Message ID | 20210103082440.34994-2-leon@kernel.org (mailing list archive) |
---|---|
State | Superseded |
Headers | show |
Series | Dynamically assign MSI-X vectors count | expand |
Context | Check | Description |
---|---|---|
netdev/tree_selection | success | Guessing tree name failed - patch did not apply |
[+cc Alex, Don] This patch does not actually *configure* the number of vectors, so the subject is not quite accurate. IIUC, this patch adds a sysfs file that can be used to configure the number of vectors. The subject should mention the sysfs connection. On Sun, Jan 03, 2021 at 10:24:37AM +0200, Leon Romanovsky wrote: > From: Leon Romanovsky <leonro@nvidia.com> > > This function is applicable for SR-IOV VFs because such devices allocate > their MSI-X table before they will run on the targeted hardware and they > can't guess the right amount of vectors. This sentence doesn't quite have enough context to make sense to me. Per PCIe r5.0, sec 9.5.1.2, I think PFs and VFs have independent MSI-X Capabilities. What is the connection between the PF MSI-X and the VF MSI-X? The MSI-X table sizes should be determined by the Table Size in the Message Control register. Apparently we write a VF's Table Size before a driver is bound to the VF? Where does that happen? "Before they run on the targeted hardware" -- do you mean before the VF is passed through to a guest virtual machine? You mention "target VM" below, which makes more sense to me. VFs don't "run"; they're not software. I apologize for not being an expert in the use of VFs. Please mention the sysfs path in the commit log. > Signed-off-by: Leon Romanovsky <leonro@nvidia.com> > --- > Documentation/ABI/testing/sysfs-bus-pci | 16 +++++++ > drivers/pci/iov.c | 57 +++++++++++++++++++++++++ > drivers/pci/msi.c | 30 +++++++++++++ > drivers/pci/pci-sysfs.c | 1 + > drivers/pci/pci.h | 1 + > include/linux/pci.h | 8 ++++ > 6 files changed, 113 insertions(+) > > diff --git a/Documentation/ABI/testing/sysfs-bus-pci b/Documentation/ABI/testing/sysfs-bus-pci > index 25c9c39770c6..30720a9e1386 100644 > --- a/Documentation/ABI/testing/sysfs-bus-pci > +++ b/Documentation/ABI/testing/sysfs-bus-pci > @@ -375,3 +375,19 @@ Description: > The value comes from the PCI kernel device state and can be one > of: "unknown", "error", "D0", D1", "D2", "D3hot", "D3cold". > The file is read only. > + > +What: /sys/bus/pci/devices/.../vf_msix_vec > +Date: December 2020 > +Contact: Leon Romanovsky <leonro@nvidia.com> > +Description: > + This file is associated with the SR-IOV VFs. It allows overwrite > + the amount of MSI-X vectors for that VF. This is needed to optimize > + performance of newly bounded devices by allocating the number of > + vectors based on the internal knowledge of targeted VM. s/allows overwrite/allows configuration of/ s/for that/for the/ s/amount of/number of/ s/bounded/bound/ What "internal knowledge" is this? AFAICT this would have to be some user-space administration knowledge, not anything internal to the kernel. > + The values accepted are: > + * > 0 - this will be number reported by the PCI VF's PCIe MSI-X capability. s/PCI// (it's obvious we're talking about PCI here) s/PCIe// (MSI-X is not PCIe-specific, and there's no need to mention it at all) > + * < 0 - not valid > + * = 0 - will reset to the device default value > + > + The file is writable if no driver is bounded. From the code, it looks more like this: The file is writable if the PF is bound to a driver that supports the ->sriov_set_msix_vec_count() callback and there is no driver bound to the VF. Please wrap all of this to fit in 80 columns like the rest of the file. > diff --git a/drivers/pci/iov.c b/drivers/pci/iov.c > index 4afd4ee4f7f0..0f8c570361fc 100644 > --- a/drivers/pci/iov.c > +++ b/drivers/pci/iov.c > @@ -31,6 +31,7 @@ int pci_iov_virtfn_devfn(struct pci_dev *dev, int vf_id) > return (dev->devfn + dev->sriov->offset + > dev->sriov->stride * vf_id) & 0xff; > } > +EXPORT_SYMBOL(pci_iov_virtfn_devfn); > > /* > * Per SR-IOV spec sec 3.3.10 and 3.3.11, First VF Offset and VF Stride may > @@ -426,6 +427,62 @@ const struct attribute_group sriov_dev_attr_group = { > .is_visible = sriov_attrs_are_visible, > }; > > +#ifdef CONFIG_PCI_MSI > +static ssize_t vf_msix_vec_show(struct device *dev, > + struct device_attribute *attr, char *buf) > +{ > + struct pci_dev *pdev = to_pci_dev(dev); > + int numb = pci_msix_vec_count(pdev); > + > + if (numb < 0) > + return numb; > + > + return sprintf(buf, "%d\n", numb); > +} > + > +static ssize_t vf_msix_vec_store(struct device *dev, > + struct device_attribute *attr, const char *buf, > + size_t count) > +{ > + struct pci_dev *vf_dev = to_pci_dev(dev); > + int val, ret; > + > + ret = kstrtoint(buf, 0, &val); > + if (ret) > + return ret; > + > + ret = pci_set_msix_vec_count(vf_dev, val); > + if (ret) > + return ret; > + > + return count; > +} > +static DEVICE_ATTR_RW(vf_msix_vec); > +#endif > + > +static struct attribute *sriov_vf_dev_attrs[] = { > +#ifdef CONFIG_PCI_MSI > + &dev_attr_vf_msix_vec.attr, > +#endif > + NULL, > +}; > + > +static umode_t sriov_vf_attrs_are_visible(struct kobject *kobj, > + struct attribute *a, int n) > +{ > + struct device *dev = kobj_to_dev(kobj); > + > + if (dev_is_pf(dev)) > + return 0; > + > + return a->mode; > +} > + > +const struct attribute_group sriov_vf_dev_attr_group = { > + .attrs = sriov_vf_dev_attrs, > + .is_visible = sriov_vf_attrs_are_visible, > +}; > + > int __weak pcibios_sriov_enable(struct pci_dev *pdev, u16 num_vfs) > { > return 0; > diff --git a/drivers/pci/msi.c b/drivers/pci/msi.c > index 3162f88fe940..0bcd705487d9 100644 > --- a/drivers/pci/msi.c > +++ b/drivers/pci/msi.c > @@ -991,6 +991,36 @@ int pci_msix_vec_count(struct pci_dev *dev) > } > EXPORT_SYMBOL(pci_msix_vec_count); > > +/** > + * pci_set_msix_vec_count - change the reported number of MSI-X vectors. Drop period at end, as other kernel doc in this file does. > + * This function is applicable for SR-IOV VFs because such devices allocate > + * their MSI-X table before they will run on the targeted hardware and they > + * can't guess the right amount of vectors. > + * @dev: VF device that is going to be changed. > + * @numb: amount of MSI-X vectors. Rewrite the "such devices allocate..." part based on the questions in the commit log. Same with "targeted hardware." s/amount of/number of/ Drop periods at end of parameter descriptions. > + **/ > +int pci_set_msix_vec_count(struct pci_dev *dev, int numb) > +{ > + struct pci_dev *pdev = pci_physfn(dev); > + > + if (!dev->msix_cap || !pdev->msix_cap) > + return -EINVAL; > + > + if (dev->driver || !pdev->driver || > + !pdev->driver->sriov_set_msix_vec_count) > + return -EOPNOTSUPP; > + > + if (numb < 0) > + /* > + * We don't support negative numbers for now, > + * but maybe in the future it will make sense. > + */ > + return -EINVAL; > + > + return pdev->driver->sriov_set_msix_vec_count(dev, numb); So we write to a VF sysfs file, get here and look up the PF, call a PF driver callback with the VF as an argument, the callback (at least for mlx5) looks up the PF from the VF, then does some mlx5-specific magic to the PF that influences the VF somehow? Help me connect the dots here. Is this required because of something peculiar to mlx5, or is something like this required for all SR-IOV devices because of the way the PCIe spec is written? > +} > +EXPORT_SYMBOL(pci_set_msix_vec_count); > + > static int __pci_enable_msix(struct pci_dev *dev, struct msix_entry *entries, > int nvec, struct irq_affinity *affd, int flags) > { > diff --git a/drivers/pci/pci-sysfs.c b/drivers/pci/pci-sysfs.c > index fb072f4b3176..0af2222643c2 100644 > --- a/drivers/pci/pci-sysfs.c > +++ b/drivers/pci/pci-sysfs.c > @@ -1557,6 +1557,7 @@ static const struct attribute_group *pci_dev_attr_groups[] = { > &pci_dev_hp_attr_group, > #ifdef CONFIG_PCI_IOV > &sriov_dev_attr_group, > + &sriov_vf_dev_attr_group, > #endif > &pci_bridge_attr_group, > &pcie_dev_attr_group, > diff --git a/drivers/pci/pci.h b/drivers/pci/pci.h > index 5c59365092fa..46396a5da2d9 100644 > --- a/drivers/pci/pci.h > +++ b/drivers/pci/pci.h > @@ -502,6 +502,7 @@ resource_size_t pci_sriov_resource_alignment(struct pci_dev *dev, int resno); > void pci_restore_iov_state(struct pci_dev *dev); > int pci_iov_bus_range(struct pci_bus *bus); > extern const struct attribute_group sriov_dev_attr_group; > +extern const struct attribute_group sriov_vf_dev_attr_group; > #else > static inline int pci_iov_init(struct pci_dev *dev) > { > diff --git a/include/linux/pci.h b/include/linux/pci.h > index b32126d26997..1acba40a1b1b 100644 > --- a/include/linux/pci.h > +++ b/include/linux/pci.h > @@ -856,6 +856,8 @@ struct module; > * e.g. drivers/net/e100.c. > * @sriov_configure: Optional driver callback to allow configuration of > * number of VFs to enable via sysfs "sriov_numvfs" file. > + * @sriov_set_msix_vec_count: Driver callback to change number of MSI-X vectors > + * exposed by the sysfs "vf_msix_vec" entry. > * @err_handler: See Documentation/PCI/pci-error-recovery.rst > * @groups: Sysfs attribute groups. > * @driver: Driver model structure. > @@ -871,6 +873,7 @@ struct pci_driver { > int (*resume)(struct pci_dev *dev); /* Device woken up */ > void (*shutdown)(struct pci_dev *dev); > int (*sriov_configure)(struct pci_dev *dev, int num_vfs); /* On PF */ > + int (*sriov_set_msix_vec_count)(struct pci_dev *vf, int msix_vec_count); /* On PF */ > const struct pci_error_handlers *err_handler; > const struct attribute_group **groups; > struct device_driver driver; > @@ -1464,6 +1467,7 @@ struct msix_entry { > int pci_msi_vec_count(struct pci_dev *dev); > void pci_disable_msi(struct pci_dev *dev); > int pci_msix_vec_count(struct pci_dev *dev); > +int pci_set_msix_vec_count(struct pci_dev *dev, int numb); This patch adds the pci_set_msix_vec_count() definition in pci/msi.c and a call in pci/iov.c. It doesn't need to be declared in include/linux/pci.h or exported. It can be declared in drivers/pci/pci.h. > void pci_disable_msix(struct pci_dev *dev); > void pci_restore_msi_state(struct pci_dev *dev); > int pci_msi_enabled(void); > @@ -2402,6 +2406,10 @@ static inline bool pci_is_thunderbolt_attached(struct pci_dev *pdev) > void pci_uevent_ers(struct pci_dev *pdev, enum pci_ers_result err_type); > #endif > > +#ifdef CONFIG_PCI_IOV > +int pci_iov_virtfn_devfn(struct pci_dev *dev, int vf_id); > +#endif pci_iov_virtfn_devfn() is already declared in this file. > /* Provide the legacy pci_dma_* API */ > #include <linux/pci-dma-compat.h> > > -- > 2.29.2 >
On 1/7/21 7:57 PM, Bjorn Helgaas wrote: > [+cc Alex, Don] > > This patch does not actually *configure* the number of vectors, so the > subject is not quite accurate. IIUC, this patch adds a sysfs file > that can be used to configure the number of vectors. The subject > should mention the sysfs connection. > > On Sun, Jan 03, 2021 at 10:24:37AM +0200, Leon Romanovsky wrote: >> From: Leon Romanovsky <leonro@nvidia.com> >> >> This function is applicable for SR-IOV VFs because such devices allocate >> their MSI-X table before they will run on the targeted hardware and they >> can't guess the right amount of vectors. > This sentence doesn't quite have enough context to make sense to me. > Per PCIe r5.0, sec 9.5.1.2, I think PFs and VFs have independent MSI-X > Capabilities. What is the connection between the PF MSI-X and the VF > MSI-X? +1... strip this commit log section and write it with correct, technical content. PFs & VF's have indep MSIX caps. Q: is this an issue where (some) mlx5's have a large msi-x capability (per VF) that may overwhelm a system's, (pci-(sub)-tree) MSI / intr capability, and this is a sysfs-based tuning knob to reduce the max number on such 'challenged' systems? -- ah; reading further below, it's based on some information gleemed from the VM's capability for intr. support. -- or maybe IOMMU (intr) support on the host system, and the VF can't exceed it or config failure in VM... whatever... its some VM cap that's being accomodated. > The MSI-X table sizes should be determined by the Table Size in the > Message Control register. Apparently we write a VF's Table Size > before a driver is bound to the VF? Where does that happen? > > "Before they run on the targeted hardware" -- do you mean before the > VF is passed through to a guest virtual machine? You mention "target > VM" below, which makes more sense to me. VFs don't "run"; they're not > software. I apologize for not being an expert in the use of VFs. > > Please mention the sysfs path in the commit log. > >> Signed-off-by: Leon Romanovsky <leonro@nvidia.com> >> --- >> Documentation/ABI/testing/sysfs-bus-pci | 16 +++++++ >> drivers/pci/iov.c | 57 +++++++++++++++++++++++++ >> drivers/pci/msi.c | 30 +++++++++++++ >> drivers/pci/pci-sysfs.c | 1 + >> drivers/pci/pci.h | 1 + >> include/linux/pci.h | 8 ++++ >> 6 files changed, 113 insertions(+) >> >> diff --git a/Documentation/ABI/testing/sysfs-bus-pci b/Documentation/ABI/testing/sysfs-bus-pci >> index 25c9c39770c6..30720a9e1386 100644 >> --- a/Documentation/ABI/testing/sysfs-bus-pci >> +++ b/Documentation/ABI/testing/sysfs-bus-pci >> @@ -375,3 +375,19 @@ Description: >> The value comes from the PCI kernel device state and can be one >> of: "unknown", "error", "D0", D1", "D2", "D3hot", "D3cold". >> The file is read only. >> + >> +What: /sys/bus/pci/devices/.../vf_msix_vec >> +Date: December 2020 >> +Contact: Leon Romanovsky <leonro@nvidia.com> >> +Description: >> + This file is associated with the SR-IOV VFs. It allows overwrite >> + the amount of MSI-X vectors for that VF. This is needed to optimize >> + performance of newly bounded devices by allocating the number of >> + vectors based on the internal knowledge of targeted VM. > s/allows overwrite/allows configuration of/ > s/for that/for the/ > s/amount of/number of/ > s/bounded/bound/ > > What "internal knowledge" is this? AFAICT this would have to be some > user-space administration knowledge, not anything internal to the > kernel. Correct; likely a libvirt VM (section of its) description; > >> + The values accepted are: >> + * > 0 - this will be number reported by the PCI VF's PCIe MSI-X capability. > s/PCI// (it's obvious we're talking about PCI here) > s/PCIe// (MSI-X is not PCIe-specific, and there's no need to mention > it at all) > >> + * < 0 - not valid >> + * = 0 - will reset to the device default value >> + >> + The file is writable if no driver is bounded. > From the code, it looks more like this: > > The file is writable if the PF is bound to a driver that supports > the ->sriov_set_msix_vec_count() callback and there is no driver > bound to the VF. > > Please wrap all of this to fit in 80 columns like the rest of the file. > >> diff --git a/drivers/pci/iov.c b/drivers/pci/iov.c >> index 4afd4ee4f7f0..0f8c570361fc 100644 >> --- a/drivers/pci/iov.c >> +++ b/drivers/pci/iov.c >> @@ -31,6 +31,7 @@ int pci_iov_virtfn_devfn(struct pci_dev *dev, int vf_id) >> return (dev->devfn + dev->sriov->offset + >> dev->sriov->stride * vf_id) & 0xff; >> } >> +EXPORT_SYMBOL(pci_iov_virtfn_devfn); >> >> /* >> * Per SR-IOV spec sec 3.3.10 and 3.3.11, First VF Offset and VF Stride may >> @@ -426,6 +427,62 @@ const struct attribute_group sriov_dev_attr_group = { >> .is_visible = sriov_attrs_are_visible, >> }; >> >> +#ifdef CONFIG_PCI_MSI >> +static ssize_t vf_msix_vec_show(struct device *dev, >> + struct device_attribute *attr, char *buf) >> +{ >> + struct pci_dev *pdev = to_pci_dev(dev); >> + int numb = pci_msix_vec_count(pdev); >> + >> + if (numb < 0) >> + return numb; >> + >> + return sprintf(buf, "%d\n", numb); >> +} >> + >> +static ssize_t vf_msix_vec_store(struct device *dev, >> + struct device_attribute *attr, const char *buf, >> + size_t count) >> +{ >> + struct pci_dev *vf_dev = to_pci_dev(dev); >> + int val, ret; >> + >> + ret = kstrtoint(buf, 0, &val); >> + if (ret) >> + return ret; >> + >> + ret = pci_set_msix_vec_count(vf_dev, val); >> + if (ret) >> + return ret; >> + >> + return count; >> +} >> +static DEVICE_ATTR_RW(vf_msix_vec); >> +#endif >> + >> +static struct attribute *sriov_vf_dev_attrs[] = { >> +#ifdef CONFIG_PCI_MSI >> + &dev_attr_vf_msix_vec.attr, >> +#endif >> + NULL, >> +}; >> + >> +static umode_t sriov_vf_attrs_are_visible(struct kobject *kobj, >> + struct attribute *a, int n) >> +{ >> + struct device *dev = kobj_to_dev(kobj); >> + >> + if (dev_is_pf(dev)) >> + return 0; >> + >> + return a->mode; >> +} >> + >> +const struct attribute_group sriov_vf_dev_attr_group = { >> + .attrs = sriov_vf_dev_attrs, >> + .is_visible = sriov_vf_attrs_are_visible, >> +}; >> + >> int __weak pcibios_sriov_enable(struct pci_dev *pdev, u16 num_vfs) >> { >> return 0; >> diff --git a/drivers/pci/msi.c b/drivers/pci/msi.c >> index 3162f88fe940..0bcd705487d9 100644 >> --- a/drivers/pci/msi.c >> +++ b/drivers/pci/msi.c >> @@ -991,6 +991,36 @@ int pci_msix_vec_count(struct pci_dev *dev) >> } >> EXPORT_SYMBOL(pci_msix_vec_count); >> >> +/** >> + * pci_set_msix_vec_count - change the reported number of MSI-X vectors. > Drop period at end, as other kernel doc in this file does. > >> + * This function is applicable for SR-IOV VFs because such devices allocate >> + * their MSI-X table before they will run on the targeted hardware and they >> + * can't guess the right amount of vectors. >> + * @dev: VF device that is going to be changed. >> + * @numb: amount of MSI-X vectors. > Rewrite the "such devices allocate..." part based on the questions in > the commit log. Same with "targeted hardware." > > s/amount of/number of/ > Drop periods at end of parameter descriptions. > >> + **/ >> +int pci_set_msix_vec_count(struct pci_dev *dev, int numb) >> +{ >> + struct pci_dev *pdev = pci_physfn(dev); >> + >> + if (!dev->msix_cap || !pdev->msix_cap) >> + return -EINVAL; >> + >> + if (dev->driver || !pdev->driver || >> + !pdev->driver->sriov_set_msix_vec_count) >> + return -EOPNOTSUPP; >> + >> + if (numb < 0) >> + /* >> + * We don't support negative numbers for now, >> + * but maybe in the future it will make sense. >> + */ >> + return -EINVAL; >> + >> + return pdev->driver->sriov_set_msix_vec_count(dev, numb); > So we write to a VF sysfs file, get here and look up the PF, call a PF > driver callback with the VF as an argument, the callback (at least for > mlx5) looks up the PF from the VF, then does some mlx5-specific magic > to the PF that influences the VF somehow? There's no PF lookup above.... it's just checking if a pdev has a driver with the desired msix-cap setting(reduction) feature. > Help me connect the dots here. Is this required because of something > peculiar to mlx5, or is something like this required for all SR-IOV > devices because of the way the PCIe spec is written? So, overall, I'm guessing the mlx5 device can have 1000's of MSIX -- say, one per send/receive/completion queue. This device capability may exceed the max number MSIX a VM can have/support (depending on guestos). So, a sysfs tunable is used to set the max MSIX available, and thus, the device puts >1 send/rcv/completion queue intr on a given MSIX. ok, time for Leon to better state what this patch does, and why it's needed on mlx5 (and may be applicable to other/future high-MSIX devices assigned to VMs (NVME?)). Hmmm, now that I said it, why is it SRIOV-centric and not pci-device centric (can pass a PF w/high number of MSIX to a VM). -Don >> +} >> +EXPORT_SYMBOL(pci_set_msix_vec_count); >> + >> static int __pci_enable_msix(struct pci_dev *dev, struct msix_entry *entries, >> int nvec, struct irq_affinity *affd, int flags) >> { >> diff --git a/drivers/pci/pci-sysfs.c b/drivers/pci/pci-sysfs.c >> index fb072f4b3176..0af2222643c2 100644 >> --- a/drivers/pci/pci-sysfs.c >> +++ b/drivers/pci/pci-sysfs.c >> @@ -1557,6 +1557,7 @@ static const struct attribute_group *pci_dev_attr_groups[] = { >> &pci_dev_hp_attr_group, >> #ifdef CONFIG_PCI_IOV >> &sriov_dev_attr_group, >> + &sriov_vf_dev_attr_group, >> #endif >> &pci_bridge_attr_group, >> &pcie_dev_attr_group, >> diff --git a/drivers/pci/pci.h b/drivers/pci/pci.h >> index 5c59365092fa..46396a5da2d9 100644 >> --- a/drivers/pci/pci.h >> +++ b/drivers/pci/pci.h >> @@ -502,6 +502,7 @@ resource_size_t pci_sriov_resource_alignment(struct pci_dev *dev, int resno); >> void pci_restore_iov_state(struct pci_dev *dev); >> int pci_iov_bus_range(struct pci_bus *bus); >> extern const struct attribute_group sriov_dev_attr_group; >> +extern const struct attribute_group sriov_vf_dev_attr_group; >> #else >> static inline int pci_iov_init(struct pci_dev *dev) >> { >> diff --git a/include/linux/pci.h b/include/linux/pci.h >> index b32126d26997..1acba40a1b1b 100644 >> --- a/include/linux/pci.h >> +++ b/include/linux/pci.h >> @@ -856,6 +856,8 @@ struct module; >> * e.g. drivers/net/e100.c. >> * @sriov_configure: Optional driver callback to allow configuration of >> * number of VFs to enable via sysfs "sriov_numvfs" file. >> + * @sriov_set_msix_vec_count: Driver callback to change number of MSI-X vectors >> + * exposed by the sysfs "vf_msix_vec" entry. >> * @err_handler: See Documentation/PCI/pci-error-recovery.rst >> * @groups: Sysfs attribute groups. >> * @driver: Driver model structure. >> @@ -871,6 +873,7 @@ struct pci_driver { >> int (*resume)(struct pci_dev *dev); /* Device woken up */ >> void (*shutdown)(struct pci_dev *dev); >> int (*sriov_configure)(struct pci_dev *dev, int num_vfs); /* On PF */ >> + int (*sriov_set_msix_vec_count)(struct pci_dev *vf, int msix_vec_count); /* On PF */ >> const struct pci_error_handlers *err_handler; >> const struct attribute_group **groups; >> struct device_driver driver; >> @@ -1464,6 +1467,7 @@ struct msix_entry { >> int pci_msi_vec_count(struct pci_dev *dev); >> void pci_disable_msi(struct pci_dev *dev); >> int pci_msix_vec_count(struct pci_dev *dev); >> +int pci_set_msix_vec_count(struct pci_dev *dev, int numb); > This patch adds the pci_set_msix_vec_count() definition in pci/msi.c > and a call in pci/iov.c. It doesn't need to be declared in > include/linux/pci.h or exported. It can be declared in > drivers/pci/pci.h. > >> void pci_disable_msix(struct pci_dev *dev); >> void pci_restore_msi_state(struct pci_dev *dev); >> int pci_msi_enabled(void); >> @@ -2402,6 +2406,10 @@ static inline bool pci_is_thunderbolt_attached(struct pci_dev *pdev) >> void pci_uevent_ers(struct pci_dev *pdev, enum pci_ers_result err_type); >> #endif >> >> +#ifdef CONFIG_PCI_IOV >> +int pci_iov_virtfn_devfn(struct pci_dev *dev, int vf_id); >> +#endif > pci_iov_virtfn_devfn() is already declared in this file. > >> /* Provide the legacy pci_dma_* API */ >> #include <linux/pci-dma-compat.h> >> >> -- >> 2.29.2 >>
On Thu, Jan 07, 2021 at 10:54:38PM -0500, Don Dutile wrote: > On 1/7/21 7:57 PM, Bjorn Helgaas wrote: > > [+cc Alex, Don] <...> > > Help me connect the dots here. Is this required because of something > > peculiar to mlx5, or is something like this required for all SR-IOV > > devices because of the way the PCIe spec is written? > So, overall, I'm guessing the mlx5 device can have 1000's of MSIX -- say, one per send/receive/completion queue. > This device capability may exceed the max number MSIX a VM can have/support (depending on guestos). > So, a sysfs tunable is used to set the max MSIX available, and thus, the device puts >1 send/rcv/completion queue intr on a given MSIX. > > ok, time for Leon to better state what this patch does, > and why it's needed on mlx5 (and may be applicable to other/future high-MSIX devices assigned to VMs (NVME?)). > Hmmm, now that I said it, why is it SRIOV-centric and not pci-device centric (can pass a PF w/high number of MSIX to a VM). Thanks Don and Bjorn, I will answer on all comments a little bit later when I will return to the office (Sunday). However it is important for me to present the use case. Our mlx5 SR-IOV devices were always capable to drive many MSI-X (upto 2K, don't catch me on exact number), however when user created VFs, the FW has no knowledge of how those VFs will be used. So FW had no choice but statically and equally assign same amount of MSI-X to all VFs. After SR-IOV VF creation, user will bind those new VFs to the VMs, but the VMs have different number of CPUs and despite HW being able to deliver all needed number of vectors (in mlx5 netdev world, number of channels == number of CPUs == number of vectors), we will be limited by already set low number of vectors. So it is not for vector reduction, but more for vector re-partition. As an example, imagine mlx5 with two VFs. One VF is bounded to VM with 200 CPUs and another is bounded to VM with 1 CPU. They need different amount of MSI-X vectors. Hope that I succeeded to explain :). Regarding why it is SR-IOV and not PCI, the amount of MSI-X vectors is read-only field in the PCI, so we can't write from pci/core toward PF device and expect HW update it. It means that if we really need it, we will need to do it after driver already loaded on that PF, so driver will forward to HW and lspci will work correctly. This will require reload of whole PCI device initialization sequence, because MSI-X table size pre-calculated very early in the init flow. Thanks > > -Don
On Fri, 8 Jan 2021 09:25:25 +0200 Leon Romanovsky <leon@kernel.org> wrote: > On Thu, Jan 07, 2021 at 10:54:38PM -0500, Don Dutile wrote: > > On 1/7/21 7:57 PM, Bjorn Helgaas wrote: > > > [+cc Alex, Don] > > <...> > > > > Help me connect the dots here. Is this required because of something > > > peculiar to mlx5, or is something like this required for all SR-IOV > > > devices because of the way the PCIe spec is written? > > So, overall, I'm guessing the mlx5 device can have 1000's of MSIX -- say, one per send/receive/completion queue. > > This device capability may exceed the max number MSIX a VM can have/support (depending on guestos). > > So, a sysfs tunable is used to set the max MSIX available, and thus, the device puts >1 send/rcv/completion queue intr on a given MSIX. > > > > ok, time for Leon to better state what this patch does, > > and why it's needed on mlx5 (and may be applicable to other/future high-MSIX devices assigned to VMs (NVME?)). > > Hmmm, now that I said it, why is it SRIOV-centric and not pci-device centric (can pass a PF w/high number of MSIX to a VM). > > Thanks Don and Bjorn, > > I will answer on all comments a little bit later when I will return > to the office (Sunday). > > However it is important for me to present the use case. > > Our mlx5 SR-IOV devices were always capable to drive many MSI-X (upto 2K, > don't catch me on exact number), however when user created VFs, the FW has > no knowledge of how those VFs will be used. So FW had no choice but statically > and equally assign same amount of MSI-X to all VFs. > > After SR-IOV VF creation, user will bind those new VFs to the VMs, but > the VMs have different number of CPUs and despite HW being able to deliver > all needed number of vectors (in mlx5 netdev world, number of channels == number > of CPUs == number of vectors), we will be limited by already set low number > of vectors. > > So it is not for vector reduction, but more for vector re-partition. > > As an example, imagine mlx5 with two VFs. One VF is bounded to VM with 200 CPUs > and another is bounded to VM with 1 CPU. They need different amount of MSI-X vectors. > > Hope that I succeeded to explain :). The idea is not unreasonable imo, but without knowing the size of the vector pool, range available per vf, or ultimately whether the vf supports this feature before we try to configure it, I don't see how userspace is expected to make use of this in the general case. If the configuration requires such specific vf vector usage and pf driver specific knowledge, I'm not sure it's fit as a generic pci-sysfs interface. Thanks, Alex
On Thu, Jan 07, 2021 at 10:54:38PM -0500, Don Dutile wrote: > On 1/7/21 7:57 PM, Bjorn Helgaas wrote: > > On Sun, Jan 03, 2021 at 10:24:37AM +0200, Leon Romanovsky wrote: > > > + **/ > > > +int pci_set_msix_vec_count(struct pci_dev *dev, int numb) > > > +{ > > > + struct pci_dev *pdev = pci_physfn(dev); > > > + > > > + if (!dev->msix_cap || !pdev->msix_cap) > > > + return -EINVAL; > > > + > > > + if (dev->driver || !pdev->driver || > > > + !pdev->driver->sriov_set_msix_vec_count) > > > + return -EOPNOTSUPP; > > > + > > > + if (numb < 0) > > > + /* > > > + * We don't support negative numbers for now, > > > + * but maybe in the future it will make sense. > > > + */ > > > + return -EINVAL; > > > + > > > + return pdev->driver->sriov_set_msix_vec_count(dev, numb); > > > > So we write to a VF sysfs file, get here and look up the PF, call a PF > > driver callback with the VF as an argument, the callback (at least for > > mlx5) looks up the PF from the VF, then does some mlx5-specific magic > > to the PF that influences the VF somehow? > > There's no PF lookup above.... it's just checking if a pdev has a > driver with the desired msix-cap setting(reduction) feature. We started with the VF (the sysfs file is attached to the VF). "pdev" is the corresponding PF; that's what I meant by "looking up the PF". Then we call the PF driver sriov_set_msix_vec_count() method. I asked because this raises questions of whether we need mutual exclusion or some other coordination between setting this for multiple VFs. Obviously it's great to answer all these in email, but at the end of the day, the rationale needs to be in the commit, either in code comments or the commit log.
On 1/8/21 4:09 PM, Bjorn Helgaas wrote: > On Thu, Jan 07, 2021 at 10:54:38PM -0500, Don Dutile wrote: >> On 1/7/21 7:57 PM, Bjorn Helgaas wrote: >>> On Sun, Jan 03, 2021 at 10:24:37AM +0200, Leon Romanovsky wrote: >>>> + **/ >>>> +int pci_set_msix_vec_count(struct pci_dev *dev, int numb) >>>> +{ >>>> + struct pci_dev *pdev = pci_physfn(dev); >>>> + >>>> + if (!dev->msix_cap || !pdev->msix_cap) >>>> + return -EINVAL; >>>> + >>>> + if (dev->driver || !pdev->driver || >>>> + !pdev->driver->sriov_set_msix_vec_count) >>>> + return -EOPNOTSUPP; >>>> + >>>> + if (numb < 0) >>>> + /* >>>> + * We don't support negative numbers for now, >>>> + * but maybe in the future it will make sense. >>>> + */ >>>> + return -EINVAL; >>>> + >>>> + return pdev->driver->sriov_set_msix_vec_count(dev, numb); >>> So we write to a VF sysfs file, get here and look up the PF, call a PF >>> driver callback with the VF as an argument, the callback (at least for >>> mlx5) looks up the PF from the VF, then does some mlx5-specific magic >>> to the PF that influences the VF somehow? >> There's no PF lookup above.... it's just checking if a pdev has a >> driver with the desired msix-cap setting(reduction) feature. > We started with the VF (the sysfs file is attached to the VF). "pdev" > is the corresponding PF; that's what I meant by "looking up the PF". > Then we call the PF driver sriov_set_msix_vec_count() method. ah, got how your statement relates to the files &/or pdev. > I asked because this raises questions of whether we need mutual > exclusion or some other coordination between setting this for multiple > VFs. > > Obviously it's great to answer all these in email, but at the end of > the day, the rationale needs to be in the commit, either in code > comments or the commit log. > I'm still not getting why this is not per-(vf)pdev -- just b/c a device has N-number of MSIX capability doesn't mean it has to all be used/configured, Setting max-MSIX for VFs in the PF's pdev means it is the same number for all VFs ... and I'm not sure that's the right solution either. It should still be (v)pdev-based, IMO. --dd
On Thu, Jan 07, 2021 at 06:57:21PM -0600, Bjorn Helgaas wrote: > [+cc Alex, Don] > > This patch does not actually *configure* the number of vectors, so the > subject is not quite accurate. IIUC, this patch adds a sysfs file > that can be used to configure the number of vectors. The subject > should mention the sysfs connection. I'll do: "PCI: Add sysfs callback to allow MSI-X table size change of SR-IOV VFs" > > On Sun, Jan 03, 2021 at 10:24:37AM +0200, Leon Romanovsky wrote: > > From: Leon Romanovsky <leonro@nvidia.com> > > > > This function is applicable for SR-IOV VFs because such devices allocate > > their MSI-X table before they will run on the targeted hardware and they > > can't guess the right amount of vectors. > > This sentence doesn't quite have enough context to make sense to me. > Per PCIe r5.0, sec 9.5.1.2, I think PFs and VFs have independent MSI-X > Capabilities. What is the connection between the PF MSI-X and the VF > MSI-X? Right, PF and VF have different capabilities, but MSI-X vectors are limited resource by the HW and the device has pool of such vectors to distribute to the VFs. The connection between PF and VF is a logical one. The PF exists and bounded to the driver, so have an ability to actually write to the HW and change VF configuration before driver bounded to it. > > The MSI-X table sizes should be determined by the Table Size in the > Message Control register. Apparently we write a VF's Table Size > before a driver is bound to the VF? Where does that happen? The table size is set by the HW when SR-IOV is enabled and VFs are created. echo num_sriov > /sys/bus/pci/devices/.../sriov_numvfs .... at this point VFs have this table set, but not used yet. The driver will read this table when it enables MSI-X: pci_enable_msix_range __pci_enable_msix_range __pci_enable_msix pci_msix_vec_count > > "Before they run on the targeted hardware" -- do you mean before the > VF is passed through to a guest virtual machine? You mention "target > VM" below, which makes more sense to me. VFs don't "run"; they're not > software. I apologize for not being an expert in the use of VFs. Yes, "target" == "VM". > > Please mention the sysfs path in the commit log. I'll do. > > > Signed-off-by: Leon Romanovsky <leonro@nvidia.com> > > --- > > Documentation/ABI/testing/sysfs-bus-pci | 16 +++++++ > > drivers/pci/iov.c | 57 +++++++++++++++++++++++++ > > drivers/pci/msi.c | 30 +++++++++++++ > > drivers/pci/pci-sysfs.c | 1 + > > drivers/pci/pci.h | 1 + > > include/linux/pci.h | 8 ++++ > > 6 files changed, 113 insertions(+) > > > > diff --git a/Documentation/ABI/testing/sysfs-bus-pci b/Documentation/ABI/testing/sysfs-bus-pci > > index 25c9c39770c6..30720a9e1386 100644 > > --- a/Documentation/ABI/testing/sysfs-bus-pci > > +++ b/Documentation/ABI/testing/sysfs-bus-pci > > @@ -375,3 +375,19 @@ Description: > > The value comes from the PCI kernel device state and can be one > > of: "unknown", "error", "D0", D1", "D2", "D3hot", "D3cold". > > The file is read only. > > + > > +What: /sys/bus/pci/devices/.../vf_msix_vec > > +Date: December 2020 > > +Contact: Leon Romanovsky <leonro@nvidia.com> > > +Description: > > + This file is associated with the SR-IOV VFs. It allows overwrite > > + the amount of MSI-X vectors for that VF. This is needed to optimize > > + performance of newly bounded devices by allocating the number of > > + vectors based on the internal knowledge of targeted VM. > > s/allows overwrite/allows configuration of/ > s/for that/for the/ > s/amount of/number of/ > s/bounded/bound/ I changed it, thanks > > What "internal knowledge" is this? AFAICT this would have to be some > user-space administration knowledge, not anything internal to the > kernel. Yes, it is not internal to the kernel, but administrator knowledge. In our case, it is orchestration software that allocates such VFs to the users. > > > + The values accepted are: > > + * > 0 - this will be number reported by the PCI VF's PCIe MSI-X capability. > > s/PCI// (it's obvious we're talking about PCI here) > s/PCIe// (MSI-X is not PCIe-specific, and there's no need to mention > it at all) Done > > > + * < 0 - not valid > > + * = 0 - will reset to the device default value > > + > > + The file is writable if no driver is bounded. > > From the code, it looks more like this: > > The file is writable if the PF is bound to a driver that supports > the ->sriov_set_msix_vec_count() callback and there is no driver > bound to the VF. I added it to the description. > > Please wrap all of this to fit in 80 columns like the rest of the file. Fixed > > > diff --git a/drivers/pci/iov.c b/drivers/pci/iov.c > > index 4afd4ee4f7f0..0f8c570361fc 100644 > > --- a/drivers/pci/iov.c > > +++ b/drivers/pci/iov.c > > @@ -31,6 +31,7 @@ int pci_iov_virtfn_devfn(struct pci_dev *dev, int vf_id) > > return (dev->devfn + dev->sriov->offset + > > dev->sriov->stride * vf_id) & 0xff; > > } > > +EXPORT_SYMBOL(pci_iov_virtfn_devfn); > > > > /* > > * Per SR-IOV spec sec 3.3.10 and 3.3.11, First VF Offset and VF Stride may > > @@ -426,6 +427,62 @@ const struct attribute_group sriov_dev_attr_group = { > > .is_visible = sriov_attrs_are_visible, > > }; > > > > +#ifdef CONFIG_PCI_MSI > > +static ssize_t vf_msix_vec_show(struct device *dev, > > + struct device_attribute *attr, char *buf) > > +{ > > + struct pci_dev *pdev = to_pci_dev(dev); > > + int numb = pci_msix_vec_count(pdev); > > + > > + if (numb < 0) > > + return numb; > > + > > + return sprintf(buf, "%d\n", numb); > > +} > > + > > +static ssize_t vf_msix_vec_store(struct device *dev, > > + struct device_attribute *attr, const char *buf, > > + size_t count) > > +{ > > + struct pci_dev *vf_dev = to_pci_dev(dev); > > + int val, ret; > > + > > + ret = kstrtoint(buf, 0, &val); > > + if (ret) > > + return ret; > > + > > + ret = pci_set_msix_vec_count(vf_dev, val); > > + if (ret) > > + return ret; > > + > > + return count; > > +} > > +static DEVICE_ATTR_RW(vf_msix_vec); > > +#endif > > + > > +static struct attribute *sriov_vf_dev_attrs[] = { > > +#ifdef CONFIG_PCI_MSI > > + &dev_attr_vf_msix_vec.attr, > > +#endif > > + NULL, > > +}; > > + > > +static umode_t sriov_vf_attrs_are_visible(struct kobject *kobj, > > + struct attribute *a, int n) > > +{ > > + struct device *dev = kobj_to_dev(kobj); > > + > > + if (dev_is_pf(dev)) > > + return 0; > > + > > + return a->mode; > > +} > > + > > +const struct attribute_group sriov_vf_dev_attr_group = { > > + .attrs = sriov_vf_dev_attrs, > > + .is_visible = sriov_vf_attrs_are_visible, > > +}; > > + > > int __weak pcibios_sriov_enable(struct pci_dev *pdev, u16 num_vfs) > > { > > return 0; > > diff --git a/drivers/pci/msi.c b/drivers/pci/msi.c > > index 3162f88fe940..0bcd705487d9 100644 > > --- a/drivers/pci/msi.c > > +++ b/drivers/pci/msi.c > > @@ -991,6 +991,36 @@ int pci_msix_vec_count(struct pci_dev *dev) > > } > > EXPORT_SYMBOL(pci_msix_vec_count); > > > > +/** > > + * pci_set_msix_vec_count - change the reported number of MSI-X vectors. > > Drop period at end, as other kernel doc in this file does. Done > > > + * This function is applicable for SR-IOV VFs because such devices allocate > > + * their MSI-X table before they will run on the targeted hardware and they > > + * can't guess the right amount of vectors. > > + * @dev: VF device that is going to be changed. > > + * @numb: amount of MSI-X vectors. > > Rewrite the "such devices allocate..." part based on the questions in > the commit log. Same with "targeted hardware." > > s/amount of/number of/ > Drop periods at end of parameter descriptions. Done > > > + **/ > > +int pci_set_msix_vec_count(struct pci_dev *dev, int numb) > > +{ > > + struct pci_dev *pdev = pci_physfn(dev); > > + > > + if (!dev->msix_cap || !pdev->msix_cap) > > + return -EINVAL; > > + > > + if (dev->driver || !pdev->driver || > > + !pdev->driver->sriov_set_msix_vec_count) > > + return -EOPNOTSUPP; > > + > > + if (numb < 0) > > + /* > > + * We don't support negative numbers for now, > > + * but maybe in the future it will make sense. > > + */ > > + return -EINVAL; > > + > > + return pdev->driver->sriov_set_msix_vec_count(dev, numb); > > So we write to a VF sysfs file, get here and look up the PF, call a PF > driver callback with the VF as an argument, the callback (at least for > mlx5) looks up the PF from the VF, then does some mlx5-specific magic > to the PF that influences the VF somehow? Right, because HW already created VFs, the PF driver is aware of them, so it simply says to the FW that specific VF should have different value in their table size. > > Help me connect the dots here. Is this required because of something > peculiar to mlx5, or is something like this required for all SR-IOV > devices because of the way the PCIe spec is written? The second one is correct, there is nothing mlx5 specific in it. This is a combination of the spec together with Linux SR-IOV implementation logic. First, PCI spec has one single bit to enable/disable all VFs at the same time without ability to dynamically add/delete. It means all SR-IOV HW in the world will do the same: split internal MSI-X pool equally or by any other same logic. This is needed so lspci right after VF created will give proper values in the MSI-X section. Second, Linux followed the spec and implemented same allocation model and separated it by the layers, at the time driver probes, it should have all PCI config ready in the PCI level. It means even change of MSI-X inside VF during driver VF init and driver reload later can be potentially problematic. > > > +} > > +EXPORT_SYMBOL(pci_set_msix_vec_count); > > + > > static int __pci_enable_msix(struct pci_dev *dev, struct msix_entry *entries, > > int nvec, struct irq_affinity *affd, int flags) > > { > > diff --git a/drivers/pci/pci-sysfs.c b/drivers/pci/pci-sysfs.c > > index fb072f4b3176..0af2222643c2 100644 > > --- a/drivers/pci/pci-sysfs.c > > +++ b/drivers/pci/pci-sysfs.c > > @@ -1557,6 +1557,7 @@ static const struct attribute_group *pci_dev_attr_groups[] = { > > &pci_dev_hp_attr_group, > > #ifdef CONFIG_PCI_IOV > > &sriov_dev_attr_group, > > + &sriov_vf_dev_attr_group, > > #endif > > &pci_bridge_attr_group, > > &pcie_dev_attr_group, > > diff --git a/drivers/pci/pci.h b/drivers/pci/pci.h > > index 5c59365092fa..46396a5da2d9 100644 > > --- a/drivers/pci/pci.h > > +++ b/drivers/pci/pci.h > > @@ -502,6 +502,7 @@ resource_size_t pci_sriov_resource_alignment(struct pci_dev *dev, int resno); > > void pci_restore_iov_state(struct pci_dev *dev); > > int pci_iov_bus_range(struct pci_bus *bus); > > extern const struct attribute_group sriov_dev_attr_group; > > +extern const struct attribute_group sriov_vf_dev_attr_group; > > #else > > static inline int pci_iov_init(struct pci_dev *dev) > > { > > diff --git a/include/linux/pci.h b/include/linux/pci.h > > index b32126d26997..1acba40a1b1b 100644 > > --- a/include/linux/pci.h > > +++ b/include/linux/pci.h > > @@ -856,6 +856,8 @@ struct module; > > * e.g. drivers/net/e100.c. > > * @sriov_configure: Optional driver callback to allow configuration of > > * number of VFs to enable via sysfs "sriov_numvfs" file. > > + * @sriov_set_msix_vec_count: Driver callback to change number of MSI-X vectors > > + * exposed by the sysfs "vf_msix_vec" entry. > > * @err_handler: See Documentation/PCI/pci-error-recovery.rst > > * @groups: Sysfs attribute groups. > > * @driver: Driver model structure. > > @@ -871,6 +873,7 @@ struct pci_driver { > > int (*resume)(struct pci_dev *dev); /* Device woken up */ > > void (*shutdown)(struct pci_dev *dev); > > int (*sriov_configure)(struct pci_dev *dev, int num_vfs); /* On PF */ > > + int (*sriov_set_msix_vec_count)(struct pci_dev *vf, int msix_vec_count); /* On PF */ > > const struct pci_error_handlers *err_handler; > > const struct attribute_group **groups; > > struct device_driver driver; > > @@ -1464,6 +1467,7 @@ struct msix_entry { > > int pci_msi_vec_count(struct pci_dev *dev); > > void pci_disable_msi(struct pci_dev *dev); > > int pci_msix_vec_count(struct pci_dev *dev); > > +int pci_set_msix_vec_count(struct pci_dev *dev, int numb); > > This patch adds the pci_set_msix_vec_count() definition in pci/msi.c > and a call in pci/iov.c. It doesn't need to be declared in > include/linux/pci.h or exported. It can be declared in > drivers/pci/pci.h. I changed it, thanks > > > void pci_disable_msix(struct pci_dev *dev); > > void pci_restore_msi_state(struct pci_dev *dev); > > int pci_msi_enabled(void); > > @@ -2402,6 +2406,10 @@ static inline bool pci_is_thunderbolt_attached(struct pci_dev *pdev) > > void pci_uevent_ers(struct pci_dev *pdev, enum pci_ers_result err_type); > > #endif > > > > +#ifdef CONFIG_PCI_IOV > > +int pci_iov_virtfn_devfn(struct pci_dev *dev, int vf_id); > > +#endif > > pci_iov_virtfn_devfn() is already declared in this file. Sorry for that. > > > /* Provide the legacy pci_dma_* API */ > > #include <linux/pci-dma-compat.h> > > > > -- > > 2.29.2 > >
On Fri, Jan 08, 2021 at 03:09:13PM -0600, Bjorn Helgaas wrote: > On Thu, Jan 07, 2021 at 10:54:38PM -0500, Don Dutile wrote: > > On 1/7/21 7:57 PM, Bjorn Helgaas wrote: > > > On Sun, Jan 03, 2021 at 10:24:37AM +0200, Leon Romanovsky wrote: > > > > > + **/ > > > > +int pci_set_msix_vec_count(struct pci_dev *dev, int numb) > > > > +{ > > > > + struct pci_dev *pdev = pci_physfn(dev); > > > > + > > > > + if (!dev->msix_cap || !pdev->msix_cap) > > > > + return -EINVAL; > > > > + > > > > + if (dev->driver || !pdev->driver || > > > > + !pdev->driver->sriov_set_msix_vec_count) > > > > + return -EOPNOTSUPP; > > > > + > > > > + if (numb < 0) > > > > + /* > > > > + * We don't support negative numbers for now, > > > > + * but maybe in the future it will make sense. > > > > + */ > > > > + return -EINVAL; > > > > + > > > > + return pdev->driver->sriov_set_msix_vec_count(dev, numb); > > > > > > So we write to a VF sysfs file, get here and look up the PF, call a PF > > > driver callback with the VF as an argument, the callback (at least for > > > mlx5) looks up the PF from the VF, then does some mlx5-specific magic > > > to the PF that influences the VF somehow? > > > > There's no PF lookup above.... it's just checking if a pdev has a > > driver with the desired msix-cap setting(reduction) feature. > > We started with the VF (the sysfs file is attached to the VF). "pdev" > is the corresponding PF; that's what I meant by "looking up the PF". > Then we call the PF driver sriov_set_msix_vec_count() method. > > I asked because this raises questions of whether we need mutual > exclusion or some other coordination between setting this for multiple > VFs. MSI-X are managed by HW and they are separated between VFs. IMHO, it will be better if SW won't do too much coordination. Thanks > > Obviously it's great to answer all these in email, but at the end of > the day, the rationale needs to be in the commit, either in code > comments or the commit log.
On Thu, Jan 07, 2021 at 10:54:38PM -0500, Don Dutile wrote: > On 1/7/21 7:57 PM, Bjorn Helgaas wrote: > > [+cc Alex, Don] > > > > This patch does not actually *configure* the number of vectors, so the > > subject is not quite accurate. IIUC, this patch adds a sysfs file > > that can be used to configure the number of vectors. The subject > > should mention the sysfs connection. > > > > On Sun, Jan 03, 2021 at 10:24:37AM +0200, Leon Romanovsky wrote: > > > From: Leon Romanovsky <leonro@nvidia.com> > > > > > > This function is applicable for SR-IOV VFs because such devices allocate > > > their MSI-X table before they will run on the targeted hardware and they > > > can't guess the right amount of vectors. > > This sentence doesn't quite have enough context to make sense to me. > > Per PCIe r5.0, sec 9.5.1.2, I think PFs and VFs have independent MSI-X > > Capabilities. What is the connection between the PF MSI-X and the VF > > MSI-X? > +1... strip this commit log section and write it with correct, technical content. > PFs & VF's have indep MSIX caps. > > Q: is this an issue where (some) mlx5's have a large msi-x capability (per VF) that may overwhelm a system's, (pci-(sub)-tree) MSI / intr capability, > and this is a sysfs-based tuning knob to reduce the max number on such 'challenged' systems? > -- ah; reading further below, it's based on some information gleemed from the VM's capability for intr. support. > -- or maybe IOMMU (intr) support on the host system, and the VF can't exceed it or config failure in VM... whatever... its some VM cap that's being accomodated. I hope that this answers. https://lore.kernel.org/linux-pci/20210110082206.GD31158@unreal/T/#md5dfc2edaaa686331ab3ce73496df7f58421c550 This feature is for MSI-X repartition and reduction. > > The MSI-X table sizes should be determined by the Table Size in the > > Message Control register. Apparently we write a VF's Table Size > > before a driver is bound to the VF? Where does that happen? > > > > "Before they run on the targeted hardware" -- do you mean before the > > VF is passed through to a guest virtual machine? You mention "target > > VM" below, which makes more sense to me. VFs don't "run"; they're not > > software. I apologize for not being an expert in the use of VFs. > > > > Please mention the sysfs path in the commit log. > > > > > Signed-off-by: Leon Romanovsky <leonro@nvidia.com> > > > --- > > > Documentation/ABI/testing/sysfs-bus-pci | 16 +++++++ > > > drivers/pci/iov.c | 57 +++++++++++++++++++++++++ > > > drivers/pci/msi.c | 30 +++++++++++++ > > > drivers/pci/pci-sysfs.c | 1 + > > > drivers/pci/pci.h | 1 + > > > include/linux/pci.h | 8 ++++ > > > 6 files changed, 113 insertions(+) > > > > > > diff --git a/Documentation/ABI/testing/sysfs-bus-pci b/Documentation/ABI/testing/sysfs-bus-pci > > > index 25c9c39770c6..30720a9e1386 100644 > > > --- a/Documentation/ABI/testing/sysfs-bus-pci > > > +++ b/Documentation/ABI/testing/sysfs-bus-pci > > > @@ -375,3 +375,19 @@ Description: > > > The value comes from the PCI kernel device state and can be one > > > of: "unknown", "error", "D0", D1", "D2", "D3hot", "D3cold". > > > The file is read only. > > > + > > > +What: /sys/bus/pci/devices/.../vf_msix_vec > > > +Date: December 2020 > > > +Contact: Leon Romanovsky <leonro@nvidia.com> > > > +Description: > > > + This file is associated with the SR-IOV VFs. It allows overwrite > > > + the amount of MSI-X vectors for that VF. This is needed to optimize > > > + performance of newly bounded devices by allocating the number of > > > + vectors based on the internal knowledge of targeted VM. > > s/allows overwrite/allows configuration of/ > > s/for that/for the/ > > s/amount of/number of/ > > s/bounded/bound/ > > > > What "internal knowledge" is this? AFAICT this would have to be some > > user-space administration knowledge, not anything internal to the > > kernel. > Correct; likely a libvirt VM (section of its) description; Right, libvirt and/or orchestration software. Thanks
On Fri, Jan 08, 2021 at 09:54:47PM -0500, Don Dutile wrote: > On 1/8/21 4:09 PM, Bjorn Helgaas wrote: > > On Thu, Jan 07, 2021 at 10:54:38PM -0500, Don Dutile wrote: > > > On 1/7/21 7:57 PM, Bjorn Helgaas wrote: > > > > On Sun, Jan 03, 2021 at 10:24:37AM +0200, Leon Romanovsky wrote: > > > > > + **/ > > > > > +int pci_set_msix_vec_count(struct pci_dev *dev, int numb) > > > > > +{ > > > > > + struct pci_dev *pdev = pci_physfn(dev); > > > > > + > > > > > + if (!dev->msix_cap || !pdev->msix_cap) > > > > > + return -EINVAL; > > > > > + > > > > > + if (dev->driver || !pdev->driver || > > > > > + !pdev->driver->sriov_set_msix_vec_count) > > > > > + return -EOPNOTSUPP; > > > > > + > > > > > + if (numb < 0) > > > > > + /* > > > > > + * We don't support negative numbers for now, > > > > > + * but maybe in the future it will make sense. > > > > > + */ > > > > > + return -EINVAL; > > > > > + > > > > > + return pdev->driver->sriov_set_msix_vec_count(dev, numb); > > > > So we write to a VF sysfs file, get here and look up the PF, call a PF > > > > driver callback with the VF as an argument, the callback (at least for > > > > mlx5) looks up the PF from the VF, then does some mlx5-specific magic > > > > to the PF that influences the VF somehow? > > > There's no PF lookup above.... it's just checking if a pdev has a > > > driver with the desired msix-cap setting(reduction) feature. > > We started with the VF (the sysfs file is attached to the VF). "pdev" > > is the corresponding PF; that's what I meant by "looking up the PF". > > Then we call the PF driver sriov_set_msix_vec_count() method. > ah, got how your statement relates to the files &/or pdev. > > > I asked because this raises questions of whether we need mutual > > exclusion or some other coordination between setting this for multiple > > VFs. > > > > Obviously it's great to answer all these in email, but at the end of > > the day, the rationale needs to be in the commit, either in code > > comments or the commit log. > > > I'm still not getting why this is not per-(vf)pdev -- just b/c a device has N-number of MSIX capability doesn't mean it has to all be used/configured, > Setting max-MSIX for VFs in the PF's pdev means it is the same number for all VFs ... and I'm not sure that's the right solution either. > It should still be (v)pdev-based, IMO. The proposed solution is per-VF, am I missing anything in this discussion? > --dd >
On Fri, Jan 08, 2021 at 09:21:45AM -0700, Alex Williamson wrote: > On Fri, 8 Jan 2021 09:25:25 +0200 > Leon Romanovsky <leon@kernel.org> wrote: > > > On Thu, Jan 07, 2021 at 10:54:38PM -0500, Don Dutile wrote: > > > On 1/7/21 7:57 PM, Bjorn Helgaas wrote: > > > > [+cc Alex, Don] > > > > <...> > > > > > > Help me connect the dots here. Is this required because of something > > > > peculiar to mlx5, or is something like this required for all SR-IOV > > > > devices because of the way the PCIe spec is written? > > > So, overall, I'm guessing the mlx5 device can have 1000's of MSIX -- say, one per send/receive/completion queue. > > > This device capability may exceed the max number MSIX a VM can have/support (depending on guestos). > > > So, a sysfs tunable is used to set the max MSIX available, and thus, the device puts >1 send/rcv/completion queue intr on a given MSIX. > > > > > > ok, time for Leon to better state what this patch does, > > > and why it's needed on mlx5 (and may be applicable to other/future high-MSIX devices assigned to VMs (NVME?)). > > > Hmmm, now that I said it, why is it SRIOV-centric and not pci-device centric (can pass a PF w/high number of MSIX to a VM). > > > > Thanks Don and Bjorn, > > > > I will answer on all comments a little bit later when I will return > > to the office (Sunday). > > > > However it is important for me to present the use case. > > > > Our mlx5 SR-IOV devices were always capable to drive many MSI-X (upto 2K, > > don't catch me on exact number), however when user created VFs, the FW has > > no knowledge of how those VFs will be used. So FW had no choice but statically > > and equally assign same amount of MSI-X to all VFs. > > > > After SR-IOV VF creation, user will bind those new VFs to the VMs, but > > the VMs have different number of CPUs and despite HW being able to deliver > > all needed number of vectors (in mlx5 netdev world, number of channels == number > > of CPUs == number of vectors), we will be limited by already set low number > > of vectors. > > > > So it is not for vector reduction, but more for vector re-partition. > > > > As an example, imagine mlx5 with two VFs. One VF is bounded to VM with 200 CPUs > > and another is bounded to VM with 1 CPU. They need different amount of MSI-X vectors. > > > > Hope that I succeeded to explain :). > > The idea is not unreasonable imo, but without knowing the size of the > vector pool, range available per vf, or ultimately whether the vf > supports this feature before we try to configure it, I don't see how > userspace is expected to make use of this in the general case. If the > configuration requires such specific vf vector usage and pf driver > specific knowledge, I'm not sure it's fit as a generic pci-sysfs > interface. Thanks, I didn't prohibit read of newly created sysfs file, but if I change the implementation to vf_msix_vec_show() to return -EOPNOTSUPP for not-supported device, the software will be able to distinguish supported/not-supported. SW will read this file: -> success -> feature supported -> failure -> feature not supported There is one extra sysfs file needed: vf_total_msix. That file will give total number of MSI-X vectors that is possible to configure. The same logic (supported/not-supported) can be applicable here as well. The feature itself will be used by orchestration software that will make decisions based on already configured values or future promises and the overall total number. The positive outcome of this scheme that driver stays lean. Thanks > > Alex >
diff --git a/Documentation/ABI/testing/sysfs-bus-pci b/Documentation/ABI/testing/sysfs-bus-pci index 25c9c39770c6..30720a9e1386 100644 --- a/Documentation/ABI/testing/sysfs-bus-pci +++ b/Documentation/ABI/testing/sysfs-bus-pci @@ -375,3 +375,19 @@ Description: The value comes from the PCI kernel device state and can be one of: "unknown", "error", "D0", D1", "D2", "D3hot", "D3cold". The file is read only. + +What: /sys/bus/pci/devices/.../vf_msix_vec +Date: December 2020 +Contact: Leon Romanovsky <leonro@nvidia.com> +Description: + This file is associated with the SR-IOV VFs. It allows overwrite + the amount of MSI-X vectors for that VF. This is needed to optimize + performance of newly bounded devices by allocating the number of + vectors based on the internal knowledge of targeted VM. + + The values accepted are: + * > 0 - this will be number reported by the PCI VF's PCIe MSI-X capability. + * < 0 - not valid + * = 0 - will reset to the device default value + + The file is writable if no driver is bounded. diff --git a/drivers/pci/iov.c b/drivers/pci/iov.c index 4afd4ee4f7f0..0f8c570361fc 100644 --- a/drivers/pci/iov.c +++ b/drivers/pci/iov.c @@ -31,6 +31,7 @@ int pci_iov_virtfn_devfn(struct pci_dev *dev, int vf_id) return (dev->devfn + dev->sriov->offset + dev->sriov->stride * vf_id) & 0xff; } +EXPORT_SYMBOL(pci_iov_virtfn_devfn); /* * Per SR-IOV spec sec 3.3.10 and 3.3.11, First VF Offset and VF Stride may @@ -426,6 +427,62 @@ const struct attribute_group sriov_dev_attr_group = { .is_visible = sriov_attrs_are_visible, }; +#ifdef CONFIG_PCI_MSI +static ssize_t vf_msix_vec_show(struct device *dev, + struct device_attribute *attr, char *buf) +{ + struct pci_dev *pdev = to_pci_dev(dev); + int numb = pci_msix_vec_count(pdev); + + if (numb < 0) + return numb; + + return sprintf(buf, "%d\n", numb); +} + +static ssize_t vf_msix_vec_store(struct device *dev, + struct device_attribute *attr, const char *buf, + size_t count) +{ + struct pci_dev *vf_dev = to_pci_dev(dev); + int val, ret; + + ret = kstrtoint(buf, 0, &val); + if (ret) + return ret; + + ret = pci_set_msix_vec_count(vf_dev, val); + if (ret) + return ret; + + return count; +} +static DEVICE_ATTR_RW(vf_msix_vec); +#endif + +static struct attribute *sriov_vf_dev_attrs[] = { +#ifdef CONFIG_PCI_MSI + &dev_attr_vf_msix_vec.attr, +#endif + NULL, +}; + +static umode_t sriov_vf_attrs_are_visible(struct kobject *kobj, + struct attribute *a, int n) +{ + struct device *dev = kobj_to_dev(kobj); + + if (dev_is_pf(dev)) + return 0; + + return a->mode; +} + +const struct attribute_group sriov_vf_dev_attr_group = { + .attrs = sriov_vf_dev_attrs, + .is_visible = sriov_vf_attrs_are_visible, +}; + int __weak pcibios_sriov_enable(struct pci_dev *pdev, u16 num_vfs) { return 0; diff --git a/drivers/pci/msi.c b/drivers/pci/msi.c index 3162f88fe940..0bcd705487d9 100644 --- a/drivers/pci/msi.c +++ b/drivers/pci/msi.c @@ -991,6 +991,36 @@ int pci_msix_vec_count(struct pci_dev *dev) } EXPORT_SYMBOL(pci_msix_vec_count); +/** + * pci_set_msix_vec_count - change the reported number of MSI-X vectors. + * This function is applicable for SR-IOV VFs because such devices allocate + * their MSI-X table before they will run on the targeted hardware and they + * can't guess the right amount of vectors. + * @dev: VF device that is going to be changed. + * @numb: amount of MSI-X vectors. + **/ +int pci_set_msix_vec_count(struct pci_dev *dev, int numb) +{ + struct pci_dev *pdev = pci_physfn(dev); + + if (!dev->msix_cap || !pdev->msix_cap) + return -EINVAL; + + if (dev->driver || !pdev->driver || + !pdev->driver->sriov_set_msix_vec_count) + return -EOPNOTSUPP; + + if (numb < 0) + /* + * We don't support negative numbers for now, + * but maybe in the future it will make sense. + */ + return -EINVAL; + + return pdev->driver->sriov_set_msix_vec_count(dev, numb); +} +EXPORT_SYMBOL(pci_set_msix_vec_count); + static int __pci_enable_msix(struct pci_dev *dev, struct msix_entry *entries, int nvec, struct irq_affinity *affd, int flags) { diff --git a/drivers/pci/pci-sysfs.c b/drivers/pci/pci-sysfs.c index fb072f4b3176..0af2222643c2 100644 --- a/drivers/pci/pci-sysfs.c +++ b/drivers/pci/pci-sysfs.c @@ -1557,6 +1557,7 @@ static const struct attribute_group *pci_dev_attr_groups[] = { &pci_dev_hp_attr_group, #ifdef CONFIG_PCI_IOV &sriov_dev_attr_group, + &sriov_vf_dev_attr_group, #endif &pci_bridge_attr_group, &pcie_dev_attr_group, diff --git a/drivers/pci/pci.h b/drivers/pci/pci.h index 5c59365092fa..46396a5da2d9 100644 --- a/drivers/pci/pci.h +++ b/drivers/pci/pci.h @@ -502,6 +502,7 @@ resource_size_t pci_sriov_resource_alignment(struct pci_dev *dev, int resno); void pci_restore_iov_state(struct pci_dev *dev); int pci_iov_bus_range(struct pci_bus *bus); extern const struct attribute_group sriov_dev_attr_group; +extern const struct attribute_group sriov_vf_dev_attr_group; #else static inline int pci_iov_init(struct pci_dev *dev) { diff --git a/include/linux/pci.h b/include/linux/pci.h index b32126d26997..1acba40a1b1b 100644 --- a/include/linux/pci.h +++ b/include/linux/pci.h @@ -856,6 +856,8 @@ struct module; * e.g. drivers/net/e100.c. * @sriov_configure: Optional driver callback to allow configuration of * number of VFs to enable via sysfs "sriov_numvfs" file. + * @sriov_set_msix_vec_count: Driver callback to change number of MSI-X vectors + * exposed by the sysfs "vf_msix_vec" entry. * @err_handler: See Documentation/PCI/pci-error-recovery.rst * @groups: Sysfs attribute groups. * @driver: Driver model structure. @@ -871,6 +873,7 @@ struct pci_driver { int (*resume)(struct pci_dev *dev); /* Device woken up */ void (*shutdown)(struct pci_dev *dev); int (*sriov_configure)(struct pci_dev *dev, int num_vfs); /* On PF */ + int (*sriov_set_msix_vec_count)(struct pci_dev *vf, int msix_vec_count); /* On PF */ const struct pci_error_handlers *err_handler; const struct attribute_group **groups; struct device_driver driver; @@ -1464,6 +1467,7 @@ struct msix_entry { int pci_msi_vec_count(struct pci_dev *dev); void pci_disable_msi(struct pci_dev *dev); int pci_msix_vec_count(struct pci_dev *dev); +int pci_set_msix_vec_count(struct pci_dev *dev, int numb); void pci_disable_msix(struct pci_dev *dev); void pci_restore_msi_state(struct pci_dev *dev); int pci_msi_enabled(void); @@ -2402,6 +2406,10 @@ static inline bool pci_is_thunderbolt_attached(struct pci_dev *pdev) void pci_uevent_ers(struct pci_dev *pdev, enum pci_ers_result err_type); #endif +#ifdef CONFIG_PCI_IOV +int pci_iov_virtfn_devfn(struct pci_dev *dev, int vf_id); +#endif + /* Provide the legacy pci_dma_* API */ #include <linux/pci-dma-compat.h>