On 9/25/2018 4:54 PM, Bjorn Helgaas wrote:
On Wed, Sep 19, 2018 at 04:30:37PM +0000, Sinan Kaya wrote:
Looking to have more control between the users of the API vs. what the API
can do internally. The new reset_type tells the PCI core about the bounds
of the request.
Signed-off-by: Sinan Kaya <okaya@xxxxxxxxxx>
---
drivers/pci/pci.c | 18 +++++++++++++++---
drivers/vfio/pci/vfio_pci.c | 6 ++++--
include/linux/pci.h | 2 +-
3 files changed, 20 insertions(+), 6 deletions(-)
diff --git a/drivers/pci/pci.c b/drivers/pci/pci.c
index b4f14419bd51..f11d29f32119 100644
--- a/drivers/pci/pci.c
+++ b/drivers/pci/pci.c
@@ -5232,13 +5232,25 @@ static int __pci_reset_bus(struct pci_bus *bus)
/**
* pci_reset_bus - Try to reset a PCI bus
* @pdev: top level PCI device to reset via slot/bus
+ * @reset_type: resets to try
*
* Same as above except return -EAGAIN if the bus cannot be locked
*/
-int pci_reset_bus(struct pci_dev *pdev)
+int pci_reset_bus(struct pci_dev *pdev, u32 reset_type)
{
- return (!pci_probe_reset_slot(pdev->slot)) ?
- __pci_reset_slot(pdev->slot) : __pci_reset_bus(pdev->bus);
+ if (reset_type & PCI_RESET_LINK) {
+ return (!pci_probe_reset_slot(pdev->slot)) ?
+ __pci_reset_slot(pdev->slot) :
+ __pci_reset_bus(pdev->bus);
+ }
+
+ if (reset_type & PCI_RESET_BUS)
+ return __pci_reset_bus(pdev->bus);
+
+ if (reset_type & PCI_RESET_SLOT)
+ return __pci_reset_slot(pdev->slot);
I don't understand this code. We have
#define PCI_RESET_LINK (PCI_RESET_SLOT | PCI_RESET_BUS)
so if either PCI_RESET_BUS or PCI_RESET_SLOT is set, we should take
the first "if" above, which always returns, and the last two should
never be reached. What am I missing?
Yup, this is broken. I need to follow up with another patchset.
I was trying to cover the case where user said,
"I need a link reset. I don't care whether it is a slot reset or bus reset
as long as we achieve a reset."
That's when PCI_RESET_LINK is used. We expect most drivers to PCI_RESET_LINK
if they are interested in a link reset and doesn't need to know about the
hotplug capability of a particular slot.
+
+ return -EINVAL;
}
EXPORT_SYMBOL_GPL(pci_reset_bus);
diff --git a/drivers/vfio/pci/vfio_pci.c b/drivers/vfio/pci/vfio_pci.c
index fe7ada997c51..0e80c72b1eaa 100644
--- a/drivers/vfio/pci/vfio_pci.c
+++ b/drivers/vfio/pci/vfio_pci.c
@@ -1015,7 +1015,8 @@ static long vfio_pci_ioctl(void *device_data,
&info, slot);
if (!ret)
/* User has access, do the reset */
- ret = pci_reset_bus(vdev->pdev);
+ ret = pci_reset_bus(vdev->pdev,
+ slot ? PCI_RESET_SLOT : PCI_RESET_BUS);
This is the only place in the whole series where a caller uses
something other than PCI_RESET_ANY. Inquiring minds want to know why.
I'm sure it's the right thing to do, it's just that we'll need to know
how to choose in future cases.
Use PCI_RESET_ANY for the existing behavior where you want one type of
reset and don't care if it is PM/function/slot/bus/specific.
Use PCI_RESET_LINK when you need link to retrain.
Use PCI_RESET_SLOT when you know that this system is hotplug capable
by calling probe functions.
Use PCI_RESET_BUS when you know that this system is not hotplug capable
and this endpoint will never be used in such a system.
I can capture this into the commit message.
hot_reset_release:
for (i--; i >= 0; i--)
@@ -1390,7 +1391,8 @@ static void vfio_pci_try_bus_reset(struct vfio_pci_device *vdev)
}
if (needs_reset)
- ret = pci_reset_bus(vdev->pdev);
+ ret = pci_reset_bus(vdev->pdev,
+ slot ? PCI_RESET_SLOT : PCI_RESET_BUS);
put_devs:
for (i = 0; i < devs.cur_index; i++) {
diff --git a/include/linux/pci.h b/include/linux/pci.h
index 4fdddcb85066..85f48e156753 100644
--- a/include/linux/pci.h
+++ b/include/linux/pci.h
@@ -1128,7 +1128,7 @@ int pci_reset_function_locked(struct pci_dev *dev, u32 reset_type);
int pci_try_reset_function(struct pci_dev *dev, u32 reset_type);
int pci_probe_reset_slot(struct pci_slot *slot);
int pci_probe_reset_bus(struct pci_bus *bus);
-int pci_reset_bus(struct pci_dev *dev);
+int pci_reset_bus(struct pci_dev *dev, u32 reset_type);
void pci_reset_secondary_bus(struct pci_dev *dev);
void pcibios_reset_secondary_bus(struct pci_dev *dev);
void pci_update_resource(struct pci_dev *dev, int resno);
--
2.18.0
