summaryrefslogtreecommitdiffstats
path: root/drivers/infiniband/sw/rxe/rxe_queue.h
diff options
context:
space:
mode:
authorBob Pearson <rpearsonhpe@gmail.com>2021-09-14 11:42:03 -0500
committerJason Gunthorpe <jgg@nvidia.com>2021-09-24 10:14:59 -0300
commitae6e843fe08d0ea8e158815809dcc20e3a1afc22 (patch)
tree723ed25b8a0edfde968f72ace5f837daf880e61d /drivers/infiniband/sw/rxe/rxe_queue.h
parent6bda39149d4b8920fdb8744090653aca3daa792d (diff)
downloadlinux-ae6e843fe08d0ea8e158815809dcc20e3a1afc22.tar.bz2
RDMA/rxe: Add memory barriers to kernel queues
Earlier patches added memory barriers to protect user space to kernel space communications. The user space queues were previously shown to have occasional memory synchonization errors which were removed by adding smp_load_acquire, smp_store_release barriers. This patch extends that to the case where queues are used between kernel space threads. This patch also extends the queue types to include kernel ULP queues which access the other end of the queues in kernel verbs calls like poll_cq and post_send/recv. Link: https://lore.kernel.org/r/20210914164206.19768-2-rpearsonhpe@gmail.com Signed-off-by: Bob Pearson <rpearsonhpe@gmail.com> Signed-off-by: Jason Gunthorpe <jgg@nvidia.com>
Diffstat (limited to 'drivers/infiniband/sw/rxe/rxe_queue.h')
-rw-r--r--drivers/infiniband/sw/rxe/rxe_queue.h292
1 files changed, 119 insertions, 173 deletions
diff --git a/drivers/infiniband/sw/rxe/rxe_queue.h b/drivers/infiniband/sw/rxe/rxe_queue.h
index 2702b0e55fc3..6227112ef7a2 100644
--- a/drivers/infiniband/sw/rxe/rxe_queue.h
+++ b/drivers/infiniband/sw/rxe/rxe_queue.h
@@ -10,34 +10,47 @@
/* for definition of shared struct rxe_queue_buf */
#include <uapi/rdma/rdma_user_rxe.h>
-/* implements a simple circular buffer that can optionally be
- * shared between user space and the kernel and can be resized
- * the requested element size is rounded up to a power of 2
- * and the number of elements in the buffer is also rounded
- * up to a power of 2. Since the queue is empty when the
- * producer and consumer indices match the maximum capacity
- * of the queue is one less than the number of element slots
+/* Implements a simple circular buffer that is shared between user
+ * and the driver and can be resized. The requested element size is
+ * rounded up to a power of 2 and the number of elements in the buffer
+ * is also rounded up to a power of 2. Since the queue is empty when
+ * the producer and consumer indices match the maximum capacity of the
+ * queue is one less than the number of element slots.
*
* Notes:
- * - Kernel space indices are always masked off to q->index_mask
- * before storing so do not need to be checked on reads.
- * - User space indices may be out of range and must be
- * masked before use when read.
- * - The kernel indices for shared queues must not be written
- * by user space so a local copy is used and a shared copy is
- * stored when the local copy changes.
+ * - The driver indices are always masked off to q->index_mask
+ * before storing so do not need to be checked on reads.
+ * - The user whether user space or kernel is generally
+ * not trusted so its parameters are masked to make sure
+ * they do not access the queue out of bounds on reads.
+ * - The driver indices for queues must not be written
+ * by user so a local copy is used and a shared copy is
+ * stored when the local copy is changed.
* - By passing the type in the parameter list separate from q
- * the compiler can eliminate the switch statement when the
- * actual queue type is known when the function is called.
- * In the performance path this is done. In less critical
- * paths just q->type is passed.
+ * the compiler can eliminate the switch statement when the
+ * actual queue type is known when the function is called at
+ * compile time.
+ * - These queues are lock free. The user and driver must protect
+ * changes to their end of the queues with locks if more than one
+ * CPU can be accessing it at the same time.
*/
-/* type of queue */
+/**
+ * enum queue_type - type of queue
+ * @QUEUE_TYPE_TO_CLIENT: Queue is written by rxe driver and
+ * read by client. Used by rxe driver only.
+ * @QUEUE_TYPE_FROM_CLIENT: Queue is written by client and
+ * read by rxe driver. Used by rxe driver only.
+ * @QUEUE_TYPE_TO_DRIVER: Queue is written by client and
+ * read by rxe driver. Used by kernel client only.
+ * @QUEUE_TYPE_FROM_DRIVER: Queue is written by rxe driver and
+ * read by client. Used by kernel client only.
+ */
enum queue_type {
- QUEUE_TYPE_KERNEL,
- QUEUE_TYPE_TO_USER,
- QUEUE_TYPE_FROM_USER,
+ QUEUE_TYPE_TO_CLIENT,
+ QUEUE_TYPE_FROM_CLIENT,
+ QUEUE_TYPE_TO_DRIVER,
+ QUEUE_TYPE_FROM_DRIVER,
};
struct rxe_queue {
@@ -69,238 +82,171 @@ struct rxe_queue *rxe_queue_init(struct rxe_dev *rxe, int *num_elem,
int rxe_queue_resize(struct rxe_queue *q, unsigned int *num_elem_p,
unsigned int elem_size, struct ib_udata *udata,
struct mminfo __user *outbuf,
- /* Protect producers while resizing queue */
- spinlock_t *producer_lock,
- /* Protect consumers while resizing queue */
- spinlock_t *consumer_lock);
+ spinlock_t *producer_lock, spinlock_t *consumer_lock);
void rxe_queue_cleanup(struct rxe_queue *queue);
-static inline int next_index(struct rxe_queue *q, int index)
+static inline u32 queue_next_index(struct rxe_queue *q, int index)
{
- return (index + 1) & q->buf->index_mask;
+ return (index + 1) & q->index_mask;
}
-static inline int queue_empty(struct rxe_queue *q, enum queue_type type)
+static inline u32 queue_get_producer(const struct rxe_queue *q,
+ enum queue_type type)
{
u32 prod;
- u32 cons;
switch (type) {
- case QUEUE_TYPE_FROM_USER:
- /* protect user space index */
+ case QUEUE_TYPE_FROM_CLIENT:
+ /* protect user index */
prod = smp_load_acquire(&q->buf->producer_index);
- cons = q->index;
break;
- case QUEUE_TYPE_TO_USER:
+ case QUEUE_TYPE_TO_CLIENT:
prod = q->index;
- /* protect user space index */
- cons = smp_load_acquire(&q->buf->consumer_index);
break;
- case QUEUE_TYPE_KERNEL:
+ case QUEUE_TYPE_FROM_DRIVER:
+ /* protect driver index */
+ prod = smp_load_acquire(&q->buf->producer_index);
+ break;
+ case QUEUE_TYPE_TO_DRIVER:
prod = q->buf->producer_index;
- cons = q->buf->consumer_index;
break;
}
- return ((prod - cons) & q->index_mask) == 0;
+ return prod;
}
-static inline int queue_full(struct rxe_queue *q, enum queue_type type)
+static inline u32 queue_get_consumer(const struct rxe_queue *q,
+ enum queue_type type)
{
- u32 prod;
u32 cons;
switch (type) {
- case QUEUE_TYPE_FROM_USER:
- /* protect user space index */
- prod = smp_load_acquire(&q->buf->producer_index);
+ case QUEUE_TYPE_FROM_CLIENT:
cons = q->index;
break;
- case QUEUE_TYPE_TO_USER:
- prod = q->index;
- /* protect user space index */
+ case QUEUE_TYPE_TO_CLIENT:
+ /* protect user index */
cons = smp_load_acquire(&q->buf->consumer_index);
break;
- case QUEUE_TYPE_KERNEL:
- prod = q->buf->producer_index;
+ case QUEUE_TYPE_FROM_DRIVER:
cons = q->buf->consumer_index;
break;
+ case QUEUE_TYPE_TO_DRIVER:
+ /* protect driver index */
+ cons = smp_load_acquire(&q->buf->consumer_index);
+ break;
}
- return ((prod + 1 - cons) & q->index_mask) == 0;
+ return cons;
}
-static inline unsigned int queue_count(const struct rxe_queue *q,
- enum queue_type type)
+static inline int queue_empty(struct rxe_queue *q, enum queue_type type)
{
- u32 prod;
- u32 cons;
-
- switch (type) {
- case QUEUE_TYPE_FROM_USER:
- /* protect user space index */
- prod = smp_load_acquire(&q->buf->producer_index);
- cons = q->index;
- break;
- case QUEUE_TYPE_TO_USER:
- prod = q->index;
- /* protect user space index */
- cons = smp_load_acquire(&q->buf->consumer_index);
- break;
- case QUEUE_TYPE_KERNEL:
- prod = q->buf->producer_index;
- cons = q->buf->consumer_index;
- break;
- }
+ u32 prod = queue_get_producer(q, type);
+ u32 cons = queue_get_consumer(q, type);
- return (prod - cons) & q->index_mask;
+ return ((prod - cons) & q->index_mask) == 0;
}
-static inline void advance_producer(struct rxe_queue *q, enum queue_type type)
+static inline int queue_full(struct rxe_queue *q, enum queue_type type)
{
- u32 prod;
+ u32 prod = queue_get_producer(q, type);
+ u32 cons = queue_get_consumer(q, type);
- switch (type) {
- case QUEUE_TYPE_FROM_USER:
- pr_warn_once("Normally kernel should not write user space index\n");
- /* protect user space index */
- prod = smp_load_acquire(&q->buf->producer_index);
- prod = (prod + 1) & q->index_mask;
- /* same */
- smp_store_release(&q->buf->producer_index, prod);
- break;
- case QUEUE_TYPE_TO_USER:
- prod = q->index;
- q->index = (prod + 1) & q->index_mask;
- q->buf->producer_index = q->index;
- break;
- case QUEUE_TYPE_KERNEL:
- prod = q->buf->producer_index;
- q->buf->producer_index = (prod + 1) & q->index_mask;
- break;
- }
+ return ((prod + 1 - cons) & q->index_mask) == 0;
}
-static inline void advance_consumer(struct rxe_queue *q, enum queue_type type)
+static inline u32 queue_count(const struct rxe_queue *q,
+ enum queue_type type)
{
- u32 cons;
+ u32 prod = queue_get_producer(q, type);
+ u32 cons = queue_get_consumer(q, type);
- switch (type) {
- case QUEUE_TYPE_FROM_USER:
- cons = q->index;
- q->index = (cons + 1) & q->index_mask;
- q->buf->consumer_index = q->index;
- break;
- case QUEUE_TYPE_TO_USER:
- pr_warn_once("Normally kernel should not write user space index\n");
- /* protect user space index */
- cons = smp_load_acquire(&q->buf->consumer_index);
- cons = (cons + 1) & q->index_mask;
- /* same */
- smp_store_release(&q->buf->consumer_index, cons);
- break;
- case QUEUE_TYPE_KERNEL:
- cons = q->buf->consumer_index;
- q->buf->consumer_index = (cons + 1) & q->index_mask;
- break;
- }
+ return (prod - cons) & q->index_mask;
}
-static inline void *producer_addr(struct rxe_queue *q, enum queue_type type)
+static inline void queue_advance_producer(struct rxe_queue *q,
+ enum queue_type type)
{
u32 prod;
switch (type) {
- case QUEUE_TYPE_FROM_USER:
- /* protect user space index */
- prod = smp_load_acquire(&q->buf->producer_index);
- prod &= q->index_mask;
+ case QUEUE_TYPE_FROM_CLIENT:
+ pr_warn("%s: attempt to advance client index\n",
+ __func__);
break;
- case QUEUE_TYPE_TO_USER:
+ case QUEUE_TYPE_TO_CLIENT:
prod = q->index;
+ prod = (prod + 1) & q->index_mask;
+ q->index = prod;
+ /* protect user index */
+ smp_store_release(&q->buf->producer_index, prod);
+ break;
+ case QUEUE_TYPE_FROM_DRIVER:
+ pr_warn("%s: attempt to advance driver index\n",
+ __func__);
break;
- case QUEUE_TYPE_KERNEL:
+ case QUEUE_TYPE_TO_DRIVER:
prod = q->buf->producer_index;
+ prod = (prod + 1) & q->index_mask;
+ q->buf->producer_index = prod;
break;
}
-
- return q->buf->data + (prod << q->log2_elem_size);
}
-static inline void *consumer_addr(struct rxe_queue *q, enum queue_type type)
+static inline void queue_advance_consumer(struct rxe_queue *q,
+ enum queue_type type)
{
u32 cons;
switch (type) {
- case QUEUE_TYPE_FROM_USER:
+ case QUEUE_TYPE_FROM_CLIENT:
cons = q->index;
+ cons = (cons + 1) & q->index_mask;
+ q->index = cons;
+ /* protect user index */
+ smp_store_release(&q->buf->consumer_index, cons);
break;
- case QUEUE_TYPE_TO_USER:
- /* protect user space index */
- cons = smp_load_acquire(&q->buf->consumer_index);
- cons &= q->index_mask;
+ case QUEUE_TYPE_TO_CLIENT:
+ pr_warn("%s: attempt to advance client index\n",
+ __func__);
break;
- case QUEUE_TYPE_KERNEL:
+ case QUEUE_TYPE_FROM_DRIVER:
cons = q->buf->consumer_index;
+ cons = (cons + 1) & q->index_mask;
+ q->buf->consumer_index = cons;
+ break;
+ case QUEUE_TYPE_TO_DRIVER:
+ pr_warn("%s: attempt to advance driver index\n",
+ __func__);
break;
}
-
- return q->buf->data + (cons << q->log2_elem_size);
}
-static inline unsigned int producer_index(struct rxe_queue *q,
- enum queue_type type)
+static inline void *queue_producer_addr(struct rxe_queue *q,
+ enum queue_type type)
{
- u32 prod;
+ u32 prod = queue_get_producer(q, type);
- switch (type) {
- case QUEUE_TYPE_FROM_USER:
- /* protect user space index */
- prod = smp_load_acquire(&q->buf->producer_index);
- prod &= q->index_mask;
- break;
- case QUEUE_TYPE_TO_USER:
- prod = q->index;
- break;
- case QUEUE_TYPE_KERNEL:
- prod = q->buf->producer_index;
- break;
- }
-
- return prod;
+ return q->buf->data + (prod << q->log2_elem_size);
}
-static inline unsigned int consumer_index(struct rxe_queue *q,
- enum queue_type type)
+static inline void *queue_consumer_addr(struct rxe_queue *q,
+ enum queue_type type)
{
- u32 cons;
-
- switch (type) {
- case QUEUE_TYPE_FROM_USER:
- cons = q->index;
- break;
- case QUEUE_TYPE_TO_USER:
- /* protect user space index */
- cons = smp_load_acquire(&q->buf->consumer_index);
- cons &= q->index_mask;
- break;
- case QUEUE_TYPE_KERNEL:
- cons = q->buf->consumer_index;
- break;
- }
+ u32 cons = queue_get_consumer(q, type);
- return cons;
+ return q->buf->data + (cons << q->log2_elem_size);
}
-static inline void *addr_from_index(struct rxe_queue *q,
- unsigned int index)
+static inline void *queue_addr_from_index(struct rxe_queue *q, u32 index)
{
return q->buf->data + ((index & q->index_mask)
- << q->buf->log2_elem_size);
+ << q->log2_elem_size);
}
-static inline unsigned int index_from_addr(const struct rxe_queue *q,
+static inline u32 queue_index_from_addr(const struct rxe_queue *q,
const void *addr)
{
return (((u8 *)addr - q->buf->data) >> q->log2_elem_size)
@@ -309,7 +255,7 @@ static inline unsigned int index_from_addr(const struct rxe_queue *q,
static inline void *queue_head(struct rxe_queue *q, enum queue_type type)
{
- return queue_empty(q, type) ? NULL : consumer_addr(q, type);
+ return queue_empty(q, type) ? NULL : queue_consumer_addr(q, type);
}
#endif /* RXE_QUEUE_H */