diff options
| author | Mike Marciniszyn <mike.marciniszyn@intel.com> | 2016-10-10 06:14:28 -0700 |
|---|---|---|
| committer | Doug Ledford <dledford@redhat.com> | 2016-11-15 16:25:59 -0500 |
| commit | 4e045572e2c2be674ed7e43cca7ca105e8a22f56 (patch) | |
| tree | 9ebe69a7c48b530e40a11a95f6bafa82abe8254e /drivers/infiniband/hw/hfi1/iowait.h | |
| parent | a909d3e636995ba7c349e2ca5dbb528154d4ac30 (diff) | |
| download | linux-4e045572e2c2be674ed7e43cca7ca105e8a22f56.tar.bz2 | |
IB/hfi1: Add unique txwait_lock for txreq events
Profiling suggests that the read_seqbegin() in
the txreq put logic is colliding with other uses
of the iowait lock.
The packet at a time use of this lock dictates a unique
lock to avoid reader/writer collisions when the number
of vTxWait events is low.
In order to support a unique lock the iowait struct embedded
in the QP is extended to remember the lock that protects the queue
head.
The QP destroy removes that QP from any wait list. It doesn't
need to know the head because of the linked list API, but it does
need to know the lock required to protect the head.
This also opens up the wait logic to have unique per resources locks
which needs to be in future refinement.
Reviewed-by: Sebastian Sanchez <sebastian.sanchez@intel.com>
Signed-off-by: Mike Marciniszyn <mike.marciniszyn@intel.com>
Signed-off-by: Dennis Dalessandro <dennis.dalessandro@intel.com>
Signed-off-by: Doug Ledford <dledford@redhat.com>
Diffstat (limited to 'drivers/infiniband/hw/hfi1/iowait.h')
| -rw-r--r-- | drivers/infiniband/hw/hfi1/iowait.h | 8 |
1 files changed, 8 insertions, 0 deletions
diff --git a/drivers/infiniband/hw/hfi1/iowait.h b/drivers/infiniband/hw/hfi1/iowait.h index 2ec6ef38d389..d9740ddea6f1 100644 --- a/drivers/infiniband/hw/hfi1/iowait.h +++ b/drivers/infiniband/hw/hfi1/iowait.h @@ -64,6 +64,7 @@ struct sdma_engine; /** * struct iowait - linkage for delayed progress/waiting * @list: used to add/insert into QP/PQ wait lists + * @lock: uses to record the list head lock * @tx_head: overflow list of sdma_txreq's * @sleep: no space callback * @wakeup: space callback wakeup @@ -91,6 +92,11 @@ struct sdma_engine; * so sleeping is not allowed. * * The wait_dma member along with the iow + * + * The lock field is used by waiters to record + * the seqlock_t that guards the list head. + * Waiters explicity know that, but the destroy + * code that unwaits QPs does not. */ struct iowait { @@ -103,6 +109,7 @@ struct iowait { unsigned seq); void (*wakeup)(struct iowait *wait, int reason); void (*sdma_drained)(struct iowait *wait); + seqlock_t *lock; struct work_struct iowork; wait_queue_head_t wait_dma; wait_queue_head_t wait_pio; @@ -141,6 +148,7 @@ static inline void iowait_init( void (*sdma_drained)(struct iowait *wait)) { wait->count = 0; + wait->lock = NULL; INIT_LIST_HEAD(&wait->list); INIT_LIST_HEAD(&wait->tx_head); INIT_WORK(&wait->iowork, func); |