. Tsirkin, 2022/02/25
- Re: [PATCH 2/4] pcie: update slot power status only is power control is enabled, Gerd Hoffmann, 2022/02/25
- Re: [PATCH 2/4] pcie: update slot power status only is power control is enabled, Michael S. Tsirkin, 2022/02/25
- Re: [PATCH 2/4] pcie: update slot power status only is power control is enabled, Igor Mammedov, 2022/02/25
- Re: [PATCH 2/4] pcie: update slot power status only is power control is enabled, Michael S. Tsirkin, 2022/02/25
- Re: [PATCH 2/4] pcie: update slot power status only is power control is enabled, Igor Mammedov, 2022/02/25
- Re: [PATCH 2/4] pcie: update slot power status only is power control is enabled,
Michael S. Tsirkin <=
- Re: [PATCH 2/4] pcie: update slot power status only is power control is enabled, Igor Mammedov, 2022/02/25
- Re: [PATCH 2/4] pcie: update slot power status only is power control is enabled, Gerd Hoffmann, 2022/02/28
- Re: [PATCH 2/4] pcie: update slot power status only is power control is enabled, Igor Mammedov, 2022/02/28
[PATCH 3/4] acpi: pcihp: disable power control on PCIe slot, Igor Mammedov, 2022/02/24
[PATCH 4/4] q35: compat: keep hotplugged PCIe device broken after migration for 6.2-older machine types, Igor Mammedov, 2022/02/24
Re: [PATCH 0/4] Fix broken PCIe device after migration, Michael S. Tsirkin, 2022/02/24
Re: [PATCH 0/4] Fix broken PCIe device after migration, Michael S. Tsirkin, 2022/02/25