| Message ID | 20171013183548.68283-9-mika.westerberg@linux.intel.com (mailing list archive) |
|---|---|
| State | New, archived |
| Delegated to: | Bjorn Helgaas |
| Headers | show |
On Fri, Oct 13, 2017 at 09:35:48PM +0300, Mika Westerberg wrote: > During surprise hot-unplug the device is not there anymore. When that > happens we read 0xffffffff from the registers and pciehp_unconfigure_device() > inadvertently thinks the device is a display device because bridge > control register returns 0xff refusing to remove it: > > pciehp 0000:00:1c.0:pcie004: Slot(0): Link Down > pciehp 0000:00:1c.0:pcie004: Slot(0): Card present > pciehp 0000:00:1c.0:pcie004: Cannot remove display device 0000:01:00.0 > > This causes the hotplug functionality to leave the hierarcy untouched > preventing further hotplug operations. > > To fix this verify presence of a device by calling pci_device_is_present() > for it before we touch it any further. > > Signed-off-by: Mika Westerberg <mika.westerberg@linux.intel.com> > --- > drivers/pci/hotplug/pciehp_pci.c | 12 +++++++++--- > 1 file changed, 9 insertions(+), 3 deletions(-) > > diff --git a/drivers/pci/hotplug/pciehp_pci.c b/drivers/pci/hotplug/pciehp_pci.c > index 2a1ca020cf5a..fb4333168e23 100644 > --- a/drivers/pci/hotplug/pciehp_pci.c > +++ b/drivers/pci/hotplug/pciehp_pci.c > @@ -100,8 +100,14 @@ int pciehp_unconfigure_device(struct slot *p_slot) > */ > list_for_each_entry_safe_reverse(dev, temp, &parent->devices, > bus_list) { > + bool present; > + > pci_dev_get(dev); > - if (dev->hdr_type == PCI_HEADER_TYPE_BRIDGE && presence) { > + > + /* Check if the device is really there anymore */ > + present = presence ? pci_device_is_present(dev) : false; > + > + if (dev->hdr_type == PCI_HEADER_TYPE_BRIDGE && present) { > pci_read_config_byte(dev, PCI_BRIDGE_CONTROL, &bctl); I don't like this fix because it's still racy. We always have to be deal with a config read that returns 0xffffffff, even if we previously checked pci_device_is_present(). The device might have disappeared in the interim. > if (bctl & PCI_BRIDGE_CTL_VGA) { > ctrl_err(ctrl, > @@ -112,7 +118,7 @@ int pciehp_unconfigure_device(struct slot *p_slot) > break; > } > } > - if (!presence) { > + if (!present) { > pci_dev_set_disconnected(dev, NULL); > if (pci_has_subordinate(dev)) > pci_walk_bus(dev->subordinate, > @@ -123,7 +129,7 @@ int pciehp_unconfigure_device(struct slot *p_slot) > * Ensure that no new Requests will be generated from > * the device. > */ > - if (presence) { > + if (present) { > pci_read_config_word(dev, PCI_COMMAND, &command); > command &= ~(PCI_COMMAND_MASTER | PCI_COMMAND_SERR); > command |= PCI_COMMAND_INTX_DISABLE; > -- > 2.14.2 >
On Fri, Oct 20, 2017 at 04:15:02PM -0500, Bjorn Helgaas wrote: > > + > > + /* Check if the device is really there anymore */ > > + present = presence ? pci_device_is_present(dev) : false; > > + > > + if (dev->hdr_type == PCI_HEADER_TYPE_BRIDGE && present) { > > pci_read_config_byte(dev, PCI_BRIDGE_CONTROL, &bctl); > > I don't like this fix because it's still racy. We always have to be deal > with a config read that returns 0xffffffff, even if we previously checked > pci_device_is_present(). The device might have disappeared in the interim. That's a fair point. I guess it is better just to check if bctl holds 0xffff before we decide it is a display device. I'll rework this patch and send an updated version separately.
diff --git a/drivers/pci/hotplug/pciehp_pci.c b/drivers/pci/hotplug/pciehp_pci.c index 2a1ca020cf5a..fb4333168e23 100644 --- a/drivers/pci/hotplug/pciehp_pci.c +++ b/drivers/pci/hotplug/pciehp_pci.c @@ -100,8 +100,14 @@ int pciehp_unconfigure_device(struct slot *p_slot) */ list_for_each_entry_safe_reverse(dev, temp, &parent->devices, bus_list) { + bool present; + pci_dev_get(dev); - if (dev->hdr_type == PCI_HEADER_TYPE_BRIDGE && presence) { + + /* Check if the device is really there anymore */ + present = presence ? pci_device_is_present(dev) : false; + + if (dev->hdr_type == PCI_HEADER_TYPE_BRIDGE && present) { pci_read_config_byte(dev, PCI_BRIDGE_CONTROL, &bctl); if (bctl & PCI_BRIDGE_CTL_VGA) { ctrl_err(ctrl, @@ -112,7 +118,7 @@ int pciehp_unconfigure_device(struct slot *p_slot) break; } } - if (!presence) { + if (!present) { pci_dev_set_disconnected(dev, NULL); if (pci_has_subordinate(dev)) pci_walk_bus(dev->subordinate, @@ -123,7 +129,7 @@ int pciehp_unconfigure_device(struct slot *p_slot) * Ensure that no new Requests will be generated from * the device. */ - if (presence) { + if (present) { pci_read_config_word(dev, PCI_COMMAND, &command); command &= ~(PCI_COMMAND_MASTER | PCI_COMMAND_SERR); command |= PCI_COMMAND_INTX_DISABLE;
During surprise hot-unplug the device is not there anymore. When that happens we read 0xffffffff from the registers and pciehp_unconfigure_device() inadvertently thinks the device is a display device because bridge control register returns 0xff refusing to remove it: pciehp 0000:00:1c.0:pcie004: Slot(0): Link Down pciehp 0000:00:1c.0:pcie004: Slot(0): Card present pciehp 0000:00:1c.0:pcie004: Cannot remove display device 0000:01:00.0 This causes the hotplug functionality to leave the hierarcy untouched preventing further hotplug operations. To fix this verify presence of a device by calling pci_device_is_present() for it before we touch it any further. Signed-off-by: Mika Westerberg <mika.westerberg@linux.intel.com> --- drivers/pci/hotplug/pciehp_pci.c | 12 +++++++++--- 1 file changed, 9 insertions(+), 3 deletions(-)