Driver will use round robin method for io submission in batches
within the high iops queues when in-flight ios on the target device
is more than 8.
If in-flight ios per SCSI device more than 8, driver will use
high iops queue else driver will use low latency reply queues.
Signed-off-by: Suganath Prabu S <suganath-prabu.subramani@xxxxxxxxxxxx>
---
drivers/scsi/mpt3sas/mpt3sas_base.c | 36 ++++++++++++++++++++++++++++-
drivers/scsi/mpt3sas/mpt3sas_base.h | 14 ++++++++++-
2 files changed, 48 insertions(+), 2 deletions(-)
diff --git a/drivers/scsi/mpt3sas/mpt3sas_base.c b/drivers/scsi/mpt3sas/mpt3sas_base.c
index 57d0e7d..74cb060 100644
--- a/drivers/scsi/mpt3sas/mpt3sas_base.c
+++ b/drivers/scsi/mpt3sas/mpt3sas_base.c
@@ -3322,6 +3322,35 @@ _base_get_msix_index(struct MPT3SAS_ADAPTER *ioc,
return ioc->cpu_msix_table[raw_smp_processor_id()];
}
+/**
+ * _base_get_high_iops_msix_index - get the msix index of
+ * high iops queues
+ * @ioc: per adapter object
+ * @scmd: scsi_cmnd object
+ *
+ * Returns: msix index of high iops reply queues.
+ * i.e. high iops reply queue on which IO request's
+ * reply should be posted by the HBA firmware.
+ */
+static inline u8
+_base_get_high_iops_msix_index(struct MPT3SAS_ADAPTER *ioc,
+ struct scsi_cmnd *scmd)
+{
+ /**
+ * Round robin the IO interrupts among the high iops
+ * reply queues in terms of batch count 16 when outstanding
+ * IOs on the target device is >=8.
+ */
+ if (atomic_read(&scmd->device->device_busy) >
+ MPT3SAS_DEVICE_HIGH_IOPS_DEPTH)
+ return base_mod64((
+ atomic64_add_return(1, &ioc->high_iops_outstanding) /
+ MPT3SAS_HIGH_IOPS_BATCH_COUNT),
+ MPT3SAS_HIGH_IOPS_REPLY_QUEUES);
+
+ return _base_get_msix_index(ioc, scmd);
+}
+
/**
* mpt3sas_base_get_smid - obtain a free smid from internal queue
* @ioc: per adapter object
@@ -6707,6 +6736,7 @@ mpt3sas_base_attach(struct MPT3SAS_ADAPTER *ioc)
ioc->build_sg_scmd = &_base_build_sg_scmd;
ioc->build_sg = &_base_build_sg;
ioc->build_zero_len_sge = &_base_build_zero_len_sge;
+ ioc->get_msix_index_for_smlio = &_base_get_msix_index;
break;
case MPI25_VERSION:
case MPI26_VERSION:
@@ -6721,7 +6751,11 @@ mpt3sas_base_attach(struct MPT3SAS_ADAPTER *ioc)
ioc->build_nvme_prp = &_base_build_nvme_prp;
ioc->build_zero_len_sge = &_base_build_zero_len_sge_ieee;
ioc->sge_size_ieee = sizeof(Mpi2IeeeSgeSimple64_t);
-
+ if (ioc->high_iops_queues)
+ ioc->get_msix_index_for_smlio =
+ &_base_get_high_iops_msix_index;
+ else
+ ioc->get_msix_index_for_smlio = &_base_get_msix_index;
break;
}
if (ioc->atomic_desc_capable) {
diff --git a/drivers/scsi/mpt3sas/mpt3sas_base.h b/drivers/scsi/mpt3sas/mpt3sas_base.h
index bbbeb88..85db1f2 100644
--- a/drivers/scsi/mpt3sas/mpt3sas_base.h
+++ b/drivers/scsi/mpt3sas/mpt3sas_base.h
@@ -356,7 +356,9 @@ struct mpt3sas_nvme_cmd {
#define VIRTUAL_IO_FAILED_RETRY (0x32010081)
/* High IOPs definitions */
+#define MPT3SAS_DEVICE_HIGH_IOPS_DEPTH 8
#define MPT3SAS_HIGH_IOPS_REPLY_QUEUES 8
+#define MPT3SAS_HIGH_IOPS_BATCH_COUNT 16
#define MPT3SAS_GEN35_MAX_MSIX_QUEUES 128
/* OEM Specific Flags will come from OEM specific header files */
@@ -928,6 +930,12 @@ typedef void (*PUT_SMID_IO_FP_HIP) (struct MPT3SAS_ADAPTER *ioc, u16 smid,
u16 funcdep);
typedef void (*PUT_SMID_DEFAULT) (struct MPT3SAS_ADAPTER *ioc, u16 smid);
typedef u32 (*BASE_READ_REG) (const volatile void __iomem *addr);
+/*
+ * To get high iops reply queue's msix index when high iops mode is enabled
+ * else get the msix index of general reply queues.
+ */
+typedef u8 (*GET_MSIX_INDEX) (struct MPT3SAS_ADAPTER *ioc,
+ struct scsi_cmnd *scmd);
/* IOC Facts and Port Facts converted from little endian to cpu */
union mpi3_version_union {
@@ -1029,6 +1037,8 @@ typedef void (*MPT3SAS_FLUSH_RUNNING_CMDS)(struct MPT3SAS_ADAPTER *ioc);
* @cpu_msix_table: table for mapping cpus to msix index
* @cpu_msix_table_sz: table size
* @total_io_cnt: Gives total IO count, used to load balance the interrupts
+ * @high_iops_outstanding: used to load balance the interrupts
+ * within high iops reply queues
* @msix_load_balance: Enables load balancing of interrupts across
* the multiple MSIXs
* @schedule_dead_ioc_flush_running_cmds: callback to flush pending commands
@@ -1152,6 +1162,7 @@ typedef void (*MPT3SAS_FLUSH_RUNNING_CMDS)(struct MPT3SAS_ADAPTER *ioc);
* crash. To avoid the above race condition we use mutex syncrhonization
* which ensures the syncrhonization between cli/sysfs_show path.
* @atomic_desc_capable: Atomic Request Descriptor support.
+ * @GET_MSIX_INDEX: Get the msix index of high iops queues.
*/
struct MPT3SAS_ADAPTER {
struct list_head list;
@@ -1211,6 +1222,7 @@ struct MPT3SAS_ADAPTER {
MPT3SAS_FLUSH_RUNNING_CMDS schedule_dead_ioc_flush_running_cmds;
u32 non_operational_loop;
atomic64_t total_io_cnt;
+ atomic64_t high_iops_outstanding;
bool msix_load_balance;
u16 thresh_hold;
u8 high_iops_queues;
@@ -1432,7 +1444,7 @@ struct MPT3SAS_ADAPTER {
PUT_SMID_IO_FP_HIP put_smid_fast_path;
PUT_SMID_IO_FP_HIP put_smid_hi_priority;
PUT_SMID_DEFAULT put_smid_default;
-
+ GET_MSIX_INDEX get_msix_index_for_smlio;
};
typedef u8 (*MPT_CALLBACK)(struct MPT3SAS_ADAPTER *ioc, u16 smid, u8 msix_index,
--
2.18.1
