提交 1094cf68 编写于 作者: J Jitendra Bhivare 提交者: Martin K. Petersen

be2iscsi: Fix IOPOLL implementation

OS not responding when running 2 port traffic on 72 CPUs system.

be2iscsi IRQs gets affined to CPU0 when irqbalancer is disabled.
be_iopoll processing completions in BLOCK_IOPOLL_SOFTIRQ hogged CPU0.

1. Use budget to exit the polling loop. beiscsi_process_cq didn't honour
   it.
2. Rearming of EQ is done only after iopoll completes.

[mkp: Fixed up blk_iopoll -> irq_poll transition]
Signed-off-by: NJitendra Bhivare <jitendra.bhivare@avagotech.com>
Reviewed-by: NHannes Reinecke <hare@suse.com>
Signed-off-by: NMartin K. Petersen <martin.petersen@oracle.com>
上级 23d7ccf1
......@@ -546,7 +546,7 @@ int beiscsi_process_mcc(struct beiscsi_hba *phba)
}
if (num)
hwi_ring_cq_db(phba, phba->ctrl.mcc_obj.cq.id, num, 1, 0);
hwi_ring_cq_db(phba, phba->ctrl.mcc_obj.cq.id, num, 1);
spin_unlock_bh(&phba->ctrl.mcc_cq_lock);
return status;
......
......@@ -1298,7 +1298,7 @@ static void beiscsi_flush_cq(struct beiscsi_hba *phba)
for (i = 0; i < phba->num_cpus; i++) {
pbe_eq = &phwi_context->be_eq[i];
irq_poll_disable(&pbe_eq->iopoll);
beiscsi_process_cq(pbe_eq);
beiscsi_process_cq(pbe_eq, BE2_MAX_NUM_CQ_PROC);
irq_poll_enable(&pbe_eq->iopoll);
}
}
......
......@@ -29,6 +29,7 @@
#include <linux/iscsi_boot_sysfs.h>
#include <linux/module.h>
#include <linux/bsg-lib.h>
#include <linux/irq_poll.h>
#include <scsi/libiscsi.h>
#include <scsi/scsi_bsg_iscsi.h>
......@@ -895,31 +896,17 @@ static irqreturn_t be_isr_mcc(int irq, void *dev_id)
static irqreturn_t be_isr_msix(int irq, void *dev_id)
{
struct beiscsi_hba *phba;
struct be_eq_entry *eqe = NULL;
struct be_queue_info *eq;
struct be_queue_info *cq;
unsigned int num_eq_processed;
struct be_eq_obj *pbe_eq;
pbe_eq = dev_id;
eq = &pbe_eq->q;
cq = pbe_eq->cq;
eqe = queue_tail_node(eq);
phba = pbe_eq->phba;
num_eq_processed = 0;
while (eqe->dw[offsetof(struct amap_eq_entry, valid) / 32]
& EQE_VALID_MASK) {
irq_poll_sched(&pbe_eq->iopoll);
AMAP_SET_BITS(struct amap_eq_entry, valid, eqe, 0);
queue_tail_inc(eq);
eqe = queue_tail_node(eq);
num_eq_processed++;
}
if (num_eq_processed)
hwi_ring_eq_db(phba, eq->id, 1, num_eq_processed, 0, 1);
/* disable interrupt till iopoll completes */
hwi_ring_eq_db(phba, eq->id, 1, 0, 0, 1);
irq_poll_sched(&pbe_eq->iopoll);
return IRQ_HANDLED;
}
......@@ -996,6 +983,7 @@ static irqreturn_t be_isr(int irq, void *dev_id)
return IRQ_NONE;
}
static int beiscsi_init_irqs(struct beiscsi_hba *phba)
{
struct pci_dev *pcidev = phba->pcidev;
......@@ -1070,7 +1058,7 @@ static int beiscsi_init_irqs(struct beiscsi_hba *phba)
void hwi_ring_cq_db(struct beiscsi_hba *phba,
unsigned int id, unsigned int num_processed,
unsigned char rearm, unsigned char event)
unsigned char rearm)
{
u32 val = 0;
......@@ -2042,7 +2030,7 @@ static void beiscsi_process_mcc_isr(struct beiscsi_hba *phba)
if (num_processed >= 32) {
hwi_ring_cq_db(phba, mcc_cq->id,
num_processed, 0, 0);
num_processed, 0);
num_processed = 0;
}
if (mcc_compl->flags & CQE_FLAGS_ASYNC_MASK) {
......@@ -2060,24 +2048,25 @@ static void beiscsi_process_mcc_isr(struct beiscsi_hba *phba)
}
if (num_processed > 0)
hwi_ring_cq_db(phba, mcc_cq->id, num_processed, 1, 0);
hwi_ring_cq_db(phba, mcc_cq->id, num_processed, 1);
}
/**
* beiscsi_process_cq()- Process the Completion Queue
* @pbe_eq: Event Q on which the Completion has come
* @budget: Max number of events to processed
*
* return
* Number of Completion Entries processed.
**/
unsigned int beiscsi_process_cq(struct be_eq_obj *pbe_eq)
unsigned int beiscsi_process_cq(struct be_eq_obj *pbe_eq, int budget)
{
struct be_queue_info *cq;
struct sol_cqe *sol;
struct dmsg_cqe *dmsg;
unsigned int total = 0;
unsigned int num_processed = 0;
unsigned int tot_nump = 0;
unsigned short code = 0, cid = 0;
uint16_t cri_index = 0;
struct beiscsi_conn *beiscsi_conn;
......@@ -2128,12 +2117,12 @@ unsigned int beiscsi_process_cq(struct be_eq_obj *pbe_eq)
beiscsi_ep = ep->dd_data;
beiscsi_conn = beiscsi_ep->conn;
if (num_processed >= 32) {
hwi_ring_cq_db(phba, cq->id,
num_processed, 0, 0);
tot_nump += num_processed;
/* replenish cq */
if (num_processed == 32) {
hwi_ring_cq_db(phba, cq->id, 32, 0);
num_processed = 0;
}
total++;
switch (code) {
case SOL_CMD_COMPLETE:
......@@ -2178,7 +2167,13 @@ unsigned int beiscsi_process_cq(struct be_eq_obj *pbe_eq)
"BM_%d : Ignoring %s[%d] on CID : %d\n",
cqe_desc[code], code, cid);
break;
case CXN_KILLED_HDR_DIGEST_ERR:
case SOL_CMD_KILLED_DATA_DIGEST_ERR:
beiscsi_log(phba, KERN_ERR,
BEISCSI_LOG_CONFIG | BEISCSI_LOG_IO,
"BM_%d : Cmd Notification %s[%d] on CID : %d\n",
cqe_desc[code], code, cid);
break;
case CMD_KILLED_INVALID_STATSN_RCVD:
case CMD_KILLED_INVALID_R2T_RCVD:
case CMD_CXN_KILLED_LUN_INVALID:
......@@ -2204,7 +2199,6 @@ unsigned int beiscsi_process_cq(struct be_eq_obj *pbe_eq)
case CXN_KILLED_PDU_SIZE_EXCEEDS_DSL:
case CXN_KILLED_BURST_LEN_MISMATCH:
case CXN_KILLED_AHS_RCVD:
case CXN_KILLED_HDR_DIGEST_ERR:
case CXN_KILLED_UNKNOWN_HDR:
case CXN_KILLED_STALE_ITT_TTT_RCVD:
case CXN_KILLED_INVALID_ITT_TTT_RCVD:
......@@ -2239,13 +2233,12 @@ unsigned int beiscsi_process_cq(struct be_eq_obj *pbe_eq)
queue_tail_inc(cq);
sol = queue_tail_node(cq);
num_processed++;
if (total == budget)
break;
}
if (num_processed > 0) {
tot_nump += num_processed;
hwi_ring_cq_db(phba, cq->id, num_processed, 1, 0);
}
return tot_nump;
hwi_ring_cq_db(phba, cq->id, num_processed, 1);
return total;
}
void beiscsi_process_all_cqs(struct work_struct *work)
......@@ -2272,7 +2265,7 @@ void beiscsi_process_all_cqs(struct work_struct *work)
spin_lock_irqsave(&phba->isr_lock, flags);
pbe_eq->todo_cq = false;
spin_unlock_irqrestore(&phba->isr_lock, flags);
beiscsi_process_cq(pbe_eq);
beiscsi_process_cq(pbe_eq, BE2_MAX_NUM_CQ_PROC);
}
/* rearm EQ for further interrupts */
......@@ -2281,20 +2274,36 @@ void beiscsi_process_all_cqs(struct work_struct *work)
static int be_iopoll(struct irq_poll *iop, int budget)
{
unsigned int ret;
unsigned int ret, num_eq_processed;
struct beiscsi_hba *phba;
struct be_eq_obj *pbe_eq;
struct be_eq_entry *eqe = NULL;
struct be_queue_info *eq;
num_eq_processed = 0;
pbe_eq = container_of(iop, struct be_eq_obj, iopoll);
ret = beiscsi_process_cq(pbe_eq);
phba = pbe_eq->phba;
eq = &pbe_eq->q;
eqe = queue_tail_node(eq);
while (eqe->dw[offsetof(struct amap_eq_entry, valid) / 32] &
EQE_VALID_MASK) {
AMAP_SET_BITS(struct amap_eq_entry, valid, eqe, 0);
queue_tail_inc(eq);
eqe = queue_tail_node(eq);
num_eq_processed++;
}
hwi_ring_eq_db(phba, eq->id, 1, num_eq_processed, 0, 1);
ret = beiscsi_process_cq(pbe_eq, budget);
pbe_eq->cq_count += ret;
if (ret < budget) {
phba = pbe_eq->phba;
irq_poll_complete(iop);
beiscsi_log(phba, KERN_INFO,
BEISCSI_LOG_CONFIG | BEISCSI_LOG_IO,
"BM_%d : rearm pbe_eq->q.id =%d\n",
pbe_eq->q.id);
"BM_%d : rearm pbe_eq->q.id =%d ret %d\n",
pbe_eq->q.id, ret);
hwi_ring_eq_db(phba, pbe_eq->q.id, 0, 0, 1, 1);
}
return ret;
......
......@@ -63,6 +63,7 @@
#define BE2_SGE 32
#define BE2_DEFPDU_HDR_SZ 64
#define BE2_DEFPDU_DATA_SZ 8192
#define BE2_MAX_NUM_CQ_PROC 512
#define MAX_CPUS 64
#define BEISCSI_MAX_NUM_CPUS 7
......@@ -848,9 +849,9 @@ void beiscsi_free_mgmt_task_handles(struct beiscsi_conn *beiscsi_conn,
void hwi_ring_cq_db(struct beiscsi_hba *phba,
unsigned int id, unsigned int num_processed,
unsigned char rearm, unsigned char event);
unsigned char rearm);
unsigned int beiscsi_process_cq(struct be_eq_obj *pbe_eq);
unsigned int beiscsi_process_cq(struct be_eq_obj *pbe_eq, int budget);
static inline bool beiscsi_error(struct beiscsi_hba *phba)
{
......
Markdown is supported
0% .
You are about to add 0 people to the discussion. Proceed with caution.
先完成此消息的编辑!
想要评论请 注册