[PATCH v2 0/6] vfio-pci: Misc enhancements

[PATCH v2 0/6] vfio-pci: Misc enhancements

[Alex Williamson]

v2:
 - Incorporate comments from Bandan and Bjorn for vfio-pci.ids option
 - Include necessary vgaarb change, already Ack'd by Dave
 - Rebase on top of my current next branch
 - Rename D3 disable parameter

There are really 3 separate features added in this series, the first
is to opt-out VGA devices from VGA arbitration if a) vfio VGA support
is disable, either via config or module optioni (new), and b) bridge
control of VGA resource routing is possible.  This means that if
multiple VGA devices are installed with the intention of using them
for device assignment without VGA, we can eliminate the effect they
have on host graphics.

The second feature is the addition of the "ids" module option, which
acts just like the option of the same name on pci-stub.  This makes it
easier to configure vfio-pci to statically claim certain devices.  By
either building vfio support into the kernel or using softdeps to
load vfio-pci before native drivers, this can make it much easier to
bind to devices which are only intended to be used through vfio, such
as those additional graphics cards.

Finally, when devices are bound to vfio-pci and unused, we can try to
put them into a low-power state.  This again feeds into that idea that
devices may be installed on the system only for use through vfio, and
that use may not be continuous.  This saves a few watts for some GPUs.
Thanks,

Alex

---

Alex Williamson (6):
      vgaarb: Stub vga_set_legacy_decoding()
      vfio-pci: Add module option to disable VGA region access
      vfio-pci: Add VGA arbiter client
      vfio-pci: Allow PCI IDs to be specified as module options
      vfio-pci: Remove warning if try-reset fails
      vfio-pci: Move idle devices to D3hot power state


 drivers/vfio/pci/vfio_pci.c |  179 +++++++++++++++++++++++++++++++++++++++----
 include/linux/vgaarb.h      |    5 +
 2 files changed, 167 insertions(+), 17 deletions(-)

[PATCH v2 1/6] vgaarb: Stub vga_set_legacy_decoding()

[Alex Williamson]

vga_set_legacy_decoding() is defined in drivers/gpu/vga/vgaarb.c,
which is only compiled with CONFIG_VGA_ARB.  A caller would
therefore get an undefined symbol if the VGA arbiter is not
enabled.

Signed-off-by: Alex Williamson <alex.williamson@redhat.com>
Acked-by: Dave Airlie <airlied@redhat.com>
---
 include/linux/vgaarb.h |    5 +++++
 1 file changed, 5 insertions(+)

diff --git a/include/linux/vgaarb.h b/include/linux/vgaarb.h
index c37bd4d..8c3b412 100644
--- a/include/linux/vgaarb.h
+++ b/include/linux/vgaarb.h
@@ -65,8 +65,13 @@ struct pci_dev;
  *     out of the arbitration process (and can be safe to take
  *     interrupts at any time.
  */
+#if defined(CONFIG_VGA_ARB)
 extern void vga_set_legacy_decoding(struct pci_dev *pdev,
 				    unsigned int decodes);
+#else
+static inline void vga_set_legacy_decoding(struct pci_dev *pdev,
+					   unsigned int decodes) { };
+#endif
 
 /**
  *     vga_get         - acquire & locks VGA resources

[PATCH v2 2/6] vfio-pci: Add module option to disable VGA region access

[Alex Williamson]

Add a module option so that we don't require a CONFIG change and
kernel rebuild to disable VGA support.  Not only can VGA support be
troublesome in itself, but by disabling it we can reduce the impact
to host devices by doing a VGA arbitration opt-out.

Signed-off-by: Alex Williamson <alex.williamson@redhat.com>
---
 drivers/vfio/pci/vfio_pci.c |   19 ++++++++++++++++---
 1 file changed, 16 insertions(+), 3 deletions(-)

diff --git a/drivers/vfio/pci/vfio_pci.c b/drivers/vfio/pci/vfio_pci.c
index 2f865d07..7053110 100644
--- a/drivers/vfio/pci/vfio_pci.c
+++ b/drivers/vfio/pci/vfio_pci.c
@@ -37,8 +37,23 @@ module_param_named(nointxmask, nointxmask, bool, S_IRUGO | S_IWUSR);
 MODULE_PARM_DESC(nointxmask,
 		  "Disable support for PCI 2.3 style INTx masking.  If this resolves problems for specific devices, report lspci -vvvxxx to linux-pci@vger.kernel.org so the device can be fixed automatically via the broken_intx_masking flag.");
 
+#ifdef CONFIG_VFIO_PCI_VGA
+static bool disable_vga;
+module_param(disable_vga, bool, S_IRUGO);
+MODULE_PARM_DESC(disable_vga, "Disable VGA resource access through vfio-pci");
+#endif
+
 static DEFINE_MUTEX(driver_lock);
 
+static inline bool vfio_vga_disabled(void)
+{
+#ifdef CONFIG_VFIO_PCI_VGA
+	return disable_vga;
+#else
+	return true;
+#endif
+}
+
 static void vfio_pci_try_bus_reset(struct vfio_pci_device *vdev);
 
 static int vfio_pci_enable(struct vfio_pci_device *vdev)
@@ -93,10 +108,8 @@ static int vfio_pci_enable(struct vfio_pci_device *vdev)
 	} else
 		vdev->msix_bar = 0xFF;
 
-#ifdef CONFIG_VFIO_PCI_VGA
-	if ((pdev->class >> 8) == PCI_CLASS_DISPLAY_VGA)
+	if (!vfio_vga_disabled() && (pdev->class >> 8) == PCI_CLASS_DISPLAY_VGA)
 		vdev->has_vga = true;
-#endif
 
 	return 0;
 }

[PATCH v2 3/6] vfio-pci: Add VGA arbiter client

[Alex Williamson]

If VFIO VGA access is disabled for the user, either by CONFIG option
or module parameter, we can often opt-out of VGA arbitration.  We can
do this when PCI bridge control of VGA routing is possible.  This
means that we must have a parent bridge and there must only be a
single VGA device below that bridge.  Fortunately this is the typical
case for discrete GPUs.

Doing this allows us to minimize the impact of additional GPUs, in
terms of VGA arbitration, when they are only used via vfio-pci for
non-VGA applications.

Signed-off-by: Alex Williamson <alex.williamson@redhat.com>
---
 drivers/vfio/pci/vfio_pci.c |   67 ++++++++++++++++++++++++++++++++++++++++---
 1 file changed, 63 insertions(+), 4 deletions(-)

diff --git a/drivers/vfio/pci/vfio_pci.c b/drivers/vfio/pci/vfio_pci.c
index 7053110..25aef05 100644
--- a/drivers/vfio/pci/vfio_pci.c
+++ b/drivers/vfio/pci/vfio_pci.c
@@ -25,6 +25,7 @@
 #include <linux/types.h>
 #include <linux/uaccess.h>
 #include <linux/vfio.h>
+#include <linux/vgaarb.h>
 
 #include "vfio_pci_private.h"
 
@@ -54,6 +55,50 @@ static inline bool vfio_vga_disabled(void)
 #endif
 }
 
+/*
+ * Our VGA arbiter participation is limited since we don't know anything
+ * about the device itself.  However, if the device is the only VGA device
+ * downstream of a bridge and VFIO VGA support is disabled, then we can
+ * safely return legacy VGA IO and memory as not decoded since the user
+ * has no way to get to it and routing can be disabled externally at the
+ * bridge.
+ */
+static unsigned int vfio_pci_set_vga_decode(void *opaque, bool single_vga)
+{
+	struct vfio_pci_device *vdev = opaque;
+	struct pci_dev *tmp = NULL, *pdev = vdev->pdev;
+	unsigned char max_busnr;
+	unsigned int decodes;
+
+	if (single_vga || !vfio_vga_disabled() || pci_is_root_bus(pdev->bus))
+		return VGA_RSRC_NORMAL_IO | VGA_RSRC_NORMAL_MEM |
+		       VGA_RSRC_LEGACY_IO | VGA_RSRC_LEGACY_MEM;
+
+	max_busnr = pci_bus_max_busnr(pdev->bus);
+	decodes = VGA_RSRC_NORMAL_IO | VGA_RSRC_NORMAL_MEM;
+
+	while ((tmp = pci_get_class(PCI_CLASS_DISPLAY_VGA << 8, tmp)) != NULL) {
+		if (tmp == pdev ||
+		    pci_domain_nr(tmp->bus) != pci_domain_nr(pdev->bus) ||
+		    pci_is_root_bus(tmp->bus))
+			continue;
+
+		if (tmp->bus->number >= pdev->bus->number &&
+		    tmp->bus->number <= max_busnr) {
+			pci_dev_put(tmp);
+			decodes |= VGA_RSRC_LEGACY_IO | VGA_RSRC_LEGACY_MEM;
+			break;
+		}
+	}
+
+	return decodes;
+}
+
+static inline bool vfio_pci_is_vga(struct pci_dev *pdev)
+{
+	return (pdev->class >> 8) == PCI_CLASS_DISPLAY_VGA;
+}
+
 static void vfio_pci_try_bus_reset(struct vfio_pci_device *vdev);
 
 static int vfio_pci_enable(struct vfio_pci_device *vdev)
@@ -108,7 +153,7 @@ static int vfio_pci_enable(struct vfio_pci_device *vdev)
 	} else
 		vdev->msix_bar = 0xFF;
 
-	if (!vfio_vga_disabled() && (pdev->class >> 8) == PCI_CLASS_DISPLAY_VGA)
+	if (!vfio_vga_disabled() && vfio_pci_is_vga(pdev))
 		vdev->has_vga = true;
 
 	return 0;
@@ -900,6 +945,12 @@ static int vfio_pci_probe(struct pci_dev *pdev, const struct pci_device_id *id)
 		kfree(vdev);
 	}
 
+	if (vfio_pci_is_vga(pdev)) {
+		vga_client_register(pdev, vdev, NULL, vfio_pci_set_vga_decode);
+		vga_set_legacy_decoding(pdev,
+					vfio_pci_set_vga_decode(vdev, false));
+	}
+
 	return ret;
 }
 
@@ -908,9 +959,17 @@ static void vfio_pci_remove(struct pci_dev *pdev)
 	struct vfio_pci_device *vdev;
 
 	vdev = vfio_del_group_dev(&pdev->dev);
-	if (vdev) {
-		iommu_group_put(pdev->dev.iommu_group);
-		kfree(vdev);
+	if (!vdev)
+		return;
+
+	iommu_group_put(pdev->dev.iommu_group);
+	kfree(vdev);
+
+	if (vfio_pci_is_vga(pdev)) {
+		vga_client_register(pdev, NULL, NULL, NULL);
+		vga_set_legacy_decoding(pdev,
+				VGA_RSRC_NORMAL_IO | VGA_RSRC_NORMAL_MEM |
+				VGA_RSRC_LEGACY_IO | VGA_RSRC_LEGACY_MEM);
 	}
 }
 

[PATCH v2 4/6] vfio-pci: Allow PCI IDs to be specified as module options

[Alex Williamson]

This copies the same support from pci-stub for exactly the same
purpose, enabling a set of PCI IDs to be automatically added to the
driver's dynamic ID table at module load time.  The code here is
pretty simple and both vfio-pci and pci-stub are fairly unique in
being meta drivers, capable of attaching to any device, so there's no
attempt made to generalize the code into pci-core.

Signed-off-by: Alex Williamson <alex.williamson@redhat.com>
---
 drivers/vfio/pci/vfio_pci.c |   49 +++++++++++++++++++++++++++++++++++++++++++
 1 file changed, 49 insertions(+)

diff --git a/drivers/vfio/pci/vfio_pci.c b/drivers/vfio/pci/vfio_pci.c
index 25aef05..43517ce 100644
--- a/drivers/vfio/pci/vfio_pci.c
+++ b/drivers/vfio/pci/vfio_pci.c
@@ -11,6 +11,8 @@
  * Author: Tom Lyon, pugs@cisco.com
  */
 
+#define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
+
 #include <linux/device.h>
 #include <linux/eventfd.h>
 #include <linux/file.h>
@@ -33,6 +35,10 @@
 #define DRIVER_AUTHOR   "Alex Williamson <alex.williamson@redhat.com>"
 #define DRIVER_DESC     "VFIO PCI - User Level meta-driver"
 
+static char ids[1024] __initdata;
+module_param_string(ids, ids, sizeof(ids), 0);
+MODULE_PARM_DESC(ids, "Initial PCI IDs to add to the vfio driver, format is \"vendor:device[:subvendor[:subdevice[:class[:class_mask]]]]\" and multiple comma separated entries can be specified");
+
 static bool nointxmask;
 module_param_named(nointxmask, nointxmask, bool, S_IRUGO | S_IWUSR);
 MODULE_PARM_DESC(nointxmask,
@@ -1105,6 +1111,47 @@ static void __exit vfio_pci_cleanup(void)
 	vfio_pci_uninit_perm_bits();
 }
 
+static void __init vfio_pci_fill_ids(void)
+{
+	char *p, *id;
+	int rc;
+
+	/* no ids passed actually */
+	if (ids[0] == '\0')
+		return;
+
+	/* add ids specified in the module parameter */
+	p = ids;
+	while ((id = strsep(&p, ","))) {
+		unsigned int vendor, device, subvendor = PCI_ANY_ID,
+			subdevice = PCI_ANY_ID, class = 0, class_mask = 0;
+		int fields;
+
+		if (!strlen(id))
+			continue;
+
+		fields = sscanf(id, "%x:%x:%x:%x:%x:%x",
+				&vendor, &device, &subvendor, &subdevice,
+				&class, &class_mask);
+
+		if (fields < 2) {
+			pr_warn("invalid id string \"%s\"\n", id);
+			continue;
+		}
+
+		rc = pci_add_dynid(&vfio_pci_driver, vendor, device,
+				   subvendor, subdevice, class, class_mask, 0);
+		if (rc)
+			pr_warn("failed to add dynamic id [%04hx:%04hx[%04hx:%04hx]] class %#08x/%08x (%d)\n",
+				vendor, device, subvendor, subdevice,
+				class, class_mask, rc);
+		else
+			pr_info("add [%04hx:%04hx[%04hx:%04hx]] class %#08x/%08x\n",
+				vendor, device, subvendor, subdevice,
+				class, class_mask);
+	}
+}
+
 static int __init vfio_pci_init(void)
 {
 	int ret;
@@ -1119,6 +1166,8 @@ static int __init vfio_pci_init(void)
 	if (ret)
 		goto out_driver;
 
+	vfio_pci_fill_ids();
+
 	return 0;
 
 out_driver:

[PATCH v2 5/6] vfio-pci: Remove warning if try-reset fails

[Alex Williamson]

As indicated in the comment, this is not entirely uncommon and
causes user concern for no reason.

Signed-off-by: Alex Williamson <alex.williamson@redhat.com>
---
 drivers/vfio/pci/vfio_pci.c |   10 ++--------
 1 file changed, 2 insertions(+), 8 deletions(-)

diff --git a/drivers/vfio/pci/vfio_pci.c b/drivers/vfio/pci/vfio_pci.c
index 43517ce..d0f1e70 100644
--- a/drivers/vfio/pci/vfio_pci.c
+++ b/drivers/vfio/pci/vfio_pci.c
@@ -217,14 +217,8 @@ static void vfio_pci_disable(struct vfio_pci_device *vdev)
 	 * Try to reset the device.  The success of this is dependent on
 	 * being able to lock the device, which is not always possible.
 	 */
-	if (vdev->reset_works) {
-		int ret = pci_try_reset_function(pdev);
-		if (ret)
-			pr_warn("%s: Failed to reset device %s (%d)\n",
-				__func__, dev_name(&pdev->dev), ret);
-		else
-			vdev->needs_reset = false;
-	}
+	if (vdev->reset_works && !pci_try_reset_function(pdev))
+		vdev->needs_reset = false;
 
 	pci_restore_state(pdev);
 out:

[PATCH v2 6/6] vfio-pci: Move idle devices to D3hot power state

[Alex Williamson]

We can save some power by putting devices that are bound to vfio-pci
but not in use by the user in the D3hot power state.  Devices get
woken into D0 when opened by the user.  Resets return the device to
D0, so we need to re-apply the low power state after a bus reset.
It's tempting to try to use D3cold, but we have no reason to inhibit
hotplug of idle devices and we might get into a loop of having the
device disappear before we have a chance to try to use it.

A new module parameter allows this feature to be disabled if there are
devices that misbehave as a result of this change.

Signed-off-by: Alex Williamson <alex.williamson@redhat.com>
---
 drivers/vfio/pci/vfio_pci.c |   36 +++++++++++++++++++++++++++++++++---
 1 file changed, 33 insertions(+), 3 deletions(-)

diff --git a/drivers/vfio/pci/vfio_pci.c b/drivers/vfio/pci/vfio_pci.c
index d0f1e70..049b9e9 100644
--- a/drivers/vfio/pci/vfio_pci.c
+++ b/drivers/vfio/pci/vfio_pci.c
@@ -50,6 +50,11 @@ module_param(disable_vga, bool, S_IRUGO);
 MODULE_PARM_DESC(disable_vga, "Disable VGA resource access through vfio-pci");
 #endif
 
+static bool disable_idle_d3;
+module_param(disable_idle_d3, bool, S_IRUGO | S_IWUSR);
+MODULE_PARM_DESC(disable_idle_d3,
+		 "Disable using the PCI D3 low power state for idle, unused devices");
+
 static DEFINE_MUTEX(driver_lock);
 
 static inline bool vfio_vga_disabled(void)
@@ -114,6 +119,8 @@ static int vfio_pci_enable(struct vfio_pci_device *vdev)
 	u16 cmd;
 	u8 msix_pos;
 
+	pci_set_power_state(pdev, PCI_D0);
+
 	/* Don't allow our initial saved state to include busmaster */
 	pci_clear_master(pdev);
 
@@ -225,6 +232,9 @@ out:
 	pci_disable_device(pdev);
 
 	vfio_pci_try_bus_reset(vdev);
+
+	if (!disable_idle_d3)
+		pci_set_power_state(pdev, PCI_D3hot);
 }
 
 static void vfio_pci_release(void *device_data)
@@ -951,6 +961,20 @@ static int vfio_pci_probe(struct pci_dev *pdev, const struct pci_device_id *id)
 					vfio_pci_set_vga_decode(vdev, false));
 	}
 
+	if (!disable_idle_d3) {
+		/*
+		 * pci-core sets the device power state to an unknown value at
+		 * bootup and after being removed from a driver.  The only
+		 * transition it allows from this unknown state is to D0, which
+		 * typically happens when a driver calls pci_enable_device().
+		 * We're not ready to enable the device yet, but we do want to
+		 * be able to get to D3.  Therefore first do a D0 transition
+		 * before going to D3.
+		 */
+		pci_set_power_state(pdev, PCI_D0);
+		pci_set_power_state(pdev, PCI_D3hot);
+	}
+
 	return ret;
 }
 
@@ -971,6 +995,9 @@ static void vfio_pci_remove(struct pci_dev *pdev)
 				VGA_RSRC_NORMAL_IO | VGA_RSRC_NORMAL_MEM |
 				VGA_RSRC_LEGACY_IO | VGA_RSRC_LEGACY_MEM);
 	}
+
+	if (!disable_idle_d3)
+		pci_set_power_state(pdev, PCI_D0);
 }
 
 static pci_ers_result_t vfio_pci_aer_err_detected(struct pci_dev *pdev,
@@ -1089,10 +1116,13 @@ static void vfio_pci_try_bus_reset(struct vfio_pci_device *vdev)
 
 put_devs:
 	for (i = 0; i < devs.cur_index; i++) {
-		if (!ret) {
-			tmp = vfio_device_data(devs.devices[i]);
+		tmp = vfio_device_data(devs.devices[i]);
+		if (!ret)
 			tmp->needs_reset = false;
-		}
+
+		if (!tmp->refcnt && !disable_idle_d3)
+			pci_set_power_state(tmp->pdev, PCI_D3hot);
+
 		vfio_device_put(devs.devices[i]);
 	}