As the vIOMMU infrastructure series part-3, this introduces a new vEVENTQ object. The existing FAULT object provides a nice notification pathway to the user space with a queue already, so let vEVENTQ reuse that.
Mimicing the HWPT structure, add a common EVENTQ structure to support its derivatives: IOMMUFD_OBJ_FAULT (existing) and IOMMUFD_OBJ_VEVENTQ (new).
An IOMMUFD_CMD_VEVENTQ_ALLOC is introduced to allocate vEVENTQ object for vIOMMUs. One vIOMMU can have multiple vEVENTQs in different types but can not support multiple vEVENTQs in the same type.
The forwarding part is fairly simple but might need to replace a physical device ID with a virtual device ID in a driver-level event data structure. So, this also adds some helpers for drivers to use.
As usual, this series comes with the selftest coverage for this new ioctl and with a real world use case in the ARM SMMUv3 driver.
This is on Github: https://github.com/nicolinc/iommufd/commits/iommufd_veventq-v8 Paring QEMU branch for testing: https://github.com/nicolinc/qemu/commits/wip/for_iommufd_veventq-v8
Changelog v8 * Add Reviewed-by from Jason and Pranjal * Fix errno returned in arm_smmu_handle_event() * Validate domain->type outside of arm_smmu_attach_prepare_vmaster() * Drop unnecessary vmaster comparison in arm_smmu_attach_commit_vmaster() v7 https://lore.kernel.org/all/cover.1740238876.git.nicolinc@nvidia.com/ * Rebase on Jason's for-next tree for latest fault.c * Add Reviewed-by * Update commit logs * Add __reserved field sanity * Skip kfree() on the static header * Replace "bool on_list" with list_is_last() * Use u32 for flags in iommufd_vevent_header * Drop casting in iommufd_viommu_get_vdev_id() * Update the bounding logic to veventq->sequence * Add missing cpu_to_le64() around STRTAB_STE_1_MEV * Reuse veventq->common.lock to fence sequence and num_events * Rename overflow to lost_events and log it in upon kmalloc failure * Correct the error handling part in iommufd_veventq_deliver_fetch() * Add an arm_smmu_clear_vmaster() to simplify identity/blocked domain attach ops * Add additional four event records to forward to user space VM, and update the uAPI doc * Reuse the existing smmu->streams_mutex lock to fence master->vmaster pointer, instead of adding a new rwsem v6 https://lore.kernel.org/all/cover.1737754129.git.nicolinc@nvidia.com/ * Drop supports_veventq viommu op * Split bug/cosmetics fixes out of the series * Drop the blocking mutex around copy_to_user() * Add veventq_depth in uAPI to limit vEVENTQ size * Revise the documentation for a clear description * Fix sparse warnings in arm_vmaster_report_event() * Rework iommufd_viommu_get_vdev_id() to return -ENOENT v.s. 0 * Allow Abort/Bypass STEs to allocate vEVENTQ and set STE.MEV for DoS mitigations v5 https://lore.kernel.org/all/cover.1736237481.git.nicolinc@nvidia.com/ * Add Reviewed-by from Baolu * Reorder the OBJ list as well * Fix alphabetical order after renaming in v4 * Add supports_veventq viommu op for vEVENTQ type validation v4 https://lore.kernel.org/all/cover.1735933254.git.nicolinc@nvidia.com/ * Rename "vIRQ" to "vEVENTQ" * Use flexible array in struct iommufd_vevent * Add the new ioctl command to union ucmd_buffer * Fix the alphabetical order in union ucmd_buffer too * Rename _TYPE_NONE to _TYPE_DEFAULT aligning with vIOMMU naming v3 https://lore.kernel.org/all/cover.1734477608.git.nicolinc@nvidia.com/ * Rebase on Will's for-joerg/arm-smmu/updates for arm_smmu_event series * Add "Reviewed-by" lines from Kevin * Fix typos in comments, kdocs, and jump tags * Add a patch to sort struct iommufd_ioctl_op * Update iommufd's userpsace-api documentation * Update uAPI kdoc to quote SMMUv3 offical spec * Drop the unused workqueue in struct iommufd_virq * Drop might_sleep() in iommufd_viommu_report_irq() helper * Add missing "break" in iommufd_viommu_get_vdev_id() helper * Shrink the scope of the vmaster's read lock in SMMUv3 driver * Pass in two arguments to iommufd_eventq_virq_handler() helper * Move "!ops || !ops->read" validation into iommufd_eventq_init() * Move "fault->ictx = ictx" closer to iommufd_ctx_get(fault->ictx) * Update commit message for arm_smmu_attach_prepare/commit_vmaster() * Keep "iommufd_fault" as-is and rename "iommufd_eventq_virq" to just "iommufd_virq" v2 https://lore.kernel.org/all/cover.1733263737.git.nicolinc@nvidia.com/ * Rebase on v6.13-rc1 * Add IOPF and vIRQ in iommufd.rst (userspace-api) * Add a proper locking in iommufd_event_virq_destroy * Add iommufd_event_virq_abort with a lockdep_assert_held * Rename "EVENT_*" to "EVENTQ_*" to describe the objects better * Reorganize flows in iommufd_eventq_virq_alloc for abort() to work * Adde struct arm_smmu_vmaster to store vSID upon attaching to a nested domain, calling a newly added iommufd_viommu_get_vdev_id helper * Adde an arm_vmaster_report_event helper in arm-smmu-v3-iommufd file to simplify the routine in arm_smmu_handle_evt() of the main driver v1 https://lore.kernel.org/all/cover.1724777091.git.nicolinc@nvidia.com/
Thanks! Nicolin
Nicolin Chen (14): iommufd/fault: Move two fault functions out of the header iommufd/fault: Add an iommufd_fault_init() helper iommufd: Abstract an iommufd_eventq from iommufd_fault iommufd: Rename fault.c to eventq.c iommufd: Add IOMMUFD_OBJ_VEVENTQ and IOMMUFD_CMD_VEVENTQ_ALLOC iommufd/viommu: Add iommufd_viommu_get_vdev_id helper iommufd/viommu: Add iommufd_viommu_report_event helper iommufd/selftest: Require vdev_id when attaching to a nested domain iommufd/selftest: Add IOMMU_TEST_OP_TRIGGER_VEVENT for vEVENTQ coverage iommufd/selftest: Add IOMMU_VEVENTQ_ALLOC test coverage Documentation: userspace-api: iommufd: Update FAULT and VEVENTQ iommu/arm-smmu-v3: Introduce struct arm_smmu_vmaster iommu/arm-smmu-v3: Report events that belong to devices attached to vIOMMU iommu/arm-smmu-v3: Set MEV bit in nested STE for DoS mitigations
drivers/iommu/iommufd/Makefile | 2 +- drivers/iommu/arm/arm-smmu-v3/arm-smmu-v3.h | 36 ++ drivers/iommu/iommufd/iommufd_private.h | 135 +++- drivers/iommu/iommufd/iommufd_test.h | 10 + include/linux/iommufd.h | 23 + include/uapi/linux/iommufd.h | 105 +++ tools/testing/selftests/iommu/iommufd_utils.h | 115 ++++ .../arm/arm-smmu-v3/arm-smmu-v3-iommufd.c | 64 ++ drivers/iommu/arm/arm-smmu-v3/arm-smmu-v3.c | 82 ++- drivers/iommu/iommufd/driver.c | 72 +++ drivers/iommu/iommufd/eventq.c | 597 ++++++++++++++++++ drivers/iommu/iommufd/fault.c | 342 ---------- drivers/iommu/iommufd/hw_pagetable.c | 6 +- drivers/iommu/iommufd/main.c | 7 + drivers/iommu/iommufd/selftest.c | 54 ++ drivers/iommu/iommufd/viommu.c | 2 + tools/testing/selftests/iommu/iommufd.c | 36 ++ .../selftests/iommu/iommufd_fail_nth.c | 7 + Documentation/userspace-api/iommufd.rst | 17 + 19 files changed, 1304 insertions(+), 408 deletions(-) create mode 100644 drivers/iommu/iommufd/eventq.c delete mode 100644 drivers/iommu/iommufd/fault.c
base-commit: 598749522d4254afb33b8a6c1bea614a95896868
There is no need to keep them in the header. The vEVENTQ version of these two functions will turn out to be a different implementation and will not share with this fault version. Thus, move them out of the header.
Reviewed-by: Jason Gunthorpe jgg@nvidia.com Reviewed-by: Kevin Tian kevin.tian@intel.com Signed-off-by: Nicolin Chen nicolinc@nvidia.com --- drivers/iommu/iommufd/iommufd_private.h | 25 ------------------------- drivers/iommu/iommufd/fault.c | 25 +++++++++++++++++++++++++ 2 files changed, 25 insertions(+), 25 deletions(-)
diff --git a/drivers/iommu/iommufd/iommufd_private.h b/drivers/iommu/iommufd/iommufd_private.h index 246297452a44..1c58f5fe17b4 100644 --- a/drivers/iommu/iommufd/iommufd_private.h +++ b/drivers/iommu/iommufd/iommufd_private.h @@ -472,31 +472,6 @@ struct iommufd_fault { struct wait_queue_head wait_queue; };
-/* Fetch the first node out of the fault->deliver list */ -static inline struct iopf_group * -iommufd_fault_deliver_fetch(struct iommufd_fault *fault) -{ - struct list_head *list = &fault->deliver; - struct iopf_group *group = NULL; - - spin_lock(&fault->lock); - if (!list_empty(list)) { - group = list_first_entry(list, struct iopf_group, node); - list_del(&group->node); - } - spin_unlock(&fault->lock); - return group; -} - -/* Restore a node back to the head of the fault->deliver list */ -static inline void iommufd_fault_deliver_restore(struct iommufd_fault *fault, - struct iopf_group *group) -{ - spin_lock(&fault->lock); - list_add(&group->node, &fault->deliver); - spin_unlock(&fault->lock); -} - struct iommufd_attach_handle { struct iommu_attach_handle handle; struct iommufd_device *idev; diff --git a/drivers/iommu/iommufd/fault.c b/drivers/iommu/iommufd/fault.c index c48d72c9668c..29e3a97c73c6 100644 --- a/drivers/iommu/iommufd/fault.c +++ b/drivers/iommu/iommufd/fault.c @@ -138,6 +138,31 @@ static void iommufd_compose_fault_message(struct iommu_fault *fault, hwpt_fault->cookie = cookie; }
+/* Fetch the first node out of the fault->deliver list */ +static struct iopf_group * +iommufd_fault_deliver_fetch(struct iommufd_fault *fault) +{ + struct list_head *list = &fault->deliver; + struct iopf_group *group = NULL; + + spin_lock(&fault->lock); + if (!list_empty(list)) { + group = list_first_entry(list, struct iopf_group, node); + list_del(&group->node); + } + spin_unlock(&fault->lock); + return group; +} + +/* Restore a node back to the head of the fault->deliver list */ +static void iommufd_fault_deliver_restore(struct iommufd_fault *fault, + struct iopf_group *group) +{ + spin_lock(&fault->lock); + list_add(&group->node, &fault->deliver); + spin_unlock(&fault->lock); +} + static ssize_t iommufd_fault_fops_read(struct file *filep, char __user *buf, size_t count, loff_t *ppos) {
On 2/26/25 01:25, Nicolin Chen wrote:
There is no need to keep them in the header. The vEVENTQ version of these two functions will turn out to be a different implementation and will not share with this fault version. Thus, move them out of the header.
Reviewed-by: Jason Gunthorpejgg@nvidia.com Reviewed-by: Kevin Tiankevin.tian@intel.com Signed-off-by: Nicolin Chennicolinc@nvidia.com
Reviewed-by: Lu Baolu baolu.lu@linux.intel.com
The infrastructure of a fault object will be shared with a new vEVENTQ object in a following change. Add an iommufd_fault_init helper and an INIT_EVENTQ_FOPS marco for a vEVENTQ allocator to use too.
Reorder the iommufd_ctx_get and refcount_inc, to keep them symmetrical with the iommufd_fault_fops_release().
Since the new vEVENTQ doesn't need "response" and its "mutex", so keep the xa_init_flags and mutex_init in their original locations.
Reviewed-by: Kevin Tian kevin.tian@intel.com Reviewed-by: Lu Baolu baolu.lu@linux.intel.com Reviewed-by: Jason Gunthorpe jgg@nvidia.com Signed-off-by: Nicolin Chen nicolinc@nvidia.com --- drivers/iommu/iommufd/fault.c | 70 +++++++++++++++++++++-------------- 1 file changed, 42 insertions(+), 28 deletions(-)
diff --git a/drivers/iommu/iommufd/fault.c b/drivers/iommu/iommufd/fault.c index 29e3a97c73c6..5d8de98732b6 100644 --- a/drivers/iommu/iommufd/fault.c +++ b/drivers/iommu/iommufd/fault.c @@ -280,20 +280,49 @@ static int iommufd_fault_fops_release(struct inode *inode, struct file *filep) return 0; }
-static const struct file_operations iommufd_fault_fops = { - .owner = THIS_MODULE, - .open = nonseekable_open, - .read = iommufd_fault_fops_read, - .write = iommufd_fault_fops_write, - .poll = iommufd_fault_fops_poll, - .release = iommufd_fault_fops_release, -}; +#define INIT_FAULT_FOPS(read_op, write_op) \ + ((const struct file_operations){ \ + .owner = THIS_MODULE, \ + .open = nonseekable_open, \ + .read = read_op, \ + .write = write_op, \ + .poll = iommufd_fault_fops_poll, \ + .release = iommufd_fault_fops_release, \ + }) + +static int iommufd_fault_init(struct iommufd_fault *fault, char *name, + struct iommufd_ctx *ictx, + const struct file_operations *fops) +{ + struct file *filep; + int fdno; + + spin_lock_init(&fault->lock); + INIT_LIST_HEAD(&fault->deliver); + init_waitqueue_head(&fault->wait_queue); + + filep = anon_inode_getfile(name, fops, fault, O_RDWR); + if (IS_ERR(filep)) + return PTR_ERR(filep); + + fault->ictx = ictx; + iommufd_ctx_get(fault->ictx); + fault->filep = filep; + refcount_inc(&fault->obj.users); + + fdno = get_unused_fd_flags(O_CLOEXEC); + if (fdno < 0) + fput(filep); + return fdno; +} + +static const struct file_operations iommufd_fault_fops = + INIT_FAULT_FOPS(iommufd_fault_fops_read, iommufd_fault_fops_write);
int iommufd_fault_alloc(struct iommufd_ucmd *ucmd) { struct iommu_fault_alloc *cmd = ucmd->cmd; struct iommufd_fault *fault; - struct file *filep; int fdno; int rc;
@@ -304,28 +333,14 @@ int iommufd_fault_alloc(struct iommufd_ucmd *ucmd) if (IS_ERR(fault)) return PTR_ERR(fault);
- fault->ictx = ucmd->ictx; - INIT_LIST_HEAD(&fault->deliver); xa_init_flags(&fault->response, XA_FLAGS_ALLOC1); mutex_init(&fault->mutex); - spin_lock_init(&fault->lock); - init_waitqueue_head(&fault->wait_queue); - - filep = anon_inode_getfile("[iommufd-pgfault]", &iommufd_fault_fops, - fault, O_RDWR); - if (IS_ERR(filep)) { - rc = PTR_ERR(filep); - goto out_abort; - }
- refcount_inc(&fault->obj.users); - iommufd_ctx_get(fault->ictx); - fault->filep = filep; - - fdno = get_unused_fd_flags(O_CLOEXEC); + fdno = iommufd_fault_init(fault, "[iommufd-pgfault]", ucmd->ictx, + &iommufd_fault_fops); if (fdno < 0) { rc = fdno; - goto out_fput; + goto out_abort; }
cmd->out_fault_id = fault->obj.id; @@ -341,8 +356,7 @@ int iommufd_fault_alloc(struct iommufd_ucmd *ucmd) return 0; out_put_fdno: put_unused_fd(fdno); -out_fput: - fput(filep); + fput(fault->filep); out_abort: iommufd_object_abort_and_destroy(ucmd->ictx, &fault->obj);
The fault object was designed exclusively for hwpt's IO page faults (PRI). But its queue implementation can be reused for other purposes too, such as hardware IRQ and event injections to user space.
Meanwhile, a fault object holds a list of faults. So it's more accurate to call it a "fault queue". Combining the reusing idea above, abstract a new iommufd_eventq as a common structure embedded into struct iommufd_fault, similar to hwpt_paging holding a common hwpt.
Add a common iommufd_eventq_ops and iommufd_eventq_init to prepare for an IOMMUFD_OBJ_VEVENTQ (vIOMMU Event Queue).
Reviewed-by: Lu Baolu baolu.lu@linux.intel.com Reviewed-by: Kevin Tian kevin.tian@intel.com Reviewed-by: Jason Gunthorpe jgg@nvidia.com Signed-off-by: Nicolin Chen nicolinc@nvidia.com --- drivers/iommu/iommufd/iommufd_private.h | 28 ++++-- drivers/iommu/iommufd/fault.c | 111 +++++++++++++----------- drivers/iommu/iommufd/hw_pagetable.c | 6 +- 3 files changed, 82 insertions(+), 63 deletions(-)
diff --git a/drivers/iommu/iommufd/iommufd_private.h b/drivers/iommu/iommufd/iommufd_private.h index 1c58f5fe17b4..44fb30af10b0 100644 --- a/drivers/iommu/iommufd/iommufd_private.h +++ b/drivers/iommu/iommufd/iommufd_private.h @@ -454,20 +454,13 @@ void iopt_remove_access(struct io_pagetable *iopt, u32 iopt_access_list_id); void iommufd_access_destroy_object(struct iommufd_object *obj);
-/* - * An iommufd_fault object represents an interface to deliver I/O page faults - * to the user space. These objects are created/destroyed by the user space and - * associated with hardware page table objects during page-table allocation. - */ -struct iommufd_fault { +struct iommufd_eventq { struct iommufd_object obj; struct iommufd_ctx *ictx; struct file *filep;
spinlock_t lock; /* protects the deliver list */ struct list_head deliver; - struct mutex mutex; /* serializes response flows */ - struct xarray response;
struct wait_queue_head wait_queue; }; @@ -480,12 +473,29 @@ struct iommufd_attach_handle { /* Convert an iommu attach handle to iommufd handle. */ #define to_iommufd_handle(hdl) container_of(hdl, struct iommufd_attach_handle, handle)
+/* + * An iommufd_fault object represents an interface to deliver I/O page faults + * to the user space. These objects are created/destroyed by the user space and + * associated with hardware page table objects during page-table allocation. + */ +struct iommufd_fault { + struct iommufd_eventq common; + struct mutex mutex; /* serializes response flows */ + struct xarray response; +}; + +static inline struct iommufd_fault * +eventq_to_fault(struct iommufd_eventq *eventq) +{ + return container_of(eventq, struct iommufd_fault, common); +} + static inline struct iommufd_fault * iommufd_get_fault(struct iommufd_ucmd *ucmd, u32 id) { return container_of(iommufd_get_object(ucmd->ictx, id, IOMMUFD_OBJ_FAULT), - struct iommufd_fault, obj); + struct iommufd_fault, common.obj); }
int iommufd_fault_alloc(struct iommufd_ucmd *ucmd); diff --git a/drivers/iommu/iommufd/fault.c b/drivers/iommu/iommufd/fault.c index 5d8de98732b6..f8e60e5879d1 100644 --- a/drivers/iommu/iommufd/fault.c +++ b/drivers/iommu/iommufd/fault.c @@ -17,6 +17,8 @@ #include "../iommu-priv.h" #include "iommufd_private.h"
+/* IOMMUFD_OBJ_FAULT Functions */ + int iommufd_fault_iopf_enable(struct iommufd_device *idev) { struct device *dev = idev->dev; @@ -73,13 +75,13 @@ void iommufd_auto_response_faults(struct iommufd_hw_pagetable *hwpt, INIT_LIST_HEAD(&free_list);
mutex_lock(&fault->mutex); - spin_lock(&fault->lock); - list_for_each_entry_safe(group, next, &fault->deliver, node) { + spin_lock(&fault->common.lock); + list_for_each_entry_safe(group, next, &fault->common.deliver, node) { if (group->attach_handle != &handle->handle) continue; list_move(&group->node, &free_list); } - spin_unlock(&fault->lock); + spin_unlock(&fault->common.lock);
list_for_each_entry_safe(group, next, &free_list, node) { list_del(&group->node); @@ -99,7 +101,9 @@ void iommufd_auto_response_faults(struct iommufd_hw_pagetable *hwpt,
void iommufd_fault_destroy(struct iommufd_object *obj) { - struct iommufd_fault *fault = container_of(obj, struct iommufd_fault, obj); + struct iommufd_eventq *eventq = + container_of(obj, struct iommufd_eventq, obj); + struct iommufd_fault *fault = eventq_to_fault(eventq); struct iopf_group *group, *next; unsigned long index;
@@ -109,7 +113,7 @@ void iommufd_fault_destroy(struct iommufd_object *obj) * accessing this pointer. Therefore, acquiring the mutex here * is unnecessary. */ - list_for_each_entry_safe(group, next, &fault->deliver, node) { + list_for_each_entry_safe(group, next, &fault->common.deliver, node) { list_del(&group->node); iopf_group_response(group, IOMMU_PAGE_RESP_INVALID); iopf_free_group(group); @@ -142,15 +146,15 @@ static void iommufd_compose_fault_message(struct iommu_fault *fault, static struct iopf_group * iommufd_fault_deliver_fetch(struct iommufd_fault *fault) { - struct list_head *list = &fault->deliver; + struct list_head *list = &fault->common.deliver; struct iopf_group *group = NULL;
- spin_lock(&fault->lock); + spin_lock(&fault->common.lock); if (!list_empty(list)) { group = list_first_entry(list, struct iopf_group, node); list_del(&group->node); } - spin_unlock(&fault->lock); + spin_unlock(&fault->common.lock); return group; }
@@ -158,16 +162,17 @@ iommufd_fault_deliver_fetch(struct iommufd_fault *fault) static void iommufd_fault_deliver_restore(struct iommufd_fault *fault, struct iopf_group *group) { - spin_lock(&fault->lock); - list_add(&group->node, &fault->deliver); - spin_unlock(&fault->lock); + spin_lock(&fault->common.lock); + list_add(&group->node, &fault->common.deliver); + spin_unlock(&fault->common.lock); }
static ssize_t iommufd_fault_fops_read(struct file *filep, char __user *buf, size_t count, loff_t *ppos) { size_t fault_size = sizeof(struct iommu_hwpt_pgfault); - struct iommufd_fault *fault = filep->private_data; + struct iommufd_eventq *eventq = filep->private_data; + struct iommufd_fault *fault = eventq_to_fault(eventq); struct iommu_hwpt_pgfault data = {}; struct iommufd_device *idev; struct iopf_group *group; @@ -216,7 +221,8 @@ static ssize_t iommufd_fault_fops_write(struct file *filep, const char __user *b size_t count, loff_t *ppos) { size_t response_size = sizeof(struct iommu_hwpt_page_response); - struct iommufd_fault *fault = filep->private_data; + struct iommufd_eventq *eventq = filep->private_data; + struct iommufd_fault *fault = eventq_to_fault(eventq); struct iommu_hwpt_page_response response; struct iopf_group *group; size_t done = 0; @@ -256,59 +262,61 @@ static ssize_t iommufd_fault_fops_write(struct file *filep, const char __user *b return done == 0 ? rc : done; }
-static __poll_t iommufd_fault_fops_poll(struct file *filep, - struct poll_table_struct *wait) +/* Common Event Queue Functions */ + +static __poll_t iommufd_eventq_fops_poll(struct file *filep, + struct poll_table_struct *wait) { - struct iommufd_fault *fault = filep->private_data; + struct iommufd_eventq *eventq = filep->private_data; __poll_t pollflags = EPOLLOUT;
- poll_wait(filep, &fault->wait_queue, wait); - spin_lock(&fault->lock); - if (!list_empty(&fault->deliver)) + poll_wait(filep, &eventq->wait_queue, wait); + spin_lock(&eventq->lock); + if (!list_empty(&eventq->deliver)) pollflags |= EPOLLIN | EPOLLRDNORM; - spin_unlock(&fault->lock); + spin_unlock(&eventq->lock);
return pollflags; }
-static int iommufd_fault_fops_release(struct inode *inode, struct file *filep) +static int iommufd_eventq_fops_release(struct inode *inode, struct file *filep) { - struct iommufd_fault *fault = filep->private_data; + struct iommufd_eventq *eventq = filep->private_data;
- refcount_dec(&fault->obj.users); - iommufd_ctx_put(fault->ictx); + refcount_dec(&eventq->obj.users); + iommufd_ctx_put(eventq->ictx); return 0; }
-#define INIT_FAULT_FOPS(read_op, write_op) \ +#define INIT_EVENTQ_FOPS(read_op, write_op) \ ((const struct file_operations){ \ .owner = THIS_MODULE, \ .open = nonseekable_open, \ .read = read_op, \ .write = write_op, \ - .poll = iommufd_fault_fops_poll, \ - .release = iommufd_fault_fops_release, \ + .poll = iommufd_eventq_fops_poll, \ + .release = iommufd_eventq_fops_release, \ })
-static int iommufd_fault_init(struct iommufd_fault *fault, char *name, - struct iommufd_ctx *ictx, - const struct file_operations *fops) +static int iommufd_eventq_init(struct iommufd_eventq *eventq, char *name, + struct iommufd_ctx *ictx, + const struct file_operations *fops) { struct file *filep; int fdno;
- spin_lock_init(&fault->lock); - INIT_LIST_HEAD(&fault->deliver); - init_waitqueue_head(&fault->wait_queue); + spin_lock_init(&eventq->lock); + INIT_LIST_HEAD(&eventq->deliver); + init_waitqueue_head(&eventq->wait_queue);
- filep = anon_inode_getfile(name, fops, fault, O_RDWR); + filep = anon_inode_getfile(name, fops, eventq, O_RDWR); if (IS_ERR(filep)) return PTR_ERR(filep);
- fault->ictx = ictx; - iommufd_ctx_get(fault->ictx); - fault->filep = filep; - refcount_inc(&fault->obj.users); + eventq->ictx = ictx; + iommufd_ctx_get(eventq->ictx); + eventq->filep = filep; + refcount_inc(&eventq->obj.users);
fdno = get_unused_fd_flags(O_CLOEXEC); if (fdno < 0) @@ -317,7 +325,7 @@ static int iommufd_fault_init(struct iommufd_fault *fault, char *name, }
static const struct file_operations iommufd_fault_fops = - INIT_FAULT_FOPS(iommufd_fault_fops_read, iommufd_fault_fops_write); + INIT_EVENTQ_FOPS(iommufd_fault_fops_read, iommufd_fault_fops_write);
int iommufd_fault_alloc(struct iommufd_ucmd *ucmd) { @@ -329,36 +337,37 @@ int iommufd_fault_alloc(struct iommufd_ucmd *ucmd) if (cmd->flags) return -EOPNOTSUPP;
- fault = iommufd_object_alloc(ucmd->ictx, fault, IOMMUFD_OBJ_FAULT); + fault = __iommufd_object_alloc(ucmd->ictx, fault, IOMMUFD_OBJ_FAULT, + common.obj); if (IS_ERR(fault)) return PTR_ERR(fault);
xa_init_flags(&fault->response, XA_FLAGS_ALLOC1); mutex_init(&fault->mutex);
- fdno = iommufd_fault_init(fault, "[iommufd-pgfault]", ucmd->ictx, - &iommufd_fault_fops); + fdno = iommufd_eventq_init(&fault->common, "[iommufd-pgfault]", + ucmd->ictx, &iommufd_fault_fops); if (fdno < 0) { rc = fdno; goto out_abort; }
- cmd->out_fault_id = fault->obj.id; + cmd->out_fault_id = fault->common.obj.id; cmd->out_fault_fd = fdno;
rc = iommufd_ucmd_respond(ucmd, sizeof(*cmd)); if (rc) goto out_put_fdno; - iommufd_object_finalize(ucmd->ictx, &fault->obj); + iommufd_object_finalize(ucmd->ictx, &fault->common.obj);
- fd_install(fdno, fault->filep); + fd_install(fdno, fault->common.filep);
return 0; out_put_fdno: put_unused_fd(fdno); - fput(fault->filep); + fput(fault->common.filep); out_abort: - iommufd_object_abort_and_destroy(ucmd->ictx, &fault->obj); + iommufd_object_abort_and_destroy(ucmd->ictx, &fault->common.obj);
return rc; } @@ -371,11 +380,11 @@ int iommufd_fault_iopf_handler(struct iopf_group *group) hwpt = group->attach_handle->domain->iommufd_hwpt; fault = hwpt->fault;
- spin_lock(&fault->lock); - list_add_tail(&group->node, &fault->deliver); - spin_unlock(&fault->lock); + spin_lock(&fault->common.lock); + list_add_tail(&group->node, &fault->common.deliver); + spin_unlock(&fault->common.lock);
- wake_up_interruptible(&fault->wait_queue); + wake_up_interruptible(&fault->common.wait_queue);
return 0; } diff --git a/drivers/iommu/iommufd/hw_pagetable.c b/drivers/iommu/iommufd/hw_pagetable.c index 7de6e914232e..006425e7f609 100644 --- a/drivers/iommu/iommufd/hw_pagetable.c +++ b/drivers/iommu/iommufd/hw_pagetable.c @@ -14,7 +14,7 @@ static void __iommufd_hwpt_destroy(struct iommufd_hw_pagetable *hwpt) iommu_domain_free(hwpt->domain);
if (hwpt->fault) - refcount_dec(&hwpt->fault->obj.users); + refcount_dec(&hwpt->fault->common.obj.users); }
void iommufd_hwpt_paging_destroy(struct iommufd_object *obj) @@ -409,8 +409,8 @@ int iommufd_hwpt_alloc(struct iommufd_ucmd *ucmd) } hwpt->fault = fault; hwpt->domain->iopf_handler = iommufd_fault_iopf_handler; - refcount_inc(&fault->obj.users); - iommufd_put_object(ucmd->ictx, &fault->obj); + refcount_inc(&fault->common.obj.users); + iommufd_put_object(ucmd->ictx, &fault->common.obj); } hwpt->domain->iommufd_hwpt = hwpt;
Rename the file, aligning with the new eventq object.
Reviewed-by: Kevin Tian kevin.tian@intel.com Reviewed-by: Lu Baolu baolu.lu@linux.intel.com Reviewed-by: Jason Gunthorpe jgg@nvidia.com Signed-off-by: Nicolin Chen nicolinc@nvidia.com --- drivers/iommu/iommufd/Makefile | 2 +- drivers/iommu/iommufd/{fault.c => eventq.c} | 0 2 files changed, 1 insertion(+), 1 deletion(-) rename drivers/iommu/iommufd/{fault.c => eventq.c} (100%)
diff --git a/drivers/iommu/iommufd/Makefile b/drivers/iommu/iommufd/Makefile index cb784da6cddc..71d692c9a8f4 100644 --- a/drivers/iommu/iommufd/Makefile +++ b/drivers/iommu/iommufd/Makefile @@ -1,7 +1,7 @@ # SPDX-License-Identifier: GPL-2.0-only iommufd-y := \ device.o \ - fault.o \ + eventq.o \ hw_pagetable.o \ io_pagetable.o \ ioas.o \ diff --git a/drivers/iommu/iommufd/fault.c b/drivers/iommu/iommufd/eventq.c similarity index 100% rename from drivers/iommu/iommufd/fault.c rename to drivers/iommu/iommufd/eventq.c
Introduce a new IOMMUFD_OBJ_VEVENTQ object for vIOMMU Event Queue that provides user space (VMM) another FD to read the vIOMMU Events.
Allow a vIOMMU object to allocate vEVENTQs, with a condition that each vIOMMU can only have one single vEVENTQ per type.
Add iommufd_veventq_alloc() with iommufd_veventq_ops for the new ioctl.
Reviewed-by: Lu Baolu baolu.lu@linux.intel.com Reviewed-by: Kevin Tian kevin.tian@intel.com Reviewed-by: Jason Gunthorpe jgg@nvidia.com Signed-off-by: Nicolin Chen nicolinc@nvidia.com --- drivers/iommu/iommufd/iommufd_private.h | 82 ++++++++++ include/linux/iommufd.h | 3 + include/uapi/linux/iommufd.h | 82 ++++++++++ drivers/iommu/iommufd/eventq.c | 209 +++++++++++++++++++++++- drivers/iommu/iommufd/main.c | 7 + drivers/iommu/iommufd/viommu.c | 2 + 6 files changed, 384 insertions(+), 1 deletion(-)
diff --git a/drivers/iommu/iommufd/iommufd_private.h b/drivers/iommu/iommufd/iommufd_private.h index 44fb30af10b0..8cda9c4672eb 100644 --- a/drivers/iommu/iommufd/iommufd_private.h +++ b/drivers/iommu/iommufd/iommufd_private.h @@ -507,6 +507,74 @@ void iommufd_fault_iopf_disable(struct iommufd_device *idev); void iommufd_auto_response_faults(struct iommufd_hw_pagetable *hwpt, struct iommufd_attach_handle *handle);
+/* An iommufd_vevent represents a vIOMMU event in an iommufd_veventq */ +struct iommufd_vevent { + struct iommufd_vevent_header header; + struct list_head node; /* for iommufd_eventq::deliver */ + ssize_t data_len; + u64 event_data[] __counted_by(data_len); +}; + +#define vevent_for_lost_events_header(vevent) \ + (vevent->header.flags & IOMMU_VEVENTQ_FLAG_LOST_EVENTS) + +/* + * An iommufd_veventq object represents an interface to deliver vIOMMU events to + * the user space. It is created/destroyed by the user space and associated with + * a vIOMMU object during the allocations. + */ +struct iommufd_veventq { + struct iommufd_eventq common; + struct iommufd_viommu *viommu; + struct list_head node; /* for iommufd_viommu::veventqs */ + struct iommufd_vevent lost_events_header; + + unsigned int type; + unsigned int depth; + + /* Use common.lock for protection */ + u32 num_events; + u32 sequence; +}; + +static inline struct iommufd_veventq * +eventq_to_veventq(struct iommufd_eventq *eventq) +{ + return container_of(eventq, struct iommufd_veventq, common); +} + +static inline struct iommufd_veventq * +iommufd_get_veventq(struct iommufd_ucmd *ucmd, u32 id) +{ + return container_of(iommufd_get_object(ucmd->ictx, id, + IOMMUFD_OBJ_VEVENTQ), + struct iommufd_veventq, common.obj); +} + +int iommufd_veventq_alloc(struct iommufd_ucmd *ucmd); +void iommufd_veventq_destroy(struct iommufd_object *obj); +void iommufd_veventq_abort(struct iommufd_object *obj); + +static inline void iommufd_vevent_handler(struct iommufd_veventq *veventq, + struct iommufd_vevent *vevent) +{ + struct iommufd_eventq *eventq = &veventq->common; + + lockdep_assert_held(&eventq->lock); + + /* + * Remove the lost_events_header and add the new node at the same time. + * Note the new node can be lost_events_header, for a sequence update. + */ + if (list_is_last(&veventq->lost_events_header.node, &eventq->deliver)) + list_del(&veventq->lost_events_header.node); + list_add_tail(&vevent->node, &eventq->deliver); + vevent->header.sequence = veventq->sequence; + veventq->sequence = (veventq->sequence + 1) & INT_MAX; + + wake_up_interruptible(&eventq->wait_queue); +} + static inline struct iommufd_viommu * iommufd_get_viommu(struct iommufd_ucmd *ucmd, u32 id) { @@ -515,6 +583,20 @@ iommufd_get_viommu(struct iommufd_ucmd *ucmd, u32 id) struct iommufd_viommu, obj); }
+static inline struct iommufd_veventq * +iommufd_viommu_find_veventq(struct iommufd_viommu *viommu, u32 type) +{ + struct iommufd_veventq *veventq, *next; + + lockdep_assert_held(&viommu->veventqs_rwsem); + + list_for_each_entry_safe(veventq, next, &viommu->veventqs, node) { + if (veventq->type == type) + return veventq; + } + return NULL; +} + int iommufd_viommu_alloc_ioctl(struct iommufd_ucmd *ucmd); void iommufd_viommu_destroy(struct iommufd_object *obj); int iommufd_vdevice_alloc_ioctl(struct iommufd_ucmd *ucmd); diff --git a/include/linux/iommufd.h b/include/linux/iommufd.h index 11110c749200..8948b1836940 100644 --- a/include/linux/iommufd.h +++ b/include/linux/iommufd.h @@ -34,6 +34,7 @@ enum iommufd_object_type { IOMMUFD_OBJ_FAULT, IOMMUFD_OBJ_VIOMMU, IOMMUFD_OBJ_VDEVICE, + IOMMUFD_OBJ_VEVENTQ, #ifdef CONFIG_IOMMUFD_TEST IOMMUFD_OBJ_SELFTEST, #endif @@ -93,6 +94,8 @@ struct iommufd_viommu { const struct iommufd_viommu_ops *ops;
struct xarray vdevs; + struct list_head veventqs; + struct rw_semaphore veventqs_rwsem;
unsigned int type; }; diff --git a/include/uapi/linux/iommufd.h b/include/uapi/linux/iommufd.h index 78747b24bd0f..2ade4839880d 100644 --- a/include/uapi/linux/iommufd.h +++ b/include/uapi/linux/iommufd.h @@ -55,6 +55,7 @@ enum { IOMMUFD_CMD_VIOMMU_ALLOC = 0x90, IOMMUFD_CMD_VDEVICE_ALLOC = 0x91, IOMMUFD_CMD_IOAS_CHANGE_PROCESS = 0x92, + IOMMUFD_CMD_VEVENTQ_ALLOC = 0x93, };
/** @@ -1014,4 +1015,85 @@ struct iommu_ioas_change_process { #define IOMMU_IOAS_CHANGE_PROCESS \ _IO(IOMMUFD_TYPE, IOMMUFD_CMD_IOAS_CHANGE_PROCESS)
+/** + * enum iommu_veventq_flag - flag for struct iommufd_vevent_header + * @IOMMU_VEVENTQ_FLAG_OVERFLOW: vEVENTQ has lost vEVENTs + */ +enum iommu_veventq_flag { + IOMMU_VEVENTQ_FLAG_LOST_EVENTS = (1U << 0), +}; + +/** + * struct iommufd_vevent_header - Virtual Event Header for a vEVENTQ Status + * @flags: Combination of enum iommu_veventq_flag + * @sequence: The sequence index of a vEVENT in the vEVENTQ, with a range of + * [0, INT_MAX] where the following index of INT_MAX is 0 + * + * Each iommufd_vevent_header reports a sequence index of the following vEVENT: + * ------------------------------------------------------------------------- + * | header0 {sequence=0} | data0 | header1 {sequence=1} | data1 |...| dataN | + * ------------------------------------------------------------------------- + * And this sequence index is expected to be monotonic to the sequence index of + * the previous vEVENT. If two adjacent sequence indexes has a delta larger than + * 1, it means that delta - 1 number of vEVENTs has lost, e.g. two lost vEVENTs: + * ------------------------------------------------------------------------- + * | ... | header3 {sequence=3} | data3 | header6 {sequence=6} | data6 | ... | + * ------------------------------------------------------------------------- + * If a vEVENT lost at the tail of the vEVENTQ and there is no following vEVENT + * providing the next sequence index, an IOMMU_VEVENTQ_FLAG_LOST_EVENTS header + * would be added to the tail, and no data would follow this header: + * --------------------------------------------------------------------------- + * |..| header3 {sequence=3} | data3 | header4 {flags=LOST_EVENTS, sequence=4} | + * --------------------------------------------------------------------------- + */ +struct iommufd_vevent_header { + __u32 flags; + __u32 sequence; +}; + +/** + * enum iommu_veventq_type - Virtual Event Queue Type + * @IOMMU_VEVENTQ_TYPE_DEFAULT: Reserved for future use + */ +enum iommu_veventq_type { + IOMMU_VEVENTQ_TYPE_DEFAULT = 0, +}; + +/** + * struct iommu_veventq_alloc - ioctl(IOMMU_VEVENTQ_ALLOC) + * @size: sizeof(struct iommu_veventq_alloc) + * @flags: Must be 0 + * @viommu: virtual IOMMU ID to associate the vEVENTQ with + * @type: Type of the vEVENTQ. Must be defined in enum iommu_veventq_type + * @veventq_depth: Maximum number of events in the vEVENTQ + * @out_veventq_id: The ID of the new vEVENTQ + * @out_veventq_fd: The fd of the new vEVENTQ. User space must close the + * successfully returned fd after using it + * @__reserved: Must be 0 + * + * Explicitly allocate a virtual event queue interface for a vIOMMU. A vIOMMU + * can have multiple FDs for different types, but is confined to one per @type. + * User space should open the @out_veventq_fd to read vEVENTs out of a vEVENTQ, + * if there are vEVENTs available. A vEVENTQ will lose events due to overflow, + * if the number of the vEVENTs hits @veventq_depth. + * + * Each vEVENT in a vEVENTQ encloses a struct iommufd_vevent_header followed by + * a type-specific data structure, in a normal case: + * ------------------------------------------------------------- + * || header0 | data0 | header1 | data1 | ... | headerN | dataN || + * ------------------------------------------------------------- + * unless a tailing IOMMU_VEVENTQ_FLAG_LOST_EVENTS header is logged (refer to + * struct iommufd_vevent_header). + */ +struct iommu_veventq_alloc { + __u32 size; + __u32 flags; + __u32 viommu_id; + __u32 type; + __u32 veventq_depth; + __u32 out_veventq_id; + __u32 out_veventq_fd; + __u32 __reserved; +}; +#define IOMMU_VEVENTQ_ALLOC _IO(IOMMUFD_TYPE, IOMMUFD_CMD_VEVENTQ_ALLOC) #endif diff --git a/drivers/iommu/iommufd/eventq.c b/drivers/iommu/iommufd/eventq.c index f8e60e5879d1..4c43ace8c725 100644 --- a/drivers/iommu/iommufd/eventq.c +++ b/drivers/iommu/iommufd/eventq.c @@ -262,13 +262,148 @@ static ssize_t iommufd_fault_fops_write(struct file *filep, const char __user *b return done == 0 ? rc : done; }
+/* IOMMUFD_OBJ_VEVENTQ Functions */ + +void iommufd_veventq_abort(struct iommufd_object *obj) +{ + struct iommufd_eventq *eventq = + container_of(obj, struct iommufd_eventq, obj); + struct iommufd_veventq *veventq = eventq_to_veventq(eventq); + struct iommufd_viommu *viommu = veventq->viommu; + struct iommufd_vevent *cur, *next; + + lockdep_assert_held_write(&viommu->veventqs_rwsem); + + list_for_each_entry_safe(cur, next, &eventq->deliver, node) { + list_del(&cur->node); + if (cur != &veventq->lost_events_header) + kfree(cur); + } + + refcount_dec(&viommu->obj.users); + list_del(&veventq->node); +} + +void iommufd_veventq_destroy(struct iommufd_object *obj) +{ + struct iommufd_veventq *veventq = eventq_to_veventq( + container_of(obj, struct iommufd_eventq, obj)); + + down_write(&veventq->viommu->veventqs_rwsem); + iommufd_veventq_abort(obj); + up_write(&veventq->viommu->veventqs_rwsem); +} + +static struct iommufd_vevent * +iommufd_veventq_deliver_fetch(struct iommufd_veventq *veventq) +{ + struct iommufd_eventq *eventq = &veventq->common; + struct list_head *list = &eventq->deliver; + struct iommufd_vevent *vevent = NULL; + + spin_lock(&eventq->lock); + if (!list_empty(list)) { + struct iommufd_vevent *next; + + next = list_first_entry(list, struct iommufd_vevent, node); + /* Make a copy of the lost_events_header for copy_to_user */ + if (next == &veventq->lost_events_header) { + vevent = kzalloc(sizeof(*vevent), GFP_ATOMIC); + if (!vevent) + goto out_unlock; + } + list_del(&next->node); + if (vevent) + memcpy(vevent, next, sizeof(*vevent)); + else + vevent = next; + } +out_unlock: + spin_unlock(&eventq->lock); + return vevent; +} + +static void iommufd_veventq_deliver_restore(struct iommufd_veventq *veventq, + struct iommufd_vevent *vevent) +{ + struct iommufd_eventq *eventq = &veventq->common; + struct list_head *list = &eventq->deliver; + + spin_lock(&eventq->lock); + if (vevent_for_lost_events_header(vevent)) { + /* Remove the copy of the lost_events_header */ + kfree(vevent); + vevent = NULL; + /* An empty list needs the lost_events_header back */ + if (list_empty(list)) + vevent = &veventq->lost_events_header; + } + if (vevent) + list_add(&vevent->node, list); + spin_unlock(&eventq->lock); +} + +static ssize_t iommufd_veventq_fops_read(struct file *filep, char __user *buf, + size_t count, loff_t *ppos) +{ + struct iommufd_eventq *eventq = filep->private_data; + struct iommufd_veventq *veventq = eventq_to_veventq(eventq); + struct iommufd_vevent_header *hdr; + struct iommufd_vevent *cur; + size_t done = 0; + int rc = 0; + + if (*ppos) + return -ESPIPE; + + while ((cur = iommufd_veventq_deliver_fetch(veventq))) { + /* Validate the remaining bytes against the header size */ + if (done >= count || sizeof(*hdr) > count - done) { + iommufd_veventq_deliver_restore(veventq, cur); + break; + } + hdr = &cur->header; + + /* If being a normal vEVENT, validate against the full size */ + if (!vevent_for_lost_events_header(cur) && + sizeof(hdr) + cur->data_len > count - done) { + iommufd_veventq_deliver_restore(veventq, cur); + break; + } + + if (copy_to_user(buf + done, hdr, sizeof(*hdr))) { + iommufd_veventq_deliver_restore(veventq, cur); + rc = -EFAULT; + break; + } + done += sizeof(*hdr); + + if (cur->data_len && + copy_to_user(buf + done, cur->event_data, cur->data_len)) { + iommufd_veventq_deliver_restore(veventq, cur); + rc = -EFAULT; + break; + } + spin_lock(&eventq->lock); + veventq->num_events--; + spin_unlock(&eventq->lock); + done += cur->data_len; + kfree(cur); + } + + return done == 0 ? rc : done; +} + /* Common Event Queue Functions */
static __poll_t iommufd_eventq_fops_poll(struct file *filep, struct poll_table_struct *wait) { struct iommufd_eventq *eventq = filep->private_data; - __poll_t pollflags = EPOLLOUT; + __poll_t pollflags = 0; + + if (eventq->obj.type == IOMMUFD_OBJ_FAULT) + pollflags |= EPOLLOUT;
poll_wait(filep, &eventq->wait_queue, wait); spin_lock(&eventq->lock); @@ -388,3 +523,75 @@ int iommufd_fault_iopf_handler(struct iopf_group *group)
return 0; } + +static const struct file_operations iommufd_veventq_fops = + INIT_EVENTQ_FOPS(iommufd_veventq_fops_read, NULL); + +int iommufd_veventq_alloc(struct iommufd_ucmd *ucmd) +{ + struct iommu_veventq_alloc *cmd = ucmd->cmd; + struct iommufd_veventq *veventq; + struct iommufd_viommu *viommu; + int fdno; + int rc; + + if (cmd->flags || cmd->__reserved || + cmd->type == IOMMU_VEVENTQ_TYPE_DEFAULT) + return -EOPNOTSUPP; + if (!cmd->veventq_depth) + return -EINVAL; + + viommu = iommufd_get_viommu(ucmd, cmd->viommu_id); + if (IS_ERR(viommu)) + return PTR_ERR(viommu); + + down_write(&viommu->veventqs_rwsem); + + if (iommufd_viommu_find_veventq(viommu, cmd->type)) { + rc = -EEXIST; + goto out_unlock_veventqs; + } + + veventq = __iommufd_object_alloc(ucmd->ictx, veventq, + IOMMUFD_OBJ_VEVENTQ, common.obj); + if (IS_ERR(veventq)) { + rc = PTR_ERR(veventq); + goto out_unlock_veventqs; + } + + veventq->type = cmd->type; + veventq->viommu = viommu; + refcount_inc(&viommu->obj.users); + veventq->depth = cmd->veventq_depth; + list_add_tail(&veventq->node, &viommu->veventqs); + veventq->lost_events_header.header.flags = + IOMMU_VEVENTQ_FLAG_LOST_EVENTS; + + fdno = iommufd_eventq_init(&veventq->common, "[iommufd-viommu-event]", + ucmd->ictx, &iommufd_veventq_fops); + if (fdno < 0) { + rc = fdno; + goto out_abort; + } + + cmd->out_veventq_id = veventq->common.obj.id; + cmd->out_veventq_fd = fdno; + + rc = iommufd_ucmd_respond(ucmd, sizeof(*cmd)); + if (rc) + goto out_put_fdno; + + iommufd_object_finalize(ucmd->ictx, &veventq->common.obj); + fd_install(fdno, veventq->common.filep); + goto out_unlock_veventqs; + +out_put_fdno: + put_unused_fd(fdno); + fput(veventq->common.filep); +out_abort: + iommufd_object_abort_and_destroy(ucmd->ictx, &veventq->common.obj); +out_unlock_veventqs: + up_write(&viommu->veventqs_rwsem); + iommufd_put_object(ucmd->ictx, &viommu->obj); + return rc; +} diff --git a/drivers/iommu/iommufd/main.c b/drivers/iommu/iommufd/main.c index b6fa9fd11bc1..3df468f64e7d 100644 --- a/drivers/iommu/iommufd/main.c +++ b/drivers/iommu/iommufd/main.c @@ -317,6 +317,7 @@ union ucmd_buffer { struct iommu_ioas_unmap unmap; struct iommu_option option; struct iommu_vdevice_alloc vdev; + struct iommu_veventq_alloc veventq; struct iommu_vfio_ioas vfio_ioas; struct iommu_viommu_alloc viommu; #ifdef CONFIG_IOMMUFD_TEST @@ -372,6 +373,8 @@ static const struct iommufd_ioctl_op iommufd_ioctl_ops[] = { IOCTL_OP(IOMMU_OPTION, iommufd_option, struct iommu_option, val64), IOCTL_OP(IOMMU_VDEVICE_ALLOC, iommufd_vdevice_alloc_ioctl, struct iommu_vdevice_alloc, virt_id), + IOCTL_OP(IOMMU_VEVENTQ_ALLOC, iommufd_veventq_alloc, + struct iommu_veventq_alloc, out_veventq_fd), IOCTL_OP(IOMMU_VFIO_IOAS, iommufd_vfio_ioas, struct iommu_vfio_ioas, __reserved), IOCTL_OP(IOMMU_VIOMMU_ALLOC, iommufd_viommu_alloc_ioctl, @@ -514,6 +517,10 @@ static const struct iommufd_object_ops iommufd_object_ops[] = { [IOMMUFD_OBJ_VDEVICE] = { .destroy = iommufd_vdevice_destroy, }, + [IOMMUFD_OBJ_VEVENTQ] = { + .destroy = iommufd_veventq_destroy, + .abort = iommufd_veventq_abort, + }, [IOMMUFD_OBJ_VIOMMU] = { .destroy = iommufd_viommu_destroy, }, diff --git a/drivers/iommu/iommufd/viommu.c b/drivers/iommu/iommufd/viommu.c index 69b88e8c7c26..01df2b985f02 100644 --- a/drivers/iommu/iommufd/viommu.c +++ b/drivers/iommu/iommufd/viommu.c @@ -59,6 +59,8 @@ int iommufd_viommu_alloc_ioctl(struct iommufd_ucmd *ucmd) viommu->ictx = ucmd->ictx; viommu->hwpt = hwpt_paging; refcount_inc(&viommu->hwpt->common.obj.users); + INIT_LIST_HEAD(&viommu->veventqs); + init_rwsem(&viommu->veventqs_rwsem); /* * It is the most likely case that a physical IOMMU is unpluggable. A * pluggable IOMMU instance (if exists) is responsible for refcounting
This is a reverse search v.s. iommufd_viommu_find_dev, as drivers may want to convert a struct device pointer (physical) to its virtual device ID for an event injection to the user space VM.
Again, this avoids exposing more core structures to the drivers, than the iommufd_viommu alone.
Reviewed-by: Lu Baolu baolu.lu@linux.intel.com Reviewed-by: Kevin Tian kevin.tian@intel.com Reviewed-by: Jason Gunthorpe jgg@nvidia.com Signed-off-by: Nicolin Chen nicolinc@nvidia.com --- include/linux/iommufd.h | 9 +++++++++ drivers/iommu/iommufd/driver.c | 24 ++++++++++++++++++++++++ 2 files changed, 33 insertions(+)
diff --git a/include/linux/iommufd.h b/include/linux/iommufd.h index 8948b1836940..05cb393aff0a 100644 --- a/include/linux/iommufd.h +++ b/include/linux/iommufd.h @@ -190,6 +190,8 @@ struct iommufd_object *_iommufd_object_alloc(struct iommufd_ctx *ictx, enum iommufd_object_type type); struct device *iommufd_viommu_find_dev(struct iommufd_viommu *viommu, unsigned long vdev_id); +int iommufd_viommu_get_vdev_id(struct iommufd_viommu *viommu, + struct device *dev, unsigned long *vdev_id); #else /* !CONFIG_IOMMUFD_DRIVER_CORE */ static inline struct iommufd_object * _iommufd_object_alloc(struct iommufd_ctx *ictx, size_t size, @@ -203,6 +205,13 @@ iommufd_viommu_find_dev(struct iommufd_viommu *viommu, unsigned long vdev_id) { return NULL; } + +static inline int iommufd_viommu_get_vdev_id(struct iommufd_viommu *viommu, + struct device *dev, + unsigned long *vdev_id) +{ + return -ENOENT; +} #endif /* CONFIG_IOMMUFD_DRIVER_CORE */
/* diff --git a/drivers/iommu/iommufd/driver.c b/drivers/iommu/iommufd/driver.c index 2d98b04ff1cb..f132b98fb899 100644 --- a/drivers/iommu/iommufd/driver.c +++ b/drivers/iommu/iommufd/driver.c @@ -49,5 +49,29 @@ struct device *iommufd_viommu_find_dev(struct iommufd_viommu *viommu, } EXPORT_SYMBOL_NS_GPL(iommufd_viommu_find_dev, "IOMMUFD");
+/* Return -ENOENT if device is not associated to the vIOMMU */ +int iommufd_viommu_get_vdev_id(struct iommufd_viommu *viommu, + struct device *dev, unsigned long *vdev_id) +{ + struct iommufd_vdevice *vdev; + unsigned long index; + int rc = -ENOENT; + + if (WARN_ON_ONCE(!vdev_id)) + return -EINVAL; + + xa_lock(&viommu->vdevs); + xa_for_each(&viommu->vdevs, index, vdev) { + if (vdev->dev == dev) { + *vdev_id = vdev->id; + rc = 0; + break; + } + } + xa_unlock(&viommu->vdevs); + return rc; +} +EXPORT_SYMBOL_NS_GPL(iommufd_viommu_get_vdev_id, "IOMMUFD"); + MODULE_DESCRIPTION("iommufd code shared with builtin modules"); MODULE_LICENSE("GPL");
Similar to iommu_report_device_fault, this allows IOMMU drivers to report vIOMMU events from threaded IRQ handlers to user space hypervisors.
Reviewed-by: Lu Baolu baolu.lu@linux.intel.com Reviewed-by: Kevin Tian kevin.tian@intel.com Reviewed-by: Jason Gunthorpe jgg@nvidia.com Signed-off-by: Nicolin Chen nicolinc@nvidia.com --- include/linux/iommufd.h | 11 ++++++++ drivers/iommu/iommufd/driver.c | 48 ++++++++++++++++++++++++++++++++++ 2 files changed, 59 insertions(+)
diff --git a/include/linux/iommufd.h b/include/linux/iommufd.h index 05cb393aff0a..60eff9272551 100644 --- a/include/linux/iommufd.h +++ b/include/linux/iommufd.h @@ -11,6 +11,7 @@ #include <linux/refcount.h> #include <linux/types.h> #include <linux/xarray.h> +#include <uapi/linux/iommufd.h>
struct device; struct file; @@ -192,6 +193,9 @@ struct device *iommufd_viommu_find_dev(struct iommufd_viommu *viommu, unsigned long vdev_id); int iommufd_viommu_get_vdev_id(struct iommufd_viommu *viommu, struct device *dev, unsigned long *vdev_id); +int iommufd_viommu_report_event(struct iommufd_viommu *viommu, + enum iommu_veventq_type type, void *event_data, + size_t data_len); #else /* !CONFIG_IOMMUFD_DRIVER_CORE */ static inline struct iommufd_object * _iommufd_object_alloc(struct iommufd_ctx *ictx, size_t size, @@ -212,6 +216,13 @@ static inline int iommufd_viommu_get_vdev_id(struct iommufd_viommu *viommu, { return -ENOENT; } + +static inline int iommufd_viommu_report_event(struct iommufd_viommu *viommu, + enum iommu_veventq_type type, + void *event_data, size_t data_len) +{ + return -EOPNOTSUPP; +} #endif /* CONFIG_IOMMUFD_DRIVER_CORE */
/* diff --git a/drivers/iommu/iommufd/driver.c b/drivers/iommu/iommufd/driver.c index f132b98fb899..75b365561c16 100644 --- a/drivers/iommu/iommufd/driver.c +++ b/drivers/iommu/iommufd/driver.c @@ -73,5 +73,53 @@ int iommufd_viommu_get_vdev_id(struct iommufd_viommu *viommu, } EXPORT_SYMBOL_NS_GPL(iommufd_viommu_get_vdev_id, "IOMMUFD");
+/* + * Typically called in driver's threaded IRQ handler. + * The @type and @event_data must be defined in include/uapi/linux/iommufd.h + */ +int iommufd_viommu_report_event(struct iommufd_viommu *viommu, + enum iommu_veventq_type type, void *event_data, + size_t data_len) +{ + struct iommufd_veventq *veventq; + struct iommufd_vevent *vevent; + int rc = 0; + + if (WARN_ON_ONCE(!data_len || !event_data)) + return -EINVAL; + + down_read(&viommu->veventqs_rwsem); + + veventq = iommufd_viommu_find_veventq(viommu, type); + if (!veventq) { + rc = -EOPNOTSUPP; + goto out_unlock_veventqs; + } + + spin_lock(&veventq->common.lock); + if (veventq->num_events == veventq->depth) { + vevent = &veventq->lost_events_header; + goto out_set_header; + } + + vevent = kmalloc(struct_size(vevent, event_data, data_len), GFP_ATOMIC); + if (!vevent) { + rc = -ENOMEM; + vevent = &veventq->lost_events_header; + goto out_set_header; + } + memcpy(vevent->event_data, event_data, data_len); + vevent->data_len = data_len; + veventq->num_events++; + +out_set_header: + iommufd_vevent_handler(veventq, vevent); + spin_unlock(&veventq->common.lock); +out_unlock_veventqs: + up_read(&viommu->veventqs_rwsem); + return rc; +} +EXPORT_SYMBOL_NS_GPL(iommufd_viommu_report_event, "IOMMUFD"); + MODULE_DESCRIPTION("iommufd code shared with builtin modules"); MODULE_LICENSE("GPL");
When attaching a device to a vIOMMU-based nested domain, vdev_id must be present. Add a piece of code hard-requesting it, preparing for a vEVENTQ support in the following patch. Then, update the TEST_F.
A HWPT-based nested domain will return a NULL new_viommu, thus no such a vDEVICE requirement.
Reviewed-by: Kevin Tian kevin.tian@intel.com Signed-off-by: Nicolin Chen nicolinc@nvidia.com --- drivers/iommu/iommufd/selftest.c | 24 ++++++++++++++++++++++++ tools/testing/selftests/iommu/iommufd.c | 5 +++++ 2 files changed, 29 insertions(+)
diff --git a/drivers/iommu/iommufd/selftest.c b/drivers/iommu/iommufd/selftest.c index d40deb0a4f06..ba84bacbce2e 100644 --- a/drivers/iommu/iommufd/selftest.c +++ b/drivers/iommu/iommufd/selftest.c @@ -161,7 +161,10 @@ enum selftest_obj_type {
struct mock_dev { struct device dev; + struct mock_viommu *viommu; + struct rw_semaphore viommu_rwsem; unsigned long flags; + unsigned long vdev_id; int id; u32 cache[MOCK_DEV_CACHE_NUM]; }; @@ -193,10 +196,30 @@ static int mock_domain_nop_attach(struct iommu_domain *domain, struct device *dev) { struct mock_dev *mdev = to_mock_dev(dev); + struct mock_viommu *new_viommu = NULL; + unsigned long vdev_id = 0; + int rc;
if (domain->dirty_ops && (mdev->flags & MOCK_FLAGS_DEVICE_NO_DIRTY)) return -EINVAL;
+ iommu_group_mutex_assert(dev); + if (domain->type == IOMMU_DOMAIN_NESTED) { + new_viommu = to_mock_nested(domain)->mock_viommu; + if (new_viommu) { + rc = iommufd_viommu_get_vdev_id(&new_viommu->core, dev, + &vdev_id); + if (rc) + return rc; + } + } + if (new_viommu != mdev->viommu) { + down_write(&mdev->viommu_rwsem); + mdev->viommu = new_viommu; + mdev->vdev_id = vdev_id; + up_write(&mdev->viommu_rwsem); + } + return 0; }
@@ -850,6 +873,7 @@ static struct mock_dev *mock_dev_create(unsigned long dev_flags) if (!mdev) return ERR_PTR(-ENOMEM);
+ init_rwsem(&mdev->viommu_rwsem); device_initialize(&mdev->dev); mdev->flags = dev_flags; mdev->dev.release = mock_dev_release; diff --git a/tools/testing/selftests/iommu/iommufd.c b/tools/testing/selftests/iommu/iommufd.c index a1b2b657999d..212e5d62e13d 100644 --- a/tools/testing/selftests/iommu/iommufd.c +++ b/tools/testing/selftests/iommu/iommufd.c @@ -2736,6 +2736,7 @@ TEST_F(iommufd_viommu, viommu_alloc_nested_iopf) uint32_t iopf_hwpt_id; uint32_t fault_id; uint32_t fault_fd; + uint32_t vdev_id;
if (self->device_id) { test_ioctl_fault_alloc(&fault_id, &fault_fd); @@ -2752,6 +2753,10 @@ TEST_F(iommufd_viommu, viommu_alloc_nested_iopf) &iopf_hwpt_id, IOMMU_HWPT_DATA_SELFTEST, &data, sizeof(data));
+ /* Must allocate vdevice before attaching to a nested hwpt */ + test_err_mock_domain_replace(ENOENT, self->stdev_id, + iopf_hwpt_id); + test_cmd_vdevice_alloc(viommu_id, dev_id, 0x99, &vdev_id); test_cmd_mock_domain_replace(self->stdev_id, iopf_hwpt_id); EXPECT_ERRNO(EBUSY, _test_ioctl_destroy(self->fd, iopf_hwpt_id));
The handler will get vDEVICE object from the given mdev and convert it to its per-vIOMMU virtual ID to mimic a real IOMMU driver.
Reviewed-by: Kevin Tian kevin.tian@intel.com Signed-off-by: Nicolin Chen nicolinc@nvidia.com --- drivers/iommu/iommufd/iommufd_test.h | 10 ++++++++++ drivers/iommu/iommufd/selftest.c | 30 ++++++++++++++++++++++++++++ 2 files changed, 40 insertions(+)
diff --git a/drivers/iommu/iommufd/iommufd_test.h b/drivers/iommu/iommufd/iommufd_test.h index a6b7a163f636..87e9165cea27 100644 --- a/drivers/iommu/iommufd/iommufd_test.h +++ b/drivers/iommu/iommufd/iommufd_test.h @@ -24,6 +24,7 @@ enum { IOMMU_TEST_OP_MD_CHECK_IOTLB, IOMMU_TEST_OP_TRIGGER_IOPF, IOMMU_TEST_OP_DEV_CHECK_CACHE, + IOMMU_TEST_OP_TRIGGER_VEVENT, };
enum { @@ -145,6 +146,9 @@ struct iommu_test_cmd { __u32 id; __u32 cache; } check_dev_cache; + struct { + __u32 dev_id; + } trigger_vevent; }; __u32 last; }; @@ -212,4 +216,10 @@ struct iommu_viommu_invalidate_selftest { __u32 cache_id; };
+#define IOMMU_VEVENTQ_TYPE_SELFTEST 0xbeefbeef + +struct iommu_viommu_event_selftest { + __u32 virt_id; +}; + #endif diff --git a/drivers/iommu/iommufd/selftest.c b/drivers/iommu/iommufd/selftest.c index ba84bacbce2e..d55dde28e9bc 100644 --- a/drivers/iommu/iommufd/selftest.c +++ b/drivers/iommu/iommufd/selftest.c @@ -1621,6 +1621,34 @@ static int iommufd_test_trigger_iopf(struct iommufd_ucmd *ucmd, return 0; }
+static int iommufd_test_trigger_vevent(struct iommufd_ucmd *ucmd, + struct iommu_test_cmd *cmd) +{ + struct iommu_viommu_event_selftest test = {}; + struct iommufd_device *idev; + struct mock_dev *mdev; + int rc = -ENOENT; + + idev = iommufd_get_device(ucmd, cmd->trigger_vevent.dev_id); + if (IS_ERR(idev)) + return PTR_ERR(idev); + mdev = to_mock_dev(idev->dev); + + down_read(&mdev->viommu_rwsem); + if (!mdev->viommu || !mdev->vdev_id) + goto out_unlock; + + test.virt_id = mdev->vdev_id; + rc = iommufd_viommu_report_event(&mdev->viommu->core, + IOMMU_VEVENTQ_TYPE_SELFTEST, &test, + sizeof(test)); +out_unlock: + up_read(&mdev->viommu_rwsem); + iommufd_put_object(ucmd->ictx, &idev->obj); + + return rc; +} + void iommufd_selftest_destroy(struct iommufd_object *obj) { struct selftest_obj *sobj = to_selftest_obj(obj); @@ -1702,6 +1730,8 @@ int iommufd_test(struct iommufd_ucmd *ucmd) cmd->dirty.flags); case IOMMU_TEST_OP_TRIGGER_IOPF: return iommufd_test_trigger_iopf(ucmd, cmd); + case IOMMU_TEST_OP_TRIGGER_VEVENT: + return iommufd_test_trigger_vevent(ucmd, cmd); default: return -EOPNOTSUPP; }
Trigger vEVENTs by feeding an idev ID and validating the returned output virt_ids whether they equal to the value that was set to the vDEVICE.
Reviewed-by: Kevin Tian kevin.tian@intel.com Signed-off-by: Nicolin Chen nicolinc@nvidia.com --- tools/testing/selftests/iommu/iommufd_utils.h | 115 ++++++++++++++++++ tools/testing/selftests/iommu/iommufd.c | 31 +++++ .../selftests/iommu/iommufd_fail_nth.c | 7 ++ 3 files changed, 153 insertions(+)
diff --git a/tools/testing/selftests/iommu/iommufd_utils.h b/tools/testing/selftests/iommu/iommufd_utils.h index d979f5b0efe8..6f2ba2fa8f76 100644 --- a/tools/testing/selftests/iommu/iommufd_utils.h +++ b/tools/testing/selftests/iommu/iommufd_utils.h @@ -9,6 +9,7 @@ #include <sys/ioctl.h> #include <stdint.h> #include <assert.h> +#include <poll.h>
#include "../kselftest_harness.h" #include "../../../../drivers/iommu/iommufd/iommufd_test.h" @@ -936,3 +937,117 @@ static int _test_cmd_vdevice_alloc(int fd, __u32 viommu_id, __u32 idev_id, EXPECT_ERRNO(_errno, \ _test_cmd_vdevice_alloc(self->fd, viommu_id, idev_id, \ virt_id, vdev_id)) + +static int _test_cmd_veventq_alloc(int fd, __u32 viommu_id, __u32 type, + __u32 *veventq_id, __u32 *veventq_fd) +{ + struct iommu_veventq_alloc cmd = { + .size = sizeof(cmd), + .type = type, + .veventq_depth = 2, + .viommu_id = viommu_id, + }; + int ret; + + ret = ioctl(fd, IOMMU_VEVENTQ_ALLOC, &cmd); + if (ret) + return ret; + if (veventq_id) + *veventq_id = cmd.out_veventq_id; + if (veventq_fd) + *veventq_fd = cmd.out_veventq_fd; + return 0; +} + +#define test_cmd_veventq_alloc(viommu_id, type, veventq_id, veventq_fd) \ + ASSERT_EQ(0, _test_cmd_veventq_alloc(self->fd, viommu_id, type, \ + veventq_id, veventq_fd)) +#define test_err_veventq_alloc(_errno, viommu_id, type, veventq_id, \ + veventq_fd) \ + EXPECT_ERRNO(_errno, \ + _test_cmd_veventq_alloc(self->fd, viommu_id, type, \ + veventq_id, veventq_fd)) + +static int _test_cmd_trigger_vevents(int fd, __u32 dev_id, __u32 nvevents) +{ + struct iommu_test_cmd trigger_vevent_cmd = { + .size = sizeof(trigger_vevent_cmd), + .op = IOMMU_TEST_OP_TRIGGER_VEVENT, + .trigger_vevent = { + .dev_id = dev_id, + }, + }; + int ret; + + while (nvevents--) { + ret = ioctl(fd, _IOMMU_TEST_CMD(IOMMU_TEST_OP_TRIGGER_VEVENT), + &trigger_vevent_cmd); + if (ret < 0) + return -1; + } + return ret; +} + +#define test_cmd_trigger_vevents(dev_id, nvevents) \ + ASSERT_EQ(0, _test_cmd_trigger_vevents(self->fd, dev_id, nvevents)) + +static int _test_cmd_read_vevents(int fd, __u32 event_fd, __u32 nvevents, + __u32 virt_id, int *prev_seq) +{ + struct pollfd pollfd = { .fd = event_fd, .events = POLLIN }; + struct iommu_viommu_event_selftest *event; + struct iommufd_vevent_header *hdr; + ssize_t bytes; + void *data; + int ret, i; + + ret = poll(&pollfd, 1, 1000); + if (ret < 0) + return -1; + + data = calloc(nvevents, sizeof(*hdr) + sizeof(*event)); + if (!data) { + errno = ENOMEM; + return -1; + } + + bytes = read(event_fd, data, + nvevents * (sizeof(*hdr) + sizeof(*event))); + if (bytes <= 0) { + errno = EFAULT; + ret = -1; + goto out_free; + } + + for (i = 0; i < nvevents; i++) { + hdr = data + i * (sizeof(*hdr) + sizeof(*event)); + + if (hdr->flags & IOMMU_VEVENTQ_FLAG_LOST_EVENTS || + hdr->sequence - *prev_seq > 1) { + *prev_seq = hdr->sequence; + errno = EOVERFLOW; + ret = -1; + goto out_free; + } + *prev_seq = hdr->sequence; + event = data + sizeof(*hdr); + if (event->virt_id != virt_id) { + errno = EINVAL; + ret = -1; + goto out_free; + } + } + + ret = 0; +out_free: + free(data); + return ret; +} + +#define test_cmd_read_vevents(event_fd, nvevents, virt_id, prev_seq) \ + ASSERT_EQ(0, _test_cmd_read_vevents(self->fd, event_fd, nvevents, \ + virt_id, prev_seq)) +#define test_err_read_vevents(_errno, event_fd, nvevents, virt_id, prev_seq) \ + EXPECT_ERRNO(_errno, \ + _test_cmd_read_vevents(self->fd, event_fd, nvevents, \ + virt_id, prev_seq)) diff --git a/tools/testing/selftests/iommu/iommufd.c b/tools/testing/selftests/iommu/iommufd.c index 212e5d62e13d..dd453aae8fed 100644 --- a/tools/testing/selftests/iommu/iommufd.c +++ b/tools/testing/selftests/iommu/iommufd.c @@ -2774,15 +2774,46 @@ TEST_F(iommufd_viommu, vdevice_alloc) uint32_t viommu_id = self->viommu_id; uint32_t dev_id = self->device_id; uint32_t vdev_id = 0; + uint32_t veventq_id; + uint32_t veventq_fd; + int prev_seq = -1;
if (dev_id) { + /* Must allocate vdevice before attaching to a nested hwpt */ + test_err_mock_domain_replace(ENOENT, self->stdev_id, + self->nested_hwpt_id); + + /* Allocate a vEVENTQ with veventq_depth=2 */ + test_cmd_veventq_alloc(viommu_id, IOMMU_VEVENTQ_TYPE_SELFTEST, + &veventq_id, &veventq_fd); + test_err_veventq_alloc(EEXIST, viommu_id, + IOMMU_VEVENTQ_TYPE_SELFTEST, NULL, NULL); /* Set vdev_id to 0x99, unset it, and set to 0x88 */ test_cmd_vdevice_alloc(viommu_id, dev_id, 0x99, &vdev_id); + test_cmd_mock_domain_replace(self->stdev_id, + self->nested_hwpt_id); + test_cmd_trigger_vevents(dev_id, 1); + test_cmd_read_vevents(veventq_fd, 1, 0x99, &prev_seq); test_err_vdevice_alloc(EEXIST, viommu_id, dev_id, 0x99, &vdev_id); + test_cmd_mock_domain_replace(self->stdev_id, self->ioas_id); test_ioctl_destroy(vdev_id); + + /* Try again with 0x88 */ test_cmd_vdevice_alloc(viommu_id, dev_id, 0x88, &vdev_id); + test_cmd_mock_domain_replace(self->stdev_id, + self->nested_hwpt_id); + /* Trigger an overflow with three events */ + test_cmd_trigger_vevents(dev_id, 3); + test_err_read_vevents(EOVERFLOW, veventq_fd, 3, 0x88, + &prev_seq); + /* Overflow must be gone after the previous reads */ + test_cmd_trigger_vevents(dev_id, 1); + test_cmd_read_vevents(veventq_fd, 1, 0x88, &prev_seq); + close(veventq_fd); + test_cmd_mock_domain_replace(self->stdev_id, self->ioas_id); test_ioctl_destroy(vdev_id); + test_ioctl_destroy(veventq_id); } else { test_err_vdevice_alloc(ENOENT, viommu_id, dev_id, 0x99, NULL); } diff --git a/tools/testing/selftests/iommu/iommufd_fail_nth.c b/tools/testing/selftests/iommu/iommufd_fail_nth.c index 64b1f8e1b0cf..99a7f7897bb2 100644 --- a/tools/testing/selftests/iommu/iommufd_fail_nth.c +++ b/tools/testing/selftests/iommu/iommufd_fail_nth.c @@ -620,6 +620,7 @@ TEST_FAIL_NTH(basic_fail_nth, device) }; struct iommu_test_hw_info info; uint32_t fault_id, fault_fd; + uint32_t veventq_id, veventq_fd; uint32_t fault_hwpt_id; uint32_t ioas_id; uint32_t ioas_id2; @@ -692,6 +693,12 @@ TEST_FAIL_NTH(basic_fail_nth, device) IOMMU_HWPT_DATA_SELFTEST, &data, sizeof(data))) return -1;
+ if (_test_cmd_veventq_alloc(self->fd, viommu_id, + IOMMU_VEVENTQ_TYPE_SELFTEST, &veventq_id, + &veventq_fd)) + return -1; + close(veventq_fd); + return 0; }
With the introduction of the new objects, update the doc to reflect that.
Reviewed-by: Lu Baolu baolu.lu@linux.intel.com Reviewed-by: Kevin Tian kevin.tian@intel.com Reviewed-by: Bagas Sanjaya bagasdotme@gmail.com Reviewed-by: Jason Gunthorpe jgg@nvidia.com Signed-off-by: Nicolin Chen nicolinc@nvidia.com --- Documentation/userspace-api/iommufd.rst | 17 +++++++++++++++++ 1 file changed, 17 insertions(+)
diff --git a/Documentation/userspace-api/iommufd.rst b/Documentation/userspace-api/iommufd.rst index 70289d6815d2..b0df15865dec 100644 --- a/Documentation/userspace-api/iommufd.rst +++ b/Documentation/userspace-api/iommufd.rst @@ -63,6 +63,13 @@ Following IOMMUFD objects are exposed to userspace: space usually has mappings from guest-level I/O virtual addresses to guest- level physical addresses.
+- IOMMUFD_FAULT, representing a software queue for an HWPT reporting IO page + faults using the IOMMU HW's PRI (Page Request Interface). This queue object + provides user space an FD to poll the page fault events and also to respond + to those events. A FAULT object must be created first to get a fault_id that + could be then used to allocate a fault-enabled HWPT via the IOMMU_HWPT_ALLOC + command by setting the IOMMU_HWPT_FAULT_ID_VALID bit in its flags field. + - IOMMUFD_OBJ_VIOMMU, representing a slice of the physical IOMMU instance, passed to or shared with a VM. It may be some HW-accelerated virtualization features and some SW resources used by the VM. For examples: @@ -109,6 +116,14 @@ Following IOMMUFD objects are exposed to userspace: vIOMMU, which is a separate ioctl call from attaching the same device to an HWPT_PAGING that the vIOMMU holds.
+- IOMMUFD_OBJ_VEVENTQ, representing a software queue for a vIOMMU to report its + events such as translation faults occurred to a nested stage-1 (excluding I/O + page faults that should go through IOMMUFD_OBJ_FAULT) and HW-specific events. + This queue object provides user space an FD to poll/read the vIOMMU events. A + vIOMMU object must be created first to get its viommu_id, which could be then + used to allocate a vEVENTQ. Each vIOMMU can support multiple types of vEVENTS, + but is confined to one vEVENTQ per vEVENTQ type. + All user-visible objects are destroyed via the IOMMU_DESTROY uAPI.
The diagrams below show relationships between user-visible objects and kernel @@ -251,8 +266,10 @@ User visible objects are backed by following datastructures: - iommufd_device for IOMMUFD_OBJ_DEVICE. - iommufd_hwpt_paging for IOMMUFD_OBJ_HWPT_PAGING. - iommufd_hwpt_nested for IOMMUFD_OBJ_HWPT_NESTED. +- iommufd_fault for IOMMUFD_OBJ_FAULT. - iommufd_viommu for IOMMUFD_OBJ_VIOMMU. - iommufd_vdevice for IOMMUFD_OBJ_VDEVICE. +- iommufd_veventq for IOMMUFD_OBJ_VEVENTQ.
Several terminologies when looking at these datastructures:
Use it to store all vSMMU-related data. The vsid (Virtual Stream ID) will be the first use case. Since the vsid reader will be the eventq handler that already holds a streams_mutex, reuse that to fenche the vmaster too.
Also add a pair of arm_smmu_attach_prepare/commit_vmaster helpers to set or unset the master->vmaster point. Put these helpers inside the existing arm_smmu_attach_prepare/commit().
For identity/blocked ops that don't call arm_smmu_attach_prepare/commit(), add a simpler arm_smmu_master_clear_vmaster helper to unset the vmaster.
Reviewed-by: Jason Gunthorpe jgg@nvidia.com Reviewed-by: Pranjal Shrivastavat praan@google.com Signed-off-by: Nicolin Chen nicolinc@nvidia.com --- drivers/iommu/arm/arm-smmu-v3/arm-smmu-v3.h | 28 ++++++++++++ .../arm/arm-smmu-v3/arm-smmu-v3-iommufd.c | 45 +++++++++++++++++++ drivers/iommu/arm/arm-smmu-v3/arm-smmu-v3.c | 18 +++++++- 3 files changed, 90 insertions(+), 1 deletion(-)
diff --git a/drivers/iommu/arm/arm-smmu-v3/arm-smmu-v3.h b/drivers/iommu/arm/arm-smmu-v3/arm-smmu-v3.h index bd9d7c85576a..36961a3579f2 100644 --- a/drivers/iommu/arm/arm-smmu-v3/arm-smmu-v3.h +++ b/drivers/iommu/arm/arm-smmu-v3/arm-smmu-v3.h @@ -799,6 +799,11 @@ struct arm_smmu_stream { struct rb_node node; };
+struct arm_smmu_vmaster { + struct arm_vsmmu *vsmmu; + unsigned long vsid; +}; + struct arm_smmu_event { u8 stall : 1, ssv : 1, @@ -824,6 +829,7 @@ struct arm_smmu_master { struct arm_smmu_device *smmu; struct device *dev; struct arm_smmu_stream *streams; + struct arm_smmu_vmaster *vmaster; /* use smmu->streams_mutex */ /* Locked by the iommu core using the group mutex */ struct arm_smmu_ctx_desc_cfg cd_table; unsigned int num_streams; @@ -972,6 +978,7 @@ struct arm_smmu_attach_state { bool disable_ats; ioasid_t ssid; /* Resulting state */ + struct arm_smmu_vmaster *vmaster; bool ats_enabled; };
@@ -1055,9 +1062,30 @@ struct iommufd_viommu *arm_vsmmu_alloc(struct device *dev, struct iommu_domain *parent, struct iommufd_ctx *ictx, unsigned int viommu_type); +int arm_smmu_attach_prepare_vmaster(struct arm_smmu_attach_state *state, + struct arm_smmu_nested_domain *nested_domain); +void arm_smmu_attach_commit_vmaster(struct arm_smmu_attach_state *state); +void arm_smmu_master_clear_vmaster(struct arm_smmu_master *master); #else #define arm_smmu_hw_info NULL #define arm_vsmmu_alloc NULL + +static inline int +arm_smmu_attach_prepare_vmaster(struct arm_smmu_attach_state *state, + struct arm_smmu_nested_domain *nested_domain) +{ + return 0; /* NOP */ +} + +static inline void +arm_smmu_attach_commit_vmaster(struct arm_smmu_attach_state *state) +{ +} + +static inline void +arm_smmu_master_clear_vmaster(struct arm_smmu_master *master) +{ +} #endif /* CONFIG_ARM_SMMU_V3_IOMMUFD */
#endif /* _ARM_SMMU_V3_H */ diff --git a/drivers/iommu/arm/arm-smmu-v3/arm-smmu-v3-iommufd.c b/drivers/iommu/arm/arm-smmu-v3/arm-smmu-v3-iommufd.c index 5aa2e7af58b4..6b712b1ab429 100644 --- a/drivers/iommu/arm/arm-smmu-v3/arm-smmu-v3-iommufd.c +++ b/drivers/iommu/arm/arm-smmu-v3/arm-smmu-v3-iommufd.c @@ -85,6 +85,51 @@ static void arm_smmu_make_nested_domain_ste( } }
+int arm_smmu_attach_prepare_vmaster(struct arm_smmu_attach_state *state, + struct arm_smmu_nested_domain *nested_domain) +{ + struct arm_smmu_vmaster *vmaster; + unsigned long vsid; + int ret; + + iommu_group_mutex_assert(state->master->dev); + + /* Skip invalid vSTE */ + if (!(nested_domain->ste[0] & cpu_to_le64(STRTAB_STE_0_V))) + return 0; + + ret = iommufd_viommu_get_vdev_id(&nested_domain->vsmmu->core, + state->master->dev, &vsid); + if (ret) + return ret; + + vmaster = kzalloc(sizeof(*vmaster), GFP_KERNEL); + if (!vmaster) + return -ENOMEM; + vmaster->vsmmu = nested_domain->vsmmu; + vmaster->vsid = vsid; + state->vmaster = vmaster; + + return 0; +} + +void arm_smmu_attach_commit_vmaster(struct arm_smmu_attach_state *state) +{ + struct arm_smmu_master *master = state->master; + + mutex_lock(&master->smmu->streams_mutex); + kfree(master->vmaster); + master->vmaster = state->vmaster; + mutex_unlock(&master->smmu->streams_mutex); +} + +void arm_smmu_master_clear_vmaster(struct arm_smmu_master *master) +{ + struct arm_smmu_attach_state state = { .master = master }; + + arm_smmu_attach_commit_vmaster(&state); +} + static int arm_smmu_attach_dev_nested(struct iommu_domain *domain, struct device *dev) { diff --git a/drivers/iommu/arm/arm-smmu-v3/arm-smmu-v3.c b/drivers/iommu/arm/arm-smmu-v3/arm-smmu-v3.c index 358072b4e293..964d2cf27d3d 100644 --- a/drivers/iommu/arm/arm-smmu-v3/arm-smmu-v3.c +++ b/drivers/iommu/arm/arm-smmu-v3/arm-smmu-v3.c @@ -2803,6 +2803,7 @@ int arm_smmu_attach_prepare(struct arm_smmu_attach_state *state, struct arm_smmu_domain *smmu_domain = to_smmu_domain_devices(new_domain); unsigned long flags; + int ret;
/* * arm_smmu_share_asid() must not see two domains pointing to the same @@ -2832,9 +2833,18 @@ int arm_smmu_attach_prepare(struct arm_smmu_attach_state *state, }
if (smmu_domain) { + if (new_domain->type == IOMMU_DOMAIN_NESTED) { + ret = arm_smmu_attach_prepare_vmaster( + state, to_smmu_nested_domain(new_domain)); + if (ret) + return ret; + } + master_domain = kzalloc(sizeof(*master_domain), GFP_KERNEL); - if (!master_domain) + if (!master_domain) { + kfree(state->vmaster); return -ENOMEM; + } master_domain->master = master; master_domain->ssid = state->ssid; if (new_domain->type == IOMMU_DOMAIN_NESTED) @@ -2861,6 +2871,7 @@ int arm_smmu_attach_prepare(struct arm_smmu_attach_state *state, spin_unlock_irqrestore(&smmu_domain->devices_lock, flags); kfree(master_domain); + kfree(state->vmaster); return -EINVAL; }
@@ -2893,6 +2904,8 @@ void arm_smmu_attach_commit(struct arm_smmu_attach_state *state)
lockdep_assert_held(&arm_smmu_asid_lock);
+ arm_smmu_attach_commit_vmaster(state); + if (state->ats_enabled && !master->ats_enabled) { arm_smmu_enable_ats(master); } else if (state->ats_enabled && master->ats_enabled) { @@ -3162,6 +3175,7 @@ static int arm_smmu_attach_dev_identity(struct iommu_domain *domain, struct arm_smmu_ste ste; struct arm_smmu_master *master = dev_iommu_priv_get(dev);
+ arm_smmu_master_clear_vmaster(master); arm_smmu_make_bypass_ste(master->smmu, &ste); arm_smmu_attach_dev_ste(domain, dev, &ste, STRTAB_STE_1_S1DSS_BYPASS); return 0; @@ -3180,7 +3194,9 @@ static int arm_smmu_attach_dev_blocked(struct iommu_domain *domain, struct device *dev) { struct arm_smmu_ste ste; + struct arm_smmu_master *master = dev_iommu_priv_get(dev);
+ arm_smmu_master_clear_vmaster(master); arm_smmu_make_abort_ste(&ste); arm_smmu_attach_dev_ste(domain, dev, &ste, STRTAB_STE_1_S1DSS_TERMINATE);
On Tue, Feb 25, 2025 at 09:25:40AM -0800, Nicolin Chen wrote:
Use it to store all vSMMU-related data. The vsid (Virtual Stream ID) will be the first use case. Since the vsid reader will be the eventq handler that already holds a streams_mutex, reuse that to fenche the vmaster too.
Also add a pair of arm_smmu_attach_prepare/commit_vmaster helpers to set or unset the master->vmaster point. Put these helpers inside the existing arm_smmu_attach_prepare/commit().
For identity/blocked ops that don't call arm_smmu_attach_prepare/commit(), add a simpler arm_smmu_master_clear_vmaster helper to unset the vmaster.
Reviewed-by: Jason Gunthorpe jgg@nvidia.com Reviewed-by: Pranjal Shrivastavat praan@google.com Signed-off-by: Nicolin Chen nicolinc@nvidia.com
drivers/iommu/arm/arm-smmu-v3/arm-smmu-v3.h | 28 ++++++++++++ .../arm/arm-smmu-v3/arm-smmu-v3-iommufd.c | 45 +++++++++++++++++++ drivers/iommu/arm/arm-smmu-v3/arm-smmu-v3.c | 18 +++++++- 3 files changed, 90 insertions(+), 1 deletion(-)
Apologies for my spelling error in [1]. It's supposed to be:
Reviewed-by: Pranjal Shrivastava praan@google.com
For all the other patches too. Correct spelling in [2].
Thanks, Praan
[1] https://lore.kernel.org/all/Z73zvIbsXzJMCaNt@google.com/ [2] https://lore.kernel.org/all/Z730M3XptvDRObBp@google.com/
Aside from the IOPF framework, iommufd provides an additional pathway to report hardware events, via the vEVENTQ of vIOMMU infrastructure.
Define an iommu_vevent_arm_smmuv3 uAPI structure, and report stage-1 events in the threaded IRQ handler. Also, add another four event record types that can be forwarded to a VM.
Reviewed-by: Kevin Tian kevin.tian@intel.com Reviewed-by: Jason Gunthorpe jgg@nvidia.com Reviewed-by: Pranjal Shrivastavat praan@google.com Signed-off-by: Nicolin Chen nicolinc@nvidia.com --- drivers/iommu/arm/arm-smmu-v3/arm-smmu-v3.h | 7 +++ include/uapi/linux/iommufd.h | 23 +++++++ .../arm/arm-smmu-v3/arm-smmu-v3-iommufd.c | 17 ++++++ drivers/iommu/arm/arm-smmu-v3/arm-smmu-v3.c | 60 +++++++++++-------- 4 files changed, 82 insertions(+), 25 deletions(-)
diff --git a/drivers/iommu/arm/arm-smmu-v3/arm-smmu-v3.h b/drivers/iommu/arm/arm-smmu-v3/arm-smmu-v3.h index 36961a3579f2..f3c5c49bf131 100644 --- a/drivers/iommu/arm/arm-smmu-v3/arm-smmu-v3.h +++ b/drivers/iommu/arm/arm-smmu-v3/arm-smmu-v3.h @@ -1066,6 +1066,7 @@ int arm_smmu_attach_prepare_vmaster(struct arm_smmu_attach_state *state, struct arm_smmu_nested_domain *nested_domain); void arm_smmu_attach_commit_vmaster(struct arm_smmu_attach_state *state); void arm_smmu_master_clear_vmaster(struct arm_smmu_master *master); +int arm_vmaster_report_event(struct arm_smmu_vmaster *vmaster, u64 *evt); #else #define arm_smmu_hw_info NULL #define arm_vsmmu_alloc NULL @@ -1086,6 +1087,12 @@ static inline void arm_smmu_master_clear_vmaster(struct arm_smmu_master *master) { } + +static inline int arm_vmaster_report_event(struct arm_smmu_vmaster *vmaster, + u64 *evt) +{ + return -EOPNOTSUPP; +} #endif /* CONFIG_ARM_SMMU_V3_IOMMUFD */
#endif /* _ARM_SMMU_V3_H */ diff --git a/include/uapi/linux/iommufd.h b/include/uapi/linux/iommufd.h index 2ade4839880d..5fc7e27804b7 100644 --- a/include/uapi/linux/iommufd.h +++ b/include/uapi/linux/iommufd.h @@ -1054,9 +1054,32 @@ struct iommufd_vevent_header { /** * enum iommu_veventq_type - Virtual Event Queue Type * @IOMMU_VEVENTQ_TYPE_DEFAULT: Reserved for future use + * @IOMMU_VEVENTQ_TYPE_ARM_SMMUV3: ARM SMMUv3 Virtual Event Queue */ enum iommu_veventq_type { IOMMU_VEVENTQ_TYPE_DEFAULT = 0, + IOMMU_VEVENTQ_TYPE_ARM_SMMUV3 = 1, +}; + +/** + * struct iommu_vevent_arm_smmuv3 - ARM SMMUv3 Virtual Event + * (IOMMU_VEVENTQ_TYPE_ARM_SMMUV3) + * @evt: 256-bit ARM SMMUv3 Event record, little-endian. + * Reported event records: (Refer to "7.3 Event records" in SMMUv3 HW Spec) + * - 0x04 C_BAD_STE + * - 0x06 F_STREAM_DISABLED + * - 0x08 C_BAD_SUBSTREAMID + * - 0x0a C_BAD_CD + * - 0x10 F_TRANSLATION + * - 0x11 F_ADDR_SIZE + * - 0x12 F_ACCESS + * - 0x13 F_PERMISSION + * + * StreamID field reports a virtual device ID. To receive a virtual event for a + * device, a vDEVICE must be allocated via IOMMU_VDEVICE_ALLOC. + */ +struct iommu_vevent_arm_smmuv3 { + __aligned_le64 evt[4]; };
/** diff --git a/drivers/iommu/arm/arm-smmu-v3/arm-smmu-v3-iommufd.c b/drivers/iommu/arm/arm-smmu-v3/arm-smmu-v3-iommufd.c index 6b712b1ab429..649e3aa39a48 100644 --- a/drivers/iommu/arm/arm-smmu-v3/arm-smmu-v3-iommufd.c +++ b/drivers/iommu/arm/arm-smmu-v3/arm-smmu-v3-iommufd.c @@ -437,4 +437,21 @@ struct iommufd_viommu *arm_vsmmu_alloc(struct device *dev, return &vsmmu->core; }
+int arm_vmaster_report_event(struct arm_smmu_vmaster *vmaster, u64 *evt) +{ + struct iommu_vevent_arm_smmuv3 vevt; + int i; + + lockdep_assert_held(&vmaster->vsmmu->smmu->streams_mutex); + + vevt.evt[0] = cpu_to_le64((evt[0] & ~EVTQ_0_SID) | + FIELD_PREP(EVTQ_0_SID, vmaster->vsid)); + for (i = 1; i < EVTQ_ENT_DWORDS; i++) + vevt.evt[i] = cpu_to_le64(evt[i]); + + return iommufd_viommu_report_event(&vmaster->vsmmu->core, + IOMMU_VEVENTQ_TYPE_ARM_SMMUV3, &vevt, + sizeof(vevt)); +} + MODULE_IMPORT_NS("IOMMUFD"); diff --git a/drivers/iommu/arm/arm-smmu-v3/arm-smmu-v3.c b/drivers/iommu/arm/arm-smmu-v3/arm-smmu-v3.c index 964d2cf27d3d..22aa5c8d1e9d 100644 --- a/drivers/iommu/arm/arm-smmu-v3/arm-smmu-v3.c +++ b/drivers/iommu/arm/arm-smmu-v3/arm-smmu-v3.c @@ -1813,8 +1813,8 @@ static void arm_smmu_decode_event(struct arm_smmu_device *smmu, u64 *raw, mutex_unlock(&smmu->streams_mutex); }
-static int arm_smmu_handle_event(struct arm_smmu_device *smmu, - struct arm_smmu_event *event) +static int arm_smmu_handle_event(struct arm_smmu_device *smmu, u64 *evt, + struct arm_smmu_event *event) { int ret = 0; u32 perm = 0; @@ -1823,6 +1823,10 @@ static int arm_smmu_handle_event(struct arm_smmu_device *smmu, struct iommu_fault *flt = &fault_evt.fault;
switch (event->id) { + case EVT_ID_BAD_STE_CONFIG: + case EVT_ID_STREAM_DISABLED_FAULT: + case EVT_ID_BAD_SUBSTREAMID_CONFIG: + case EVT_ID_BAD_CD_CONFIG: case EVT_ID_TRANSLATION_FAULT: case EVT_ID_ADDR_SIZE_FAULT: case EVT_ID_ACCESS_FAULT: @@ -1832,31 +1836,30 @@ static int arm_smmu_handle_event(struct arm_smmu_device *smmu, return -EOPNOTSUPP; }
- if (!event->stall) - return -EOPNOTSUPP; - - if (event->read) - perm |= IOMMU_FAULT_PERM_READ; - else - perm |= IOMMU_FAULT_PERM_WRITE; + if (event->stall) { + if (event->read) + perm |= IOMMU_FAULT_PERM_READ; + else + perm |= IOMMU_FAULT_PERM_WRITE;
- if (event->instruction) - perm |= IOMMU_FAULT_PERM_EXEC; + if (event->instruction) + perm |= IOMMU_FAULT_PERM_EXEC;
- if (event->privileged) - perm |= IOMMU_FAULT_PERM_PRIV; + if (event->privileged) + perm |= IOMMU_FAULT_PERM_PRIV;
- flt->type = IOMMU_FAULT_PAGE_REQ; - flt->prm = (struct iommu_fault_page_request) { - .flags = IOMMU_FAULT_PAGE_REQUEST_LAST_PAGE, - .grpid = event->stag, - .perm = perm, - .addr = event->iova, - }; + flt->type = IOMMU_FAULT_PAGE_REQ; + flt->prm = (struct iommu_fault_page_request){ + .flags = IOMMU_FAULT_PAGE_REQUEST_LAST_PAGE, + .grpid = event->stag, + .perm = perm, + .addr = event->iova, + };
- if (event->ssv) { - flt->prm.flags |= IOMMU_FAULT_PAGE_REQUEST_PASID_VALID; - flt->prm.pasid = event->ssid; + if (event->ssv) { + flt->prm.flags |= IOMMU_FAULT_PAGE_REQUEST_PASID_VALID; + flt->prm.pasid = event->ssid; + } }
mutex_lock(&smmu->streams_mutex); @@ -1866,7 +1869,14 @@ static int arm_smmu_handle_event(struct arm_smmu_device *smmu, goto out_unlock; }
- ret = iommu_report_device_fault(master->dev, &fault_evt); + if (event->stall) { + ret = iommu_report_device_fault(master->dev, &fault_evt); + } else { + if (master->vmaster && !event->s2) + ret = arm_vmaster_report_event(master->vmaster, evt); + else + ret = -EOPNOTSUPP; /* Unhandled events should be pinned */ + } out_unlock: mutex_unlock(&smmu->streams_mutex); return ret; @@ -1944,7 +1954,7 @@ static irqreturn_t arm_smmu_evtq_thread(int irq, void *dev) do { while (!queue_remove_raw(q, evt)) { arm_smmu_decode_event(smmu, evt, &event); - if (arm_smmu_handle_event(smmu, &event)) + if (arm_smmu_handle_event(smmu, evt, &event)) arm_smmu_dump_event(smmu, evt, &event, &rs);
put_device(event.dev);
On Tue, Feb 25, 2025 at 09:25:41AM -0800, Nicolin Chen wrote:
Aside from the IOPF framework, iommufd provides an additional pathway to report hardware events, via the vEVENTQ of vIOMMU infrastructure.
Define an iommu_vevent_arm_smmuv3 uAPI structure, and report stage-1 events in the threaded IRQ handler. Also, add another four event record types that can be forwarded to a VM.
Reviewed-by: Kevin Tian kevin.tian@intel.com Reviewed-by: Jason Gunthorpe jgg@nvidia.com Reviewed-by: Pranjal Shrivastavat praan@google.com Signed-off-by: Nicolin Chen nicolinc@nvidia.com
drivers/iommu/arm/arm-smmu-v3/arm-smmu-v3.h | 7 +++ include/uapi/linux/iommufd.h | 23 +++++++ .../arm/arm-smmu-v3/arm-smmu-v3-iommufd.c | 17 ++++++ drivers/iommu/arm/arm-smmu-v3/arm-smmu-v3.c | 60 +++++++++++-------- 4 files changed, 82 insertions(+), 25 deletions(-)
Apologies for my spelling error in [1]. It's supposed to be:
Reviewed-by: Pranjal Shrivastava praan@google.com
Correct spelling in [2].
Thanks, Praan
[1] https://lore.kernel.org/all/Z73zvIbsXzJMCaNt@google.com/ [2] https://lore.kernel.org/all/Z730M3XptvDRObBp@google.com/
There is a DoS concern on the shared hardware event queue among devices passed through to VMs, that too many translation failures that belong to VMs could overflow the shared hardware event queue if those VMs or their VMMs don't handle/recover the devices properly.
The MEV bit in the STE allows to configure the SMMU HW to merge similar event records, though there is no guarantee. Set it in a nested STE for DoS mitigations.
In the future, we might want to enable the MEV for non-nested cases too such as domain->type == IOMMU_DOMAIN_UNMANAGED or even IOMMU_DOMAIN_DMA.
Reviewed-by: Jason Gunthorpe jgg@nvidia.com Reviewed-by: Pranjal Shrivastavat praan@google.com Signed-off-by: Nicolin Chen nicolinc@nvidia.com --- drivers/iommu/arm/arm-smmu-v3/arm-smmu-v3.h | 1 + drivers/iommu/arm/arm-smmu-v3/arm-smmu-v3-iommufd.c | 2 ++ drivers/iommu/arm/arm-smmu-v3/arm-smmu-v3.c | 4 ++-- 3 files changed, 5 insertions(+), 2 deletions(-)
diff --git a/drivers/iommu/arm/arm-smmu-v3/arm-smmu-v3.h b/drivers/iommu/arm/arm-smmu-v3/arm-smmu-v3.h index f3c5c49bf131..bc4f536f72ce 100644 --- a/drivers/iommu/arm/arm-smmu-v3/arm-smmu-v3.h +++ b/drivers/iommu/arm/arm-smmu-v3/arm-smmu-v3.h @@ -266,6 +266,7 @@ static inline u32 arm_smmu_strtab_l2_idx(u32 sid) #define STRTAB_STE_1_S1COR GENMASK_ULL(5, 4) #define STRTAB_STE_1_S1CSH GENMASK_ULL(7, 6)
+#define STRTAB_STE_1_MEV (1UL << 19) #define STRTAB_STE_1_S2FWB (1UL << 25) #define STRTAB_STE_1_S1STALLD (1UL << 27)
diff --git a/drivers/iommu/arm/arm-smmu-v3/arm-smmu-v3-iommufd.c b/drivers/iommu/arm/arm-smmu-v3/arm-smmu-v3-iommufd.c index 649e3aa39a48..8e8ea3702ce5 100644 --- a/drivers/iommu/arm/arm-smmu-v3/arm-smmu-v3-iommufd.c +++ b/drivers/iommu/arm/arm-smmu-v3/arm-smmu-v3-iommufd.c @@ -43,6 +43,8 @@ static void arm_smmu_make_nested_cd_table_ste( target->data[0] |= nested_domain->ste[0] & ~cpu_to_le64(STRTAB_STE_0_CFG); target->data[1] |= nested_domain->ste[1]; + /* Merge events for DoS mitigations on eventq */ + target->data[1] |= cpu_to_le64(STRTAB_STE_1_MEV); }
/* diff --git a/drivers/iommu/arm/arm-smmu-v3/arm-smmu-v3.c b/drivers/iommu/arm/arm-smmu-v3/arm-smmu-v3.c index 22aa5c8d1e9d..3fcb1089a7c7 100644 --- a/drivers/iommu/arm/arm-smmu-v3/arm-smmu-v3.c +++ b/drivers/iommu/arm/arm-smmu-v3/arm-smmu-v3.c @@ -1052,7 +1052,7 @@ void arm_smmu_get_ste_used(const __le64 *ent, __le64 *used_bits) cpu_to_le64(STRTAB_STE_1_S1DSS | STRTAB_STE_1_S1CIR | STRTAB_STE_1_S1COR | STRTAB_STE_1_S1CSH | STRTAB_STE_1_S1STALLD | STRTAB_STE_1_STRW | - STRTAB_STE_1_EATS); + STRTAB_STE_1_EATS | STRTAB_STE_1_MEV); used_bits[2] |= cpu_to_le64(STRTAB_STE_2_S2VMID);
/* @@ -1068,7 +1068,7 @@ void arm_smmu_get_ste_used(const __le64 *ent, __le64 *used_bits) if (cfg & BIT(1)) { used_bits[1] |= cpu_to_le64(STRTAB_STE_1_S2FWB | STRTAB_STE_1_EATS | - STRTAB_STE_1_SHCFG); + STRTAB_STE_1_SHCFG | STRTAB_STE_1_MEV); used_bits[2] |= cpu_to_le64(STRTAB_STE_2_S2VMID | STRTAB_STE_2_VTCR | STRTAB_STE_2_S2AA64 | STRTAB_STE_2_S2ENDI |
On Tue, Feb 25, 2025 at 09:25:42AM -0800, Nicolin Chen wrote:
There is a DoS concern on the shared hardware event queue among devices passed through to VMs, that too many translation failures that belong to VMs could overflow the shared hardware event queue if those VMs or their VMMs don't handle/recover the devices properly.
The MEV bit in the STE allows to configure the SMMU HW to merge similar event records, though there is no guarantee. Set it in a nested STE for DoS mitigations.
In the future, we might want to enable the MEV for non-nested cases too such as domain->type == IOMMU_DOMAIN_UNMANAGED or even IOMMU_DOMAIN_DMA.
Reviewed-by: Jason Gunthorpe jgg@nvidia.com Reviewed-by: Pranjal Shrivastavat praan@google.com Signed-off-by: Nicolin Chen nicolinc@nvidia.com
drivers/iommu/arm/arm-smmu-v3/arm-smmu-v3.h | 1 + drivers/iommu/arm/arm-smmu-v3/arm-smmu-v3-iommufd.c | 2 ++ drivers/iommu/arm/arm-smmu-v3/arm-smmu-v3.c | 4 ++-- 3 files changed, 5 insertions(+), 2 deletions(-)
Apologies for my spelling error in [1]. It's supposed to be:
Reviewed-by: Pranjal Shrivastava praan@google.com
Correct spelling in [2].
Thanks, Praan
[1] https://lore.kernel.org/all/Z73zvIbsXzJMCaNt@google.com/ [2] https://lore.kernel.org/all/Z730M3XptvDRObBp@google.com/
On Wed, 26 Feb 2025 at 01:33, Nicolin Chen nicolinc@nvidia.com wrote:
As the vIOMMU infrastructure series part-3, this introduces a new vEVENTQ object. The existing FAULT object provides a nice notification pathway to the user space with a queue already, so let vEVENTQ reuse that.
Mimicing the HWPT structure, add a common EVENTQ structure to support its derivatives: IOMMUFD_OBJ_FAULT (existing) and IOMMUFD_OBJ_VEVENTQ (new).
An IOMMUFD_CMD_VEVENTQ_ALLOC is introduced to allocate vEVENTQ object for vIOMMUs. One vIOMMU can have multiple vEVENTQs in different types but can not support multiple vEVENTQs in the same type.
The forwarding part is fairly simple but might need to replace a physical device ID with a virtual device ID in a driver-level event data structure. So, this also adds some helpers for drivers to use.
As usual, this series comes with the selftest coverage for this new ioctl and with a real world use case in the ARM SMMUv3 driver.
This is on Github: https://github.com/nicolinc/iommufd/commits/iommufd_veventq-v8 Paring QEMU branch for testing: https://github.com/nicolinc/qemu/commits/wip/for_iommufd_veventq-v8
Thanks Nico
Tested-by: Zhangfei Gao zhangfei.gao@linaro.org
After rebase on your branch https://github.com/Linaro/linux-kernel-warpdrive/tree/iommufd_veventq-v8 https://github.com/Linaro/qemu/tree/for_iommufd_veventq-v8
Tested with multi-device in guests, with io page faults happen.
Thanks
On Sat, Mar 01, 2025 at 12:16:00PM +0800, Zhangfei Gao wrote:
On Wed, 26 Feb 2025 at 01:33, Nicolin Chen nicolinc@nvidia.com wrote:
As the vIOMMU infrastructure series part-3, this introduces a new vEVENTQ object. The existing FAULT object provides a nice notification pathway to the user space with a queue already, so let vEVENTQ reuse that.
Mimicing the HWPT structure, add a common EVENTQ structure to support its derivatives: IOMMUFD_OBJ_FAULT (existing) and IOMMUFD_OBJ_VEVENTQ (new).
An IOMMUFD_CMD_VEVENTQ_ALLOC is introduced to allocate vEVENTQ object for vIOMMUs. One vIOMMU can have multiple vEVENTQs in different types but can not support multiple vEVENTQs in the same type.
The forwarding part is fairly simple but might need to replace a physical device ID with a virtual device ID in a driver-level event data structure. So, this also adds some helpers for drivers to use.
As usual, this series comes with the selftest coverage for this new ioctl and with a real world use case in the ARM SMMUv3 driver.
This is on Github: https://github.com/nicolinc/iommufd/commits/iommufd_veventq-v8 Paring QEMU branch for testing: https://github.com/nicolinc/qemu/commits/wip/for_iommufd_veventq-v8
Thanks Nico
Tested-by: Zhangfei Gao zhangfei.gao@linaro.org
After rebase on your branch https://github.com/Linaro/linux-kernel-warpdrive/tree/iommufd_veventq-v8 https://github.com/Linaro/qemu/tree/for_iommufd_veventq-v8
Tested with multi-device in guests, with io page faults happen.
Thanks for testing!
Nicolin
linux-kselftest-mirror@lists.linaro.org
-
Baolu Lu -
Nicolin Chen -
Pranjal Shrivastava -
Zhangfei Gao