isci: cleanup/optimize pool implementation
The circ_buf macros are ~6% faster, as measured by perf, because they take
advantage of power-of-two math assumptions i.e. no test and branch for
rollover. Their semantics are clearer than the hidden side effects in pool.h
(like sci_pool_get() which hides an assignment).
Signed-off-by: Dan Williams <dan.j.williams@intel.com>
diff --git a/drivers/scsi/isci/host.c b/drivers/scsi/isci/host.c
index 41a7c50..343655b 100644
--- a/drivers/scsi/isci/host.c
+++ b/drivers/scsi/isci/host.c
@@ -52,6 +52,7 @@
* (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
* OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
*/
+#include <linux/circ_buf.h>
#include <linux/device.h>
#include <scsi/sas.h>
#include "host.h"
@@ -1054,6 +1055,33 @@
spin_unlock_irqrestore(&ihost->scic_lock, flags);
}
+static void isci_tci_free(struct isci_host *ihost, u16 tci)
+{
+ u16 tail = ihost->tci_tail & (SCI_MAX_IO_REQUESTS-1);
+
+ ihost->tci_pool[tail] = tci;
+ ihost->tci_tail = tail + 1;
+}
+
+static u16 isci_tci_alloc(struct isci_host *ihost)
+{
+ u16 head = ihost->tci_head & (SCI_MAX_IO_REQUESTS-1);
+ u16 tci = ihost->tci_pool[head];
+
+ ihost->tci_head = head + 1;
+ return tci;
+}
+
+static u16 isci_tci_active(struct isci_host *ihost)
+{
+ return CIRC_CNT(ihost->tci_head, ihost->tci_tail, SCI_MAX_IO_REQUESTS);
+}
+
+static u16 isci_tci_space(struct isci_host *ihost)
+{
+ return CIRC_SPACE(ihost->tci_head, ihost->tci_tail, SCI_MAX_IO_REQUESTS);
+}
+
static enum sci_status scic_controller_start(struct scic_sds_controller *scic,
u32 timeout)
{
@@ -1069,9 +1097,11 @@
}
/* Build the TCi free pool */
- sci_pool_initialize(scic->tci_pool);
+ BUILD_BUG_ON(SCI_MAX_IO_REQUESTS > 1 << sizeof(ihost->tci_pool[0]) * 8);
+ ihost->tci_head = 0;
+ ihost->tci_tail = 0;
for (index = 0; index < scic->task_context_entries; index++)
- sci_pool_put(scic->tci_pool, index);
+ isci_tci_free(ihost, index);
/* Build the RNi free pool */
scic_sds_remote_node_table_initialize(
@@ -3063,18 +3093,17 @@
* currently available tags to be allocated. All return other values indicate a
* legitimate tag.
*/
-u16 scic_controller_allocate_io_tag(
- struct scic_sds_controller *scic)
+u16 scic_controller_allocate_io_tag(struct scic_sds_controller *scic)
{
- u16 task_context;
- u16 sequence_count;
+ struct isci_host *ihost = scic_to_ihost(scic);
+ u16 tci;
+ u16 seq;
- if (!sci_pool_empty(scic->tci_pool)) {
- sci_pool_get(scic->tci_pool, task_context);
+ if (isci_tci_space(ihost)) {
+ tci = isci_tci_alloc(ihost);
+ seq = scic->io_request_sequence[tci];
- sequence_count = scic->io_request_sequence[task_context];
-
- return scic_sds_io_tag_construct(sequence_count, task_context);
+ return scic_sds_io_tag_construct(seq, tci);
}
return SCI_CONTROLLER_INVALID_IO_TAG;
@@ -3105,10 +3134,10 @@
* tags. SCI_FAILURE_INVALID_IO_TAG This value is returned if the supplied tag
* is not a valid IO tag value.
*/
-enum sci_status scic_controller_free_io_tag(
- struct scic_sds_controller *scic,
- u16 io_tag)
+enum sci_status scic_controller_free_io_tag(struct scic_sds_controller *scic,
+ u16 io_tag)
{
+ struct isci_host *ihost = scic_to_ihost(scic);
u16 sequence;
u16 index;
@@ -3117,18 +3146,16 @@
sequence = scic_sds_io_tag_get_sequence(io_tag);
index = scic_sds_io_tag_get_index(io_tag);
- if (!sci_pool_full(scic->tci_pool)) {
- if (sequence == scic->io_request_sequence[index]) {
- scic_sds_io_sequence_increment(
- scic->io_request_sequence[index]);
+ /* prevent tail from passing head */
+ if (isci_tci_active(ihost) == 0)
+ return SCI_FAILURE_INVALID_IO_TAG;
- sci_pool_put(scic->tci_pool, index);
+ if (sequence == scic->io_request_sequence[index]) {
+ scic_sds_io_sequence_increment(scic->io_request_sequence[index]);
- return SCI_SUCCESS;
- }
+ isci_tci_free(ihost, index);
+
+ return SCI_SUCCESS;
}
-
return SCI_FAILURE_INVALID_IO_TAG;
}
-
-