vfio/pci: Back guest interrupts from Interrupt Message Store (IMS)
Subsystems (3)
v3 cover vfio/pci: Back guest interrupts from Interrupt Message Store (IMS)
Changes since RFC V2:
- RFC V2: https://lore.kernel.org/lkml/[email protected]/
- Still submiting this as RFC series. I believe that this now matches the expectatations raised during earlier reviews. If you agree this is
the right direction then I can drop the RFC prefix on next submission. If you do not agree then please do let me know where I missed expectations.
- First patch (PCI/MSI: Provide stubs for IMS functions)
has been submitted upstream separately and is queued for inclusion during the next merge window. I do still include it in this series to avoid the noise about issues that bots will find when checking this series without it included.
https://lore.kernel.org/lkml/169757242009.3135.5502383859327174030.tip-bot2@tip-bot2/
- Eliminated duplicate code between the PCI passthrough device backend and the IMS backend through more abstraction within the interrupt management frontend. (Kevin)
- Emulated interrupts are now managed by the interrupt management frontend and no longer unique to IMS. (Jason and Kevin)
- Since being an emulated interrupt is a persistent property there is
a new functional change to PCI interrupt management in that per-interrupt contexts (managed by frontend) are now persistent (they remain allocated until device release).
- Most of the patches from RFC V2 look the same with more patches
added to support the additional abstraction needed to eliminate the duplicate code. The IMS support was refactored to benefit from the
new abstraction. Please refer to individual patches for specific changes.
Changes since RFC V1:
- RFC V1: https://lore.kernel.org/lkml/[email protected]/
- This is a complete rewrite based on feedback from Jason and Kevin. Primarily the transition is to make IMS a new backend of MSI-X
emulation: VFIO PCI transitions to be an interrupt management frontend with existing interrupt management for PCI passthrough devices as a backend and IMS interrupt management introduced as a new backend.
The first part of the series splits VFIO PCI interrupt
management into a "frontend" and "backend" with the existing PCI interrupt management as its first backend. The second part of the series adds IMS interrupt management as a new interrupt management backend.
This is a significant change from RFC V1 as well as in the impact of the changes on existing VFIO PCI. This was done in response to feedback that I hope I understood as intended. If I did not get it right, please do point out to me where I went astray and I'd be
happy to rewrite. Of course, suggestions for improvement will
be much appreciated.
Hi Everybody,
With Interrupt Message Store (IMS) support introduced in commit 0194425af0c8 ("PCI/MSI: Provide IMS (Interrupt Message Store) support") a device can create a secondary interrupt domain that works side by side with MSI-X on the same device. IMS allows for implementation-specific interrupt storage that is managed by the implementation specific interrupt chip associated with the IMS domain at the time it (the IMS domain) is created for the device via pci_create_ims_domain().
An example usage of IMS is for devices that can have their resources assigned to guests with varying granularity. For example, an accelerator device may support many workqueues and a single workqueue can be composed into a virtual device for use by a guest. Using IMS interrupts for the guest preserves MSI-X for host usage while allowing a significantly larger number of interrupt vectors than allowed by MSI-X. All while enabling usage of the same device driver within the host and guest.
This series introduces IMS support to VFIO PCI for use by virtual devices that support MSI-X interrupts that are backed by IMS interrupts on the host. Specifically, that means that when the virtual device's VFIO_DEVICE_SET_IRQS ioctl() receives a "trigger interrupt" (VFIO_IRQ_SET_ACTION_TRIGGER) for a MSI-X index then VFIO PCI IMS allocates/frees an IMS interrupt on the host.
VFIO PCI assumes that it is managing interrupts of a passthrough PCI device. Split VFIO PCI into a "frontend" and "backend" to support interrupt management for virtual devices that are not passthrough PCI devices. The VFIO PCI frontend directs guest requests to the appropriate backend. Existing interrupt management for passthrough PCI devices is the first backend, guest MSI-X interrupts backed by IMS interrupts on the host is the new backend (VFIO PCI IMS).
An IMS interrupt is allocated via pci_ims_alloc_irq() that requires an implementation specific cookie that is opaque to VFIO PCI IMS. This can be a PASID, queue ID, pointer etc. During initialization VFIO PCI IMS learns which PCI device to operate on and what the default cookie should be for any new interrupt allocation. VFIO PCI IMS can also associate a unique cookie with each vector.
Guests may access a virtual device via both 'direct-path', where the guest interacts directly with the underlying hardware, and 'intercepted path', where the virtual device emulates operations. VFIO PCI supports emulated interrupts (better naming suggestions are welcome) to handle 'intercepted path' operations where completion interrupts are signaled from the virtual device, not the underlying hardware.
This has been tested with a yet to be published VFIO driver for the Intel Data Accelerators (IDXD) present in Intel Xeon CPUs.
While this series contains a working implementation it is presented as an RFC with the goal to obtain feedback on whether VFIO PCI IMS is appropriate for inclusion into VFIO and whether it is (or could be adapted to be) appropriate for support of other planned IMS usages you may be aware of.
Any feedback will be greatly appreciated.
Reinette
Reinette Chatre (26):
PCI/MSI: Provide stubs for IMS functions
vfio/pci: Move PCI specific check from wrapper to PCI function vfio/pci: Use unsigned int instead of unsigned
vfio/pci: Make core interrupt callbacks accessible to all virtual
devices
vfio/pci: Split PCI interrupt management into front and backend vfio/pci: Separate MSI and MSI-X handling
vfio/pci: Move interrupt eventfd to interrupt context vfio/pci: Move mutex acquisition into function
vfio/pci: Move per-interrupt contexts to generic interrupt struct vfio/pci: Move IRQ type to generic interrupt context vfio/pci: Provide interrupt context to irq_is() and is_irq_none() vfio/pci: Provide interrupt context to generic ops vfio/pci: Provide interrupt context to vfio_msi_enable() and
vfio_msi_disable()
vfio/pci: Let interrupt management backend interpret interrupt index vfio/pci: Move generic code to frontend
vfio/pci: Split interrupt context initialization vfio/pci: Make vfio_pci_set_irqs_ioctl() available vfio/pci: Preserve per-interrupt contexts
vfio/pci: Store Linux IRQ number in per-interrupt context vfio/pci: Separate frontend and backend code during interrupt
enable/disable
vfio/pci: Replace backend specific calls with callbacks vfio/pci: Introduce backend specific context initializer vfio/pci: Support emulated interrupts
vfio/pci: Add core IMS support
vfio/pci: Add accessor for IMS index
vfio/pci: Support IMS cookie modification
drivers/vfio/pci/vfio_pci_config.c | 2 +- drivers/vfio/pci/vfio_pci_core.c | 50 +- drivers/vfio/pci/vfio_pci_intrs.c | 758 ++++++++++++++++++++++++----- drivers/vfio/pci/vfio_pci_priv.h | 2 +- include/linux/pci.h | 34 +- include/linux/vfio_pci_core.h | 87 +++- 6 files changed, 756 insertions(+), 177 deletions(-)
base-commit: 611da07b89fdd53f140d7b33013f255bf0ed8f34
Signature · 1 line
2.34.1
RFC PATCH v3 1/26 PCI/MSI: Provide stubs for IMS functions
The IMS related functions (pci_create_ims_domain(), pci_ims_alloc_irq(), and pci_ims_free_irq()) are not declared when CONFIG_PCI_MSI is disabled.
Provide definitions of these functions for use when callers are compiled with CONFIG_PCI_MSI disabled.
---
Patch has been submitted separately and is queued for inclusion.
https://lore.kernel.org/lkml/169757242009.3135.5502383859327174030.tip-bot2@tip-bot2/
It is included in this series in support of automated testing by bots picking series from this submission.
include/linux/pci.h | 34 ++++++++++++++++++++++++++-------- 1 file changed, 26 insertions(+), 8 deletions(-)
include/linux/pci.h +26 -8
@@ -1624,6 +1624,8 @@ struct msix_entry {u16 entry;/* Driver uses to specify entry, OS writes */ }; +struct msi_domain_template ; +#ifdef CONFIG_PCI_MSI int pci_msi_vec_count (struct pci_dev *dev);void pci_disable_msi (struct pci_dev *dev);
@@ -1656,6 +1658,11 @@ void pci_msix_free_irq(struct pci_dev *pdev, struct msi_map map);void pci_free_irq_vectors (struct pci_dev *dev);int pci_irq_vector (struct pci_dev *dev,unsigned int nr);const struct cpumask *pci_irq_get_affinity (struct pci_dev *pdev,int vec); +bool pci_create_ims_domain (struct pci_dev *pdev,const struct msi_domain_template *template, +unsigned int hwsize,void *data); +struct msi_map pci_ims_alloc_irq (struct pci_dev *pdev,union msi_instance_cookie *icookie, +const struct irq_affinity_desc *affdesc); +void pci_ims_free_irq (struct pci_dev *pdev,struct msi_map map);#else static inline int pci_msi_vec_count (struct pci_dev *dev) {return -ENOSYS ; }
@@ -1719,6 +1726,25 @@ static inline const struct cpumask *pci_irq_get_affinity(struct pci_dev *pdev, {return cpu_possible_mask; } + +static inline bool pci_create_ims_domain (struct pci_dev *pdev, +const struct msi_domain_template *template, +unsigned int hwsize,void *data) +{return false; } + +static inline struct msi_map pci_ims_alloc_irq (struct pci_dev *pdev, +union msi_instance_cookie *icookie, +const struct irq_affinity_desc *affdesc) +{ +struct msi_map map = { .index = -ENOSYS , }; + +return map; +} + +static inline void pci_ims_free_irq (struct pci_dev *pdev,struct msi_map map) +{ +} +#endif /**
@@ -2616,14 +2642,6 @@ 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 -struct msi_domain_template ; - -bool pci_create_ims_domain (struct pci_dev *pdev,const struct msi_domain_template *template, -unsigned int hwsize,void *data); -struct msi_map pci_ims_alloc_irq (struct pci_dev *pdev,union msi_instance_cookie *icookie, -const struct irq_affinity_desc *affdesc); -void pci_ims_free_irq (struct pci_dev *pdev,struct msi_map map); -#include <linux/dma-mapping.h> #define pci_printk(level, pdev, fmt, arg...) \
Signature · 1 line
2.34.1
RFC PATCH v3 2/26 vfio/pci: Move PCI specific check from wrapper to PCI function
In reply to Reinette Chatrevfio_pci_set_irqs_ioctl() uses a PCI device specific check to determine if PCI specific vfio_pci_set_err_trigger() should be called.
Move the PCI device specific check into PCI specific vfio_pci_set_err_trigger() to make it easier for vfio_pci_set_irqs_ioctl() to become a frontend for interrupt backends for PCI devices as well as virtual devices.
---
No changes since RFC V2.
drivers/vfio/pci/vfio_pci_intrs.c | 6 ++++-- 1 file changed, 4 insertions(+), 2 deletions(-)
drivers/vfio/pci/vfio_pci_intrs.c +4 -2
@@ -758,6 +758,9 @@ static int vfio_pci_set_err_trigger(struct vfio_pci_core_device *vdev, unsigned index, unsigned start, unsigned count, uint32_t flags, void *data) { + if (!pci_is_pcie(vdev->pdev)) + return -ENOTTY; +if (index !=VFIO_PCI_ERR_IRQ_INDEX || start !=0 || count >1 )return -EINVAL ;
@@ -813,8 +816,7 @@ int vfio_pci_set_irqs_ioctl(struct vfio_pci_core_device *vdev, uint32_t flags,case VFIO_PCI_ERR_IRQ_INDEX :switch (flags &VFIO_IRQ_SET_ACTION_TYPE_MASK ) {case VFIO_IRQ_SET_ACTION_TRIGGER : -if (pci_is_pcie (vdev->pdev)) - func = vfio_pci_set_err_trigger; + func = vfio_pci_set_err_trigger;break ; }break ;
Signature · 1 line
2.34.1
RFC PATCH v3 3/26 vfio/pci: Use unsigned int instead of unsigned
In reply to Reinette Chatrecheckpatch.pl warns about usage of bare unsigned.
Change unsigned to unsigned int as a preparatory change to avoid checkpatch.pl producing several warnings as the work adding support for backends to VFIO interrupt management progress.
---
Changes since RFC V2:
- Include vfio_msi_set_block() in changes.
Note to maintainers:
After this change checkpatch.pl still has a few complaints about existing code using int32_t instead of s32. This was not changed and these warnings remain.
drivers/vfio/pci/vfio_pci_intrs.c | 42 ++++++++++++++++++------------- 1 file changed, 24 insertions(+), 18 deletions(-)
drivers/vfio/pci/vfio_pci_intrs.c +24 -18
@@ -503,8 +503,9 @@ static int vfio_msi_set_vector_signal(struct vfio_pci_core_device *vdev,return ret; } -static int vfio_msi_set_block (struct vfio_pci_core_device *vdev,unsigned start, -unsigned count,int32_t *fds,bool msix) +static int vfio_msi_set_block (struct vfio_pci_core_device *vdev, +unsigned int start,unsigned int count, +int32_t *fds,bool msix) {unsigned int i, j;int ret =0 ;
@@ -553,8 +554,9 @@ static void vfio_msi_disable(struct vfio_pci_core_device *vdev, bool msix)* IOCTL support */ static int vfio_pci_set_intx_unmask (struct vfio_pci_core_device *vdev, -unsigned index,unsigned start, -unsigned count,uint32_t flags,void *data) +unsigned int index,unsigned int start, +unsigned int count,uint32_t flags, +void *data) {if (!is_intx (vdev) || start !=0 || count !=1 )return -EINVAL ;
@@ -584,8 +586,8 @@ static int vfio_pci_set_intx_unmask(struct vfio_pci_core_device *vdev, }static int vfio_pci_set_intx_mask (struct vfio_pci_core_device *vdev, -unsigned index,unsigned start, -unsigned count,uint32_t flags,void *data) +unsigned int index,unsigned int start, +unsigned int count,uint32_t flags,void *data) {if (!is_intx (vdev) || start !=0 || count !=1 )return -EINVAL ;
@@ -604,8 +606,9 @@ static int vfio_pci_set_intx_mask(struct vfio_pci_core_device *vdev, }static int vfio_pci_set_intx_trigger (struct vfio_pci_core_device *vdev, -unsigned index,unsigned start, -unsigned count,uint32_t flags,void *data) +unsigned int index,unsigned int start, +unsigned int count,uint32_t flags, +void *data) {if (is_intx (vdev) && !count && (flags &VFIO_IRQ_SET_DATA_NONE )) {vfio_intx_disable (vdev);
@@ -647,8 +650,9 @@ static int vfio_pci_set_intx_trigger(struct vfio_pci_core_device *vdev, }static int vfio_pci_set_msi_trigger (struct vfio_pci_core_device *vdev, -unsigned index,unsigned start, -unsigned count,uint32_t flags,void *data) +unsigned int index,unsigned int start, +unsigned int count,uint32_t flags, +void *data) {struct vfio_pci_irq_ctx *ctx;unsigned int i;
@@ -755,8 +759,9 @@ static int vfio_pci_set_ctx_trigger_single(struct eventfd_ctx **ctx, }static int vfio_pci_set_err_trigger (struct vfio_pci_core_device *vdev, -unsigned index,unsigned start, -unsigned count,uint32_t flags,void *data) +unsigned int index,unsigned int start, +unsigned int count,uint32_t flags, +void *data) {if (!pci_is_pcie (vdev->pdev))return -ENOTTY ;
@@ -769,8 +774,9 @@ static int vfio_pci_set_err_trigger(struct vfio_pci_core_device *vdev, }static int vfio_pci_set_req_trigger (struct vfio_pci_core_device *vdev, -unsigned index,unsigned start, -unsigned count,uint32_t flags,void *data) +unsigned int index,unsigned int start, +unsigned int count,uint32_t flags, +void *data) {if (index !=VFIO_PCI_REQ_IRQ_INDEX || start !=0 || count >1 )return -EINVAL ;
@@ -780,11 +786,11 @@ static int vfio_pci_set_req_trigger(struct vfio_pci_core_device *vdev, }int vfio_pci_set_irqs_ioctl (struct vfio_pci_core_device *vdev,uint32_t flags, -unsigned index,unsigned start,unsigned count, -void *data) +unsigned int index,unsigned int start, +unsigned int count,void *data) { -int (*func)(struct vfio_pci_core_device *vdev,unsigned index, -unsigned start,unsigned count,uint32_t flags, +int (*func)(struct vfio_pci_core_device *vdev,unsigned int index, +unsigned int start,unsigned int count,uint32_t flags,void *data) =NULL ;switch (index) {
Signature · 1 line
2.34.1
RFC PATCH v3 4/26 vfio/pci: Make core interrupt callbacks accessible to all virtual devices
In reply to Reinette ChatreThe functions handling actions on interrupts for a virtual PCI device triggered by VFIO_DEVICE_SET_IRQS ioctl() expect to act on a passthrough PCI device represented by a struct vfio_pci_core_device.
A virtual device can support MSI-X while not being a passthrough PCI device and thus not be represented by a struct vfio_pci_core_device.
To support MSI-X in virtual devices it needs to be possible for their drivers to interact with the MSI-X interrupt management and thus the interrupt management should not require struct vfio_pci_core_device.
Introduce struct vfio_pci_intr_ctx that will contain a virtual device's interrupt context to be managed by an interrupt management backend. The first supported backend is the existing PCI device interrupt management. Modify the core VFIO PCI interrupt management functions to expect this structure. As a backend managing interrupts of passthrough PCI devices the existing VFIO PCI functions continue to operate on an actual PCI device represented by
struct vfio_pci_core_device that is provided via a private pointer.
More members will be added to struct vfio_pci_intr_ctx as members unique to interrupt context are transitioned from struct vfio_pci_core_device.
---
Changes since RFC V2:
- Improve changelog and comments.
drivers/vfio/pci/vfio_pci_core.c | 7 ++++--- drivers/vfio/pci/vfio_pci_intrs.c | 29 ++++++++++++++++++++--------- drivers/vfio/pci/vfio_pci_priv.h | 2 +- include/linux/vfio_pci_core.h | 9 +++++++++ 4 files changed, 34 insertions(+), 13 deletions(-)
drivers/vfio/pci/vfio_pci_core.c +4 -3
@@ -594,7 +594,7 @@ void vfio_pci_core_disable(struct vfio_pci_core_device *vdev)/* Stop the device from further DMA */ pci_clear_master (pdev); -vfio_pci_set_irqs_ioctl (vdev,VFIO_IRQ_SET_DATA_NONE | +vfio_pci_set_irqs_ioctl (&vdev->intr_ctx,VFIO_IRQ_SET_DATA_NONE |VFIO_IRQ_SET_ACTION_TRIGGER , vdev->irq_type,0 ,0 ,NULL );
@@ -1216,8 +1216,8 @@ static int vfio_pci_ioctl_set_irqs(struct vfio_pci_core_device *vdev,mutex_lock (&vdev->igate); - ret =vfio_pci_set_irqs_ioctl (vdev, hdr.flags, hdr.index, hdr.start, - hdr.count, data); + ret =vfio_pci_set_irqs_ioctl (&vdev->intr_ctx, hdr.flags, hdr.index, + hdr.start, hdr.count, data);mutex_unlock (&vdev->igate);kfree (data);
@@ -2166,6 +2166,7 @@ int vfio_pci_core_init_dev(struct vfio_device *core_vdev)INIT_LIST_HEAD (&vdev->sriov_pfs_item);init_rwsem (&vdev->memory_lock);xa_init (&vdev->ctx); + vdev->intr_ctx.priv = vdev;return 0 ; }
drivers/vfio/pci/vfio_pci_intrs.c +20 -9
@@ -553,11 +553,13 @@ static void vfio_msi_disable(struct vfio_pci_core_device *vdev, bool msix)/* * IOCTL support */ -static int vfio_pci_set_intx_unmask (struct vfio_pci_core_device *vdev, +static int vfio_pci_set_intx_unmask (struct vfio_pci_intr_ctx *intr_ctx,unsigned int index,unsigned int start,unsigned int count,uint32_t flags,void *data) { +struct vfio_pci_core_device *vdev = intr_ctx->priv; +if (!is_intx (vdev) || start !=0 || count !=1 )return -EINVAL ;
@@ -585,10 +587,12 @@ static int vfio_pci_set_intx_unmask(struct vfio_pci_core_device *vdev,return 0 ; } -static int vfio_pci_set_intx_mask (struct vfio_pci_core_device *vdev, +static int vfio_pci_set_intx_mask (struct vfio_pci_intr_ctx *intr_ctx,unsigned int index,unsigned int start,unsigned int count,uint32_t flags,void *data) { +struct vfio_pci_core_device *vdev = intr_ctx->priv; +if (!is_intx (vdev) || start !=0 || count !=1 )return -EINVAL ;
@@ -605,11 +609,13 @@ static int vfio_pci_set_intx_mask(struct vfio_pci_core_device *vdev,return 0 ; } -static int vfio_pci_set_intx_trigger (struct vfio_pci_core_device *vdev, +static int vfio_pci_set_intx_trigger (struct vfio_pci_intr_ctx *intr_ctx,unsigned int index,unsigned int start,unsigned int count,uint32_t flags,void *data) { +struct vfio_pci_core_device *vdev = intr_ctx->priv; +if (is_intx (vdev) && !count && (flags &VFIO_IRQ_SET_DATA_NONE )) {vfio_intx_disable (vdev);return 0 ;
@@ -649,11 +655,12 @@ static int vfio_pci_set_intx_trigger(struct vfio_pci_core_device *vdev,return 0 ; } -static int vfio_pci_set_msi_trigger (struct vfio_pci_core_device *vdev, +static int vfio_pci_set_msi_trigger (struct vfio_pci_intr_ctx *intr_ctx,unsigned int index,unsigned int start,unsigned int count,uint32_t flags,void *data) { +struct vfio_pci_core_device *vdev = intr_ctx->priv;struct vfio_pci_irq_ctx *ctx;unsigned int i;bool msix = (index ==VFIO_PCI_MSIX_IRQ_INDEX ) ? true : false;
@@ -758,11 +765,13 @@ static int vfio_pci_set_ctx_trigger_single(struct eventfd_ctx **ctx,return -EINVAL ; } -static int vfio_pci_set_err_trigger (struct vfio_pci_core_device *vdev, +static int vfio_pci_set_err_trigger (struct vfio_pci_intr_ctx *intr_ctx,unsigned int index,unsigned int start,unsigned int count,uint32_t flags,void *data) { +struct vfio_pci_core_device *vdev = intr_ctx->priv; +if (!pci_is_pcie (vdev->pdev))return -ENOTTY ;
@@ -773,11 +782,13 @@ static int vfio_pci_set_err_trigger(struct vfio_pci_core_device *vdev, count, flags, data); } -static int vfio_pci_set_req_trigger (struct vfio_pci_core_device *vdev, +static int vfio_pci_set_req_trigger (struct vfio_pci_intr_ctx *intr_ctx,unsigned int index,unsigned int start,unsigned int count,uint32_t flags,void *data) { +struct vfio_pci_core_device *vdev = intr_ctx->priv; +if (index !=VFIO_PCI_REQ_IRQ_INDEX || start !=0 || count >1 )return -EINVAL ;
@@ -785,11 +796,11 @@ static int vfio_pci_set_req_trigger(struct vfio_pci_core_device *vdev, count, flags, data); } -int vfio_pci_set_irqs_ioctl (struct vfio_pci_core_device *vdev,uint32_t flags, +int vfio_pci_set_irqs_ioctl (struct vfio_pci_intr_ctx *intr_ctx,uint32_t flags,unsigned int index,unsigned int start,unsigned int count,void *data) { -int (*func)(struct vfio_pci_core_device *vdev,unsigned int index, +int (*func)(struct vfio_pci_intr_ctx *intr_ctx,unsigned int index,unsigned int start,unsigned int count,uint32_t flags,void *data) =NULL ;
@@ -838,5 +849,5 @@ int vfio_pci_set_irqs_ioctl(struct vfio_pci_core_device *vdev, uint32_t flags,if (!func)return -ENOTTY ; -return func (vdev, index, start, count, flags, data); +return func (intr_ctx, index, start, count, flags, data); }
drivers/vfio/pci/vfio_pci_priv.h +1 -1
@@ -26,7 +26,7 @@ struct vfio_pci_ioeventfd {bool vfio_pci_intx_mask (struct vfio_pci_core_device *vdev);void vfio_pci_intx_unmask (struct vfio_pci_core_device *vdev); -int vfio_pci_set_irqs_ioctl (struct vfio_pci_core_device *vdev,uint32_t flags, +int vfio_pci_set_irqs_ioctl (struct vfio_pci_intr_ctx *intr_ctx,uint32_t flags,unsigned index,unsigned start,unsigned count,void *data);
include/linux/vfio_pci_core.h +9 -0
@@ -49,6 +49,14 @@ struct vfio_pci_region {u32 flags; }; +/* +* Interrupt context of virtual PCI device +* @priv: Private data of interrupt management backend +*/ +struct vfio_pci_intr_ctx { +void *priv; +}; +struct vfio_pci_core_device {struct vfio_device vdev;struct pci_dev *pdev;
@@ -96,6 +104,7 @@ struct vfio_pci_core_device {struct mutex vma_lock;struct list_head vma_list;struct rw_semaphore memory_lock; +struct vfio_pci_intr_ctx intr_ctx; };/* Will be exported for vfio pci drivers usage */
Signature · 1 line
2.34.1
RFC PATCH v3 5/26 vfio/pci: Split PCI interrupt management into front and backend
In reply to Reinette ChatreVFIO PCI interrupt management supports passthrough PCI devices with an interrupt in the guest backed by the same type of interrupt on the PCI device.
Interrupt management can be more flexible. An interrupt in the guest may be backed by a different type of interrupt on the host, for example MSI-X in guest can be backed by IMS on the host, or not backed by a device interrupt at all when the interrupt is emulated by the virtual device driver.
The main entry to guest interrupt management is via the VFIO_DEVICE_SET_IRQS ioctl(). By default the work is passed to interrupt management for PCI devices with the PCI specific functions called directly.
Make the ioctl() configurable to support different interrupt management backends. This is accomplished by introducing interrupt context specific callbacks that are initialized by the virtual device driver and then triggered via the ioctl().
The introduction of virtual device driver specific callbacks require its initialization. Create a dedicated interrupt context initialization function to avoid mixing more interrupt context initialization with general virtual device driver initialization.
---
Changes since RFC V2:
- Improve changelog and comments.
- Make vfio_pci_intr_ops static.
drivers/vfio/pci/vfio_pci_core.c | 2 +- drivers/vfio/pci/vfio_pci_intrs.c | 35 +++++++++++++++++++++++++------ include/linux/vfio_pci_core.h | 25 ++++++++++++++++++++++ 3 files changed, 55 insertions(+), 7 deletions(-)
drivers/vfio/pci/vfio_pci_core.c +1 -1
@@ -2166,7 +2166,7 @@ int vfio_pci_core_init_dev(struct vfio_device *core_vdev)INIT_LIST_HEAD (&vdev->sriov_pfs_item);init_rwsem (&vdev->memory_lock);xa_init (&vdev->ctx); - vdev->intr_ctx.priv = vdev; +vfio_pci_init_intr_ctx (vdev, &vdev->intr_ctx);return 0 ; }
drivers/vfio/pci/vfio_pci_intrs.c +29 -6
@@ -796,6 +796,23 @@ static int vfio_pci_set_req_trigger(struct vfio_pci_intr_ctx *intr_ctx, count, flags, data); } +static struct vfio_pci_intr_ops vfio_pci_intr_ops = { + .set_intx_mask = vfio_pci_set_intx_mask, + .set_intx_unmask = vfio_pci_set_intx_unmask, + .set_intx_trigger = vfio_pci_set_intx_trigger, + .set_msi_trigger = vfio_pci_set_msi_trigger, + .set_err_trigger = vfio_pci_set_err_trigger, + .set_req_trigger = vfio_pci_set_req_trigger, +}; + +void vfio_pci_init_intr_ctx (struct vfio_pci_core_device *vdev, +struct vfio_pci_intr_ctx *intr_ctx) +{ + intr_ctx->ops = &vfio_pci_intr_ops; + intr_ctx->priv = vdev; +} +EXPORT_SYMBOL_GPL (vfio_pci_init_intr_ctx); +int vfio_pci_set_irqs_ioctl (struct vfio_pci_intr_ctx *intr_ctx,uint32_t flags,unsigned int index,unsigned int start,unsigned int count,void *data)
@@ -808,13 +825,16 @@ int vfio_pci_set_irqs_ioctl(struct vfio_pci_intr_ctx *intr_ctx, uint32_t flags,case VFIO_PCI_INTX_IRQ_INDEX :switch (flags &VFIO_IRQ_SET_ACTION_TYPE_MASK ) {case VFIO_IRQ_SET_ACTION_MASK : - func = vfio_pci_set_intx_mask; +if (intr_ctx->ops->set_intx_mask) + func = intr_ctx->ops->set_intx_mask;break ;case VFIO_IRQ_SET_ACTION_UNMASK : - func = vfio_pci_set_intx_unmask; +if (intr_ctx->ops->set_intx_unmask) + func = intr_ctx->ops->set_intx_unmask;break ;case VFIO_IRQ_SET_ACTION_TRIGGER : - func = vfio_pci_set_intx_trigger; +if (intr_ctx->ops->set_intx_trigger) + func = intr_ctx->ops->set_intx_trigger;break ; }break ;
@@ -826,21 +846,24 @@ int vfio_pci_set_irqs_ioctl(struct vfio_pci_intr_ctx *intr_ctx, uint32_t flags,/* XXX Need masking support exported */ break ;case VFIO_IRQ_SET_ACTION_TRIGGER : - func = vfio_pci_set_msi_trigger; +if (intr_ctx->ops->set_msi_trigger) + func = intr_ctx->ops->set_msi_trigger;break ; }break ;case VFIO_PCI_ERR_IRQ_INDEX :switch (flags &VFIO_IRQ_SET_ACTION_TYPE_MASK ) {case VFIO_IRQ_SET_ACTION_TRIGGER : - func = vfio_pci_set_err_trigger; +if (intr_ctx->ops->set_err_trigger) + func = intr_ctx->ops->set_err_trigger;break ; }break ;case VFIO_PCI_REQ_IRQ_INDEX :switch (flags &VFIO_IRQ_SET_ACTION_TYPE_MASK ) {case VFIO_IRQ_SET_ACTION_TRIGGER : - func = vfio_pci_set_req_trigger; +if (intr_ctx->ops->set_req_trigger) + func = intr_ctx->ops->set_req_trigger;break ; }break ;
include/linux/vfio_pci_core.h +25 -0
@@ -51,12 +51,35 @@ struct vfio_pci_region {/* * Interrupt context of virtual PCI device +* @ops: Interrupt management backend functions * @priv: Private data of interrupt management backend */ struct vfio_pci_intr_ctx { +const struct vfio_pci_intr_ops *ops;void *priv; }; +struct vfio_pci_intr_ops { +int (*set_intx_mask)(struct vfio_pci_intr_ctx *intr_ctx, +unsigned int index,unsigned int start, +unsigned int count,uint32_t flags,void *data); +int (*set_intx_unmask)(struct vfio_pci_intr_ctx *intr_ctx, +unsigned int index,unsigned int start, +unsigned int count,uint32_t flags,void *data); +int (*set_intx_trigger)(struct vfio_pci_intr_ctx *intr_ctx, +unsigned int index,unsigned int start, +unsigned int count,uint32_t flags,void *data); +int (*set_msi_trigger)(struct vfio_pci_intr_ctx *intr_ctx, +unsigned int index,unsigned int start, +unsigned int count,uint32_t flags,void *data); +int (*set_err_trigger)(struct vfio_pci_intr_ctx *intr_ctx, +unsigned int index,unsigned int start, +unsigned int count,uint32_t flags,void *data); +int (*set_req_trigger)(struct vfio_pci_intr_ctx *intr_ctx, +unsigned int index,unsigned int start, +unsigned int count,uint32_t flags,void *data); +}; +struct vfio_pci_core_device {struct vfio_device vdev;struct pci_dev *pdev;
@@ -124,6 +147,8 @@ int vfio_pci_core_sriov_configure(struct vfio_pci_core_device *vdev, int nr_virtfn);long vfio_pci_core_ioctl (struct vfio_device *core_vdev,unsigned int cmd,unsigned long arg); +void vfio_pci_init_intr_ctx (struct vfio_pci_core_device *vdev, +struct vfio_pci_intr_ctx *intr_ctx);int vfio_pci_core_ioctl_feature (struct vfio_device *device,u32 flags,void __user *arg,size_t argsz); ssize_t vfio_pci_core_read(struct vfio_device *core_vdev, char __user *buf,
Signature · 1 line
2.34.1
RFC PATCH v3 6/26 vfio/pci: Separate MSI and MSI-X handling
In reply to Reinette ChatreVFIO PCI interrupt management uses a single entry for both MSI and MSI-X management with the called functions using a boolean when necessary to distinguish between MSI and MSI-X. This remains unchanged.
Virtual device interrupt management should not be required to use the same callback for both MSI and MSI-X. It may be possible for a virtual device to not support MSI at all and only provide MSI-X interrupt management.
Separate the MSI and MSI-X interrupt management by allowing different callbacks for each interrupt type. For PCI devices the callback remains the same.
---
No changes since RFC V2.
drivers/vfio/pci/vfio_pci_intrs.c | 14 +++++++++++++- include/linux/vfio_pci_core.h | 3 +++ 2 files changed, 16 insertions(+), 1 deletion(-)
drivers/vfio/pci/vfio_pci_intrs.c +13 -1
@@ -801,6 +801,7 @@ static struct vfio_pci_intr_ops vfio_pci_intr_ops = { .set_intx_unmask = vfio_pci_set_intx_unmask, .set_intx_trigger = vfio_pci_set_intx_trigger, .set_msi_trigger = vfio_pci_set_msi_trigger, + .set_msix_trigger = vfio_pci_set_msi_trigger, .set_err_trigger = vfio_pci_set_err_trigger, .set_req_trigger = vfio_pci_set_req_trigger, };
@@ -839,7 +840,6 @@ int vfio_pci_set_irqs_ioctl(struct vfio_pci_intr_ctx *intr_ctx, uint32_t flags, }break ;case VFIO_PCI_MSI_IRQ_INDEX : -case VFIO_PCI_MSIX_IRQ_INDEX :switch (flags &VFIO_IRQ_SET_ACTION_TYPE_MASK ) {case VFIO_IRQ_SET_ACTION_MASK :case VFIO_IRQ_SET_ACTION_UNMASK :
@@ -851,6 +851,18 @@ int vfio_pci_set_irqs_ioctl(struct vfio_pci_intr_ctx *intr_ctx, uint32_t flags,break ; }break ; +case VFIO_PCI_MSIX_IRQ_INDEX : +switch (flags &VFIO_IRQ_SET_ACTION_TYPE_MASK ) { +case VFIO_IRQ_SET_ACTION_MASK : +case VFIO_IRQ_SET_ACTION_UNMASK : +/* XXX Need masking support exported */ +break ; +case VFIO_IRQ_SET_ACTION_TRIGGER : +if (intr_ctx->ops->set_msix_trigger) + func = intr_ctx->ops->set_msix_trigger; +break ; + } +break ;case VFIO_PCI_ERR_IRQ_INDEX :switch (flags &VFIO_IRQ_SET_ACTION_TYPE_MASK ) {case VFIO_IRQ_SET_ACTION_TRIGGER :
include/linux/vfio_pci_core.h +3 -0
@@ -72,6 +72,9 @@ struct vfio_pci_intr_ops {int (*set_msi_trigger)(struct vfio_pci_intr_ctx *intr_ctx,unsigned int index,unsigned int start,unsigned int count,uint32_t flags,void *data); +int (*set_msix_trigger)(struct vfio_pci_intr_ctx *intr_ctx, +unsigned int index,unsigned int start, +unsigned int count,uint32_t flags,void *data);int (*set_err_trigger)(struct vfio_pci_intr_ctx *intr_ctx,unsigned int index,unsigned int start,unsigned int count,uint32_t flags,void *data);
Signature · 1 line
2.34.1
RFC PATCH v3 7/26 vfio/pci: Move interrupt eventfd to interrupt context
In reply to Reinette ChatreThe eventfds associated with device request notification and error IRQ are managed by VFIO PCI interrupt management as triggered by the VFIO_DEVICE_SET_IRQS ioctl().
Move these eventfds as well as their mutex to the generic and dedicated interrupt management context struct vfio_pci_intr_ctx to enable another interrupt management backend to manage these eventfd.
igate mutex protects eventfd modification. With the eventfd within the bigger scoped interrupt context the mutex scope is also expanded to mean that all members of struct vfio_pci_intr_ctx are protected by it.
This move results in the vfio_pci_set_req_trigger() to no longer require a struct vfio_pci_core_device, operating just on the generic struct vfio_pci_intr_ctx, and thus available for direct use by other interrupt management backends.
This introduces the first interrupt context related cleanup call. Create vfio_pci_release_intr_ctx() to match existing vfio_pci_init_intr_ctx().
---
Changes since RFC V2:
- Improve changelog.
drivers/vfio/pci/vfio_pci_core.c | 39 +++++++++++++++---------------- drivers/vfio/pci/vfio_pci_intrs.c | 13 +++++++---- include/linux/vfio_pci_core.h | 10 +++++--- 3 files changed, 35 insertions(+), 27 deletions(-)
drivers/vfio/pci/vfio_pci_core.c +19 -20
@@ -700,16 +700,16 @@ void vfio_pci_core_close_device(struct vfio_device *core_vdev)#endif vfio_pci_core_disable (vdev); -mutex_lock (&vdev->igate); -if (vdev->err_trigger) { -eventfd_ctx_put (vdev->err_trigger); - vdev->err_trigger =NULL ; +mutex_lock (&vdev->intr_ctx.igate); +if (vdev->intr_ctx.err_trigger) { +eventfd_ctx_put (vdev->intr_ctx.err_trigger); + vdev->intr_ctx.err_trigger =NULL ; } -if (vdev->req_trigger) { -eventfd_ctx_put (vdev->req_trigger); - vdev->req_trigger =NULL ; +if (vdev->intr_ctx.req_trigger) { +eventfd_ctx_put (vdev->intr_ctx.req_trigger); + vdev->intr_ctx.req_trigger =NULL ; } -mutex_unlock (&vdev->igate); +mutex_unlock (&vdev->intr_ctx.igate); }EXPORT_SYMBOL_GPL (vfio_pci_core_close_device);
@@ -1214,12 +1214,12 @@ static int vfio_pci_ioctl_set_irqs(struct vfio_pci_core_device *vdev,return PTR_ERR (data); } -mutex_lock (&vdev->igate); +mutex_lock (&vdev->intr_ctx.igate); ret =vfio_pci_set_irqs_ioctl (&vdev->intr_ctx, hdr.flags, hdr.index, hdr.start, hdr.count, data); -mutex_unlock (&vdev->igate); +mutex_unlock (&vdev->intr_ctx.igate);kfree (data);return ret;
@@ -1876,20 +1876,20 @@ void vfio_pci_core_request(struct vfio_device *core_vdev, unsigned int count) container_of(core_vdev, struct vfio_pci_core_device, vdev);struct pci_dev *pdev = vdev->pdev; -mutex_lock (&vdev->igate); +mutex_lock (&vdev->intr_ctx.igate); -if (vdev->req_trigger) { +if (vdev->intr_ctx.req_trigger) {if (!(count %10 ))pci_notice_ratelimited (pdev,"Relaying device request to user (#%u)\n" , count); -eventfd_signal (vdev->req_trigger,1 ); +eventfd_signal (vdev->intr_ctx.req_trigger,1 ); }else if (count ==0 ) {pci_warn (pdev,"No device request channel registered, blocked until released by user\n" ); } -mutex_unlock (&vdev->igate); +mutex_unlock (&vdev->intr_ctx.igate); }EXPORT_SYMBOL_GPL (vfio_pci_core_request);
@@ -2156,7 +2156,6 @@ int vfio_pci_core_init_dev(struct vfio_device *core_vdev) vdev->pdev =to_pci_dev (core_vdev->dev); vdev->irq_type =VFIO_PCI_NUM_IRQS ; -mutex_init (&vdev->igate);spin_lock_init (&vdev->irqlock);mutex_init (&vdev->ioeventfds_lock);INIT_LIST_HEAD (&vdev->dummy_resources_list);
@@ -2177,7 +2176,7 @@ void vfio_pci_core_release_dev(struct vfio_device *core_vdev)struct vfio_pci_core_device *vdev = container_of(core_vdev, struct vfio_pci_core_device, vdev); -mutex_destroy (&vdev->igate); +vfio_pci_release_intr_ctx (&vdev->intr_ctx);mutex_destroy (&vdev->ioeventfds_lock);mutex_destroy (&vdev->vma_lock);kfree (vdev->region);
@@ -2300,12 +2299,12 @@ pci_ers_result_t vfio_pci_core_aer_err_detected(struct pci_dev *pdev, {struct vfio_pci_core_device *vdev =dev_get_drvdata (&pdev->dev); -mutex_lock (&vdev->igate); +mutex_lock (&vdev->intr_ctx.igate); -if (vdev->err_trigger) -eventfd_signal (vdev->err_trigger,1 ); +if (vdev->intr_ctx.err_trigger) +eventfd_signal (vdev->intr_ctx.err_trigger,1 ); -mutex_unlock (&vdev->igate); +mutex_unlock (&vdev->intr_ctx.igate);return PCI_ERS_RESULT_CAN_RECOVER ; }
drivers/vfio/pci/vfio_pci_intrs.c +9 -4
@@ -778,7 +778,7 @@ static int vfio_pci_set_err_trigger(struct vfio_pci_intr_ctx *intr_ctx,if (index !=VFIO_PCI_ERR_IRQ_INDEX || start !=0 || count >1 )return -EINVAL ; -return vfio_pci_set_ctx_trigger_single (&vdev->err_trigger, +return vfio_pci_set_ctx_trigger_single (&intr_ctx->err_trigger, count, flags, data); }
@@ -787,12 +787,10 @@ static int vfio_pci_set_req_trigger(struct vfio_pci_intr_ctx *intr_ctx, unsigned int count, uint32_t flags, void *data) { - struct vfio_pci_core_device *vdev = intr_ctx->priv; - if (index != VFIO_PCI_REQ_IRQ_INDEX || start != 0 || count > 1) return -EINVAL; -return vfio_pci_set_ctx_trigger_single (&vdev->req_trigger, +return vfio_pci_set_ctx_trigger_single (&intr_ctx->req_trigger, count, flags, data); }
@@ -811,9 +809,16 @@ void vfio_pci_init_intr_ctx(struct vfio_pci_core_device *vdev, { intr_ctx->ops = &vfio_pci_intr_ops; intr_ctx->priv = vdev; +mutex_init (&intr_ctx->igate); }EXPORT_SYMBOL_GPL (vfio_pci_init_intr_ctx); +void vfio_pci_release_intr_ctx (struct vfio_pci_intr_ctx *intr_ctx) +{ +mutex_destroy (&intr_ctx->igate); +} +EXPORT_SYMBOL_GPL (vfio_pci_release_intr_ctx); +int vfio_pci_set_irqs_ioctl (struct vfio_pci_intr_ctx *intr_ctx,uint32_t flags,unsigned int index,unsigned int start,unsigned int count,void *data)
include/linux/vfio_pci_core.h +7 -3
@@ -53,10 +53,16 @@ struct vfio_pci_region {* Interrupt context of virtual PCI device * @ops: Interrupt management backend functions * @priv: Private data of interrupt management backend +* @igate: Protects members of struct vfio_pci_intr_ctx +* @err_trigger: Eventfd associated with error reporting IRQ +* @req_trigger: Eventfd associated with device request notification */ struct vfio_pci_intr_ctx {const struct vfio_pci_intr_ops *ops;void *priv; +struct mutex igate; +struct eventfd_ctx *err_trigger; +struct eventfd_ctx *req_trigger; };struct vfio_pci_intr_ops {
@@ -92,7 +98,6 @@ struct vfio_pci_core_device {u8 *vconfig;struct perm_bits *msi_perm;spinlock_t irqlock; -struct mutex igate;struct xarray ctx;int irq_type;int num_regions;
@@ -117,8 +122,6 @@ struct vfio_pci_core_device {struct pci_saved_state *pci_saved_state;struct pci_saved_state *pm_save;int ioeventfds_nr; -struct eventfd_ctx *err_trigger; -struct eventfd_ctx *req_trigger;struct eventfd_ctx *pm_wake_eventfd_ctx;struct list_head dummy_resources_list;struct mutex ioeventfds_lock;
@@ -152,6 +155,7 @@ long vfio_pci_core_ioctl(struct vfio_device *core_vdev, unsigned int cmd, unsigned long arg);void vfio_pci_init_intr_ctx (struct vfio_pci_core_device *vdev,struct vfio_pci_intr_ctx *intr_ctx); +void vfio_pci_release_intr_ctx (struct vfio_pci_intr_ctx *intr_ctx);int vfio_pci_core_ioctl_feature (struct vfio_device *device,u32 flags,void __user *arg,size_t argsz); ssize_t vfio_pci_core_read(struct vfio_device *core_vdev, char __user *buf,
Signature · 1 line
2.34.1
RFC PATCH v3 8/26 vfio/pci: Move mutex acquisition into function
In reply to Reinette Chatrevfio_pci_set_irqs_ioctl() is the entrypoint for interrupt management via the VFIO_DEVICE_SET_IRQS ioctl(). vfio_pci_set_irqs_ioctl() can be called from a virtual device driver after its callbacks have been configured to support the needed interrupt management.
The igate mutex is obtained before vfio_pci_set_irqs_ioctl() to protect against concurrent changes to interrupt context. It should not be necessary for all users of vfio_pci_set_irqs_ioctl() to remember to take the mutex.
Acquire and release the mutex within vfio_pci_set_irqs_ioctl().
---
Changes since RFC V2:
- Improve changelog.
drivers/vfio/pci/vfio_pci_core.c | 2 -- drivers/vfio/pci/vfio_pci_intrs.c | 10 ++++++++-- 2 files changed, 8 insertions(+), 4 deletions(-)
drivers/vfio/pci/vfio_pci_core.c +0 -2
@@ -1214,12 +1214,10 @@ static int vfio_pci_ioctl_set_irqs(struct vfio_pci_core_device *vdev,return PTR_ERR (data); } -mutex_lock (&vdev->intr_ctx.igate); ret =vfio_pci_set_irqs_ioctl (&vdev->intr_ctx, hdr.flags, hdr.index, hdr.start, hdr.count, data); -mutex_unlock (&vdev->intr_ctx.igate);kfree (data);return ret;
drivers/vfio/pci/vfio_pci_intrs.c +8 -2
@@ -826,7 +826,9 @@ int vfio_pci_set_irqs_ioctl(struct vfio_pci_intr_ctx *intr_ctx, uint32_t flags,int (*func)(struct vfio_pci_intr_ctx *intr_ctx,unsigned int index,unsigned int start,unsigned int count,uint32_t flags,void *data) =NULL ; +int ret = -ENOTTY ; +mutex_lock (&intr_ctx->igate);switch (index) {case VFIO_PCI_INTX_IRQ_INDEX :switch (flags &VFIO_IRQ_SET_ACTION_TYPE_MASK ) {
@@ -887,7 +889,11 @@ int vfio_pci_set_irqs_ioctl(struct vfio_pci_intr_ctx *intr_ctx, uint32_t flags, }if (!func) -return -ENOTTY ; + goto out_unlock; + + ret =func (intr_ctx, index, start, count, flags, data); -return func (intr_ctx, index, start, count, flags, data); +out_unlock: +mutex_unlock (&intr_ctx->igate); +return ret; }
Signature · 1 line
2.34.1
RFC PATCH v3 9/26 vfio/pci: Move per-interrupt contexts to generic interrupt struct
In reply to Reinette ChatreVFIO PCI interrupt management maintains per-interrupt context within an xarray using the interrupt vector as index.
Move the per-interrupt context to the generic interrupt context in struct vfio_pci_intr_ctx to enable the per-interrupt context to be managed by different backends.
---
Changes since RFC V2:
- Improve changelog.
drivers/vfio/pci/vfio_pci_core.c | 1 - drivers/vfio/pci/vfio_pci_intrs.c | 9 +++++---- include/linux/vfio_pci_core.h | 3 ++- 3 files changed, 7 insertions(+), 6 deletions(-)
drivers/vfio/pci/vfio_pci_core.c +0 -1
@@ -2162,7 +2162,6 @@ int vfio_pci_core_init_dev(struct vfio_device *core_vdev)INIT_LIST_HEAD (&vdev->vma_list);INIT_LIST_HEAD (&vdev->sriov_pfs_item);init_rwsem (&vdev->memory_lock); -xa_init (&vdev->ctx);vfio_pci_init_intr_ctx (vdev, &vdev->intr_ctx);return 0 ;
drivers/vfio/pci/vfio_pci_intrs.c +5 -4
@@ -52,13 +52,13 @@ staticstruct vfio_pci_irq_ctx *vfio_irq_ctx_get (struct vfio_pci_core_device *vdev,unsigned long index) { -return xa_load (&vdev->ctx, index); +return xa_load (&vdev->intr_ctx.ctx, index); }static void vfio_irq_ctx_free (struct vfio_pci_core_device *vdev,struct vfio_pci_irq_ctx *ctx,unsigned long index) { -xa_erase (&vdev->ctx, index); +xa_erase (&vdev->intr_ctx.ctx, index);kfree (ctx); }
@@ -72,7 +72,7 @@ vfio_irq_ctx_alloc(struct vfio_pci_core_device *vdev, unsigned long index)if (!ctx)return NULL ; - ret =xa_insert (&vdev->ctx, index, ctx,GFP_KERNEL_ACCOUNT ); + ret =xa_insert (&vdev->intr_ctx.ctx, index, ctx,GFP_KERNEL_ACCOUNT );if (ret) {kfree (ctx);return NULL ;
@@ -530,7 +530,7 @@ static void vfio_msi_disable(struct vfio_pci_core_device *vdev, bool msix)unsigned long i;u16 cmd; -xa_for_each (&vdev->ctx, i, ctx) { +xa_for_each (&vdev->intr_ctx.ctx, i, ctx) {vfio_virqfd_disable (&ctx->unmask);vfio_virqfd_disable (&ctx->mask);vfio_msi_set_vector_signal (vdev, i,-1 , msix);
@@ -810,6 +810,7 @@ void vfio_pci_init_intr_ctx(struct vfio_pci_core_device *vdev, intr_ctx->ops = &vfio_pci_intr_ops; intr_ctx->priv = vdev;mutex_init (&intr_ctx->igate); +xa_init (&intr_ctx->ctx); }EXPORT_SYMBOL_GPL (vfio_pci_init_intr_ctx);
include/linux/vfio_pci_core.h +2 -1
@@ -56,6 +56,7 @@ struct vfio_pci_region {* @igate: Protects members of struct vfio_pci_intr_ctx * @err_trigger: Eventfd associated with error reporting IRQ * @req_trigger: Eventfd associated with device request notification +* @ctx: Per-interrupt context indexed by vector */ struct vfio_pci_intr_ctx {const struct vfio_pci_intr_ops *ops;
@@ -63,6 +64,7 @@ struct vfio_pci_intr_ctx {struct mutex igate;struct eventfd_ctx *err_trigger;struct eventfd_ctx *req_trigger; +struct xarray ctx; };struct vfio_pci_intr_ops {
@@ -98,7 +100,6 @@ struct vfio_pci_core_device {u8 *vconfig;struct perm_bits *msi_perm;spinlock_t irqlock; -struct xarray ctx;int irq_type;int num_regions;struct vfio_pci_region *region;
Signature · 1 line
2.34.1
RFC PATCH v3 10/26 vfio/pci: Move IRQ type to generic interrupt context
In reply to Reinette ChatreThe type of interrupts within the guest is not unique to PCI devices and needed for other virtual devices supporting interrupts.
Move interrupt type to the generic interrupt context struct vfio_pci_intr_ctx.
---
Question for maintainers:irq_type is accessed in ioctl() flow as well as other flows. It is not clear to me how it is protected against concurrent access. Should accesses outside of ioctl() flow take the mutex?
No changes since RFC V2.
drivers/vfio/pci/vfio_pci_config.c | 2 +- drivers/vfio/pci/vfio_pci_core.c | 5 ++--- drivers/vfio/pci/vfio_pci_intrs.c | 21 +++++++++++---------- include/linux/vfio_pci_core.h | 3 ++- 4 files changed, 16 insertions(+), 15 deletions(-)
drivers/vfio/pci/vfio_pci_config.c +1 -1
@@ -1168,7 +1168,7 @@ static int vfio_msi_config_write(struct vfio_pci_core_device *vdev, int pos, flags =le16_to_cpu (*pflags);/* MSI is enabled via ioctl */ -if (vdev->irq_type !=VFIO_PCI_MSI_IRQ_INDEX ) +if (vdev->intr_ctx.irq_type !=VFIO_PCI_MSI_IRQ_INDEX ) flags &= ~PCI_MSI_FLAGS_ENABLE ;/* Check queue size */
drivers/vfio/pci/vfio_pci_core.c +2 -3
@@ -427,7 +427,7 @@ static int vfio_pci_core_runtime_suspend(struct device *dev)* vfio_pci_intx_mask() will return false and in that case, INTx * should not be unmasked in the runtime resume. */ - vdev->pm_intx_masked = ((vdev->irq_type ==VFIO_PCI_INTX_IRQ_INDEX ) && + vdev->pm_intx_masked = ((vdev->intr_ctx.irq_type ==VFIO_PCI_INTX_IRQ_INDEX ) &&vfio_pci_intx_mask (vdev));return 0 ;
@@ -596,7 +596,7 @@ void vfio_pci_core_disable(struct vfio_pci_core_device *vdev)vfio_pci_set_irqs_ioctl (&vdev->intr_ctx,VFIO_IRQ_SET_DATA_NONE |VFIO_IRQ_SET_ACTION_TRIGGER , - vdev->irq_type,0 ,0 ,NULL ); + vdev->intr_ctx.irq_type,0 ,0 ,NULL );/* Device closed, don't need mutex here */ list_for_each_entry_safe(ioeventfd, ioeventfd_tmp,
@@ -2153,7 +2153,6 @@ int vfio_pci_core_init_dev(struct vfio_device *core_vdev) container_of(core_vdev, struct vfio_pci_core_device, vdev); vdev->pdev =to_pci_dev (core_vdev->dev); - vdev->irq_type =VFIO_PCI_NUM_IRQS ;spin_lock_init (&vdev->irqlock);mutex_init (&vdev->ioeventfds_lock);INIT_LIST_HEAD (&vdev->dummy_resources_list);
drivers/vfio/pci/vfio_pci_intrs.c +11 -10
@@ -33,19 +33,19 @@ struct vfio_pci_irq_ctx {static bool irq_is (struct vfio_pci_core_device *vdev,int type) { -return vdev->irq_type == type; +return vdev->intr_ctx.irq_type == type; }static bool is_intx (struct vfio_pci_core_device *vdev) { -return vdev->irq_type ==VFIO_PCI_INTX_IRQ_INDEX ; +return vdev->intr_ctx.irq_type ==VFIO_PCI_INTX_IRQ_INDEX ; }static bool is_irq_none (struct vfio_pci_core_device *vdev) { -return !(vdev->irq_type ==VFIO_PCI_INTX_IRQ_INDEX || - vdev->irq_type ==VFIO_PCI_MSI_IRQ_INDEX || - vdev->irq_type ==VFIO_PCI_MSIX_IRQ_INDEX ); +return !(vdev->intr_ctx.irq_type ==VFIO_PCI_INTX_IRQ_INDEX || + vdev->intr_ctx.irq_type ==VFIO_PCI_MSI_IRQ_INDEX || + vdev->intr_ctx.irq_type ==VFIO_PCI_MSIX_IRQ_INDEX ); } static
@@ -255,7 +255,7 @@ static int vfio_intx_enable(struct vfio_pci_core_device *vdev)if (vdev->pci_2_3)pci_intx (vdev->pdev, !ctx->masked); - vdev->irq_type =VFIO_PCI_INTX_IRQ_INDEX ; + vdev->intr_ctx.irq_type =VFIO_PCI_INTX_IRQ_INDEX ;return 0 ; }
@@ -331,7 +331,7 @@ static void vfio_intx_disable(struct vfio_pci_core_device *vdev)vfio_virqfd_disable (&ctx->mask); }vfio_intx_set_signal (vdev,-1 ); - vdev->irq_type =VFIO_PCI_NUM_IRQS ; + vdev->intr_ctx.irq_type =VFIO_PCI_NUM_IRQS ;vfio_irq_ctx_free (vdev, ctx,0 ); }
@@ -367,7 +367,7 @@ static int vfio_msi_enable(struct vfio_pci_core_device *vdev, int nvec, bool msi }vfio_pci_memory_unlock_and_restore (vdev, cmd); - vdev->irq_type = msix ?VFIO_PCI_MSIX_IRQ_INDEX : + vdev->intr_ctx.irq_type = msix ?VFIO_PCI_MSIX_IRQ_INDEX :VFIO_PCI_MSI_IRQ_INDEX ;if (!msix) {
@@ -547,7 +547,7 @@ static void vfio_msi_disable(struct vfio_pci_core_device *vdev, bool msix)if (vdev->nointx)pci_intx (pdev,0 ); - vdev->irq_type =VFIO_PCI_NUM_IRQS ; + vdev->intr_ctx.irq_type =VFIO_PCI_NUM_IRQS ; }/*
@@ -677,7 +677,7 @@ static int vfio_pci_set_msi_trigger(struct vfio_pci_intr_ctx *intr_ctx,int32_t *fds = data;int ret; -if (vdev->irq_type == index) +if (vdev->intr_ctx.irq_type == index)return vfio_msi_set_block (vdev, start, count, fds, msix);
@@ -807,6 +807,7 @@ static struct vfio_pci_intr_ops vfio_pci_intr_ops = {void vfio_pci_init_intr_ctx (struct vfio_pci_core_device *vdev,struct vfio_pci_intr_ctx *intr_ctx) { + intr_ctx->irq_type =VFIO_PCI_NUM_IRQS ; intr_ctx->ops = &vfio_pci_intr_ops; intr_ctx->priv = vdev;mutex_init (&intr_ctx->igate);
include/linux/vfio_pci_core.h +2 -1
@@ -57,6 +57,7 @@ struct vfio_pci_region {* @err_trigger: Eventfd associated with error reporting IRQ * @req_trigger: Eventfd associated with device request notification * @ctx: Per-interrupt context indexed by vector +* @irq_type: Type of interrupt from guest perspective */ struct vfio_pci_intr_ctx {const struct vfio_pci_intr_ops *ops;
@@ -65,6 +66,7 @@ struct vfio_pci_intr_ctx {struct eventfd_ctx *err_trigger;struct eventfd_ctx *req_trigger;struct xarray ctx; +int irq_type; };struct vfio_pci_intr_ops {
@@ -100,7 +102,6 @@ struct vfio_pci_core_device {u8 *vconfig;struct perm_bits *msi_perm;spinlock_t irqlock; -int irq_type;int num_regions;struct vfio_pci_region *region;u8 msi_qmax;
Signature · 1 line
2.34.1
RFC PATCH v3 11/26 vfio/pci: Provide interrupt context to irq_is() and is_irq_none()
In reply to Reinette ChatreThe IRQ type moved to the interrupt context, struct vfio_pci_intr_ctx.
Let the tests on the IRQ type use the interrupt context directly without any assumption about the containing structure. Doing so makes these generic utilities available to all interrupt management backends.
---
Changes since RFC V2:
- New patch
drivers/vfio/pci/vfio_pci_intrs.c | 24 ++++++++++++------------ 1 file changed, 12 insertions(+), 12 deletions(-)
drivers/vfio/pci/vfio_pci_intrs.c +12 -12
@@ -31,9 +31,9 @@ struct vfio_pci_irq_ctx {struct irq_bypass_producer producer; }; -static bool irq_is (struct vfio_pci_core_device *vdev,int type) +static bool irq_is (struct vfio_pci_intr_ctx *intr_ctx,int type) { -return vdev->intr_ctx.irq_type == type; +return intr_ctx->irq_type == type; }static bool is_intx (struct vfio_pci_core_device *vdev)
@@ -41,11 +41,11 @@ static bool is_intx(struct vfio_pci_core_device *vdev)return vdev->intr_ctx.irq_type ==VFIO_PCI_INTX_IRQ_INDEX ; } -static bool is_irq_none (struct vfio_pci_core_device *vdev) +static bool is_irq_none (struct vfio_pci_intr_ctx *intr_ctx) { -return !(vdev->intr_ctx.irq_type ==VFIO_PCI_INTX_IRQ_INDEX || - vdev->intr_ctx.irq_type ==VFIO_PCI_MSI_IRQ_INDEX || - vdev->intr_ctx.irq_type ==VFIO_PCI_MSIX_IRQ_INDEX ); +return !(intr_ctx->irq_type ==VFIO_PCI_INTX_IRQ_INDEX || + intr_ctx->irq_type ==VFIO_PCI_MSI_IRQ_INDEX || + intr_ctx->irq_type ==VFIO_PCI_MSIX_IRQ_INDEX ); } static
@@ -235,7 +235,7 @@ static int vfio_intx_enable(struct vfio_pci_core_device *vdev) {struct vfio_pci_irq_ctx *ctx; -if (!is_irq_none (vdev)) +if (!is_irq_none (&vdev->intr_ctx))return -EINVAL ;if (!vdev->pdev->irq)
@@ -353,7 +353,7 @@ static int vfio_msi_enable(struct vfio_pci_core_device *vdev, int nvec, bool msiint ret;u16 cmd; -if (!is_irq_none (vdev)) +if (!is_irq_none (&vdev->intr_ctx))return -EINVAL ;/* return the number of supported vectors if we can't get all: */
@@ -621,7 +621,7 @@ static int vfio_pci_set_intx_trigger(struct vfio_pci_intr_ctx *intr_ctx,return 0 ; } -if (!(is_intx (vdev) ||is_irq_none (vdev)) || start !=0 || count !=1 ) +if (!(is_intx (vdev) ||is_irq_none (intr_ctx)) || start !=0 || count !=1 )return -EINVAL ;if (flags &VFIO_IRQ_SET_DATA_EVENTFD ) {
@@ -665,12 +665,12 @@ static int vfio_pci_set_msi_trigger(struct vfio_pci_intr_ctx *intr_ctx,unsigned int i;bool msix = (index ==VFIO_PCI_MSIX_IRQ_INDEX ) ? true : false; -if (irq_is (vdev, index) && !count && (flags &VFIO_IRQ_SET_DATA_NONE )) { +if (irq_is (intr_ctx, index) && !count && (flags &VFIO_IRQ_SET_DATA_NONE )) {vfio_msi_disable (vdev, msix);return 0 ; } -if (!(irq_is (vdev, index) ||is_irq_none (vdev))) +if (!(irq_is (intr_ctx, index) ||is_irq_none (intr_ctx)))return -EINVAL ;if (flags &VFIO_IRQ_SET_DATA_EVENTFD ) {
@@ -692,7 +692,7 @@ static int vfio_pci_set_msi_trigger(struct vfio_pci_intr_ctx *intr_ctx,return ret; } -if (!irq_is (vdev, index)) +if (!irq_is (intr_ctx, index))return -EINVAL ;for (i = start; i < start + count; i++) {
Signature · 1 line
2.34.1
RFC PATCH v3 12/26 vfio/pci: Provide interrupt context to generic ops
In reply to Reinette ChatreThe functions operating on the per-interrupt context were originally created to support management of PCI device interrupts where the interrupt context was maintained within the virtual PCI device's struct vfio_pci_core_device. Now that the per-interrupt context has been moved to a more generic struct vfio_pci_intr_ctx these utilities can be changed to expect the generic structure instead. This enables these utilities to be used in other interrupt management backends.
---
No changes since RFC V2.
drivers/vfio/pci/vfio_pci_intrs.c | 41 ++++++++++++++++--------------- 1 file changed, 21 insertions(+), 20 deletions(-)
drivers/vfio/pci/vfio_pci_intrs.c +21 -20
@@ -49,21 +49,21 @@ static bool is_irq_none(struct vfio_pci_intr_ctx *intr_ctx) }static -struct vfio_pci_irq_ctx *vfio_irq_ctx_get (struct vfio_pci_core_device *vdev, +struct vfio_pci_irq_ctx *vfio_irq_ctx_get (struct vfio_pci_intr_ctx *intr_ctx,unsigned long index) { -return xa_load (&vdev->intr_ctx.ctx, index); +return xa_load (&intr_ctx->ctx, index); } -static void vfio_irq_ctx_free (struct vfio_pci_core_device *vdev, +static void vfio_irq_ctx_free (struct vfio_pci_intr_ctx *intr_ctx,struct vfio_pci_irq_ctx *ctx,unsigned long index) { -xa_erase (&vdev->intr_ctx.ctx, index); +xa_erase (&intr_ctx->ctx, index);kfree (ctx); }static struct vfio_pci_irq_ctx * -vfio_irq_ctx_alloc (struct vfio_pci_core_device *vdev,unsigned long index) +vfio_irq_ctx_alloc (struct vfio_pci_intr_ctx *intr_ctx,unsigned long index) {struct vfio_pci_irq_ctx *ctx;int ret;
@@ -72,7 +72,7 @@ vfio_irq_ctx_alloc(struct vfio_pci_core_device *vdev, unsigned long index)if (!ctx)return NULL ; - ret =xa_insert (&vdev->intr_ctx.ctx, index, ctx,GFP_KERNEL_ACCOUNT ); + ret =xa_insert (&intr_ctx->ctx, index, ctx,GFP_KERNEL_ACCOUNT );if (ret) {kfree (ctx);return NULL ;
@@ -91,7 +91,7 @@ static void vfio_send_intx_eventfd(void *opaque, void *unused)if (likely (is_intx (vdev) && !vdev->virq_disabled)) {struct vfio_pci_irq_ctx *ctx; - ctx =vfio_irq_ctx_get (vdev,0 ); + ctx =vfio_irq_ctx_get (&vdev->intr_ctx,0 );if (WARN_ON_ONCE (!ctx))return ;eventfd_signal (ctx->trigger,1 );
@@ -120,7 +120,7 @@ bool vfio_pci_intx_mask(struct vfio_pci_core_device *vdev) goto out_unlock; } - ctx =vfio_irq_ctx_get (vdev,0 ); + ctx =vfio_irq_ctx_get (&vdev->intr_ctx,0 );if (WARN_ON_ONCE (!ctx)) goto out_unlock;
@@ -169,7 +169,7 @@ static int vfio_pci_intx_unmask_handler(void *opaque, void *unused) goto out_unlock; } - ctx =vfio_irq_ctx_get (vdev,0 ); + ctx =vfio_irq_ctx_get (&vdev->intr_ctx,0 );if (WARN_ON_ONCE (!ctx)) goto out_unlock;
@@ -207,7 +207,7 @@ static irqreturn_t vfio_intx_handler(int irq, void *dev_id)unsigned long flags;int ret =IRQ_NONE ; - ctx =vfio_irq_ctx_get (vdev,0 ); + ctx =vfio_irq_ctx_get (&vdev->intr_ctx,0 );if (WARN_ON_ONCE (!ctx))return ret;
@@ -241,7 +241,7 @@ static int vfio_intx_enable(struct vfio_pci_core_device *vdev)if (!vdev->pdev->irq)return -ENODEV ; - ctx =vfio_irq_ctx_alloc (vdev,0 ); + ctx =vfio_irq_ctx_alloc (&vdev->intr_ctx,0 );if (!ctx)return -ENOMEM ;
@@ -269,7 +269,7 @@ static int vfio_intx_set_signal(struct vfio_pci_core_device *vdev, int fd)unsigned long flags;int ret; - ctx =vfio_irq_ctx_get (vdev,0 ); + ctx =vfio_irq_ctx_get (&vdev->intr_ctx,0 );if (WARN_ON_ONCE (!ctx))return -EINVAL ;
@@ -324,7 +324,7 @@ static void vfio_intx_disable(struct vfio_pci_core_device *vdev) {struct vfio_pci_irq_ctx *ctx; - ctx =vfio_irq_ctx_get (vdev,0 ); + ctx =vfio_irq_ctx_get (&vdev->intr_ctx,0 );WARN_ON_ONCE (!ctx);if (ctx) {vfio_virqfd_disable (&ctx->unmask);
@@ -332,7 +332,7 @@ static void vfio_intx_disable(struct vfio_pci_core_device *vdev) }vfio_intx_set_signal (vdev,-1 ); vdev->intr_ctx.irq_type =VFIO_PCI_NUM_IRQS ; -vfio_irq_ctx_free (vdev, ctx,0 ); +vfio_irq_ctx_free (&vdev->intr_ctx, ctx,0 ); }/*
@@ -421,7 +421,7 @@ static int vfio_msi_set_vector_signal(struct vfio_pci_core_device *vdev,int irq = -EINVAL , ret;u16 cmd; - ctx =vfio_irq_ctx_get (vdev, vector); + ctx =vfio_irq_ctx_get (&vdev->intr_ctx, vector);if (ctx) {irq_bypass_unregister_producer (&ctx->producer);
@@ -432,7 +432,7 @@ static int vfio_msi_set_vector_signal(struct vfio_pci_core_device *vdev,/* Interrupt stays allocated, will be freed at MSI-X disable. */ kfree (ctx->name);eventfd_ctx_put (ctx->trigger); -vfio_irq_ctx_free (vdev, ctx, vector); +vfio_irq_ctx_free (&vdev->intr_ctx, ctx, vector); }if (fd <0 )
@@ -445,7 +445,7 @@ static int vfio_msi_set_vector_signal(struct vfio_pci_core_device *vdev,return irq; } - ctx =vfio_irq_ctx_alloc (vdev, vector); + ctx =vfio_irq_ctx_alloc (&vdev->intr_ctx, vector);if (!ctx)return -ENOMEM ;
@@ -499,7 +499,7 @@ static int vfio_msi_set_vector_signal(struct vfio_pci_core_device *vdev, out_free_name:kfree (ctx->name); out_free_ctx: -vfio_irq_ctx_free (vdev, ctx, vector); +vfio_irq_ctx_free (&vdev->intr_ctx, ctx, vector);return ret; }
@@ -570,7 +570,8 @@ static int vfio_pci_set_intx_unmask(struct vfio_pci_intr_ctx *intr_ctx,if (unmask)vfio_pci_intx_unmask (vdev); }else if (flags &VFIO_IRQ_SET_DATA_EVENTFD ) { -struct vfio_pci_irq_ctx *ctx =vfio_irq_ctx_get (vdev,0 ); +struct vfio_pci_irq_ctx *ctx =vfio_irq_ctx_get (&vdev->intr_ctx, +0 );int32_t fd = *(int32_t *)data;if (WARN_ON_ONCE (!ctx))
@@ -696,7 +697,7 @@ static int vfio_pci_set_msi_trigger(struct vfio_pci_intr_ctx *intr_ctx,return -EINVAL ;for (i = start; i < start + count; i++) { - ctx =vfio_irq_ctx_get (vdev, i); + ctx =vfio_irq_ctx_get (&vdev->intr_ctx, i);if (!ctx)continue ;if (flags &VFIO_IRQ_SET_DATA_NONE ) {
Signature · 1 line
2.34.1
RFC PATCH v3 14/26 vfio/pci: Let interrupt management backend interpret interrupt index
In reply to Reinette Chatrevfio_pci_set_msi_trigger() and vfio_msi_set_block() are generic and can be shared by different interrupt backends. This implies that these functions should not interpret user provided parameters but instead pass them to the backend specific code for interpretation.
Instead of assuming that only MSI or MSI-X can be provided via the index and passing a boolean based on what was received, pass the actual index to backend for interpretation.
---
Changes since RFC V2:
- New patch
drivers/vfio/pci/vfio_pci_intrs.c | 38 +++++++++++++++++-------------- 1 file changed, 21 insertions(+), 17 deletions(-)
drivers/vfio/pci/vfio_pci_intrs.c +21 -17
@@ -346,17 +346,20 @@ static irqreturn_t vfio_msihandler(int irq, void *arg)return IRQ_HANDLED ; } -static int vfio_msi_enable (struct vfio_pci_intr_ctx *intr_ctx,int nvec,bool msix) +static int vfio_msi_enable (struct vfio_pci_intr_ctx *intr_ctx,int nvec, +unsigned int index) {struct vfio_pci_core_device *vdev = intr_ctx->priv;struct pci_dev *pdev = vdev->pdev; -unsigned int flag = msix ?PCI_IRQ_MSIX :PCI_IRQ_MSI ; +unsigned int flag;int ret;u16 cmd;if (!is_irq_none (intr_ctx))return -EINVAL ; + flag = (index ==VFIO_PCI_MSIX_IRQ_INDEX ) ?PCI_IRQ_MSIX :PCI_IRQ_MSI ; +/* return the number of supported vectors if we can't get all: */ cmd =vfio_pci_memory_lock_and_enable (vdev); ret =pci_alloc_irq_vectors (pdev,1 , nvec, flag);
@@ -368,10 +371,9 @@ static int vfio_msi_enable(struct vfio_pci_intr_ctx *intr_ctx, int nvec, bool ms }vfio_pci_memory_unlock_and_restore (vdev, cmd); - intr_ctx->irq_type = msix ?VFIO_PCI_MSIX_IRQ_INDEX : -VFIO_PCI_MSI_IRQ_INDEX ; + intr_ctx->irq_type = index; -if (!msix) { +if (index ==VFIO_PCI_MSI_IRQ_INDEX ) {/* * Compute the virtual hardware field for max msi vectors - * it is the log base 2 of the number of vectors.
@@ -414,8 +416,10 @@ static int vfio_msi_alloc_irq(struct vfio_pci_core_device *vdev, }static int vfio_msi_set_vector_signal (struct vfio_pci_core_device *vdev, -unsigned int vector,int fd,bool msix) +unsigned int vector,int fd, +unsigned int index) { +bool msix = (index ==VFIO_PCI_MSIX_IRQ_INDEX ) ? true : false;struct pci_dev *pdev = vdev->pdev;struct vfio_pci_irq_ctx *ctx;struct eventfd_ctx *trigger;
@@ -506,25 +510,26 @@ static int vfio_msi_set_vector_signal(struct vfio_pci_core_device *vdev,static int vfio_msi_set_block (struct vfio_pci_core_device *vdev,unsigned int start,unsigned int count, -int32_t *fds,bool msix) +int32_t *fds,unsigned int index) {unsigned int i, j;int ret =0 ;for (i =0 , j = start; i < count && !ret; i++, j++) {int fd = fds ? fds[i] :-1 ; - ret =vfio_msi_set_vector_signal (vdev, j, fd, msix); + ret =vfio_msi_set_vector_signal (vdev, j, fd, index); }if (ret) {for (i = start; i < j; i++) -vfio_msi_set_vector_signal (vdev, i,-1 , msix); +vfio_msi_set_vector_signal (vdev, i,-1 , index); }return ret; } -static void vfio_msi_disable (struct vfio_pci_intr_ctx *intr_ctx,bool msix) +static void vfio_msi_disable (struct vfio_pci_intr_ctx *intr_ctx, +unsigned int index) {struct vfio_pci_core_device *vdev = intr_ctx->priv;struct pci_dev *pdev = vdev->pdev;
@@ -535,7 +540,7 @@ static void vfio_msi_disable(struct vfio_pci_intr_ctx *intr_ctx, bool msix)xa_for_each (&intr_ctx->ctx, i, ctx) {vfio_virqfd_disable (&ctx->unmask);vfio_virqfd_disable (&ctx->mask); -vfio_msi_set_vector_signal (vdev, i,-1 , msix); +vfio_msi_set_vector_signal (vdev, i,-1 , index); } cmd =vfio_pci_memory_lock_and_enable (vdev);
@@ -666,10 +671,9 @@ static int vfio_pci_set_msi_trigger(struct vfio_pci_intr_ctx *intr_ctx,struct vfio_pci_core_device *vdev = intr_ctx->priv;struct vfio_pci_irq_ctx *ctx;unsigned int i; -bool msix = (index ==VFIO_PCI_MSIX_IRQ_INDEX ) ? true : false;if (irq_is (intr_ctx, index) && !count && (flags &VFIO_IRQ_SET_DATA_NONE )) { -vfio_msi_disable (intr_ctx, msix); +vfio_msi_disable (intr_ctx, index);return 0 ; }
@@ -682,15 +686,15 @@ static int vfio_pci_set_msi_trigger(struct vfio_pci_intr_ctx *intr_ctx,if (vdev->intr_ctx.irq_type == index)return vfio_msi_set_block (vdev, start, count, - fds, msix); + fds, index); - ret =vfio_msi_enable (intr_ctx, start + count, msix); + ret =vfio_msi_enable (intr_ctx, start + count, index);if (ret)return ret; - ret =vfio_msi_set_block (vdev, start, count, fds, msix); + ret =vfio_msi_set_block (vdev, start, count, fds, index);if (ret) -vfio_msi_disable (intr_ctx, msix); +vfio_msi_disable (intr_ctx, index);return ret; }
Signature · 1 line
2.34.1
RFC PATCH v3 13/26 vfio/pci: Provide interrupt context to vfio_msi_enable() and vfio_msi_disable()
In reply to Reinette Chatrevfio_msi_enable() and vfio_msi_disable() perform the PCI specific operations to allocate and free interrupts on the device that will back the guest interrupts. This makes these functions backend specific calls that should be called by the interrupt management frontend.
Pass the interrupt context as parameter to vfio_msi_enable() and vfio_msi_disable() so that they can be called by a generic frontend and make it possible for other backends to provide their own vfio_msi_enable() and vfio_msi_disable().
---
Changes since RFC V2:
- New patch
drivers/vfio/pci/vfio_pci_intrs.c | 20 +++++++++++--------- 1 file changed, 11 insertions(+), 9 deletions(-)
drivers/vfio/pci/vfio_pci_intrs.c +11 -9
@@ -346,14 +346,15 @@ static irqreturn_t vfio_msihandler(int irq, void *arg)return IRQ_HANDLED ; } -static int vfio_msi_enable (struct vfio_pci_core_device *vdev,int nvec,bool msix) +static int vfio_msi_enable (struct vfio_pci_intr_ctx *intr_ctx,int nvec,bool msix) { +struct vfio_pci_core_device *vdev = intr_ctx->priv;struct pci_dev *pdev = vdev->pdev;unsigned int flag = msix ?PCI_IRQ_MSIX :PCI_IRQ_MSI ;int ret;u16 cmd; -if (!is_irq_none (&vdev->intr_ctx)) +if (!is_irq_none (intr_ctx))return -EINVAL ;/* return the number of supported vectors if we can't get all: */
@@ -367,7 +368,7 @@ static int vfio_msi_enable(struct vfio_pci_core_device *vdev, int nvec, bool msi }vfio_pci_memory_unlock_and_restore (vdev, cmd); - vdev->intr_ctx.irq_type = msix ?VFIO_PCI_MSIX_IRQ_INDEX : + intr_ctx->irq_type = msix ?VFIO_PCI_MSIX_IRQ_INDEX :VFIO_PCI_MSI_IRQ_INDEX ;if (!msix) {
@@ -523,14 +524,15 @@ static int vfio_msi_set_block(struct vfio_pci_core_device *vdev,return ret; } -static void vfio_msi_disable (struct vfio_pci_core_device *vdev,bool msix) +static void vfio_msi_disable (struct vfio_pci_intr_ctx *intr_ctx,bool msix) { +struct vfio_pci_core_device *vdev = intr_ctx->priv;struct pci_dev *pdev = vdev->pdev;struct vfio_pci_irq_ctx *ctx;unsigned long i;u16 cmd; -xa_for_each (&vdev->intr_ctx.ctx, i, ctx) { +xa_for_each (&intr_ctx->ctx, i, ctx) {vfio_virqfd_disable (&ctx->unmask);vfio_virqfd_disable (&ctx->mask);vfio_msi_set_vector_signal (vdev, i,-1 , msix);
@@ -547,7 +549,7 @@ static void vfio_msi_disable(struct vfio_pci_core_device *vdev, bool msix)if (vdev->nointx)pci_intx (pdev,0 ); - vdev->intr_ctx.irq_type =VFIO_PCI_NUM_IRQS ; + intr_ctx->irq_type =VFIO_PCI_NUM_IRQS ; }/*
@@ -667,7 +669,7 @@ static int vfio_pci_set_msi_trigger(struct vfio_pci_intr_ctx *intr_ctx,bool msix = (index ==VFIO_PCI_MSIX_IRQ_INDEX ) ? true : false;if (irq_is (intr_ctx, index) && !count && (flags &VFIO_IRQ_SET_DATA_NONE )) { -vfio_msi_disable (vdev, msix); +vfio_msi_disable (intr_ctx, msix);return 0 ; }
@@ -682,13 +684,13 @@ static int vfio_pci_set_msi_trigger(struct vfio_pci_intr_ctx *intr_ctx,return vfio_msi_set_block (vdev, start, count, fds, msix); - ret =vfio_msi_enable (vdev, start + count, msix); + ret =vfio_msi_enable (intr_ctx, start + count, msix);if (ret)return ret; ret =vfio_msi_set_block (vdev, start, count, fds, msix);if (ret) -vfio_msi_disable (vdev, msix); +vfio_msi_disable (intr_ctx, msix);return ret; }
Signature · 1 line
2.34.1
RFC PATCH v3 16/26 vfio/pci: Split interrupt context initialization
In reply to Reinette Chatrestruct vfio_pci_intr_ctx is the context associated with interrupts of a virtual device. The interrupt context is initialized with backend specific data required by the particular interrupt management backend as well as common initialization required by all interrupt management backends.
Split interrupt context initialization into common and interrupt management backend specific calls. The entrypoint will be the initialization of a particular interrupt management backend which in turn calls the common initialization.
---
No changes since RFC V2.
drivers/vfio/pci/vfio_pci_intrs.c | 18 ++++++++++++++---- 1 file changed, 14 insertions(+), 4 deletions(-)
drivers/vfio/pci/vfio_pci_intrs.c +14 -4
@@ -801,6 +801,18 @@ static int vfio_pci_set_req_trigger(struct vfio_pci_intr_ctx *intr_ctx, count, flags, data); } +static void _vfio_pci_init_intr_ctx (struct vfio_pci_intr_ctx *intr_ctx) +{ + intr_ctx->irq_type =VFIO_PCI_NUM_IRQS ; +mutex_init (&intr_ctx->igate); +xa_init (&intr_ctx->ctx); +} + +static void _vfio_pci_release_intr_ctx (struct vfio_pci_intr_ctx *intr_ctx) +{ +mutex_destroy (&intr_ctx->igate); +} +static struct vfio_pci_intr_ops vfio_pci_intr_ops = { .set_intx_mask = vfio_pci_set_intx_mask, .set_intx_unmask = vfio_pci_set_intx_unmask,
@@ -814,17 +826,15 @@ static struct vfio_pci_intr_ops vfio_pci_intr_ops = {void vfio_pci_init_intr_ctx (struct vfio_pci_core_device *vdev,struct vfio_pci_intr_ctx *intr_ctx) { - intr_ctx->irq_type =VFIO_PCI_NUM_IRQS ; +_vfio_pci_init_intr_ctx (intr_ctx); intr_ctx->ops = &vfio_pci_intr_ops; intr_ctx->priv = vdev; -mutex_init (&intr_ctx->igate); -xa_init (&intr_ctx->ctx); }EXPORT_SYMBOL_GPL (vfio_pci_init_intr_ctx);void vfio_pci_release_intr_ctx (struct vfio_pci_intr_ctx *intr_ctx) { -mutex_destroy (&intr_ctx->igate); +_vfio_pci_release_intr_ctx (intr_ctx); }EXPORT_SYMBOL_GPL (vfio_pci_release_intr_ctx);
Signature · 1 line
2.34.1
RFC PATCH v3 15/26 vfio/pci: Move generic code to frontend
In reply to Reinette Chatrevfio_pci_set_msi_trigger() and vfio_msi_set_block() are generic and thus appropriate to be frontend code. This means that they should operate on the interrupt context, not backend specific data.
Provide the interrupt context as parameter to vfio_pci_set_msi_trigger() and vfio_msi_set_block() and remove all references to the PCI interrupt management data from these functions. This enables these functions to form part of the interrupt management frontend shared by different interrupt management backends.
---
Changes since RFC V2:
- New patch
drivers/vfio/pci/vfio_pci_intrs.c | 28 ++++++++++++++-------------- 1 file changed, 14 insertions(+), 14 deletions(-)
drivers/vfio/pci/vfio_pci_intrs.c +14 -14
@@ -415,18 +415,19 @@ static int vfio_msi_alloc_irq(struct vfio_pci_core_device *vdev,return map.index <0 ? map.index : map.virq; } -static int vfio_msi_set_vector_signal (struct vfio_pci_core_device *vdev, +static int vfio_msi_set_vector_signal (struct vfio_pci_intr_ctx *intr_ctx,unsigned int vector,int fd,unsigned int index) {bool msix = (index ==VFIO_PCI_MSIX_IRQ_INDEX ) ? true : false; +struct vfio_pci_core_device *vdev = intr_ctx->priv;struct pci_dev *pdev = vdev->pdev;struct vfio_pci_irq_ctx *ctx;struct eventfd_ctx *trigger;int irq = -EINVAL , ret;u16 cmd; - ctx =vfio_irq_ctx_get (&vdev->intr_ctx, vector); + ctx =vfio_irq_ctx_get (intr_ctx, vector);if (ctx) {irq_bypass_unregister_producer (&ctx->producer);
@@ -437,7 +438,7 @@ static int vfio_msi_set_vector_signal(struct vfio_pci_core_device *vdev,/* Interrupt stays allocated, will be freed at MSI-X disable. */ kfree (ctx->name);eventfd_ctx_put (ctx->trigger); -vfio_irq_ctx_free (&vdev->intr_ctx, ctx, vector); +vfio_irq_ctx_free (intr_ctx, ctx, vector); }if (fd <0 )
@@ -450,7 +451,7 @@ static int vfio_msi_set_vector_signal(struct vfio_pci_core_device *vdev,return irq; } - ctx =vfio_irq_ctx_alloc (&vdev->intr_ctx, vector); + ctx =vfio_irq_ctx_alloc (intr_ctx, vector);if (!ctx)return -ENOMEM ;
@@ -504,11 +505,11 @@ static int vfio_msi_set_vector_signal(struct vfio_pci_core_device *vdev, out_free_name:kfree (ctx->name); out_free_ctx: -vfio_irq_ctx_free (&vdev->intr_ctx, ctx, vector); +vfio_irq_ctx_free (intr_ctx, ctx, vector);return ret; } -static int vfio_msi_set_block (struct vfio_pci_core_device *vdev, +static int vfio_msi_set_block (struct vfio_pci_intr_ctx *intr_ctx,unsigned int start,unsigned int count,int32_t *fds,unsigned int index) {
@@ -517,12 +518,12 @@ static int vfio_msi_set_block(struct vfio_pci_core_device *vdev,for (i =0 , j = start; i < count && !ret; i++, j++) {int fd = fds ? fds[i] :-1 ; - ret =vfio_msi_set_vector_signal (vdev, j, fd, index); + ret =vfio_msi_set_vector_signal (intr_ctx, j, fd, index); }if (ret) {for (i = start; i < j; i++) -vfio_msi_set_vector_signal (vdev, i,-1 , index); +vfio_msi_set_vector_signal (intr_ctx, i,-1 , index); }return ret;
@@ -540,7 +541,7 @@ static void vfio_msi_disable(struct vfio_pci_intr_ctx *intr_ctx,xa_for_each (&intr_ctx->ctx, i, ctx) {vfio_virqfd_disable (&ctx->unmask);vfio_virqfd_disable (&ctx->mask); -vfio_msi_set_vector_signal (vdev, i,-1 , index); +vfio_msi_set_vector_signal (intr_ctx, i,-1 , index); } cmd =vfio_pci_memory_lock_and_enable (vdev);
@@ -668,7 +669,6 @@ static int vfio_pci_set_msi_trigger(struct vfio_pci_intr_ctx *intr_ctx, unsigned int count, uint32_t flags, void *data) { - struct vfio_pci_core_device *vdev = intr_ctx->priv; struct vfio_pci_irq_ctx *ctx;unsigned int i;
@@ -684,15 +684,15 @@ static int vfio_pci_set_msi_trigger(struct vfio_pci_intr_ctx *intr_ctx,int32_t *fds = data;int ret; -if (vdev->intr_ctx.irq_type == index) -return vfio_msi_set_block (vdev, start, count, +if (intr_ctx->irq_type == index) +return vfio_msi_set_block (intr_ctx, start, count, fds, index); ret =vfio_msi_enable (intr_ctx, start + count, index);if (ret)return ret; - ret =vfio_msi_set_block (vdev, start, count, fds, index); + ret =vfio_msi_set_block (intr_ctx, start, count, fds, index);if (ret)vfio_msi_disable (intr_ctx, index);
@@ -703,7 +703,7 @@ static int vfio_pci_set_msi_trigger(struct vfio_pci_intr_ctx *intr_ctx,return -EINVAL ;for (i = start; i < start + count; i++) { - ctx =vfio_irq_ctx_get (&vdev->intr_ctx, i); + ctx =vfio_irq_ctx_get (intr_ctx, i);if (!ctx)continue ;if (flags &VFIO_IRQ_SET_DATA_NONE ) {
Signature · 1 line
2.34.1
RFC PATCH v3 17/26 vfio/pci: Make vfio_pci_set_irqs_ioctl() available
In reply to Reinette Chatrevfio_pci_set_irqs_ioctl() is now a generic entrypoint that can be configured to support different interrupt management backend.
Export vfio_pci_set_irqs_ioctl() for use by other virtual device drivers.
---
Changes since RFC V2:
- Improve changelog.
drivers/vfio/pci/vfio_pci_intrs.c | 1 + include/linux/vfio_pci_core.h | 3 +++ 2 files changed, 4 insertions(+)
drivers/vfio/pci/vfio_pci_intrs.c +1 -0
@@ -916,3 +916,4 @@ int vfio_pci_set_irqs_ioctl(struct vfio_pci_intr_ctx *intr_ctx, uint32_t flags,mutex_unlock (&intr_ctx->igate);return ret; } +EXPORT_SYMBOL_GPL (vfio_pci_set_irqs_ioctl);
include/linux/vfio_pci_core.h +3 -0
@@ -158,6 +158,9 @@ long vfio_pci_core_ioctl(struct vfio_device *core_vdev, unsigned int cmd,void vfio_pci_init_intr_ctx (struct vfio_pci_core_device *vdev,struct vfio_pci_intr_ctx *intr_ctx);void vfio_pci_release_intr_ctx (struct vfio_pci_intr_ctx *intr_ctx); +int vfio_pci_set_irqs_ioctl (struct vfio_pci_intr_ctx *intr_ctx,uint32_t flags, +unsigned int index,unsigned int start, +unsigned int count,void *data);int vfio_pci_core_ioctl_feature (struct vfio_device *device,u32 flags,void __user *arg,size_t argsz); ssize_t vfio_pci_core_read(struct vfio_device *core_vdev, char __user *buf,
Signature · 1 line
2.34.1
RFC PATCH v3 18/26 vfio/pci: Preserve per-interrupt contexts
In reply to Reinette ChatreInterrupt management for PCI passthrough devices create a new per-interrupt context every time an interrupt is allocated, freeing it when the interrupt is freed.
The per-interrupt context contains the properties of a particular interrupt. Without a property that guides interrupt allocation and free it is acceptable to always create a new per-interrupt context.
Maintain per-interrupt context across interrupt allocate and free events in preparation for per-interrupt properties that guide interrupt allocation and free. Examples of such properties are:
(a) whether the interrupt is emulated or not, which guides whether the backend should indeed allocate and/or free an interrupt, (b) an instance cookie associated with the interrupt that needs to be provided to interrupt allocation when the interrupt is backed by IMS.
This means that existence of per-interrupt context no longer implies a valid trigger, pointers to freed memory should be cleared, and a new per-interrupt context cannot be assumed needing allocation when an interrupt is allocated.
---
Changes since RFC V2:
- New patch
drivers/vfio/pci/vfio_pci_intrs.c | 41 ++++++++++++++++++++++--------- 1 file changed, 29 insertions(+), 12 deletions(-)
drivers/vfio/pci/vfio_pci_intrs.c +29 -12
@@ -429,7 +429,7 @@ static int vfio_msi_set_vector_signal(struct vfio_pci_intr_ctx *intr_ctx, ctx =vfio_irq_ctx_get (intr_ctx, vector); -if (ctx) { +if (ctx && ctx->trigger) {irq_bypass_unregister_producer (&ctx->producer); irq =pci_irq_vector (pdev, vector); cmd =vfio_pci_memory_lock_and_enable (vdev);
@@ -437,8 +437,9 @@ static int vfio_msi_set_vector_signal(struct vfio_pci_intr_ctx *intr_ctx,vfio_pci_memory_unlock_and_restore (vdev, cmd);/* Interrupt stays allocated, will be freed at MSI-X disable. */ kfree (ctx->name); + ctx->name =NULL ;eventfd_ctx_put (ctx->trigger); -vfio_irq_ctx_free (intr_ctx, ctx, vector); + ctx->trigger =NULL ; }if (fd <0 )
@@ -451,16 +452,17 @@ static int vfio_msi_set_vector_signal(struct vfio_pci_intr_ctx *intr_ctx,return irq; } - ctx =vfio_irq_ctx_alloc (intr_ctx, vector); -if (!ctx) -return -ENOMEM ; +/* Per-interrupt context remain allocated. */ +if (!ctx) { + ctx =vfio_irq_ctx_alloc (intr_ctx, vector); +if (!ctx) +return -ENOMEM ; + } ctx->name =kasprintf (GFP_KERNEL_ACCOUNT ,"vfio-msi%s[%d](%s)" , msix ?"x" :"" , vector,pci_name (pdev)); -if (!ctx->name) { - ret = -ENOMEM ; - goto out_free_ctx; - } +if (!ctx->name) +return -ENOMEM ; trigger =eventfd_ctx_fdget (fd);if (IS_ERR (trigger)) {
@@ -504,8 +506,7 @@ static int vfio_msi_set_vector_signal(struct vfio_pci_intr_ctx *intr_ctx,eventfd_ctx_put (trigger); out_free_name:kfree (ctx->name); -out_free_ctx: -vfio_irq_ctx_free (intr_ctx, ctx, vector); + ctx->name =NULL ;return ret; }
@@ -704,7 +705,7 @@ static int vfio_pci_set_msi_trigger(struct vfio_pci_intr_ctx *intr_ctx,for (i = start; i < start + count; i++) { ctx =vfio_irq_ctx_get (intr_ctx, i); -if (!ctx) +if (!ctx || !ctx->trigger)continue ;if (flags &VFIO_IRQ_SET_DATA_NONE ) {eventfd_signal (ctx->trigger,1 );
@@ -810,6 +811,22 @@ static void _vfio_pci_init_intr_ctx(struct vfio_pci_intr_ctx *intr_ctx)static void _vfio_pci_release_intr_ctx (struct vfio_pci_intr_ctx *intr_ctx) { +struct vfio_pci_irq_ctx *ctx; +unsigned long i; + +/* +* Per-interrupt context remains allocated after interrupt is +* freed. Per-interrupt context need to be freed separately. +*/ +mutex_lock (&intr_ctx->igate); +xa_for_each (&intr_ctx->ctx, i, ctx) { +WARN_ON_ONCE (ctx->trigger); +WARN_ON_ONCE (ctx->name); +xa_erase (&intr_ctx->ctx, i); +kfree (ctx); + } +mutex_unlock (&intr_ctx->igate); +mutex_destroy (&intr_ctx->igate); }
Signature · 1 line
2.34.1
RFC PATCH v3 19/26 vfio/pci: Store Linux IRQ number in per-interrupt context
In reply to Reinette ChatreThe Linux IRQ number is a property shared among all interrupt backends but not all interrupt management backends have a simple query for it. pci_irq_vector() can be used to obtain the Linux IRQ number of a MSI-X interrupt but there is no such query for IMS interrupts.
The Linux IRQ number is needed during interrupt free as well as during register of IRQ bypass producer. It is unnecessary to query the Linux IRQ number at each stage, the number can be stored at the time the interrupt is allocated and obtained from its per-interrupt context when needed.
---
Changes since RFC V2:
- New patch
drivers/vfio/pci/vfio_pci_intrs.c | 10 +++++++--- 1 file changed, 7 insertions(+), 3 deletions(-)
drivers/vfio/pci/vfio_pci_intrs.c +7 -3
@@ -29,6 +29,7 @@ struct vfio_pci_irq_ctx {char *name;bool masked;struct irq_bypass_producer producer; +int virq; };static bool irq_is (struct vfio_pci_intr_ctx *intr_ctx,int type)
@@ -431,10 +432,11 @@ static int vfio_msi_set_vector_signal(struct vfio_pci_intr_ctx *intr_ctx,if (ctx && ctx->trigger) {irq_bypass_unregister_producer (&ctx->producer); - irq =pci_irq_vector (pdev, vector); + irq = ctx->virq; cmd =vfio_pci_memory_lock_and_enable (vdev); -free_irq (irq, ctx->trigger); +free_irq (ctx->virq, ctx->trigger);vfio_pci_memory_unlock_and_restore (vdev, cmd); + ctx->virq =0 ;/* Interrupt stays allocated, will be freed at MSI-X disable. */ kfree (ctx->name); ctx->name =NULL ;
@@ -488,8 +490,10 @@ static int vfio_msi_set_vector_signal(struct vfio_pci_intr_ctx *intr_ctx,if (ret) goto out_put_eventfd_ctx; + ctx->virq = irq; + ctx->producer.token = trigger; - ctx->producer.irq = irq; + ctx->producer.irq = ctx->virq; ret =irq_bypass_register_producer (&ctx->producer);if (unlikely (ret)) { dev_info(&pdev->dev,
Signature · 1 line
2.34.1
RFC PATCH v3 21/26 vfio/pci: Replace backend specific calls with callbacks
In reply to Reinette ChatreThe backend specific code needed to manage the interrupts are isolated into separate functions. With the backend specific code isolated into functions, these functions
can be turned into callbacks for other backends to use.
---
Changes since RFC V2:
- New patch
drivers/vfio/pci/vfio_pci_intrs.c | 17 +++++++++++------ include/linux/vfio_pci_core.h | 15 +++++++++++++++ 2 files changed, 26 insertions(+), 6 deletions(-)
drivers/vfio/pci/vfio_pci_intrs.c +11 -6
@@ -490,7 +490,7 @@ static int vfio_msi_set_vector_signal(struct vfio_pci_intr_ctx *intr_ctx,if (ctx && ctx->trigger) {irq_bypass_unregister_producer (&ctx->producer); -vfio_msi_free_interrupt (intr_ctx, ctx, vector); + intr_ctx->ops->msi_free_interrupt (intr_ctx, ctx, vector);kfree (ctx->name); ctx->name =NULL ;eventfd_ctx_put (ctx->trigger);
@@ -507,7 +507,7 @@ static int vfio_msi_set_vector_signal(struct vfio_pci_intr_ctx *intr_ctx,return -ENOMEM ; } - ctx->name =vfio_msi_device_name (intr_ctx, vector, index); + ctx->name = intr_ctx->ops->msi_device_name (intr_ctx, vector, index);if (!ctx->name)return -ENOMEM ;
@@ -519,7 +519,7 @@ static int vfio_msi_set_vector_signal(struct vfio_pci_intr_ctx *intr_ctx, ctx->trigger = trigger; - ret =vfio_msi_request_interrupt (intr_ctx, ctx, vector, index); + ret = intr_ctx->ops->msi_request_interrupt (intr_ctx, ctx, vector, index);if (ret) goto out_put_eventfd_ctx;
@@ -708,7 +708,7 @@ static int vfio_pci_set_msi_trigger(struct vfio_pci_intr_ctx *intr_ctx,unsigned int i;if (irq_is (intr_ctx, index) && !count && (flags &VFIO_IRQ_SET_DATA_NONE )) { -vfio_msi_disable (intr_ctx, index); + intr_ctx->ops->msi_disable (intr_ctx, index);return 0 ; }
@@ -723,13 +723,13 @@ static int vfio_pci_set_msi_trigger(struct vfio_pci_intr_ctx *intr_ctx,return vfio_msi_set_block (intr_ctx, start, count, fds, index); - ret =vfio_msi_enable (intr_ctx, start + count, index); + ret = intr_ctx->ops->msi_enable (intr_ctx, start + count, index);if (ret)return ret; ret =vfio_msi_set_block (intr_ctx, start, count, fds, index);if (ret) -vfio_msi_disable (intr_ctx, index); + intr_ctx->ops->msi_disable (intr_ctx, index);return ret; }
@@ -872,6 +872,11 @@ static struct vfio_pci_intr_ops vfio_pci_intr_ops = { .set_msix_trigger = vfio_pci_set_msi_trigger, .set_err_trigger = vfio_pci_set_err_trigger, .set_req_trigger = vfio_pci_set_req_trigger, + .msi_enable = vfio_msi_enable, + .msi_disable = vfio_msi_disable, + .msi_request_interrupt = vfio_msi_request_interrupt, + .msi_free_interrupt = vfio_msi_free_interrupt, + .msi_device_name = vfio_msi_device_name, }; void vfio_pci_init_intr_ctx(struct vfio_pci_core_device *vdev,
include/linux/vfio_pci_core.h +15 -0
@@ -69,6 +69,8 @@ struct vfio_pci_intr_ctx {int irq_type; }; +struct vfio_pci_irq_ctx ; +struct vfio_pci_intr_ops { int (*set_intx_mask)(struct vfio_pci_intr_ctx *intr_ctx, unsigned int index, unsigned int start,
@@ -91,6 +93,19 @@ struct vfio_pci_intr_ops {int (*set_req_trigger)(struct vfio_pci_intr_ctx *intr_ctx,unsigned int index,unsigned int start,unsigned int count,uint32_t flags,void *data); +int (*msi_enable)(struct vfio_pci_intr_ctx *intr_ctx,int nvec, +unsigned int index); +void (*msi_disable)(struct vfio_pci_intr_ctx *intr_ctx, +unsigned int index); +int (*msi_request_interrupt)(struct vfio_pci_intr_ctx *intr_ctx, +struct vfio_pci_irq_ctx *ctx, +unsigned int vector, +unsigned int index); +void (*msi_free_interrupt)(struct vfio_pci_intr_ctx *intr_ctx, +struct vfio_pci_irq_ctx *ctx, +unsigned int vector); +char *(*msi_device_name)(struct vfio_pci_intr_ctx *intr_ctx, +unsigned int vector,unsigned int index); };struct vfio_pci_core_device {
Signature · 1 line
2.34.1
RFC PATCH v3 20/26 vfio/pci: Separate frontend and backend code during interrupt enable/disable
In reply to Reinette Chatrevfio_msi_set_vector_signal() contains a mix of generic and backend specific code.
Separate the backend specific code into functions that can be replaced by backend-specific callbacks.
The dev_info() used in error message is replaced by a pr_info() that prints the device name generated by the backend specific code intended to be used during request_irq().
---
Changes since RFC V2:
- New patch
drivers/vfio/pci/vfio_pci_intrs.c | 110 +++++++++++++++++++----------- 1 file changed, 70 insertions(+), 40 deletions(-)
drivers/vfio/pci/vfio_pci_intrs.c +70 -40
@@ -416,28 +416,81 @@ static int vfio_msi_alloc_irq(struct vfio_pci_core_device *vdev,return map.index <0 ? map.index : map.virq; } -static int vfio_msi_set_vector_signal (struct vfio_pci_intr_ctx *intr_ctx, -unsigned int vector,int fd, +static void vfio_msi_free_interrupt (struct vfio_pci_intr_ctx *intr_ctx, +struct vfio_pci_irq_ctx *ctx, +unsigned int vector) +{ +struct vfio_pci_core_device *vdev = intr_ctx->priv; +u16 cmd; + + cmd =vfio_pci_memory_lock_and_enable (vdev); +free_irq (ctx->virq, ctx->trigger); +vfio_pci_memory_unlock_and_restore (vdev, cmd); + ctx->virq =0 ; +/* Interrupt stays allocated, will be freed at MSI-X disable. */ +} + +static int vfio_msi_request_interrupt (struct vfio_pci_intr_ctx *intr_ctx, +struct vfio_pci_irq_ctx *ctx, +unsigned int vector,unsigned int index) +{ +bool msix = (index ==VFIO_PCI_MSIX_IRQ_INDEX ) ? true : false; +struct vfio_pci_core_device *vdev = intr_ctx->priv; +int irq, ret; +u16 cmd; + +/* Interrupt stays allocated, will be freed at MSI-X disable. */ + irq =vfio_msi_alloc_irq (vdev, vector, msix); +if (irq <0 ) +return irq; + +/* +* If the vector was previously allocated, refresh the on-device +* message data before enabling in case it had been cleared or +* corrupted (e.g. due to backdoor resets) since writing. +*/ + cmd =vfio_pci_memory_lock_and_enable (vdev); +if (msix) { +struct msi_msg msg; + +get_cached_msi_msg (irq, &msg); +pci_write_msi_msg (irq, &msg); + } + + ret =request_irq (irq, vfio_msihandler,0 , ctx->name, ctx->trigger); +vfio_pci_memory_unlock_and_restore (vdev, cmd); + + ctx->virq = irq; + +return ret; +} + +static char *vfio_msi_device_name (struct vfio_pci_intr_ctx *intr_ctx, +unsigned int vector, +unsigned int index) {bool msix = (index ==VFIO_PCI_MSIX_IRQ_INDEX ) ? true : false;struct vfio_pci_core_device *vdev = intr_ctx->priv;struct pci_dev *pdev = vdev->pdev; + +return kasprintf (GFP_KERNEL_ACCOUNT ,"vfio-msi%s[%d](%s)" , + msix ?"x" :"" , vector,pci_name (pdev)); +} + +static int vfio_msi_set_vector_signal (struct vfio_pci_intr_ctx *intr_ctx, +unsigned int vector,int fd, +unsigned int index) +{struct vfio_pci_irq_ctx *ctx;struct eventfd_ctx *trigger; -int irq = -EINVAL , ret; -u16 cmd; +int ret; ctx =vfio_irq_ctx_get (intr_ctx, vector);if (ctx && ctx->trigger) {irq_bypass_unregister_producer (&ctx->producer); - irq = ctx->virq; - cmd =vfio_pci_memory_lock_and_enable (vdev); -free_irq (ctx->virq, ctx->trigger); -vfio_pci_memory_unlock_and_restore (vdev, cmd); - ctx->virq =0 ; -/* Interrupt stays allocated, will be freed at MSI-X disable. */ +vfio_msi_free_interrupt (intr_ctx, ctx, vector);kfree (ctx->name); ctx->name =NULL ;eventfd_ctx_put (ctx->trigger);
@@ -447,13 +500,6 @@ static int vfio_msi_set_vector_signal(struct vfio_pci_intr_ctx *intr_ctx,if (fd <0 )return 0 ; -if (irq == -EINVAL ) { -/* Interrupt stays allocated, will be freed at MSI-X disable. */ - irq =vfio_msi_alloc_irq (vdev, vector, msix); -if (irq <0 ) -return irq; - } -/* Per-interrupt context remain allocated. */ if (!ctx) { ctx =vfio_irq_ctx_alloc (intr_ctx, vector);
@@ -461,8 +507,7 @@ static int vfio_msi_set_vector_signal(struct vfio_pci_intr_ctx *intr_ctx,return -ENOMEM ; } - ctx->name =kasprintf (GFP_KERNEL_ACCOUNT ,"vfio-msi%s[%d](%s)" , - msix ?"x" :"" , vector,pci_name (pdev)); + ctx->name =vfio_msi_device_name (intr_ctx, vector, index);if (!ctx->name)return -ENOMEM ;
@@ -472,42 +517,27 @@ static int vfio_msi_set_vector_signal(struct vfio_pci_intr_ctx *intr_ctx, goto out_free_name; } -/* -* If the vector was previously allocated, refresh the on-device -* message data before enabling in case it had been cleared or -* corrupted (e.g. due to backdoor resets) since writing. -*/ - cmd =vfio_pci_memory_lock_and_enable (vdev); -if (msix) { -struct msi_msg msg; - -get_cached_msi_msg (irq, &msg); -pci_write_msi_msg (irq, &msg); - } + ctx->trigger = trigger; - ret =request_irq (irq, vfio_msihandler,0 , ctx->name, trigger); -vfio_pci_memory_unlock_and_restore (vdev, cmd); + ret =vfio_msi_request_interrupt (intr_ctx, ctx, vector, index);if (ret) goto out_put_eventfd_ctx; - ctx->virq = irq; - ctx->producer.token = trigger; ctx->producer.irq = ctx->virq; ret =irq_bypass_register_producer (&ctx->producer);if (unlikely (ret)) { -dev_info (&pdev->dev, -"irq bypass producer (token %p) registration fails: %d\n" , - ctx->producer.token, ret); +pr_info ("%s irq bypass producer (token %p) registration fails: %d\n" , + ctx->name, ctx->producer.token, ret); ctx->producer.token =NULL ; } - ctx->trigger = trigger;return 0 ; out_put_eventfd_ctx: -eventfd_ctx_put (trigger); +eventfd_ctx_put (ctx->trigger); + ctx->trigger =NULL ; out_free_name:kfree (ctx->name); ctx->name =NULL ;
Signature · 1 line
2.34.1
RFC PATCH v3 22/26 vfio/pci: Introduce backend specific context initializer
In reply to Reinette ChatreThe per-interrupt context may contain backend specific data.
Call a backend provided initializer on per-interrupt context creation.
---
Changes since RFC V2:
- New patch
drivers/vfio/pci/vfio_pci_intrs.c | 8 ++++++++ include/linux/vfio_pci_core.h | 2 ++ 2 files changed, 10 insertions(+)
drivers/vfio/pci/vfio_pci_intrs.c +8 -0
@@ -73,6 +73,14 @@ vfio_irq_ctx_alloc(struct vfio_pci_intr_ctx *intr_ctx, unsigned long index)if (!ctx)return NULL ; +if (intr_ctx->ops->init_irq_ctx) { + ret = intr_ctx->ops->init_irq_ctx (intr_ctx, ctx); +if (ret <0 ) { +kfree (ctx); +return NULL ; + } + } + ret =xa_insert (&intr_ctx->ctx, index, ctx,GFP_KERNEL_ACCOUNT );if (ret) {kfree (ctx);
include/linux/vfio_pci_core.h +2 -0
@@ -106,6 +106,8 @@ struct vfio_pci_intr_ops { unsigned int vector);char *(*msi_device_name)(struct vfio_pci_intr_ctx *intr_ctx,unsigned int vector,unsigned int index); +int (*init_irq_ctx)(struct vfio_pci_intr_ctx *intr_ctx, +struct vfio_pci_irq_ctx *ctx); };struct vfio_pci_core_device {
Signature · 1 line
2.34.1
RFC PATCH v3 23/26 vfio/pci: Support emulated interrupts
In reply to Reinette ChatreAccess from a guest to a virtual device may be either 'direct-path', where the guest interacts directly with the underlying hardware, or 'intercepted path' where the virtual device emulates operations.
Support emulated interrupts that can be used to handle 'intercepted path' operations. For example, a virtual device may use 'intercepted path' for configuration. Doing so, configuration requests intercepted by the virtual device driver are handled within the virtual device driver with completion signaled to the guest without interacting with the underlying hardware.
Add vfio_pci_set_emulated() and vfio_pci_send_signal() to the VFIO PCI API. vfio_pci_set_emulated() configures a range of interrupts to be emulated.
Any range of interrupts can be configured as emulated as long as no interrupt has previously been allocated at that vector. The virtual device driver uses vfio_pci_send_signal() to trigger interrupts in the guest.
---
Changes since RFC V2:
- Remove the backend "supports_emulated" flag. All backends now support emulated interrupts.
- Move emulated interrupt enabling from IMS backend to frontend.
drivers/vfio/pci/vfio_pci_intrs.c | 87 ++++++++++++++++++++++++++++++- include/linux/vfio_pci_core.h | 3 ++ 2 files changed, 88 insertions(+), 2 deletions(-)
drivers/vfio/pci/vfio_pci_intrs.c +85 -2
@@ -23,6 +23,7 @@#include "vfio_pci_priv.h" struct vfio_pci_irq_ctx { +bool emulated:1 ;struct eventfd_ctx *trigger;struct virqfd *unmask;struct virqfd *mask;
@@ -497,8 +498,10 @@ static int vfio_msi_set_vector_signal(struct vfio_pci_intr_ctx *intr_ctx, ctx =vfio_irq_ctx_get (intr_ctx, vector);if (ctx && ctx->trigger) { -irq_bypass_unregister_producer (&ctx->producer); - intr_ctx->ops->msi_free_interrupt (intr_ctx, ctx, vector); +if (!ctx->emulated) { +irq_bypass_unregister_producer (&ctx->producer); + intr_ctx->ops->msi_free_interrupt (intr_ctx, ctx, vector); + }kfree (ctx->name); ctx->name =NULL ;eventfd_ctx_put (ctx->trigger);
@@ -527,6 +530,9 @@ static int vfio_msi_set_vector_signal(struct vfio_pci_intr_ctx *intr_ctx, ctx->trigger = trigger; +if (ctx->emulated) +return 0 ; + ret = intr_ctx->ops->msi_request_interrupt (intr_ctx, ctx, vector, index);if (ret) goto out_put_eventfd_ctx;
@@ -902,6 +908,83 @@ void vfio_pci_release_intr_ctx(struct vfio_pci_intr_ctx *intr_ctx) }EXPORT_SYMBOL_GPL (vfio_pci_release_intr_ctx); +/* +* vfio_pci_send_signal() - Send signal to the eventfd. +* @intr_ctx: Interrupt context. +* @vector: Vector for which interrupt will be signaled. +* +* Trigger signal to guest for emulated interrupts. +*/ +void vfio_pci_send_signal (struct vfio_pci_intr_ctx *intr_ctx,unsigned int vector) +{ +struct vfio_pci_irq_ctx *ctx; + +mutex_lock (&intr_ctx->igate); + + ctx =vfio_irq_ctx_get (intr_ctx, vector); + +if (WARN_ON_ONCE (!ctx || !ctx->emulated || !ctx->trigger)) + goto out_unlock; + +eventfd_signal (ctx->trigger,1 ); + +out_unlock: +mutex_unlock (&intr_ctx->igate); +} +EXPORT_SYMBOL_GPL (vfio_pci_send_signal); + +/* +* vfio_pci_set_emulated() - Set range of interrupts that will be emulated. +* @intr_ctx: Interrupt context. +* @start: First emulated interrupt vector. +* @count: Number of emulated interrupts starting from @start. +* +* Emulated interrupts will not be backed by hardware interrupts but +* instead triggered by virtual device driver. +* +* Return: error code on failure (-EBUSY if the vector is not available, +* -ENOMEM on allocation failure), 0 on success. No partial success, on +* success entire range was set as emulated, on failure no interrupt in +* range was set as emulated. +*/ +int vfio_pci_set_emulated (struct vfio_pci_intr_ctx *intr_ctx, +unsigned int start,unsigned int count) +{ +struct vfio_pci_irq_ctx *ctx; +unsigned long i, j; +int ret = -EINVAL ; + +mutex_lock (&intr_ctx->igate); + +for (i = start; i < start + count; i++) { + ctx =kzalloc (sizeof (*ctx),GFP_KERNEL_ACCOUNT ); +if (!ctx) { + ret = -ENOMEM ; + goto out_err; + } + ctx->emulated = true; + ret =xa_insert (&intr_ctx->ctx, i, ctx,GFP_KERNEL_ACCOUNT ); +if (ret) { +kfree (ctx); + goto out_err; + } + } + +mutex_unlock (&intr_ctx->igate); +return 0 ; + +out_err: +for (j = start; j < i; j++) { + ctx =vfio_irq_ctx_get (intr_ctx, j); +vfio_irq_ctx_free (intr_ctx, ctx, j); + } + +mutex_unlock (&intr_ctx->igate); + +return ret; +} +EXPORT_SYMBOL_GPL (vfio_pci_set_emulated); +int vfio_pci_set_irqs_ioctl (struct vfio_pci_intr_ctx *intr_ctx,uint32_t flags,unsigned int index,unsigned int start,unsigned int count,void *data)
include/linux/vfio_pci_core.h +3 -0
@@ -178,6 +178,9 @@ void vfio_pci_release_intr_ctx(struct vfio_pci_intr_ctx *intr_ctx);int vfio_pci_set_irqs_ioctl (struct vfio_pci_intr_ctx *intr_ctx,uint32_t flags,unsigned int index,unsigned int start,unsigned int count,void *data); +void vfio_pci_send_signal (struct vfio_pci_intr_ctx *intr_ctx,unsigned int vector); +int vfio_pci_set_emulated (struct vfio_pci_intr_ctx *intr_ctx, +unsigned int start,unsigned int count);int vfio_pci_core_ioctl_feature (struct vfio_device *device,u32 flags,void __user *arg,size_t argsz); ssize_t vfio_pci_core_read(struct vfio_device *core_vdev, char __user *buf,
Signature · 1 line
2.34.1
RFC PATCH v3 24/26 vfio/pci: Add core IMS support
In reply to Reinette ChatreAdd a new interrupt management backend enabling a guest MSI-X interrupt to be backed by an IMS interrupt on the host.
An IMS interrupt is allocated via pci_ims_alloc_irq() and requires an implementation specific cookie that is opaque to the IMS backend. This can be a PASID, queue ID, pointer etc. During initialization the IMS backend learns which PCI device to operate on (and thus which interrupt domain to allocate from) and what the default cookie should be for any new interrupt allocation.
A virtual device driver starts by initializing the backend using new vfio_pci_ims_init_intr_ctx(), cleanup using new vfio_pci_ims_release_intr_ctx(). Once initialized the virtual device driver can call vfio_pci_set_irqs_ioctl() to handle the VFIO_DEVICE_SET_IRQS ioctl() after it has validated the parameters to be appropriate for the particular device.
To support the IMS backend the core utilities need to be aware which interrupt context it interacts with. New ims_backed_irq enables this and is false for the PCI passthrough backend and true for the IMS backend.
---
Changes since RFC V2:
- Improve changelog.
- Refactored implementation to use new callbacks for interrupt enable/disable and allocate/free to eliminate code duplication. (Kevin)
- Make vfio_pci_ims_intr_ops static.
drivers/vfio/pci/vfio_pci_intrs.c | 178 ++++++++++++++++++++++++++++++ include/linux/vfio_pci_core.h | 7 ++ 2 files changed, 185 insertions(+)
drivers/vfio/pci/vfio_pci_intrs.c +178 -0
@@ -22,6 +22,21 @@#include "vfio_pci_priv.h" +/* +* Interrupt Message Store (IMS) private interrupt context data +* @vdev: Virtual device. Used for name of device in +* request_irq(). +* @pdev: PCI device owning the IMS domain from where +* interrupts are allocated. +* @default_cookie: Default cookie used for IMS interrupts without unique +* cookie. +*/ +struct vfio_pci_ims { +struct vfio_device *vdev; +struct pci_dev *pdev; +union msi_instance_cookie default_cookie; +}; +struct vfio_pci_irq_ctx {bool emulated:1 ;struct eventfd_ctx *trigger;
@@ -31,6 +46,8 @@ struct vfio_pci_irq_ctx {bool masked;struct irq_bypass_producer producer;int virq; +int ims_id; +union msi_instance_cookie icookie; };static bool irq_is (struct vfio_pci_intr_ctx *intr_ctx,int type)
@@ -899,6 +916,7 @@ void vfio_pci_init_intr_ctx(struct vfio_pci_core_device *vdev,_vfio_pci_init_intr_ctx (intr_ctx); intr_ctx->ops = &vfio_pci_intr_ops; intr_ctx->priv = vdev; + intr_ctx->ims_backed_irq = false; }EXPORT_SYMBOL_GPL (vfio_pci_init_intr_ctx);
@@ -985,6 +1003,166 @@ int vfio_pci_set_emulated(struct vfio_pci_intr_ctx *intr_ctx, }EXPORT_SYMBOL_GPL (vfio_pci_set_emulated); +/* Guest MSI-X interrupts backed by IMS host interrupts */ + +/* +* Free the IMS interrupt associated with @ctx. +* +* For an IMS interrupt the interrupt is freed from the underlying +* PCI device's IMS domain. +*/ +static void vfio_pci_ims_irq_free (struct vfio_pci_intr_ctx *intr_ctx, +struct vfio_pci_irq_ctx *ctx) +{ +struct vfio_pci_ims *ims = intr_ctx->priv; +struct msi_map irq_map = {}; + + irq_map.index = ctx->ims_id; + irq_map.virq = ctx->virq; +pci_ims_free_irq (ims->pdev, irq_map); + ctx->ims_id = -EINVAL ; + ctx->virq =0 ; +} + +/* +* Allocate a host IMS interrupt for @ctx. +* +* For an IMS interrupt the interrupt is allocated from the underlying +* PCI device's IMS domain. +*/ +static int vfio_pci_ims_irq_alloc (struct vfio_pci_intr_ctx *intr_ctx, +struct vfio_pci_irq_ctx *ctx) +{ +struct vfio_pci_ims *ims = intr_ctx->priv; +struct msi_map irq_map = {}; + + irq_map =pci_ims_alloc_irq (ims->pdev, &ctx->icookie,NULL ); +if (irq_map.index <0 ) +return irq_map.index; + + ctx->ims_id = irq_map.index; + ctx->virq = irq_map.virq; + +return 0 ; +} + +static void vfio_ims_free_interrupt (struct vfio_pci_intr_ctx *intr_ctx, +struct vfio_pci_irq_ctx *ctx, +unsigned int vector) +{ +free_irq (ctx->virq, ctx->trigger); +vfio_pci_ims_irq_free (intr_ctx, ctx); +} + +static int vfio_ims_request_interrupt (struct vfio_pci_intr_ctx *intr_ctx, +struct vfio_pci_irq_ctx *ctx, +unsigned int vector, +unsigned int index) +{ +int ret; + + ret =vfio_pci_ims_irq_alloc (intr_ctx, ctx); +if (ret <0 ) +return ret; + + ret =request_irq (ctx->virq, vfio_msihandler,0 , ctx->name, + ctx->trigger); +if (ret <0 ) { +vfio_pci_ims_irq_free (intr_ctx, ctx); +return ret; + } + +return 0 ; +} + +static char *vfio_ims_device_name (struct vfio_pci_intr_ctx *intr_ctx, +unsigned int vector, +unsigned int index) +{ +struct vfio_pci_ims *ims = intr_ctx->priv; +struct device *dev = &ims->vdev->device; + +return kasprintf (GFP_KERNEL ,"vfio-ims[%d](%s)" , vector,dev_name (dev)); +} + +static void vfio_ims_disable (struct vfio_pci_intr_ctx *intr_ctx, +unsigned int index) +{ +struct vfio_pci_irq_ctx *ctx; +unsigned long i; + +xa_for_each (&intr_ctx->ctx, i, ctx) +vfio_msi_set_vector_signal (intr_ctx, i,-1 , index); +} + +/* +* The virtual device driver is responsible for enabling IMS by creating +* the IMS domaim from where interrupts will be allocated dynamically. +* IMS thus has to be enabled by the time an ioctl() arrives. +*/ +static int vfio_ims_enable (struct vfio_pci_intr_ctx *intr_ctx,int nvec, +unsigned int index) +{ +return -EINVAL ; +} + +static int vfio_ims_init_irq_ctx (struct vfio_pci_intr_ctx *intr_ctx, +struct vfio_pci_irq_ctx *ctx) +{ +struct vfio_pci_ims *ims = intr_ctx->priv; + + ctx->icookie = ims->default_cookie; + +return 0 ; +} + +static struct vfio_pci_intr_ops vfio_pci_ims_intr_ops = { + .set_msix_trigger = vfio_pci_set_msi_trigger, + .set_req_trigger = vfio_pci_set_req_trigger, + .msi_enable = vfio_ims_enable, + .msi_disable = vfio_ims_disable, + .msi_request_interrupt = vfio_ims_request_interrupt, + .msi_free_interrupt = vfio_ims_free_interrupt, + .msi_device_name = vfio_ims_device_name, + .init_irq_ctx = vfio_ims_init_irq_ctx, +}; + +int vfio_pci_ims_init_intr_ctx (struct vfio_device *vdev, +struct vfio_pci_intr_ctx *intr_ctx, +struct pci_dev *pdev, +union msi_instance_cookie *default_cookie) +{ +struct vfio_pci_ims *ims; + + ims =kzalloc (sizeof (*ims),GFP_KERNEL_ACCOUNT ); +if (!ims) +return -ENOMEM ; + + ims->pdev = pdev; + ims->default_cookie = *default_cookie; + ims->vdev = vdev; + +_vfio_pci_init_intr_ctx (intr_ctx); + + intr_ctx->ops = &vfio_pci_ims_intr_ops; + intr_ctx->priv = ims; + intr_ctx->ims_backed_irq = true; + intr_ctx->irq_type =VFIO_PCI_MSIX_IRQ_INDEX ; + +return 0 ; +} +EXPORT_SYMBOL_GPL (vfio_pci_ims_init_intr_ctx); + +void vfio_pci_ims_release_intr_ctx (struct vfio_pci_intr_ctx *intr_ctx) +{ +struct vfio_pci_ims *ims = intr_ctx->priv; + +_vfio_pci_release_intr_ctx (intr_ctx); +kfree (ims); + intr_ctx->irq_type =VFIO_PCI_NUM_IRQS ; +} +EXPORT_SYMBOL_GPL (vfio_pci_ims_release_intr_ctx); +int vfio_pci_set_irqs_ioctl (struct vfio_pci_intr_ctx *intr_ctx,uint32_t flags,unsigned int index,unsigned int start,unsigned int count,void *data)
include/linux/vfio_pci_core.h +7 -0
@@ -58,6 +58,7 @@ struct vfio_pci_region {* @req_trigger: Eventfd associated with device request notification * @ctx: Per-interrupt context indexed by vector * @irq_type: Type of interrupt from guest perspective +* @ims_backed_irq: Interrupts managed by IMS backend */ struct vfio_pci_intr_ctx {const struct vfio_pci_intr_ops *ops;
@@ -67,6 +68,7 @@ struct vfio_pci_intr_ctx {struct eventfd_ctx *req_trigger;struct xarray ctx;int irq_type; + bool ims_backed_irq:1; };struct vfio_pci_irq_ctx ;
@@ -181,6 +183,11 @@ int vfio_pci_set_irqs_ioctl(struct vfio_pci_intr_ctx *intr_ctx, uint32_t flags,void vfio_pci_send_signal (struct vfio_pci_intr_ctx *intr_ctx,unsigned int vector);int vfio_pci_set_emulated (struct vfio_pci_intr_ctx *intr_ctx,unsigned int start,unsigned int count); +int vfio_pci_ims_init_intr_ctx (struct vfio_device *vdev, +struct vfio_pci_intr_ctx *intr_ctx, +struct pci_dev *pdev, +union msi_instance_cookie *default_cookie); +void vfio_pci_ims_release_intr_ctx (struct vfio_pci_intr_ctx *intr_ctx);int vfio_pci_core_ioctl_feature (struct vfio_device *device,u32 flags,void __user *arg,size_t argsz); ssize_t vfio_pci_core_read(struct vfio_device *core_vdev, char __user *buf,
Signature · 1 line
2.34.1
RFC PATCH v3 25/26 vfio/pci: Add accessor for IMS index
In reply to Reinette ChatreA virtual device driver needs to facilitate translation between the guest's MSI-X interrupt and the backing IMS interrupt with which the physical device is programmed. For example, the guest may need to obtain the IMS index from the virtual device driver that it needs to program into descriptors submitted to the device to ensure that the completion interrupts are generated correctly.
Introduce vfio_pci_ims_hwirq() to the IMS backend as a helper that returns the IMS interrupt index backing a provided MSI-X interrupt index belonging to a guest.
---
No changes since RFC V2.
drivers/vfio/pci/vfio_pci_intrs.c | 25 +++++++++++++++++++++++++ include/linux/vfio_pci_core.h | 1 + 2 files changed, 26 insertions(+)
drivers/vfio/pci/vfio_pci_intrs.c +25 -0
@@ -1163,6 +1163,31 @@ void vfio_pci_ims_release_intr_ctx(struct vfio_pci_intr_ctx *intr_ctx) }EXPORT_SYMBOL_GPL (vfio_pci_ims_release_intr_ctx); +/* +* Return IMS index of IMS interrupt backing MSI-X interrupt @vector +*/ +int vfio_pci_ims_hwirq (struct vfio_pci_intr_ctx *intr_ctx,unsigned int vector) +{ +struct vfio_pci_irq_ctx *ctx; +int id = -EINVAL ; + +mutex_lock (&intr_ctx->igate); + +if (!intr_ctx->ims_backed_irq) + goto out_unlock; + + ctx =vfio_irq_ctx_get (intr_ctx, vector); +if (!ctx || ctx->emulated) + goto out_unlock; + + id = ctx->ims_id; + +out_unlock: +mutex_unlock (&intr_ctx->igate); +return id; +} +EXPORT_SYMBOL_GPL (vfio_pci_ims_hwirq); +int vfio_pci_set_irqs_ioctl (struct vfio_pci_intr_ctx *intr_ctx,uint32_t flags,unsigned int index,unsigned int start,unsigned int count,void *data)
include/linux/vfio_pci_core.h +1 -0
@@ -180,6 +180,7 @@ void vfio_pci_release_intr_ctx(struct vfio_pci_intr_ctx *intr_ctx);int vfio_pci_set_irqs_ioctl (struct vfio_pci_intr_ctx *intr_ctx,uint32_t flags,unsigned int index,unsigned int start,unsigned int count,void *data); +int vfio_pci_ims_hwirq (struct vfio_pci_intr_ctx *intr_ctx,unsigned int vector);void vfio_pci_send_signal (struct vfio_pci_intr_ctx *intr_ctx,unsigned int vector);int vfio_pci_set_emulated (struct vfio_pci_intr_ctx *intr_ctx,unsigned int start,unsigned int count);
Signature · 1 line
2.34.1
RFC PATCH v3 26/26 vfio/pci: Support IMS cookie modification
In reply to Reinette ChatreIMS supports an implementation specific cookie that is associated with each interrupt. By default the IMS interrupt allocation backend will assign a default cookie to a new interrupt instance.
Add support for a virtual device driver to set the interrupt instance specific cookie. For example, the virtual device driver may intercept the guest's MMIO write that configuresa a new PASID for a particular interrupt. Calling vfio_pci_ims_set_cookie() with the new PASID value as IMS cookie enables subsequent interrupts to be allocated with accurate data.
---
No changes since RFC V2.
drivers/vfio/pci/vfio_pci_intrs.c | 53 +++++++++++++++++++++++++++++++ include/linux/vfio_pci_core.h | 3 ++ 2 files changed, 56 insertions(+)
drivers/vfio/pci/vfio_pci_intrs.c +53 -0
@@ -1188,6 +1188,59 @@ int vfio_pci_ims_hwirq(struct vfio_pci_intr_ctx *intr_ctx, unsigned int vector) }EXPORT_SYMBOL_GPL (vfio_pci_ims_hwirq); +/* +* vfio_pci_ims_set_cookie() - Set unique cookie for vector. +* @intr_ctx: Interrupt context. +* @vector: Vector. +* @icookie: New cookie for @vector. +* +* When new IMS interrupt is allocated for @vector it will be +* assigned @icookie. +*/ +int vfio_pci_ims_set_cookie (struct vfio_pci_intr_ctx *intr_ctx, +unsigned int vector, +union msi_instance_cookie *icookie) +{ +struct vfio_pci_irq_ctx *ctx; +int ret = -EINVAL ; + +mutex_lock (&intr_ctx->igate); + +if (!intr_ctx->ims_backed_irq) + goto out_unlock; + + ctx =vfio_irq_ctx_get (intr_ctx, vector); +if (ctx) { +if (WARN_ON_ONCE (ctx->emulated)) { + ret = -EINVAL ; + goto out_unlock; + } + ctx->icookie = *icookie; + ret =0 ; + goto out_unlock; + } + + ctx =kzalloc (sizeof (*ctx),GFP_KERNEL_ACCOUNT ); +if (!ctx) { + ret = -ENOMEM ; + goto out_unlock; + } + + ctx->icookie = *icookie; + ret =xa_insert (&intr_ctx->ctx, vector, ctx,GFP_KERNEL_ACCOUNT ); +if (ret) { +kfree (ctx); + goto out_unlock; + } + + ret =0 ; + +out_unlock: +mutex_unlock (&intr_ctx->igate); +return ret; +} +EXPORT_SYMBOL_GPL (vfio_pci_ims_set_cookie); +int vfio_pci_set_irqs_ioctl (struct vfio_pci_intr_ctx *intr_ctx,uint32_t flags,unsigned int index,unsigned int start,unsigned int count,void *data)
include/linux/vfio_pci_core.h +3 -0
@@ -181,6 +181,9 @@ int vfio_pci_set_irqs_ioctl(struct vfio_pci_intr_ctx *intr_ctx, uint32_t flags, unsigned int index, unsigned int start, unsigned int count, void *data);int vfio_pci_ims_hwirq (struct vfio_pci_intr_ctx *intr_ctx,unsigned int vector); +int vfio_pci_ims_set_cookie (struct vfio_pci_intr_ctx *intr_ctx, +unsigned int vector, +union msi_instance_cookie *icookie);void vfio_pci_send_signal (struct vfio_pci_intr_ctx *intr_ctx,unsigned int vector);int vfio_pci_set_emulated (struct vfio_pci_intr_ctx *intr_ctx,unsigned int start,unsigned int count);
Signature · 1 line
2.34.1
Quoted 12 lines
From: Chatre, Reinette <[email protected]>
Sent: Saturday, October 28, 2023 1:01 AMChanges since RFC V2:
- RFC V2:
https://lore.kernel.org/lkml/[email protected]
/
- Still submiting this as RFC series. I believe that this now matches the expectatations raised during earlier reviews. If you agree this is the right direction then I can drop the RFC prefix on next submission. If you do not agree then please do let me know where I missed expectations.
Overall this matches my expectation. Let's wait for Alex/Jason's thoughts before moving to next-level refinement.
btw as commented to last version, if this is the agreed direction probably next version can be split into two parts: part1 contains the new framework and converts intel vgpu driver to use it, then part2 for ims specific logic.
this way part1 can be verified and merged as a integral part. 😊
Quoted 13 lines · Kevin
Quoted 12 lines
From: Chatre, Reinette <[email protected]>
Sent: Saturday, October 28, 2023 1:01 AMChanges since RFC V2:
- RFC V2:
https://lore.kernel.org/lkml/[email protected]
/
- Still submiting this as RFC series. I believe that this now matches the expectatations raised during earlier reviews. If you agree this is the right direction then I can drop the RFC prefix on next submission. If you do not agree then please do let me know where I missed expectations.
Quoted 2 lines
Overall this matches my expectation. Let's wait for Alex/Jason's thoughts before moving to next-level refinement.
It feels like there's a lot of gratuitous change without any clear purpose. We create an ops structure so that a variant/mdev driver can make use of the vfio-pci-core set_irqs ioctl piecemeal, but then the two entry points that are actually implemented by the ims version are the same as the core version, so the ops appear to be at the wrong level. The use of the priv pointer for the core callbacks looks like it's just trying to justify the existence of the opaque pointer, it should really just be using container_of(). We drill down into various support functions for MSI (ie. enable, disable, request_interrupt, free_interrupt, device name), but INTx is largely ignored, where we haven't even kept is_intx() consistent with the other helpers.
Without an in-tree user of this code, we're just chopping up code for no real purpose. There's no reason that a variant driver requiring IMS couldn't initially implement their own SET_IRQS ioctl. Doing that might lead to a more organic solution where we create interfaces where they're actually needed. The existing mdev sample drivers should also be considered in any schemes to refactor the core code into a generic SET_IRQS helper for devices exposing a vfio-pci API. Thanks,
Alex
Quoted 23 lines
From: Alex Williamson <[email protected]>
Sent: Thursday, November 2, 2023 2:07 AMQuoted 6 lines · Kevin
Quoted 6 lines
From: Chatre, Reinette <[email protected]>
Sent: Saturday, October 28, 2023 1:01 AMChanges since RFC V2:
- RFC V2:https://lore.kernel.org/lkml/[email protected]
Quoted 9 lines
Quoted 6 lines
/
- Still submiting this as RFC series. I believe that this now matches the expectatations raised during earlier reviews. If you agree this is the right direction then I can drop the RFC prefix on next submission. If you do not agree then please do let me know where I missed expectations.Overall this matches my expectation. Let's wait for Alex/Jason's thoughts before moving to next-level refinement.
Quoted 11 lines
It feels like there's a lot of gratuitous change without any clear purpose. We create an ops structure so that a variant/mdev driver can make use of the vfio-pci-core
set_irqsioctl piecemeal, but then the two entry points that are actually implemented by the ims version are the same as the core version, so the ops appear to be at the wrong level. The use of the priv pointer for the core callbacks looks like it's just trying to justify the existence of the opaque pointer, it should really just be usingcontainer_of(). We drill down into various support functions for MSI (ie. enable, disable,request_interrupt,free_interrupt, device name), but INTx is largely ignored, where we haven't even keptis_intx()consistent with the other helpers.
All above are good points. The main interest of this series is to share MSI frontend interface with various backends (PCI MSI/X, IMS, and purely emulated). From this angle the current ops abstraction does sound to sit in a wrong level. But if counting your suggestion to also refactor mdev sample driver (e.g. mtty emulates INTx) then there might be a different outcome.
Quoted 4 lines
Without an in-tree user of this code, we're just chopping up code for no real purpose. There's no reason that a variant driver requiring IMS couldn't initially implement their own
SET_IRQSioctl. Doing that
this is an interesting idea. We haven't seen a real usage which wants such MSI emulation on IMS for variant drivers. but if the code is simple enough to demonstrate the 1st user of IMS it might not be a bad choice. There are additional trap-emulation required in the device MMIO bar (mostly copying MSI permission entry which contains PASID info to the corresponding IMS entry). At a glance that area is 4k-aligned so should be doable.
let's explore more into this option.
Quoted 5 lines
might lead to a more organic solution where we create interfaces where they're actually needed. The existing mdev sample drivers should also be considered in any schemes to refactor the core code into a generic
SET_IRQShelper for devices exposing a vfio-pci API. Thanks,
In this case we'll have mtty to demonstrate an emulated INTx backend and intel vgpu to contain an emulated MSI backend.
and moving forward all drivers with a vfio-pci API should share a same frontend interface.
Quoted 8 lines
From: Tian, Kevin
Sent: Thursday, November 2, 2023 10:52 AMQuoted 4 lines
Without an in-tree user of this code, we're just chopping up code for no real purpose. There's no reason that a variant driver requiring IMS couldn't initially implement their own
SET_IRQSioctl. Doing that
Quoted 8 lines
this is an interesting idea. We haven't seen a real usage which wants such MSI emulation on IMS for variant drivers. but if the code is simple enough to demonstrate the 1st user of IMS it might not be a bad choice. There are additional trap-emulation required in the device MMIO bar (mostly copying MSI permission entry which contains PASID info to the corresponding IMS entry). At a glance that area is 4k-aligned so should be doable.
misread the spec. the MSI-X permission table which provides auxiliary data to MSI-X table is not 4k-aligned. It sits in the 1st 4k page together with many other registers. emulation of them could be simple with a native read/write handler but not sure whether any of them may sit in a hot path to affect perf due to trap...
Quoted 17 lines · Kevin
Quoted 8 lines
From: Tian, Kevin
Sent: Thursday, November 2, 2023 10:52 AMQuoted 4 lines
Without an in-tree user of this code, we're just chopping up code for no real purpose. There's no reason that a variant driver requiring IMS couldn't initially implement their own
SET_IRQSioctl. Doing thatQuoted 8 lines
this is an interesting idea. We haven't seen a real usage which wants such MSI emulation on IMS for variant drivers. but if the code is simple enough to demonstrate the 1st user of IMS it might not be a bad choice. There are additional trap-emulation required in the device MMIO bar (mostly copying MSI permission entry which contains PASID info to the corresponding IMS entry). At a glance that area is 4k-aligned so should be doable.
Quoted 6 lines
misread the spec. the MSI-X permission table which provides auxiliary data to MSI-X table is not 4k-aligned. It sits in the 1st 4k page together with many other registers. emulation of them could be simple with a native read/write handler but not sure whether any of them may sit in a hot path to affect perf due to trap...
I'm not sure if you're referring to a specific device spec or the PCI spec, but the PCI spec has long included an implementation note suggesting alignment of the MSI-X vector table and pba and separation from CSRs, and I see this is now even more strongly worded in the 6.0 spec.
Note though that for QEMU, these are emulated in the VMM and not written through to the device. The result of writes to the vector table in the VMM are translated to vector use/unuse operations, which we see at the kernel level through SET_IRQS ioctl calls. Are you expecting to get PASID information written by the guest through the emulated vector table? That would entail something more than a simple IMS backend to MSI-X frontend. Thanks,
Alex
Quoted 36 lines
From: Alex Williamson <[email protected]>
Sent: Friday, November 3, 2023 5:14 AMQuoted 17 lines · Kevin
Quoted 8 lines
From: Tian, Kevin
Sent: Thursday, November 2, 2023 10:52 AMQuoted 4 lines
Without an in-tree user of this code, we're just chopping up code for no real purpose. There's no reason that a variant driver requiring IMS couldn't initially implement their own
SET_IRQSioctl. Doing thatQuoted 8 lines
this is an interesting idea. We haven't seen a real usage which wants such MSI emulation on IMS for variant drivers. but if the code is simple enough to demonstrate the 1st user of IMS it might not be a bad choice. There are additional trap-emulation required in the device MMIO bar (mostly copying MSI permission entry which contains PASID info to the corresponding IMS entry). At a glance that area is 4k-aligned so should be doable.
Quoted 6 lines
misread the spec. the MSI-X permission table which provides auxiliary data to MSI-X table is not 4k-aligned. It sits in the 1st 4k page together with many other registers. emulation of them could be simple with a native read/write handler but not sure whether any of them may sit in a hot path to affect perf due to trap...
I'm not sure if you're referring to a specific device spec or the PCI spec, but the PCI spec has long included an implementation note suggesting alignment of the MSI-X vector table and pba and separation from CSRs, and I see this is now even more strongly worded in the 6.0 spec.
Quoted 8 lines
Note though that for QEMU, these are emulated in the VMM and not written through to the device. The result of writes to the vector table in the VMM are translated to vector use/unuse operations, which we see at the kernel level through
SET_IRQSioctl calls. Are you expecting to get PASID information written by the guest through the emulated vector table? That would entail something more than a simple IMS backend to MSI-X frontend. Thanks,
I was referring to IDXD device spec. Basically it allows a process to submit a descriptor which contains a completion interrupt handle. The handle is the index of a MSI-X entry or IMS entry allocated by the idxd driver. To mark the association between application and related handles the driver records the PASID of the application in an auxiliary structure for MSI-X (called MSI-X permission table) or directly in the IMS entry. This additional info includes whether an MSI-X/IMS entry has PASID enabled and if yes what is the PASID value to be checked against the descriptor.
As you said virtualizing MSI-X table itself is via SET_IRQS and it's 4k aligned. Then we also need to capture guest updates to the MSI-X permission table and copy the PASID information into the corresponding IMS entry when using the IMS backend. It's MSI-X permission table not 4k aligned then trapping it will affect adjacent registers.
My quick check in idxd spec doesn't reveal an real impact in perf critical path. Most registers are configuration/control registers accessed at driver init time and a few interrupt registers related to errors or administrative purpose.
Quoted 62 lines · Kevin
Quoted 36 lines
From: Alex Williamson <[email protected]>
Sent: Friday, November 3, 2023 5:14 AMQuoted 17 lines · Kevin
Quoted 8 lines
From: Tian, Kevin
Sent: Thursday, November 2, 2023 10:52 AMQuoted 4 lines
Without an in-tree user of this code, we're just chopping up code for no real purpose. There's no reason that a variant driver requiring IMS couldn't initially implement their own
SET_IRQSioctl. Doing thatQuoted 8 lines
this is an interesting idea. We haven't seen a real usage which wants such MSI emulation on IMS for variant drivers. but if the code is simple enough to demonstrate the 1st user of IMS it might not be a bad choice. There are additional trap-emulation required in the device MMIO bar (mostly copying MSI permission entry which contains PASID info to the corresponding IMS entry). At a glance that area is 4k-aligned so should be doable.
Quoted 6 lines
misread the spec. the MSI-X permission table which provides auxiliary data to MSI-X table is not 4k-aligned. It sits in the 1st 4k page together with many other registers. emulation of them could be simple with a native read/write handler but not sure whether any of them may sit in a hot path to affect perf due to trap...
I'm not sure if you're referring to a specific device spec or the PCI spec, but the PCI spec has long included an implementation note suggesting alignment of the MSI-X vector table and pba and separation from CSRs, and I see this is now even more strongly worded in the 6.0 spec.
Quoted 8 lines
Note though that for QEMU, these are emulated in the VMM and not written through to the device. The result of writes to the vector table in the VMM are translated to vector use/unuse operations, which we see at the kernel level through
SET_IRQSioctl calls. Are you expecting to get PASID information written by the guest through the emulated vector table? That would entail something more than a simple IMS backend to MSI-X frontend. Thanks,I was referring to IDXD device spec. Basically it allows a process to submit a descriptor which contains a completion interrupt handle. The handle is the index of a MSI-X entry or IMS entry allocated by the idxd driver. To mark the association between application and related handles the driver records the PASID of the application in an auxiliary structure for MSI-X (called MSI-X permission table) or directly in the IMS entry. This additional info includes whether an MSI-X/IMS entry has PASID enabled and if yes what is the PASID value to be checked against the descriptor.
As you said virtualizing MSI-X table itself is via
SET_IRQSand it's 4k aligned. Then we also need to capture guest updates to the MSI-X permission table and copy the PASID information into the corresponding IMS entry when using the IMS backend. It's MSI-X permission table not 4k aligned then trapping it will affect adjacent registers.
Quoted 4 lines
My quick check in idxd spec doesn't reveal an real impact in perf critical path. Most registers are configuration/control registers accessed at driver init time and a few interrupt registers related to errors or administrative purpose.
Right, it looks like you'll need to trap writes to the MSI-X Permissions Table via a sparse mmap capability to avoid assumptions whether it lives on the same page as the MSI-X vector table or PBA. Ideally the hardware folks have considered this to avoid any conflict with latency sensitive registers.
The variant driver would use this for collecting the meta data relative to the IMS interrupt, but this is all tangential to whether we preemptively slice up vfio-pci-core's SET_IRQS ioctl or the iDXD driver implements its own.
And just to be clear, I don't expect the iDXD variant driver to go to extraordinary lengths to duplicate the core ioctl, we can certainly refactor and export things where it makes sense, but I think it likely makes more sense for the variant driver to implement the shell of the ioctl rather than trying to multiplex the entire core ioctl with an ops structure that's so intimately tied to the core implementation and focused only on the MSI-X code paths. Thanks,
Alex
Quoted 81 lines
From: Alex Williamson <[email protected]>
Sent: Friday, November 3, 2023 11:51 PMQuoted 6 lines · Kevin
Quoted 6 lines
From: Alex Williamson <[email protected]>
Sent: Friday, November 3, 2023 5:14 AMOn Thu, 2 Nov 2023 03:14:09 +0000
"Tian, Kevin" <[email protected]> wrote:Quoted 6 lines
Quoted 6 lines
Quoted 6 lines
Quoted 6 lines
From: Tian, Kevin
Sent: Thursday, November 2, 2023 10:52 AMQuoted 3 lines
Without an in-tree user of this code, we're just chopping up code for no real purpose. There's no reason that a variant driver requiring
IMS
Quoted 7 lines
Quoted 7 lines
Quoted 7 lines
Quoted 7 lines
Quoted 1 line
couldn't initially implement their own
SET_IRQSioctl. Doing thatthis is an interesting idea. We haven't seen a real usage which wants such MSI emulation on IMS for variant drivers. but if the code is simple enough to demonstrate the 1st user of IMS it might not be a bad choice. There are additional trap-emulation required in the device MMIO bar (mostly copying MSI permission entry which
contains
Quoted 43 lines
Quoted 17 lines
Quoted 10 lines
Quoted 3 lines
PASID info to the corresponding IMS entry). At a glance that area is 4k-aligned so should be doable.
misread the spec. the MSI-X permission table which provides auxiliary data to MSI-X table is not 4k-aligned. It sits in the 1st 4k page together with many other registers. emulation of them could be simple with a native read/write handler but not sure whether any of them may sit in a hot path to affect perf due to trap...
I'm not sure if you're referring to a specific device spec or the PCI spec, but the PCI spec has long included an implementation note suggesting alignment of the MSI-X vector table and pba and separation from CSRs, and I see this is now even more strongly worded in the 6.0 spec.
Quoted 8 lines
Note though that for QEMU, these are emulated in the VMM and not written through to the device. The result of writes to the vector table in the VMM are translated to vector use/unuse operations, which we see at the kernel level through
SET_IRQSioctl calls. Are you expecting to get PASID information written by the guest through the emulated vector table? That would entail something more than a simple IMS backend to MSI-X frontend. Thanks,I was referring to IDXD device spec. Basically it allows a process to submit a descriptor which contains a completion interrupt handle. The handle is the index of a MSI-X entry or IMS entry allocated by the idxd driver. To mark the association between application and related handles the driver records the PASID of the application in an auxiliary structure for MSI-X (called MSI-X permission table) or directly in the IMS entry. This additional info includes whether an MSI-X/IMS entry has PASID enabled and if yes what is the PASID value to be checked against the descriptor.
As you said virtualizing MSI-X table itself is via
SET_IRQSand it's 4k aligned. Then we also need to capture guest updates to the MSI-X permission table and copy the PASID information into the corresponding IMS entry when using the IMS backend. It's MSI-X permission table not 4k aligned then trapping it will affect adjacent registers.Quoted 4 lines
My quick check in idxd spec doesn't reveal an real impact in perf critical path. Most registers are configuration/control registers accessed at driver init time and a few interrupt registers related to errors or administrative purpose.
Right, it looks like you'll need to trap writes to the MSI-X Permissions Table via a sparse mmap capability to avoid assumptions whether it lives on the same page as the MSI-X vector table or PBA. Ideally the hardware folks have considered this to avoid any conflict with latency sensitive registers.
Quoted 4 lines
The variant driver would use this for collecting the meta data relative to the IMS interrupt, but this is all tangential to whether we preemptively slice up vfio-pci-core's
SET_IRQSioctl or the iDXD driver implements its own.
Agree
Quoted 9 lines
And just to be clear, I don't expect the iDXD variant driver to go to extraordinary lengths to duplicate the core ioctl, we can certainly refactor and export things where it makes sense, but I think it likely makes more sense for the variant driver to implement the shell of the ioctl rather than trying to multiplex the entire core ioctl with an ops structure that's so intimately tied to the core implementation and focused only on the MSI-X code paths. Thanks,
btw I'll let Reinette to decide whether she wants to implement such a variant driver or waits until idxd siov driver is ready to demonstrate the usage. One concern with the variant driver approach is lacking of a real-world usage (e.g. what IMS brings when guest only wants MSI-X on an assigned PF). Though it may provide a shorter path to enable the IMS backend support, also comes with the long-term maintenance burden.
Hi Alex and Kevin,
Quoted 98 lines · Kevin
Quoted 81 lines
From: Alex Williamson <[email protected]>
Sent: Friday, November 3, 2023 11:51 PMQuoted 6 lines · Kevin
Quoted 6 lines
From: Alex Williamson <[email protected]>
Sent: Friday, November 3, 2023 5:14 AMOn Thu, 2 Nov 2023 03:14:09 +0000
"Tian, Kevin" <[email protected]> wrote:Quoted 6 lines
Quoted 6 lines
Quoted 6 lines
Quoted 6 lines
From: Tian, Kevin
Sent: Thursday, November 2, 2023 10:52 AMQuoted 3 lines
Without an in-tree user of this code, we're just chopping up code for no real purpose. There's no reason that a variant driver requiring
IMS
Quoted 7 lines
Quoted 7 lines
Quoted 7 lines
Quoted 7 lines
Quoted 1 line
couldn't initially implement their own
SET_IRQSioctl. Doing thatthis is an interesting idea. We haven't seen a real usage which wants such MSI emulation on IMS for variant drivers. but if the code is simple enough to demonstrate the 1st user of IMS it might not be a bad choice. There are additional trap-emulation required in the device MMIO bar (mostly copying MSI permission entry which
contains
Quoted 43 lines
Quoted 17 lines
Quoted 10 lines
Quoted 3 lines
PASID info to the corresponding IMS entry). At a glance that area is 4k-aligned so should be doable.
misread the spec. the MSI-X permission table which provides auxiliary data to MSI-X table is not 4k-aligned. It sits in the 1st 4k page together with many other registers. emulation of them could be simple with a native read/write handler but not sure whether any of them may sit in a hot path to affect perf due to trap...
I'm not sure if you're referring to a specific device spec or the PCI spec, but the PCI spec has long included an implementation note suggesting alignment of the MSI-X vector table and pba and separation from CSRs, and I see this is now even more strongly worded in the 6.0 spec.
Quoted 8 lines
Note though that for QEMU, these are emulated in the VMM and not written through to the device. The result of writes to the vector table in the VMM are translated to vector use/unuse operations, which we see at the kernel level through
SET_IRQSioctl calls. Are you expecting to get PASID information written by the guest through the emulated vector table? That would entail something more than a simple IMS backend to MSI-X frontend. Thanks,I was referring to IDXD device spec. Basically it allows a process to submit a descriptor which contains a completion interrupt handle. The handle is the index of a MSI-X entry or IMS entry allocated by the idxd driver. To mark the association between application and related handles the driver records the PASID of the application in an auxiliary structure for MSI-X (called MSI-X permission table) or directly in the IMS entry. This additional info includes whether an MSI-X/IMS entry has PASID enabled and if yes what is the PASID value to be checked against the descriptor.
As you said virtualizing MSI-X table itself is via
SET_IRQSand it's 4k aligned. Then we also need to capture guest updates to the MSI-X permission table and copy the PASID information into the corresponding IMS entry when using the IMS backend. It's MSI-X permission table not 4k aligned then trapping it will affect adjacent registers.Quoted 4 lines
My quick check in idxd spec doesn't reveal an real impact in perf critical path. Most registers are configuration/control registers accessed at driver init time and a few interrupt registers related to errors or administrative purpose.
Right, it looks like you'll need to trap writes to the MSI-X Permissions Table via a sparse mmap capability to avoid assumptions whether it lives on the same page as the MSI-X vector table or PBA. Ideally the hardware folks have considered this to avoid any conflict with latency sensitive registers.
Quoted 4 lines
The variant driver would use this for collecting the meta data relative to the IMS interrupt, but this is all tangential to whether we preemptively slice up vfio-pci-core's
SET_IRQSioctl or the iDXD driver implements its own.Agree
Quoted 9 lines
And just to be clear, I don't expect the iDXD variant driver to go to extraordinary lengths to duplicate the core ioctl, we can certainly refactor and export things where it makes sense, but I think it likely makes more sense for the variant driver to implement the shell of the ioctl rather than trying to multiplex the entire core ioctl with an ops structure that's so intimately tied to the core implementation and focused only on the MSI-X code paths. Thanks,
Quoted 7 lines
btw I'll let Reinette to decide whether she wants to implement such a variant driver or waits until idxd siov driver is ready to demonstrate the usage. One concern with the variant driver approach is lacking of a real-world usage (e.g. what IMS brings when guest only wants MSI-X on an assigned PF). Though it may provide a shorter path to enable the IMS backend support, also comes with the long-term maintenance burden.
Thank you very much for your guidance and advice.
I'd be happy to implement what is required for this work. Unfortunately I am not confident that I understand what is meant with "variant driver".
I initially understood "variant driver" to mean the planned IDXD virtual device driver (the "IDXD VDCM" driver) that will assign IDXD resources to guests with varying granularity and back the guest MSI-X interrupts of these virtual devices with IMS interrupts on the host. From Kevin I understand "variant driver" is a new sample driver for an IDXD assigned PF, backing guest MSI-X interrupts with IMS interrupts on the host.
The IDXD VDCM driver is in progress. If a new variant driver needs to be created then I still need to do some research in how to accomplish it. Even so, it is not clear to me who the users of this new driver would be. If the requirement is to demonstrate this VFIO IMS usage then we could perhaps wait until the IDXD VDCM driver is ready and thus not have to deal with additional maintenance burden.
In the mean time there are items that I do understand better and will work on right away:
- Do not have ops span the SET_IRQS ioctl()
- Use container_of() instead of opaque pointer
- Do not ignore INTx, consider the mdev sample driver when refactoring this code.
- Consider the Intel vgpu driver as user of new emulated interrupt interface.
Reinette
Quoted 122 lines · On Tue, 7 Nov 2023 11:48:28 -0800 Reinette Chatre <[email protected]>…
Hi Alex and Kevin,
Quoted 98 lines · Kevin
Quoted 81 lines
From: Alex Williamson <[email protected]>
Sent: Friday, November 3, 2023 11:51 PMQuoted 6 lines · Kevin
Quoted 6 lines
From: Alex Williamson <[email protected]>
Sent: Friday, November 3, 2023 5:14 AMOn Thu, 2 Nov 2023 03:14:09 +0000
"Tian, Kevin" <[email protected]> wrote:Quoted 6 lines
Quoted 6 lines
Quoted 6 lines
Quoted 6 lines
From: Tian, Kevin
Sent: Thursday, November 2, 2023 10:52 AMQuoted 3 lines
Without an in-tree user of this code, we're just chopping up code for no real purpose. There's no reason that a variant driver requiring
IMS
Quoted 7 lines
Quoted 7 lines
Quoted 7 lines
Quoted 7 lines
Quoted 1 line
couldn't initially implement their own
SET_IRQSioctl. Doing thatthis is an interesting idea. We haven't seen a real usage which wants such MSI emulation on IMS for variant drivers. but if the code is simple enough to demonstrate the 1st user of IMS it might not be a bad choice. There are additional trap-emulation required in the device MMIO bar (mostly copying MSI permission entry which
contains
Quoted 43 lines
Quoted 17 lines
Quoted 10 lines
Quoted 3 lines
PASID info to the corresponding IMS entry). At a glance that area is 4k-aligned so should be doable.
misread the spec. the MSI-X permission table which provides auxiliary data to MSI-X table is not 4k-aligned. It sits in the 1st 4k page together with many other registers. emulation of them could be simple with a native read/write handler but not sure whether any of them may sit in a hot path to affect perf due to trap...
I'm not sure if you're referring to a specific device spec or the PCI spec, but the PCI spec has long included an implementation note suggesting alignment of the MSI-X vector table and pba and separation from CSRs, and I see this is now even more strongly worded in the 6.0 spec.
Quoted 8 lines
Note though that for QEMU, these are emulated in the VMM and not written through to the device. The result of writes to the vector table in the VMM are translated to vector use/unuse operations, which we see at the kernel level through
SET_IRQSioctl calls. Are you expecting to get PASID information written by the guest through the emulated vector table? That would entail something more than a simple IMS backend to MSI-X frontend. Thanks,I was referring to IDXD device spec. Basically it allows a process to submit a descriptor which contains a completion interrupt handle. The handle is the index of a MSI-X entry or IMS entry allocated by the idxd driver. To mark the association between application and related handles the driver records the PASID of the application in an auxiliary structure for MSI-X (called MSI-X permission table) or directly in the IMS entry. This additional info includes whether an MSI-X/IMS entry has PASID enabled and if yes what is the PASID value to be checked against the descriptor.
As you said virtualizing MSI-X table itself is via
SET_IRQSand it's 4k aligned. Then we also need to capture guest updates to the MSI-X permission table and copy the PASID information into the corresponding IMS entry when using the IMS backend. It's MSI-X permission table not 4k aligned then trapping it will affect adjacent registers.Quoted 4 lines
My quick check in idxd spec doesn't reveal an real impact in perf critical path. Most registers are configuration/control registers accessed at driver init time and a few interrupt registers related to errors or administrative purpose.
Right, it looks like you'll need to trap writes to the MSI-X Permissions Table via a sparse mmap capability to avoid assumptions whether it lives on the same page as the MSI-X vector table or PBA. Ideally the hardware folks have considered this to avoid any conflict with latency sensitive registers.
Quoted 4 lines
The variant driver would use this for collecting the meta data relative to the IMS interrupt, but this is all tangential to whether we preemptively slice up vfio-pci-core's
SET_IRQSioctl or the iDXD driver implements its own.Agree
Quoted 9 lines
And just to be clear, I don't expect the iDXD variant driver to go to extraordinary lengths to duplicate the core ioctl, we can certainly refactor and export things where it makes sense, but I think it likely makes more sense for the variant driver to implement the shell of the ioctl rather than trying to multiplex the entire core ioctl with an ops structure that's so intimately tied to the core implementation and focused only on the MSI-X code paths. Thanks,
Quoted 7 lines
btw I'll let Reinette to decide whether she wants to implement such a variant driver or waits until idxd siov driver is ready to demonstrate the usage. One concern with the variant driver approach is lacking of a real-world usage (e.g. what IMS brings when guest only wants MSI-X on an assigned PF). Though it may provide a shorter path to enable the IMS backend support, also comes with the long-term maintenance burden.
Thank you very much for your guidance and advice.
I'd be happy to implement what is required for this work. Unfortunately I am not confident that I understand what is meant with "variant driver".
I initially understood "variant driver" to mean the planned IDXD virtual device driver (the "IDXD VDCM" driver) that will assign IDXD resources to guests with varying granularity and back the guest MSI-X interrupts of these virtual devices with IMS interrupts on the host. From Kevin I understand "variant driver" is a new sample driver for an IDXD assigned PF, backing guest MSI-X interrupts with IMS interrupts on the host.
Quoted 6 lines
The IDXD VDCM driver is in progress. If a new variant driver needs to be created then I still need to do some research in how to accomplish it. Even so, it is not clear to me who the users of this new driver would be. If the requirement is to demonstrate this VFIO IMS usage then we could perhaps wait until the IDXD VDCM driver is ready and thus not have to deal with additional maintenance burden.
A vfio-pci variant driver is specifically a driver that leverages portions of vfio-pci-core for implementing vfio_device_ops and binds to a PCI device. It might actually be the wrong term here, but I jumped to that since the series tries to generalize portions of one of the vfio-pci-core code paths. You might very well be intending to use this with something more like an mdev driver, which is fine.
That also sort of illustrates the point though that this series is taking a pretty broad approach to slicing up vfio-pci-core's SET_IRQS ioctl code path, enabling support for IMS backed interrupts, but in effect complicating the whole thing without any actual consumer to justify the complication. Meanwhile I think the goal is to reduce complication to a driver that doesn't exist yet. So it currently seems like a poor trade-off.
This driver that doesn't exist yet could implement its own SET_IRQS ioctl that backs MSI-X with IMS as a starting point. Presumably we expect multiple drivers to require this behavior, so common code makes sense, but the rest of us in the community can't really evaluate how much it makes sense to slice the common code without seeing that implementation and how it might leverage, if not directly use, the existing core code.
The sample drivers come into play in that if we were to make the vfio-pci-core SET_IRQS path usable in this generic way, then any driver implementing SET_IRQS for a PCI device should be able to take advantage of it, including those that already exist in samples/vfio-mdev/. I don't think anyone is requesting an iDXD sample driver, the real driver should be sufficient.
Quoted 8 lines
In the mean time there are items that I do understand better and will work on right away:
- Do not have ops span theSET_IRQSioctl()
- Usecontainer_of()instead of opaque pointer
- Do not ignore INTx, consider the mdev sample driver when refactoring this code.
- Consider the Intel vgpu driver as user of new emulated interrupt interface.
I think looking at kvmgt and the existing sample drivers to find commonality that actually provides simplification would be a good start, but I don't anticipate implementing IMS backed MSI-X in common code unless we're at least imminently able to make use of it. Thanks,
Alex
Quoted 6 lines · Alex Williamson
A vfio-pci variant driver is specifically a driver that leverages portions of vfio-pci-core for implementing
vfio_device_opsand binds to a PCI device. It might actually be the wrong term here, but I jumped to that since the series tries to generalize portions of one of the vfio-pci-core code paths. You might very well be intending to use this with something more like an mdev driver, which is fine.
IDXD will be a SIOV device and we need to have a serious talk about how SIOV device lifecycle will work..
Quoted 7 lines
That also sort of illustrates the point though that this series is taking a pretty broad approach to slicing up vfio-pci-core's
SET_IRQSioctl code path, enabling support for IMS backed interrupts, but in effect complicating the whole thing without any actual consumer to justify the complication. Meanwhile I think the goal is to reduce complication to a driver that doesn't exist yet. So it currently seems like a poor trade-off.
I think we need to see some draft of the IDXD driver to really understand this
Quoted 7 lines
This driver that doesn't exist yet could implement its own
SET_IRQSioctl that backs MSI-X with IMS as a starting point. Presumably we expect multiple drivers to require this behavior, so common code makes sense, but the rest of us in the community can't really evaluate how much it makes sense to slice the common code without seeing that implementation and how it might leverage, if not directly use, the existing core code.
I've been seeing a general interest in taking something that is not MSI-X (eg "IMS" for IDXD) and converting it into MSI-X for the vPCI function. I think this will be a durable need in this space.
Ideally it will be overtaken by simply teaching the guest, vfio and the hypervisor interrupt logic how to directly generate interrupts with a guest controlled addr/data pair without requiring MSI-X trapping. That is the fundamental reason why this has to be done this convoluted way.
Jason
Quoted 17 lines
From: Jason Gunthorpe <[email protected]>
Sent: Wednesday, November 8, 2023 10:50 AMQuoted 7 lines · Alex Williamson
This driver that doesn't exist yet could implement its own
SET_IRQSioctl that backs MSI-X with IMS as a starting point. Presumably we expect multiple drivers to require this behavior, so common code makes sense, but the rest of us in the community can't really evaluate how much it makes sense to slice the common code without seeing that implementation and how it might leverage, if not directly use, the existing core code.I've been seeing a general interest in taking something that is not MSI-X (eg "IMS" for IDXD) and converting it into MSI-X for the vPCI function. I think this will be a durable need in this space.
Quoted 6 lines
Ideally it will be overtaken by simply teaching the guest, vfio and the hypervisor interrupt logic how to directly generate interrupts with a guest controlled addr/data pair without requiring MSI-X trapping. That is the fundamental reason why this has to be done this convoluted way.
Even with that a legacy guest which doesn't support such enlightened way still needs this convoluted way. 😊
and for SIOV anyway the trap cannot be eliminated given the interrupt storage is shared by all vdev's.
Hi Alex,
Quoted 8 lines · Alex Williamson
That also sort of illustrates the point though that this series is taking a pretty broad approach to slicing up vfio-pci-core's
SET_IRQSioctl code path, enabling support for IMS backed interrupts, but in effect complicating the whole thing without any actual consumer to justify the complication. Meanwhile I think the goal is to reduce complication to a driver that doesn't exist yet. So it currently seems like a poor trade-off.
Quoted 7 lines
This driver that doesn't exist yet could implement its own
SET_IRQSioctl that backs MSI-X with IMS as a starting point. Presumably we expect multiple drivers to require this behavior, so common code makes sense, but the rest of us in the community can't really evaluate how much it makes sense to slice the common code without seeing that implementation and how it might leverage, if not directly use, the existing core code.
I understand. I'm hearing the same from you and Jason. I plan to work on addressing your feedback but will only share it when it can be accompanied by a draft of the IDXD VDCM driver. Please let me know if you prefer a different approach.
Reinette