| Message ID | 166336088796.3597940.14973499936692558556.stgit@omen (mailing list archive) |
|---|---|
| State | Accepted |
| Delegated to: | Bjorn Helgaas |
| Headers | show |
| Series | [v2] PCI: Expose PCIe Resizable BAR support via sysfs | expand |
Am 16.09.22 um 22:44 schrieb Alex Williamson: > This proposes a simple sysfs interface to Resizable BAR support, > largely for the purposes of assigning such devices to a VM through > VFIO. Resizable BARs present a difficult feature to expose to a VM > through emulation, as resizing a BAR is done on the host. It can > fail, and often does, but we have no means via emulation of a PCIe > REBAR capability to handle the error cases. > > A vfio-pci specific ioctl interface is also cumbersome as there are > often multiple devices within the same bridge aperture and handling > them is a challenge. In the interface proposed here, expanding a > BAR potentially requires such devices to be soft-removed during the > resize operation and rescanned after, in order for all the necessary > resources to be released. A pci-sysfs interface is also more > universal than a vfio specific interface. > > Please see the ABI documentation update for usage. > > Cc: Christian König <christian.koenig@amd.com> > Cc: Krzysztof Wilczyński <kw@linux.com> > Signed-off-by: Alex Williamson <alex.williamson@redhat.com> I'm not an expert for that sysfs stuff, but the rest looks totally sane to me. Reviewed-by: Christian König <christian.koenig@amd.com> > --- > > v2: > - Convert to static attributes with is_visible callback > - Include aperture driver removal for console drivers > - Remove and recreate resourceN attributes > - Expand ABI description > - Drop 2nd field in show attribute > > Documentation/ABI/testing/sysfs-bus-pci | 33 +++++++++ > drivers/pci/pci-sysfs.c | 108 +++++++++++++++++++++++++++++++ > 2 files changed, 141 insertions(+) > > diff --git a/Documentation/ABI/testing/sysfs-bus-pci b/Documentation/ABI/testing/sysfs-bus-pci > index 6fc2c2efe8ab..ba9a5482436f 100644 > --- a/Documentation/ABI/testing/sysfs-bus-pci > +++ b/Documentation/ABI/testing/sysfs-bus-pci > @@ -457,3 +457,36 @@ Description: > > The file is writable if the PF is bound to a driver that > implements ->sriov_set_msix_vec_count(). > + > +What: /sys/bus/pci/devices/.../resourceN_resize > +Date: September 2022 > +Contact: Alex Williamson <alex.williamson@redhat.com> > +Description: > + These files provide an interface to PCIe Resizable BAR support. > + A file is created for each BAR resource (N) supported by the > + PCIe Resizable BAR extended capability of the device. Reading > + each file exposes the bitmap of available resources sizes: > + > + # cat resource1_resize > + 00000000000001c0 > + > + The bitmap represents supported resources sizes for the BAR, > + where bit0 = 1MB, bit1 = 2MB, bit2 = 4MB, etc. In the above > + example the devices supports 64MB, 128MB, and 256MB BAR sizes. > + > + When writing the file, the user provides the bit position of > + the desired resource size, for example: > + > + # echo 7 > resource1_resize > + > + This indicates to set the size value corresponding to bit 7, > + 128MB. The resulting size is 2 ^ (bit# + 20). This definition > + matches the PCIe specification of this capability. > + > + In order to make use of resouce resizing, all PCI drivers must > + be unbound from the device and peer devices under the same > + parent bridge may need to be soft removed. In the case of > + VGA devices, writing a resize value will remove low level > + console drivers from the device. Raw users of pci-sysfs > + resourceN attributes must be terminated prior to resizing. > + Success of the resizing operation is not a guaranteed. > diff --git a/drivers/pci/pci-sysfs.c b/drivers/pci/pci-sysfs.c > index 9ac92e6a2397..f0298a8b08d9 100644 > --- a/drivers/pci/pci-sysfs.c > +++ b/drivers/pci/pci-sysfs.c > @@ -28,6 +28,7 @@ > #include <linux/pm_runtime.h> > #include <linux/msi.h> > #include <linux/of.h> > +#include <linux/aperture.h> > #include "pci.h" > > static int sysfs_initialized; /* = 0 */ > @@ -1379,6 +1380,112 @@ static const struct attribute_group pci_dev_reset_attr_group = { > .is_visible = pci_dev_reset_attr_is_visible, > }; > > +#define pci_dev_resource_resize_attr(n) \ > +static ssize_t resource##n##_resize_show(struct device *dev, \ > + struct device_attribute *attr, \ > + char * buf) \ > +{ \ > + struct pci_dev *pdev = to_pci_dev(dev); \ > + ssize_t ret; \ > + \ > + pci_config_pm_runtime_get(pdev); \ > + \ > + ret = sysfs_emit(buf, "%016llx\n", \ > + (u64)pci_rebar_get_possible_sizes(pdev, n)); \ > + \ > + pci_config_pm_runtime_put(pdev); \ > + \ > + return ret; \ > +} \ > + \ > +static ssize_t resource##n##_resize_store(struct device *dev, \ > + struct device_attribute *attr,\ > + const char *buf, size_t count)\ > +{ \ > + struct pci_dev *pdev = to_pci_dev(dev); \ > + unsigned long size, flags; \ > + int ret, i; \ > + u16 cmd; \ > + \ > + if (kstrtoul(buf, 0, &size) < 0) \ > + return -EINVAL; \ > + \ > + device_lock(dev); \ > + if (dev->driver) { \ > + ret = -EBUSY; \ > + goto unlock; \ > + } \ > + \ > + pci_config_pm_runtime_get(pdev); \ > + \ > + if ((pdev->class >> 8) == PCI_CLASS_DISPLAY_VGA) { \ > + ret = aperture_remove_conflicting_pci_devices(pdev, \ > + "resourceN_resize"); \ > + if (ret) \ > + goto pm_put; \ > + } \ > + \ > + pci_read_config_word(pdev, PCI_COMMAND, &cmd); \ > + pci_write_config_word(pdev, PCI_COMMAND, \ > + cmd & ~PCI_COMMAND_MEMORY); \ > + \ > + flags = pci_resource_flags(pdev, n); \ > + \ > + pci_remove_resource_files(pdev); \ > + \ > + for (i = 0; i < PCI_STD_NUM_BARS; i++) { \ > + if (pci_resource_len(pdev, i) && \ > + pci_resource_flags(pdev, i) == flags) \ > + pci_release_resource(pdev, i); \ > + } \ > + \ > + ret = pci_resize_resource(pdev, n, size); \ > + \ > + pci_assign_unassigned_bus_resources(pdev->bus); \ > + \ > + if (pci_create_resource_files(pdev)) \ > + pci_warn(pdev, "Failed to recreate resource files after BAR resizing\n");\ > + \ > + pci_write_config_word(pdev, PCI_COMMAND, cmd); \ > +pm_put: \ > + pci_config_pm_runtime_put(pdev); \ > +unlock: \ > + device_unlock(dev); \ > + \ > + return ret ? ret : count; \ > +} \ > +static DEVICE_ATTR_RW(resource##n##_resize) > + > +pci_dev_resource_resize_attr(0); > +pci_dev_resource_resize_attr(1); > +pci_dev_resource_resize_attr(2); > +pci_dev_resource_resize_attr(3); > +pci_dev_resource_resize_attr(4); > +pci_dev_resource_resize_attr(5); > + > +static struct attribute *resource_resize_attrs[] = { > + &dev_attr_resource0_resize.attr, > + &dev_attr_resource1_resize.attr, > + &dev_attr_resource2_resize.attr, > + &dev_attr_resource3_resize.attr, > + &dev_attr_resource4_resize.attr, > + &dev_attr_resource5_resize.attr, > + NULL, > +}; > + > +static umode_t resource_resize_is_visible(struct kobject *kobj, > + struct attribute *a, int n) > +{ > + struct pci_dev *pdev = to_pci_dev(kobj_to_dev(kobj)); > + > + return pci_rebar_get_current_size(pdev, n) < 0 ? 0 : a->mode; > +} > + > +static const struct attribute_group pci_dev_resource_resize_group = { > + .attrs = resource_resize_attrs, > + .is_visible = resource_resize_is_visible, > +}; > + > int __must_check pci_create_sysfs_dev_files(struct pci_dev *pdev) > { > if (!sysfs_initialized) > @@ -1500,6 +1607,7 @@ const struct attribute_group *pci_dev_groups[] = { > #ifdef CONFIG_ACPI > &pci_dev_acpi_attr_group, > #endif > + &pci_dev_resource_resize_group, > NULL, > }; > > >
On Fri, Sep 16, 2022 at 02:44:48PM -0600, Alex Williamson wrote: > This proposes a simple sysfs interface to Resizable BAR support, > largely for the purposes of assigning such devices to a VM through > VFIO. Resizable BARs present a difficult feature to expose to a VM > through emulation, as resizing a BAR is done on the host. It can > fail, and often does, but we have no means via emulation of a PCIe > REBAR capability to handle the error cases. > > A vfio-pci specific ioctl interface is also cumbersome as there are > often multiple devices within the same bridge aperture and handling > them is a challenge. In the interface proposed here, expanding a > BAR potentially requires such devices to be soft-removed during the > resize operation and rescanned after, in order for all the necessary > resources to be released. A pci-sysfs interface is also more > universal than a vfio specific interface. > > Please see the ABI documentation update for usage. > > Cc: Christian König <christian.koenig@amd.com> > Cc: Krzysztof Wilczyński <kw@linux.com> > Signed-off-by: Alex Williamson <alex.williamson@redhat.com> Applied with Christian's Reviewed-by to pci/rebar for v6.1, thanks, Alex! > --- > > v2: > - Convert to static attributes with is_visible callback > - Include aperture driver removal for console drivers > - Remove and recreate resourceN attributes > - Expand ABI description > - Drop 2nd field in show attribute > > Documentation/ABI/testing/sysfs-bus-pci | 33 +++++++++ > drivers/pci/pci-sysfs.c | 108 +++++++++++++++++++++++++++++++ > 2 files changed, 141 insertions(+) > > diff --git a/Documentation/ABI/testing/sysfs-bus-pci b/Documentation/ABI/testing/sysfs-bus-pci > index 6fc2c2efe8ab..ba9a5482436f 100644 > --- a/Documentation/ABI/testing/sysfs-bus-pci > +++ b/Documentation/ABI/testing/sysfs-bus-pci > @@ -457,3 +457,36 @@ Description: > > The file is writable if the PF is bound to a driver that > implements ->sriov_set_msix_vec_count(). > + > +What: /sys/bus/pci/devices/.../resourceN_resize > +Date: September 2022 > +Contact: Alex Williamson <alex.williamson@redhat.com> > +Description: > + These files provide an interface to PCIe Resizable BAR support. > + A file is created for each BAR resource (N) supported by the > + PCIe Resizable BAR extended capability of the device. Reading > + each file exposes the bitmap of available resources sizes: > + > + # cat resource1_resize > + 00000000000001c0 > + > + The bitmap represents supported resources sizes for the BAR, > + where bit0 = 1MB, bit1 = 2MB, bit2 = 4MB, etc. In the above > + example the devices supports 64MB, 128MB, and 256MB BAR sizes. > + > + When writing the file, the user provides the bit position of > + the desired resource size, for example: > + > + # echo 7 > resource1_resize > + > + This indicates to set the size value corresponding to bit 7, > + 128MB. The resulting size is 2 ^ (bit# + 20). This definition > + matches the PCIe specification of this capability. > + > + In order to make use of resouce resizing, all PCI drivers must > + be unbound from the device and peer devices under the same > + parent bridge may need to be soft removed. In the case of > + VGA devices, writing a resize value will remove low level > + console drivers from the device. Raw users of pci-sysfs > + resourceN attributes must be terminated prior to resizing. > + Success of the resizing operation is not a guaranteed. > diff --git a/drivers/pci/pci-sysfs.c b/drivers/pci/pci-sysfs.c > index 9ac92e6a2397..f0298a8b08d9 100644 > --- a/drivers/pci/pci-sysfs.c > +++ b/drivers/pci/pci-sysfs.c > @@ -28,6 +28,7 @@ > #include <linux/pm_runtime.h> > #include <linux/msi.h> > #include <linux/of.h> > +#include <linux/aperture.h> > #include "pci.h" > > static int sysfs_initialized; /* = 0 */ > @@ -1379,6 +1380,112 @@ static const struct attribute_group pci_dev_reset_attr_group = { > .is_visible = pci_dev_reset_attr_is_visible, > }; > > +#define pci_dev_resource_resize_attr(n) \ > +static ssize_t resource##n##_resize_show(struct device *dev, \ > + struct device_attribute *attr, \ > + char * buf) \ > +{ \ > + struct pci_dev *pdev = to_pci_dev(dev); \ > + ssize_t ret; \ > + \ > + pci_config_pm_runtime_get(pdev); \ > + \ > + ret = sysfs_emit(buf, "%016llx\n", \ > + (u64)pci_rebar_get_possible_sizes(pdev, n)); \ > + \ > + pci_config_pm_runtime_put(pdev); \ > + \ > + return ret; \ > +} \ > + \ > +static ssize_t resource##n##_resize_store(struct device *dev, \ > + struct device_attribute *attr,\ > + const char *buf, size_t count)\ > +{ \ > + struct pci_dev *pdev = to_pci_dev(dev); \ > + unsigned long size, flags; \ > + int ret, i; \ > + u16 cmd; \ > + \ > + if (kstrtoul(buf, 0, &size) < 0) \ > + return -EINVAL; \ > + \ > + device_lock(dev); \ > + if (dev->driver) { \ > + ret = -EBUSY; \ > + goto unlock; \ > + } \ > + \ > + pci_config_pm_runtime_get(pdev); \ > + \ > + if ((pdev->class >> 8) == PCI_CLASS_DISPLAY_VGA) { \ > + ret = aperture_remove_conflicting_pci_devices(pdev, \ > + "resourceN_resize"); \ > + if (ret) \ > + goto pm_put; \ > + } \ > + \ > + pci_read_config_word(pdev, PCI_COMMAND, &cmd); \ > + pci_write_config_word(pdev, PCI_COMMAND, \ > + cmd & ~PCI_COMMAND_MEMORY); \ > + \ > + flags = pci_resource_flags(pdev, n); \ > + \ > + pci_remove_resource_files(pdev); \ > + \ > + for (i = 0; i < PCI_STD_NUM_BARS; i++) { \ > + if (pci_resource_len(pdev, i) && \ > + pci_resource_flags(pdev, i) == flags) \ > + pci_release_resource(pdev, i); \ > + } \ > + \ > + ret = pci_resize_resource(pdev, n, size); \ > + \ > + pci_assign_unassigned_bus_resources(pdev->bus); \ > + \ > + if (pci_create_resource_files(pdev)) \ > + pci_warn(pdev, "Failed to recreate resource files after BAR resizing\n");\ > + \ > + pci_write_config_word(pdev, PCI_COMMAND, cmd); \ > +pm_put: \ > + pci_config_pm_runtime_put(pdev); \ > +unlock: \ > + device_unlock(dev); \ > + \ > + return ret ? ret : count; \ > +} \ > +static DEVICE_ATTR_RW(resource##n##_resize) > + > +pci_dev_resource_resize_attr(0); > +pci_dev_resource_resize_attr(1); > +pci_dev_resource_resize_attr(2); > +pci_dev_resource_resize_attr(3); > +pci_dev_resource_resize_attr(4); > +pci_dev_resource_resize_attr(5); > + > +static struct attribute *resource_resize_attrs[] = { > + &dev_attr_resource0_resize.attr, > + &dev_attr_resource1_resize.attr, > + &dev_attr_resource2_resize.attr, > + &dev_attr_resource3_resize.attr, > + &dev_attr_resource4_resize.attr, > + &dev_attr_resource5_resize.attr, > + NULL, > +}; > + > +static umode_t resource_resize_is_visible(struct kobject *kobj, > + struct attribute *a, int n) > +{ > + struct pci_dev *pdev = to_pci_dev(kobj_to_dev(kobj)); > + > + return pci_rebar_get_current_size(pdev, n) < 0 ? 0 : a->mode; > +} > + > +static const struct attribute_group pci_dev_resource_resize_group = { > + .attrs = resource_resize_attrs, > + .is_visible = resource_resize_is_visible, > +}; > + > int __must_check pci_create_sysfs_dev_files(struct pci_dev *pdev) > { > if (!sysfs_initialized) > @@ -1500,6 +1607,7 @@ const struct attribute_group *pci_dev_groups[] = { > #ifdef CONFIG_ACPI > &pci_dev_acpi_attr_group, > #endif > + &pci_dev_resource_resize_group, > NULL, > }; > > >
diff --git a/Documentation/ABI/testing/sysfs-bus-pci b/Documentation/ABI/testing/sysfs-bus-pci index 6fc2c2efe8ab..ba9a5482436f 100644 --- a/Documentation/ABI/testing/sysfs-bus-pci +++ b/Documentation/ABI/testing/sysfs-bus-pci @@ -457,3 +457,36 @@ Description: The file is writable if the PF is bound to a driver that implements ->sriov_set_msix_vec_count(). + +What: /sys/bus/pci/devices/.../resourceN_resize +Date: September 2022 +Contact: Alex Williamson <alex.williamson@redhat.com> +Description: + These files provide an interface to PCIe Resizable BAR support. + A file is created for each BAR resource (N) supported by the + PCIe Resizable BAR extended capability of the device. Reading + each file exposes the bitmap of available resources sizes: + + # cat resource1_resize + 00000000000001c0 + + The bitmap represents supported resources sizes for the BAR, + where bit0 = 1MB, bit1 = 2MB, bit2 = 4MB, etc. In the above + example the devices supports 64MB, 128MB, and 256MB BAR sizes. + + When writing the file, the user provides the bit position of + the desired resource size, for example: + + # echo 7 > resource1_resize + + This indicates to set the size value corresponding to bit 7, + 128MB. The resulting size is 2 ^ (bit# + 20). This definition + matches the PCIe specification of this capability. + + In order to make use of resouce resizing, all PCI drivers must + be unbound from the device and peer devices under the same + parent bridge may need to be soft removed. In the case of + VGA devices, writing a resize value will remove low level + console drivers from the device. Raw users of pci-sysfs + resourceN attributes must be terminated prior to resizing. + Success of the resizing operation is not a guaranteed. diff --git a/drivers/pci/pci-sysfs.c b/drivers/pci/pci-sysfs.c index 9ac92e6a2397..f0298a8b08d9 100644 --- a/drivers/pci/pci-sysfs.c +++ b/drivers/pci/pci-sysfs.c @@ -28,6 +28,7 @@ #include <linux/pm_runtime.h> #include <linux/msi.h> #include <linux/of.h> +#include <linux/aperture.h> #include "pci.h" static int sysfs_initialized; /* = 0 */ @@ -1379,6 +1380,112 @@ static const struct attribute_group pci_dev_reset_attr_group = { .is_visible = pci_dev_reset_attr_is_visible, }; +#define pci_dev_resource_resize_attr(n) \ +static ssize_t resource##n##_resize_show(struct device *dev, \ + struct device_attribute *attr, \ + char * buf) \ +{ \ + struct pci_dev *pdev = to_pci_dev(dev); \ + ssize_t ret; \ + \ + pci_config_pm_runtime_get(pdev); \ + \ + ret = sysfs_emit(buf, "%016llx\n", \ + (u64)pci_rebar_get_possible_sizes(pdev, n)); \ + \ + pci_config_pm_runtime_put(pdev); \ + \ + return ret; \ +} \ + \ +static ssize_t resource##n##_resize_store(struct device *dev, \ + struct device_attribute *attr,\ + const char *buf, size_t count)\ +{ \ + struct pci_dev *pdev = to_pci_dev(dev); \ + unsigned long size, flags; \ + int ret, i; \ + u16 cmd; \ + \ + if (kstrtoul(buf, 0, &size) < 0) \ + return -EINVAL; \ + \ + device_lock(dev); \ + if (dev->driver) { \ + ret = -EBUSY; \ + goto unlock; \ + } \ + \ + pci_config_pm_runtime_get(pdev); \ + \ + if ((pdev->class >> 8) == PCI_CLASS_DISPLAY_VGA) { \ + ret = aperture_remove_conflicting_pci_devices(pdev, \ + "resourceN_resize"); \ + if (ret) \ + goto pm_put; \ + } \ + \ + pci_read_config_word(pdev, PCI_COMMAND, &cmd); \ + pci_write_config_word(pdev, PCI_COMMAND, \ + cmd & ~PCI_COMMAND_MEMORY); \ + \ + flags = pci_resource_flags(pdev, n); \ + \ + pci_remove_resource_files(pdev); \ + \ + for (i = 0; i < PCI_STD_NUM_BARS; i++) { \ + if (pci_resource_len(pdev, i) && \ + pci_resource_flags(pdev, i) == flags) \ + pci_release_resource(pdev, i); \ + } \ + \ + ret = pci_resize_resource(pdev, n, size); \ + \ + pci_assign_unassigned_bus_resources(pdev->bus); \ + \ + if (pci_create_resource_files(pdev)) \ + pci_warn(pdev, "Failed to recreate resource files after BAR resizing\n");\ + \ + pci_write_config_word(pdev, PCI_COMMAND, cmd); \ +pm_put: \ + pci_config_pm_runtime_put(pdev); \ +unlock: \ + device_unlock(dev); \ + \ + return ret ? ret : count; \ +} \ +static DEVICE_ATTR_RW(resource##n##_resize) + +pci_dev_resource_resize_attr(0); +pci_dev_resource_resize_attr(1); +pci_dev_resource_resize_attr(2); +pci_dev_resource_resize_attr(3); +pci_dev_resource_resize_attr(4); +pci_dev_resource_resize_attr(5); + +static struct attribute *resource_resize_attrs[] = { + &dev_attr_resource0_resize.attr, + &dev_attr_resource1_resize.attr, + &dev_attr_resource2_resize.attr, + &dev_attr_resource3_resize.attr, + &dev_attr_resource4_resize.attr, + &dev_attr_resource5_resize.attr, + NULL, +}; + +static umode_t resource_resize_is_visible(struct kobject *kobj, + struct attribute *a, int n) +{ + struct pci_dev *pdev = to_pci_dev(kobj_to_dev(kobj)); + + return pci_rebar_get_current_size(pdev, n) < 0 ? 0 : a->mode; +} + +static const struct attribute_group pci_dev_resource_resize_group = { + .attrs = resource_resize_attrs, + .is_visible = resource_resize_is_visible, +}; + int __must_check pci_create_sysfs_dev_files(struct pci_dev *pdev) { if (!sysfs_initialized) @@ -1500,6 +1607,7 @@ const struct attribute_group *pci_dev_groups[] = { #ifdef CONFIG_ACPI &pci_dev_acpi_attr_group, #endif + &pci_dev_resource_resize_group, NULL, };
This proposes a simple sysfs interface to Resizable BAR support, largely for the purposes of assigning such devices to a VM through VFIO. Resizable BARs present a difficult feature to expose to a VM through emulation, as resizing a BAR is done on the host. It can fail, and often does, but we have no means via emulation of a PCIe REBAR capability to handle the error cases. A vfio-pci specific ioctl interface is also cumbersome as there are often multiple devices within the same bridge aperture and handling them is a challenge. In the interface proposed here, expanding a BAR potentially requires such devices to be soft-removed during the resize operation and rescanned after, in order for all the necessary resources to be released. A pci-sysfs interface is also more universal than a vfio specific interface. Please see the ABI documentation update for usage. Cc: Christian König <christian.koenig@amd.com> Cc: Krzysztof Wilczyński <kw@linux.com> Signed-off-by: Alex Williamson <alex.williamson@redhat.com> --- v2: - Convert to static attributes with is_visible callback - Include aperture driver removal for console drivers - Remove and recreate resourceN attributes - Expand ABI description - Drop 2nd field in show attribute Documentation/ABI/testing/sysfs-bus-pci | 33 +++++++++ drivers/pci/pci-sysfs.c | 108 +++++++++++++++++++++++++++++++ 2 files changed, 141 insertions(+)