IB/qib: Improve SDMA performance
1. The code accepts chunks of messages, and splits the chunk into
packets when converting packets into sdma queue entries. Adjacent
packets will use user buffer pages smartly to avoid pinning the
same page multiple times.
2. Instead of discarding all the work when SDMA queue is full, the
work is saved in a pending queue. Whenever there are enough SDMA
queue free entries, pending queue is directly put onto SDMA queue.
3. An interrupt handler is used to progress this pending queue.
Reviewed-by: Mike Marciniszyn <mike.marciniszyn@intel.com>
Signed-off-by: CQ Tang <cq.tang@intel.com>
Signed-off-by: Mike Marciniszyn <mike.marciniszyn@intel.com>
[ Fixed up sparse warnings. - Roland ]
Signed-off-by: Roland Dreier <roland@purestorage.com>
diff --git a/drivers/infiniband/hw/qib/qib.h b/drivers/infiniband/hw/qib/qib.h
index 4a9af79..ae3e4fe 100644
--- a/drivers/infiniband/hw/qib/qib.h
+++ b/drivers/infiniband/hw/qib/qib.h
@@ -576,11 +576,13 @@
/* read/write using lock */
spinlock_t sdma_lock ____cacheline_aligned_in_smp;
struct list_head sdma_activelist;
+ struct list_head sdma_userpending;
u64 sdma_descq_added;
u64 sdma_descq_removed;
u16 sdma_descq_tail;
u16 sdma_descq_head;
u8 sdma_generation;
+ u8 sdma_intrequest;
struct tasklet_struct sdma_sw_clean_up_task
____cacheline_aligned_in_smp;
@@ -1326,6 +1328,8 @@
void qib_teardown_sdma(struct qib_pportdata *);
void __qib_sdma_intr(struct qib_pportdata *);
void qib_sdma_intr(struct qib_pportdata *);
+void qib_user_sdma_send_desc(struct qib_pportdata *dd,
+ struct list_head *pktlist);
int qib_sdma_verbs_send(struct qib_pportdata *, struct qib_sge_state *,
u32, struct qib_verbs_txreq *);
/* ppd->sdma_lock should be locked before calling this. */