Groups | Search | Server Info | Keyboard shortcuts | Login | Register [http] [https] [nntp] [nntps]
Groups > linux.kernel > #1268953 > unrolled thread
| Started by | Christoph Hellwig <hch@lst.de> |
|---|---|
| First post | 2015-11-13 15:00 +0100 |
| Last post | 2015-11-13 15:00 +0100 |
| Articles | 14 — 3 participants |
Back to article view | Back to linux.kernel
add a proper completion queue abstraction Christoph Hellwig <hch@lst.de> - 2015-11-13 15:00 +0100
[PATCH 7/9] IB/iser: Use a dedicated descriptor for login Christoph Hellwig <hch@lst.de> - 2015-11-13 15:00 +0100
[PATCH 1/9] move blk_iopoll to limit and make it generally available Christoph Hellwig <hch@lst.de> - 2015-11-13 15:00 +0100
Re: [PATCH 1/9] move blk_iopoll to limit and make it generally available Or Gerlitz <gerlitz.or@gmail.com> - 2015-11-13 16:30 +0100
Re: [PATCH 1/9] move blk_iopoll to limit and make it generally available Christoph Hellwig <hch@lst.de> - 2015-11-14 08:10 +0100
Re: [PATCH 1/9] move blk_iopoll to limit and make it generally available Sagi Grimberg <sagig@dev.mellanox.co.il> - 2015-11-15 09:50 +0100
Re: [PATCH 1/9] move blk_iopoll to limit and make it generally available Or Gerlitz <gerlitz.or@gmail.com> - 2015-11-15 10:10 +0100
Re: [PATCH 1/9] move blk_iopoll to limit and make it generally available Sagi Grimberg <sagig@dev.mellanox.co.il> - 2015-11-15 14:20 +0100
Re: [PATCH 1/9] move blk_iopoll to limit and make it generally available Christoph Hellwig <hch@lst.de> - 2015-11-15 14:00 +0100
Re: [PATCH 1/9] move blk_iopoll to limit and make it generally available Christoph Hellwig <hch@lst.de> - 2015-11-14 08:10 +0100
[PATCH 6/9] srp: use the new CQ API Christoph Hellwig <hch@lst.de> - 2015-11-13 15:00 +0100
[PATCH 4/9] srpt: chain RDMA READ/WRITE requests Christoph Hellwig <hch@lst.de> - 2015-11-13 15:00 +0100
[PATCH 9/9] IB/iser: Convert to CQ abstraction Christoph Hellwig <hch@lst.de> - 2015-11-13 15:00 +0100
[PATCH 8/9] IB/iser: Use helper for container_of Christoph Hellwig <hch@lst.de> - 2015-11-13 15:00 +0100
| From | Christoph Hellwig <hch@lst.de> |
|---|---|
| Date | 2015-11-13 15:00 +0100 |
| Subject | add a proper completion queue abstraction |
| Message-ID | <qun3s-77o-17@gated-at.bofh.it> |
This series adds a new RDMA core abstraction that insulated the ULPs from the nitty gritty details of CQ polling. See the individual patches for more details. Note that this series should be applied on top of my "IB: merge struct ib_device_attr into struct ib_device" patch. A git tree is also available: http://git.infradead.org/users/hch/rdma.git/shortlog/refs/heads/rdma-cq git://git.infradead.org/users/hch/rdma.git rdma-cq -- To unsubscribe from this list: send the line "unsubscribe linux-kernel" in the body of a message to majordomo@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html Please read the FAQ at http://www.tux.org/lkml/
[toc] | [next] | [standalone]
| From | Christoph Hellwig <hch@lst.de> |
|---|---|
| Date | 2015-11-13 15:00 +0100 |
| Subject | [PATCH 7/9] IB/iser: Use a dedicated descriptor for login |
| Message-ID | <qun3v-77o-57@gated-at.bofh.it> |
| In reply to | #1268953 |
From: Sagi Grimberg <sagig@mellanox.com>
Makes better sense and we'll need it later with CQ
abstraction.
iser switch login bufs to void
Signed-off-by: Sagi Grimberg <sagig@mellanox.com>
Signed-off-by: Christoph Hellwig <hch@lst.de>
---
drivers/infiniband/ulp/iser/iscsi_iser.h | 30 +++++--
drivers/infiniband/ulp/iser/iser_initiator.c | 128 +++++++++++++--------------
drivers/infiniband/ulp/iser/iser_verbs.c | 14 +--
3 files changed, 89 insertions(+), 83 deletions(-)
diff --git a/drivers/infiniband/ulp/iser/iscsi_iser.h b/drivers/infiniband/ulp/iser/iscsi_iser.h
index 502063b..5648409 100644
--- a/drivers/infiniband/ulp/iser/iscsi_iser.h
+++ b/drivers/infiniband/ulp/iser/iscsi_iser.h
@@ -326,6 +326,25 @@ struct iser_rx_desc {
char pad[ISER_RX_PAD_SIZE];
} __attribute__((packed));
+
+/**
+ * struct iser_login_desc - iSER login descriptor
+ *
+ * @req: pointer to login request buffer
+ * @resp: pointer to login response buffer
+ * @req_dma: DMA address of login request buffer
+ * @rsp_dma: DMA address of login response buffer
+ * @sge: IB sge for login post recv
+ */
+struct iser_login_desc {
+ void *req;
+ void *rsp;
+ u64 req_dma;
+ u64 rsp_dma;
+ struct ib_sge sge;
+} __attribute__((packed));
+
+
struct iser_conn;
struct ib_conn;
struct iscsi_iser_task;
@@ -512,11 +531,7 @@ struct ib_conn {
* @up_completion: connection establishment completed
* (state is ISER_CONN_UP)
* @conn_list: entry in ig conn list
- * @login_buf: login data buffer (stores login parameters)
- * @login_req_buf: login request buffer
- * @login_req_dma: login request buffer dma address
- * @login_resp_buf: login response buffer
- * @login_resp_dma: login response buffer dma address
+ * @login_desc: login descriptor
* @rx_desc_head: head of rx_descs cyclic buffer
* @rx_descs: rx buffers array (cyclic buffer)
* @num_rx_descs: number of rx descriptors
@@ -539,10 +554,7 @@ struct iser_conn {
struct completion ib_completion;
struct completion up_completion;
struct list_head conn_list;
-
- char *login_buf;
- char *login_req_buf, *login_resp_buf;
- u64 login_req_dma, login_resp_dma;
+ struct iser_login_desc login_desc;
unsigned int rx_desc_head;
struct iser_rx_desc *rx_descs;
u32 num_rx_descs;
diff --git a/drivers/infiniband/ulp/iser/iser_initiator.c b/drivers/infiniband/ulp/iser/iser_initiator.c
index ffd00c4..21f28c8 100644
--- a/drivers/infiniband/ulp/iser/iser_initiator.c
+++ b/drivers/infiniband/ulp/iser/iser_initiator.c
@@ -174,73 +174,63 @@ static void iser_create_send_desc(struct iser_conn *iser_conn,
static void iser_free_login_buf(struct iser_conn *iser_conn)
{
struct iser_device *device = iser_conn->ib_conn.device;
+ struct iser_login_desc *desc = &iser_conn->login_desc;
- if (!iser_conn->login_buf)
+ if (!desc->req)
return;
- if (iser_conn->login_req_dma)
- ib_dma_unmap_single(device->ib_device,
- iser_conn->login_req_dma,
- ISCSI_DEF_MAX_RECV_SEG_LEN, DMA_TO_DEVICE);
+ ib_dma_unmap_single(device->ib_device, desc->req_dma,
+ ISCSI_DEF_MAX_RECV_SEG_LEN, DMA_TO_DEVICE);
- if (iser_conn->login_resp_dma)
- ib_dma_unmap_single(device->ib_device,
- iser_conn->login_resp_dma,
- ISER_RX_LOGIN_SIZE, DMA_FROM_DEVICE);
+ ib_dma_unmap_single(device->ib_device, desc->rsp_dma,
+ ISER_RX_LOGIN_SIZE, DMA_FROM_DEVICE);
- kfree(iser_conn->login_buf);
+ kfree(desc->req);
+ kfree(desc->rsp);
/* make sure we never redo any unmapping */
- iser_conn->login_req_dma = 0;
- iser_conn->login_resp_dma = 0;
- iser_conn->login_buf = NULL;
+ desc->req = NULL;
+ desc->rsp = NULL;
}
static int iser_alloc_login_buf(struct iser_conn *iser_conn)
{
struct iser_device *device = iser_conn->ib_conn.device;
- int req_err, resp_err;
-
- BUG_ON(device == NULL);
-
- iser_conn->login_buf = kmalloc(ISCSI_DEF_MAX_RECV_SEG_LEN +
- ISER_RX_LOGIN_SIZE, GFP_KERNEL);
- if (!iser_conn->login_buf)
- goto out_err;
-
- iser_conn->login_req_buf = iser_conn->login_buf;
- iser_conn->login_resp_buf = iser_conn->login_buf +
- ISCSI_DEF_MAX_RECV_SEG_LEN;
-
- iser_conn->login_req_dma = ib_dma_map_single(device->ib_device,
- iser_conn->login_req_buf,
- ISCSI_DEF_MAX_RECV_SEG_LEN,
- DMA_TO_DEVICE);
-
- iser_conn->login_resp_dma = ib_dma_map_single(device->ib_device,
- iser_conn->login_resp_buf,
- ISER_RX_LOGIN_SIZE,
- DMA_FROM_DEVICE);
-
- req_err = ib_dma_mapping_error(device->ib_device,
- iser_conn->login_req_dma);
- resp_err = ib_dma_mapping_error(device->ib_device,
- iser_conn->login_resp_dma);
-
- if (req_err || resp_err) {
- if (req_err)
- iser_conn->login_req_dma = 0;
- if (resp_err)
- iser_conn->login_resp_dma = 0;
- goto free_login_buf;
- }
+ struct iser_login_desc *desc = &iser_conn->login_desc;
+
+ desc->req = kmalloc(ISCSI_DEF_MAX_RECV_SEG_LEN, GFP_KERNEL);
+ if (!desc->req)
+ return -ENOMEM;
+
+ desc->req_dma = ib_dma_map_single(device->ib_device, desc->req,
+ ISCSI_DEF_MAX_RECV_SEG_LEN,
+ DMA_TO_DEVICE);
+ if (ib_dma_mapping_error(device->ib_device,
+ desc->req_dma))
+ goto free_req;
+
+ desc->rsp = kmalloc(ISER_RX_LOGIN_SIZE, GFP_KERNEL);
+ if (!desc->rsp)
+ goto unmap_req;
+
+ desc->rsp_dma = ib_dma_map_single(device->ib_device, desc->rsp,
+ ISER_RX_LOGIN_SIZE,
+ DMA_FROM_DEVICE);
+ if (ib_dma_mapping_error(device->ib_device,
+ desc->rsp_dma))
+ goto free_rsp;
+
return 0;
-free_login_buf:
- iser_free_login_buf(iser_conn);
+free_rsp:
+ kfree(desc->rsp);
+unmap_req:
+ ib_dma_unmap_single(device->ib_device, desc->req_dma,
+ ISCSI_DEF_MAX_RECV_SEG_LEN,
+ DMA_TO_DEVICE);
+free_req:
+ kfree(desc->req);
-out_err:
- iser_err("unable to alloc or map login buf\n");
return -ENOMEM;
}
@@ -520,25 +510,25 @@ int iser_send_control(struct iscsi_conn *conn,
data_seg_len = ntoh24(task->hdr->dlength);
if (data_seg_len > 0) {
+ struct iser_login_desc *desc = &iser_conn->login_desc;
struct ib_sge *tx_dsg = &mdesc->tx_sg[1];
+
if (task != conn->login_task) {
iser_err("data present on non login task!!!\n");
goto send_control_error;
}
- ib_dma_sync_single_for_cpu(device->ib_device,
- iser_conn->login_req_dma, task->data_count,
- DMA_TO_DEVICE);
+ ib_dma_sync_single_for_cpu(device->ib_device, desc->req_dma,
+ task->data_count, DMA_TO_DEVICE);
- memcpy(iser_conn->login_req_buf, task->data, task->data_count);
+ memcpy(desc->req, task->data, task->data_count);
- ib_dma_sync_single_for_device(device->ib_device,
- iser_conn->login_req_dma, task->data_count,
- DMA_TO_DEVICE);
+ ib_dma_sync_single_for_device(device->ib_device, desc->req_dma,
+ task->data_count, DMA_TO_DEVICE);
- tx_dsg->addr = iser_conn->login_req_dma;
- tx_dsg->length = task->data_count;
- tx_dsg->lkey = device->pd->local_dma_lkey;
+ tx_dsg->addr = desc->req_dma;
+ tx_dsg->length = task->data_count;
+ tx_dsg->lkey = device->pd->local_dma_lkey;
mdesc->num_sge = 2;
}
@@ -572,27 +562,31 @@ void iser_rcv_completion(struct iser_rx_desc *rx_desc,
struct iser_conn *iser_conn = container_of(ib_conn, struct iser_conn,
ib_conn);
struct iscsi_hdr *hdr;
+ char *data;
u64 rx_dma;
int rx_buflen, outstanding, count, err;
/* differentiate between login to all other PDUs */
- if ((char *)rx_desc == iser_conn->login_resp_buf) {
- rx_dma = iser_conn->login_resp_dma;
+ if (rx_desc == (void *)&iser_conn->login_desc) {
+ rx_dma = iser_conn->login_desc.rsp_dma;
rx_buflen = ISER_RX_LOGIN_SIZE;
+ hdr = iser_conn->login_desc.rsp + sizeof(struct iser_hdr);
+ data = iser_conn->login_desc.rsp + ISER_HEADERS_LEN;
} else {
rx_dma = rx_desc->dma_addr;
rx_buflen = ISER_RX_PAYLOAD_SIZE;
+ hdr = &rx_desc->iscsi_header;
+ data = rx_desc->data;
}
ib_dma_sync_single_for_cpu(ib_conn->device->ib_device, rx_dma,
rx_buflen, DMA_FROM_DEVICE);
- hdr = &rx_desc->iscsi_header;
iser_dbg("op 0x%x itt 0x%x dlen %d\n", hdr->opcode,
hdr->itt, (int)(rx_xfer_len - ISER_HEADERS_LEN));
- iscsi_iser_recv(iser_conn->iscsi_conn, hdr, rx_desc->data,
+ iscsi_iser_recv(iser_conn->iscsi_conn, hdr, data,
rx_xfer_len - ISER_HEADERS_LEN);
ib_dma_sync_single_for_device(ib_conn->device->ib_device, rx_dma,
@@ -604,7 +598,7 @@ void iser_rcv_completion(struct iser_rx_desc *rx_desc,
* for the posted rx bufs refcount to become zero handles everything */
ib_conn->post_recv_buf_count--;
- if (rx_dma == iser_conn->login_resp_dma)
+ if (rx_desc == (void *)&iser_conn->login_desc)
return;
outstanding = ib_conn->post_recv_buf_count;
diff --git a/drivers/infiniband/ulp/iser/iser_verbs.c b/drivers/infiniband/ulp/iser/iser_verbs.c
index bf29ddf..ee4cebc 100644
--- a/drivers/infiniband/ulp/iser/iser_verbs.c
+++ b/drivers/infiniband/ulp/iser/iser_verbs.c
@@ -1041,17 +1041,17 @@ int iser_post_recvl(struct iser_conn *iser_conn)
{
struct ib_recv_wr rx_wr, *rx_wr_failed;
struct ib_conn *ib_conn = &iser_conn->ib_conn;
- struct ib_sge sge;
+ struct iser_login_desc *desc = &iser_conn->login_desc;
int ib_ret;
- sge.addr = iser_conn->login_resp_dma;
- sge.length = ISER_RX_LOGIN_SIZE;
- sge.lkey = ib_conn->device->pd->local_dma_lkey;
+ desc->sge.addr = desc->rsp_dma;
+ desc->sge.length = ISER_RX_LOGIN_SIZE;
+ desc->sge.lkey = ib_conn->device->pd->local_dma_lkey;
- rx_wr.wr_id = (uintptr_t)iser_conn->login_resp_buf;
- rx_wr.sg_list = &sge;
+ rx_wr.wr_id = (uintptr_t)desc;
+ rx_wr.sg_list = &desc->sge;
rx_wr.num_sge = 1;
- rx_wr.next = NULL;
+ rx_wr.next = NULL;
ib_conn->post_recv_buf_count++;
ib_ret = ib_post_recv(ib_conn->qp, &rx_wr, &rx_wr_failed);
--
1.9.1
--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at http://vger.kernel.org/majordomo-info.html
Please read the FAQ at http://www.tux.org/lkml/
[toc] | [prev] | [next] | [standalone]
| From | Christoph Hellwig <hch@lst.de> |
|---|---|
| Date | 2015-11-13 15:00 +0100 |
| Subject | [PATCH 1/9] move blk_iopoll to limit and make it generally available |
| Message-ID | <qun3v-77o-61@gated-at.bofh.it> |
| In reply to | #1268953 |
The new name is irq_poll as iopoll is already taken. Better suggestions
welcome.
Signed-off-by: Christoph Hellwig <hch@lst.de>
---
Documentation/kernel-per-CPU-kthreads.txt | 2 +-
block/Makefile | 2 +-
block/blk-iopoll.c | 224 ------------------------------
drivers/scsi/Kconfig | 1 +
drivers/scsi/be2iscsi/Kconfig | 1 +
drivers/scsi/be2iscsi/be.h | 4 +-
drivers/scsi/be2iscsi/be_iscsi.c | 4 +-
drivers/scsi/be2iscsi/be_main.c | 24 ++--
drivers/scsi/ipr.c | 28 ++--
drivers/scsi/ipr.h | 4 +-
include/linux/blk-iopoll.h | 46 ------
include/linux/interrupt.h | 2 +-
include/linux/irq_poll.h | 46 ++++++
include/trace/events/irq.h | 2 +-
lib/Kconfig | 5 +
lib/Makefile | 1 +
lib/irq_poll.c | 221 +++++++++++++++++++++++++++++
tools/lib/traceevent/event-parse.c | 2 +-
tools/perf/util/trace-event-parse.c | 2 +-
19 files changed, 313 insertions(+), 308 deletions(-)
delete mode 100644 block/blk-iopoll.c
delete mode 100644 include/linux/blk-iopoll.h
create mode 100644 include/linux/irq_poll.h
create mode 100644 lib/irq_poll.c
diff --git a/Documentation/kernel-per-CPU-kthreads.txt b/Documentation/kernel-per-CPU-kthreads.txt
index f4cbfe0..edec3a3 100644
--- a/Documentation/kernel-per-CPU-kthreads.txt
+++ b/Documentation/kernel-per-CPU-kthreads.txt
@@ -90,7 +90,7 @@ BLOCK_SOFTIRQ: Do all of the following:
from being initiated from tasks that might run on the CPU to
be de-jittered. (It is OK to force this CPU offline and then
bring it back online before you start your application.)
-BLOCK_IOPOLL_SOFTIRQ: Do all of the following:
+IRQ_POLL_SOFTIRQ: Do all of the following:
1. Force block-device interrupts onto some other CPU.
2. Initiate any block I/O and block-I/O polling on other CPUs.
3. Once your application has started, prevent CPU-hotplug operations
diff --git a/block/Makefile b/block/Makefile
index 00ecc97..e850474 100644
--- a/block/Makefile
+++ b/block/Makefile
@@ -5,7 +5,7 @@
obj-$(CONFIG_BLOCK) := bio.o elevator.o blk-core.o blk-tag.o blk-sysfs.o \
blk-flush.o blk-settings.o blk-ioc.o blk-map.o \
blk-exec.o blk-merge.o blk-softirq.o blk-timeout.o \
- blk-iopoll.o blk-lib.o blk-mq.o blk-mq-tag.o \
+ blk-lib.o blk-mq.o blk-mq-tag.o \
blk-mq-sysfs.o blk-mq-cpu.o blk-mq-cpumap.o ioctl.o \
genhd.o scsi_ioctl.o partition-generic.o ioprio.o \
partitions/
diff --git a/block/blk-iopoll.c b/block/blk-iopoll.c
deleted file mode 100644
index 0736729..0000000
--- a/block/blk-iopoll.c
+++ /dev/null
@@ -1,224 +0,0 @@
-/*
- * Functions related to interrupt-poll handling in the block layer. This
- * is similar to NAPI for network devices.
- */
-#include <linux/kernel.h>
-#include <linux/module.h>
-#include <linux/init.h>
-#include <linux/bio.h>
-#include <linux/blkdev.h>
-#include <linux/interrupt.h>
-#include <linux/cpu.h>
-#include <linux/blk-iopoll.h>
-#include <linux/delay.h>
-
-#include "blk.h"
-
-static unsigned int blk_iopoll_budget __read_mostly = 256;
-
-static DEFINE_PER_CPU(struct list_head, blk_cpu_iopoll);
-
-/**
- * blk_iopoll_sched - Schedule a run of the iopoll handler
- * @iop: The parent iopoll structure
- *
- * Description:
- * Add this blk_iopoll structure to the pending poll list and trigger the
- * raise of the blk iopoll softirq. The driver must already have gotten a
- * successful return from blk_iopoll_sched_prep() before calling this.
- **/
-void blk_iopoll_sched(struct blk_iopoll *iop)
-{
- unsigned long flags;
-
- local_irq_save(flags);
- list_add_tail(&iop->list, this_cpu_ptr(&blk_cpu_iopoll));
- __raise_softirq_irqoff(BLOCK_IOPOLL_SOFTIRQ);
- local_irq_restore(flags);
-}
-EXPORT_SYMBOL(blk_iopoll_sched);
-
-/**
- * __blk_iopoll_complete - Mark this @iop as un-polled again
- * @iop: The parent iopoll structure
- *
- * Description:
- * See blk_iopoll_complete(). This function must be called with interrupts
- * disabled.
- **/
-void __blk_iopoll_complete(struct blk_iopoll *iop)
-{
- list_del(&iop->list);
- smp_mb__before_atomic();
- clear_bit_unlock(IOPOLL_F_SCHED, &iop->state);
-}
-EXPORT_SYMBOL(__blk_iopoll_complete);
-
-/**
- * blk_iopoll_complete - Mark this @iop as un-polled again
- * @iop: The parent iopoll structure
- *
- * Description:
- * If a driver consumes less than the assigned budget in its run of the
- * iopoll handler, it'll end the polled mode by calling this function. The
- * iopoll handler will not be invoked again before blk_iopoll_sched_prep()
- * is called.
- **/
-void blk_iopoll_complete(struct blk_iopoll *iop)
-{
- unsigned long flags;
-
- local_irq_save(flags);
- __blk_iopoll_complete(iop);
- local_irq_restore(flags);
-}
-EXPORT_SYMBOL(blk_iopoll_complete);
-
-static void blk_iopoll_softirq(struct softirq_action *h)
-{
- struct list_head *list = this_cpu_ptr(&blk_cpu_iopoll);
- int rearm = 0, budget = blk_iopoll_budget;
- unsigned long start_time = jiffies;
-
- local_irq_disable();
-
- while (!list_empty(list)) {
- struct blk_iopoll *iop;
- int work, weight;
-
- /*
- * If softirq window is exhausted then punt.
- */
- if (budget <= 0 || time_after(jiffies, start_time)) {
- rearm = 1;
- break;
- }
-
- local_irq_enable();
-
- /* Even though interrupts have been re-enabled, this
- * access is safe because interrupts can only add new
- * entries to the tail of this list, and only ->poll()
- * calls can remove this head entry from the list.
- */
- iop = list_entry(list->next, struct blk_iopoll, list);
-
- weight = iop->weight;
- work = 0;
- if (test_bit(IOPOLL_F_SCHED, &iop->state))
- work = iop->poll(iop, weight);
-
- budget -= work;
-
- local_irq_disable();
-
- /*
- * Drivers must not modify the iopoll state, if they
- * consume their assigned weight (or more, some drivers can't
- * easily just stop processing, they have to complete an
- * entire mask of commands).In such cases this code
- * still "owns" the iopoll instance and therefore can
- * move the instance around on the list at-will.
- */
- if (work >= weight) {
- if (blk_iopoll_disable_pending(iop))
- __blk_iopoll_complete(iop);
- else
- list_move_tail(&iop->list, list);
- }
- }
-
- if (rearm)
- __raise_softirq_irqoff(BLOCK_IOPOLL_SOFTIRQ);
-
- local_irq_enable();
-}
-
-/**
- * blk_iopoll_disable - Disable iopoll on this @iop
- * @iop: The parent iopoll structure
- *
- * Description:
- * Disable io polling and wait for any pending callbacks to have completed.
- **/
-void blk_iopoll_disable(struct blk_iopoll *iop)
-{
- set_bit(IOPOLL_F_DISABLE, &iop->state);
- while (test_and_set_bit(IOPOLL_F_SCHED, &iop->state))
- msleep(1);
- clear_bit(IOPOLL_F_DISABLE, &iop->state);
-}
-EXPORT_SYMBOL(blk_iopoll_disable);
-
-/**
- * blk_iopoll_enable - Enable iopoll on this @iop
- * @iop: The parent iopoll structure
- *
- * Description:
- * Enable iopoll on this @iop. Note that the handler run will not be
- * scheduled, it will only mark it as active.
- **/
-void blk_iopoll_enable(struct blk_iopoll *iop)
-{
- BUG_ON(!test_bit(IOPOLL_F_SCHED, &iop->state));
- smp_mb__before_atomic();
- clear_bit_unlock(IOPOLL_F_SCHED, &iop->state);
-}
-EXPORT_SYMBOL(blk_iopoll_enable);
-
-/**
- * blk_iopoll_init - Initialize this @iop
- * @iop: The parent iopoll structure
- * @weight: The default weight (or command completion budget)
- * @poll_fn: The handler to invoke
- *
- * Description:
- * Initialize this blk_iopoll structure. Before being actively used, the
- * driver must call blk_iopoll_enable().
- **/
-void blk_iopoll_init(struct blk_iopoll *iop, int weight, blk_iopoll_fn *poll_fn)
-{
- memset(iop, 0, sizeof(*iop));
- INIT_LIST_HEAD(&iop->list);
- iop->weight = weight;
- iop->poll = poll_fn;
- set_bit(IOPOLL_F_SCHED, &iop->state);
-}
-EXPORT_SYMBOL(blk_iopoll_init);
-
-static int blk_iopoll_cpu_notify(struct notifier_block *self,
- unsigned long action, void *hcpu)
-{
- /*
- * If a CPU goes away, splice its entries to the current CPU
- * and trigger a run of the softirq
- */
- if (action == CPU_DEAD || action == CPU_DEAD_FROZEN) {
- int cpu = (unsigned long) hcpu;
-
- local_irq_disable();
- list_splice_init(&per_cpu(blk_cpu_iopoll, cpu),
- this_cpu_ptr(&blk_cpu_iopoll));
- __raise_softirq_irqoff(BLOCK_IOPOLL_SOFTIRQ);
- local_irq_enable();
- }
-
- return NOTIFY_OK;
-}
-
-static struct notifier_block blk_iopoll_cpu_notifier = {
- .notifier_call = blk_iopoll_cpu_notify,
-};
-
-static __init int blk_iopoll_setup(void)
-{
- int i;
-
- for_each_possible_cpu(i)
- INIT_LIST_HEAD(&per_cpu(blk_cpu_iopoll, i));
-
- open_softirq(BLOCK_IOPOLL_SOFTIRQ, blk_iopoll_softirq);
- register_hotcpu_notifier(&blk_iopoll_cpu_notifier);
- return 0;
-}
-subsys_initcall(blk_iopoll_setup);
diff --git a/drivers/scsi/Kconfig b/drivers/scsi/Kconfig
index d2f480b..cea683e 100644
--- a/drivers/scsi/Kconfig
+++ b/drivers/scsi/Kconfig
@@ -1103,6 +1103,7 @@ config SCSI_IPR
tristate "IBM Power Linux RAID adapter support"
depends on PCI && SCSI && ATA
select FW_LOADER
+ select IRQ_POLL
---help---
This driver supports the IBM Power Linux family RAID adapters.
This includes IBM pSeries 5712, 5703, 5709, and 570A, as well
diff --git a/drivers/scsi/be2iscsi/Kconfig b/drivers/scsi/be2iscsi/Kconfig
index 4e7cad2..bad5f32 100644
--- a/drivers/scsi/be2iscsi/Kconfig
+++ b/drivers/scsi/be2iscsi/Kconfig
@@ -3,6 +3,7 @@ config BE2ISCSI
depends on PCI && SCSI && NET
select SCSI_ISCSI_ATTRS
select ISCSI_BOOT_SYSFS
+ select IRQ_POLL
help
This driver implements the iSCSI functionality for Emulex
diff --git a/drivers/scsi/be2iscsi/be.h b/drivers/scsi/be2iscsi/be.h
index 77f992e..a41c643 100644
--- a/drivers/scsi/be2iscsi/be.h
+++ b/drivers/scsi/be2iscsi/be.h
@@ -20,7 +20,7 @@
#include <linux/pci.h>
#include <linux/if_vlan.h>
-#include <linux/blk-iopoll.h>
+#include <linux/irq_poll.h>
#define FW_VER_LEN 32
#define MCC_Q_LEN 128
#define MCC_CQ_LEN 256
@@ -101,7 +101,7 @@ struct be_eq_obj {
struct beiscsi_hba *phba;
struct be_queue_info *cq;
struct work_struct work_cqs; /* Work Item */
- struct blk_iopoll iopoll;
+ struct irq_poll iopoll;
};
struct be_mcc_obj {
diff --git a/drivers/scsi/be2iscsi/be_iscsi.c b/drivers/scsi/be2iscsi/be_iscsi.c
index b7087ba..022e87b 100644
--- a/drivers/scsi/be2iscsi/be_iscsi.c
+++ b/drivers/scsi/be2iscsi/be_iscsi.c
@@ -1292,9 +1292,9 @@ static void beiscsi_flush_cq(struct beiscsi_hba *phba)
for (i = 0; i < phba->num_cpus; i++) {
pbe_eq = &phwi_context->be_eq[i];
- blk_iopoll_disable(&pbe_eq->iopoll);
+ irq_poll_disable(&pbe_eq->iopoll);
beiscsi_process_cq(pbe_eq);
- blk_iopoll_enable(&pbe_eq->iopoll);
+ irq_poll_enable(&pbe_eq->iopoll);
}
}
diff --git a/drivers/scsi/be2iscsi/be_main.c b/drivers/scsi/be2iscsi/be_main.c
index 2e6abe7..7f0fdbe 100644
--- a/drivers/scsi/be2iscsi/be_main.c
+++ b/drivers/scsi/be2iscsi/be_main.c
@@ -910,8 +910,8 @@ static irqreturn_t be_isr_msix(int irq, void *dev_id)
num_eq_processed = 0;
while (eqe->dw[offsetof(struct amap_eq_entry, valid) / 32]
& EQE_VALID_MASK) {
- if (!blk_iopoll_sched_prep(&pbe_eq->iopoll))
- blk_iopoll_sched(&pbe_eq->iopoll);
+ if (!irq_poll_sched_prep(&pbe_eq->iopoll))
+ irq_poll_sched(&pbe_eq->iopoll);
AMAP_SET_BITS(struct amap_eq_entry, valid, eqe, 0);
queue_tail_inc(eq);
@@ -972,8 +972,8 @@ static irqreturn_t be_isr(int irq, void *dev_id)
spin_unlock_irqrestore(&phba->isr_lock, flags);
num_mcceq_processed++;
} else {
- if (!blk_iopoll_sched_prep(&pbe_eq->iopoll))
- blk_iopoll_sched(&pbe_eq->iopoll);
+ if (!irq_poll_sched_prep(&pbe_eq->iopoll))
+ irq_poll_sched(&pbe_eq->iopoll);
num_ioeq_processed++;
}
AMAP_SET_BITS(struct amap_eq_entry, valid, eqe, 0);
@@ -2293,7 +2293,7 @@ void beiscsi_process_all_cqs(struct work_struct *work)
hwi_ring_eq_db(phba, pbe_eq->q.id, 0, 0, 1, 1);
}
-static int be_iopoll(struct blk_iopoll *iop, int budget)
+static int be_iopoll(struct irq_poll *iop, int budget)
{
unsigned int ret;
struct beiscsi_hba *phba;
@@ -2304,7 +2304,7 @@ static int be_iopoll(struct blk_iopoll *iop, int budget)
pbe_eq->cq_count += ret;
if (ret < budget) {
phba = pbe_eq->phba;
- blk_iopoll_complete(iop);
+ irq_poll_complete(iop);
beiscsi_log(phba, KERN_INFO,
BEISCSI_LOG_CONFIG | BEISCSI_LOG_IO,
"BM_%d : rearm pbe_eq->q.id =%d\n",
@@ -5261,7 +5261,7 @@ static void beiscsi_quiesce(struct beiscsi_hba *phba,
for (i = 0; i < phba->num_cpus; i++) {
pbe_eq = &phwi_context->be_eq[i];
- blk_iopoll_disable(&pbe_eq->iopoll);
+ irq_poll_disable(&pbe_eq->iopoll);
}
if (unload_state == BEISCSI_CLEAN_UNLOAD) {
@@ -5547,9 +5547,9 @@ static void beiscsi_eeh_resume(struct pci_dev *pdev)
for (i = 0; i < phba->num_cpus; i++) {
pbe_eq = &phwi_context->be_eq[i];
- blk_iopoll_init(&pbe_eq->iopoll, be_iopoll_budget,
+ irq_poll_init(&pbe_eq->iopoll, be_iopoll_budget,
be_iopoll);
- blk_iopoll_enable(&pbe_eq->iopoll);
+ irq_poll_enable(&pbe_eq->iopoll);
}
i = (phba->msix_enabled) ? i : 0;
@@ -5720,9 +5720,9 @@ static int beiscsi_dev_probe(struct pci_dev *pcidev,
for (i = 0; i < phba->num_cpus; i++) {
pbe_eq = &phwi_context->be_eq[i];
- blk_iopoll_init(&pbe_eq->iopoll, be_iopoll_budget,
+ irq_poll_init(&pbe_eq->iopoll, be_iopoll_budget,
be_iopoll);
- blk_iopoll_enable(&pbe_eq->iopoll);
+ irq_poll_enable(&pbe_eq->iopoll);
}
i = (phba->msix_enabled) ? i : 0;
@@ -5763,7 +5763,7 @@ free_blkenbld:
destroy_workqueue(phba->wq);
for (i = 0; i < phba->num_cpus; i++) {
pbe_eq = &phwi_context->be_eq[i];
- blk_iopoll_disable(&pbe_eq->iopoll);
+ irq_poll_disable(&pbe_eq->iopoll);
}
free_twq:
beiscsi_clean_port(phba);
diff --git a/drivers/scsi/ipr.c b/drivers/scsi/ipr.c
index b62836d..6b98e75 100644
--- a/drivers/scsi/ipr.c
+++ b/drivers/scsi/ipr.c
@@ -3638,7 +3638,7 @@ static struct device_attribute ipr_ioa_reset_attr = {
.store = ipr_store_reset_adapter
};
-static int ipr_iopoll(struct blk_iopoll *iop, int budget);
+static int ipr_iopoll(struct irq_poll *iop, int budget);
/**
* ipr_show_iopoll_weight - Show ipr polling mode
* @dev: class device struct
@@ -3681,34 +3681,34 @@ static ssize_t ipr_store_iopoll_weight(struct device *dev,
int i;
if (!ioa_cfg->sis64) {
- dev_info(&ioa_cfg->pdev->dev, "blk-iopoll not supported on this adapter\n");
+ dev_info(&ioa_cfg->pdev->dev, "irq_poll not supported on this adapter\n");
return -EINVAL;
}
if (kstrtoul(buf, 10, &user_iopoll_weight))
return -EINVAL;
if (user_iopoll_weight > 256) {
- dev_info(&ioa_cfg->pdev->dev, "Invalid blk-iopoll weight. It must be less than 256\n");
+ dev_info(&ioa_cfg->pdev->dev, "Invalid irq_poll weight. It must be less than 256\n");
return -EINVAL;
}
if (user_iopoll_weight == ioa_cfg->iopoll_weight) {
- dev_info(&ioa_cfg->pdev->dev, "Current blk-iopoll weight has the same weight\n");
+ dev_info(&ioa_cfg->pdev->dev, "Current irq_poll weight has the same weight\n");
return strlen(buf);
}
if (ioa_cfg->iopoll_weight && ioa_cfg->sis64 && ioa_cfg->nvectors > 1) {
for (i = 1; i < ioa_cfg->hrrq_num; i++)
- blk_iopoll_disable(&ioa_cfg->hrrq[i].iopoll);
+ irq_poll_disable(&ioa_cfg->hrrq[i].iopoll);
}
spin_lock_irqsave(shost->host_lock, lock_flags);
ioa_cfg->iopoll_weight = user_iopoll_weight;
if (ioa_cfg->iopoll_weight && ioa_cfg->sis64 && ioa_cfg->nvectors > 1) {
for (i = 1; i < ioa_cfg->hrrq_num; i++) {
- blk_iopoll_init(&ioa_cfg->hrrq[i].iopoll,
+ irq_poll_init(&ioa_cfg->hrrq[i].iopoll,
ioa_cfg->iopoll_weight, ipr_iopoll);
- blk_iopoll_enable(&ioa_cfg->hrrq[i].iopoll);
+ irq_poll_enable(&ioa_cfg->hrrq[i].iopoll);
}
}
spin_unlock_irqrestore(shost->host_lock, lock_flags);
@@ -5569,7 +5569,7 @@ static int ipr_process_hrrq(struct ipr_hrr_queue *hrr_queue, int budget,
return num_hrrq;
}
-static int ipr_iopoll(struct blk_iopoll *iop, int budget)
+static int ipr_iopoll(struct irq_poll *iop, int budget)
{
struct ipr_ioa_cfg *ioa_cfg;
struct ipr_hrr_queue *hrrq;
@@ -5585,7 +5585,7 @@ static int ipr_iopoll(struct blk_iopoll *iop, int budget)
completed_ops = ipr_process_hrrq(hrrq, budget, &doneq);
if (completed_ops < budget)
- blk_iopoll_complete(iop);
+ irq_poll_complete(iop);
spin_unlock_irqrestore(hrrq->lock, hrrq_flags);
list_for_each_entry_safe(ipr_cmd, temp, &doneq, queue) {
@@ -5693,8 +5693,8 @@ static irqreturn_t ipr_isr_mhrrq(int irq, void *devp)
if (ioa_cfg->iopoll_weight && ioa_cfg->sis64 && ioa_cfg->nvectors > 1) {
if ((be32_to_cpu(*hrrq->hrrq_curr) & IPR_HRRQ_TOGGLE_BIT) ==
hrrq->toggle_bit) {
- if (!blk_iopoll_sched_prep(&hrrq->iopoll))
- blk_iopoll_sched(&hrrq->iopoll);
+ if (!irq_poll_sched_prep(&hrrq->iopoll))
+ irq_poll_sched(&hrrq->iopoll);
spin_unlock_irqrestore(hrrq->lock, hrrq_flags);
return IRQ_HANDLED;
}
@@ -10285,9 +10285,9 @@ static int ipr_probe(struct pci_dev *pdev, const struct pci_device_id *dev_id)
if (ioa_cfg->iopoll_weight && ioa_cfg->sis64 && ioa_cfg->nvectors > 1) {
for (i = 1; i < ioa_cfg->hrrq_num; i++) {
- blk_iopoll_init(&ioa_cfg->hrrq[i].iopoll,
+ irq_poll_init(&ioa_cfg->hrrq[i].iopoll,
ioa_cfg->iopoll_weight, ipr_iopoll);
- blk_iopoll_enable(&ioa_cfg->hrrq[i].iopoll);
+ irq_poll_enable(&ioa_cfg->hrrq[i].iopoll);
}
}
@@ -10316,7 +10316,7 @@ static void ipr_shutdown(struct pci_dev *pdev)
if (ioa_cfg->iopoll_weight && ioa_cfg->sis64 && ioa_cfg->nvectors > 1) {
ioa_cfg->iopoll_weight = 0;
for (i = 1; i < ioa_cfg->hrrq_num; i++)
- blk_iopoll_disable(&ioa_cfg->hrrq[i].iopoll);
+ irq_poll_disable(&ioa_cfg->hrrq[i].iopoll);
}
while (ioa_cfg->in_reset_reload) {
diff --git a/drivers/scsi/ipr.h b/drivers/scsi/ipr.h
index e4fb17a..022fc3c 100644
--- a/drivers/scsi/ipr.h
+++ b/drivers/scsi/ipr.h
@@ -32,7 +32,7 @@
#include <linux/libata.h>
#include <linux/list.h>
#include <linux/kref.h>
-#include <linux/blk-iopoll.h>
+#include <linux/irq_poll.h>
#include <scsi/scsi.h>
#include <scsi/scsi_cmnd.h>
@@ -510,7 +510,7 @@ struct ipr_hrr_queue {
u8 allow_cmds:1;
u8 removing_ioa:1;
- struct blk_iopoll iopoll;
+ struct irq_poll iopoll;
};
/* Command packet structure */
diff --git a/include/linux/blk-iopoll.h b/include/linux/blk-iopoll.h
deleted file mode 100644
index 77ae77c..0000000
--- a/include/linux/blk-iopoll.h
+++ /dev/null
@@ -1,46 +0,0 @@
-#ifndef BLK_IOPOLL_H
-#define BLK_IOPOLL_H
-
-struct blk_iopoll;
-typedef int (blk_iopoll_fn)(struct blk_iopoll *, int);
-
-struct blk_iopoll {
- struct list_head list;
- unsigned long state;
- unsigned long data;
- int weight;
- int max;
- blk_iopoll_fn *poll;
-};
-
-enum {
- IOPOLL_F_SCHED = 0,
- IOPOLL_F_DISABLE = 1,
-};
-
-/*
- * Returns 0 if we successfully set the IOPOLL_F_SCHED bit, indicating
- * that we were the first to acquire this iop for scheduling. If this iop
- * is currently disabled, return "failure".
- */
-static inline int blk_iopoll_sched_prep(struct blk_iopoll *iop)
-{
- if (!test_bit(IOPOLL_F_DISABLE, &iop->state))
- return test_and_set_bit(IOPOLL_F_SCHED, &iop->state);
-
- return 1;
-}
-
-static inline int blk_iopoll_disable_pending(struct blk_iopoll *iop)
-{
- return test_bit(IOPOLL_F_DISABLE, &iop->state);
-}
-
-extern void blk_iopoll_sched(struct blk_iopoll *);
-extern void blk_iopoll_init(struct blk_iopoll *, int, blk_iopoll_fn *);
-extern void blk_iopoll_complete(struct blk_iopoll *);
-extern void __blk_iopoll_complete(struct blk_iopoll *);
-extern void blk_iopoll_enable(struct blk_iopoll *);
-extern void blk_iopoll_disable(struct blk_iopoll *);
-
-#endif
diff --git a/include/linux/interrupt.h b/include/linux/interrupt.h
index ad16809..7ff98c2 100644
--- a/include/linux/interrupt.h
+++ b/include/linux/interrupt.h
@@ -412,7 +412,7 @@ enum
NET_TX_SOFTIRQ,
NET_RX_SOFTIRQ,
BLOCK_SOFTIRQ,
- BLOCK_IOPOLL_SOFTIRQ,
+ IRQ_POLL_SOFTIRQ,
TASKLET_SOFTIRQ,
SCHED_SOFTIRQ,
HRTIMER_SOFTIRQ, /* Unused, but kept as tools rely on the
diff --git a/include/linux/irq_poll.h b/include/linux/irq_poll.h
new file mode 100644
index 0000000..0cf7c26
--- /dev/null
+++ b/include/linux/irq_poll.h
@@ -0,0 +1,46 @@
+#ifndef IRQ_POLL_H
+#define IRQ_POLL_H
+
+struct irq_poll;
+typedef int (irq_poll_fn)(struct irq_poll *, int);
+
+struct irq_poll {
+ struct list_head list;
+ unsigned long state;
+ unsigned long data;
+ int weight;
+ int max;
+ irq_poll_fn *poll;
+};
+
+enum {
+ IRQ_POLL_F_SCHED = 0,
+ IRQ_POLL_F_DISABLE = 1,
+};
+
+/*
+ * Returns 0 if we successfully set the IRQ_POLL_F_SCHED bit, indicating
+ * that we were the first to acquire this iop for scheduling. If this iop
+ * is currently disabled, return "failure".
+ */
+static inline int irq_poll_sched_prep(struct irq_poll *iop)
+{
+ if (!test_bit(IRQ_POLL_F_DISABLE, &iop->state))
+ return test_and_set_bit(IRQ_POLL_F_SCHED, &iop->state);
+
+ return 1;
+}
+
+static inline int irq_poll_disable_pending(struct irq_poll *iop)
+{
+ return test_bit(IRQ_POLL_F_DISABLE, &iop->state);
+}
+
+extern void irq_poll_sched(struct irq_poll *);
+extern void irq_poll_init(struct irq_poll *, int, irq_poll_fn *);
+extern void irq_poll_complete(struct irq_poll *);
+extern void __irq_poll_complete(struct irq_poll *);
+extern void irq_poll_enable(struct irq_poll *);
+extern void irq_poll_disable(struct irq_poll *);
+
+#endif
diff --git a/include/trace/events/irq.h b/include/trace/events/irq.h
index ff8f6c0..f95f25e 100644
--- a/include/trace/events/irq.h
+++ b/include/trace/events/irq.h
@@ -15,7 +15,7 @@ struct softirq_action;
softirq_name(NET_TX) \
softirq_name(NET_RX) \
softirq_name(BLOCK) \
- softirq_name(BLOCK_IOPOLL) \
+ softirq_name(IRQ_POLL) \
softirq_name(TASKLET) \
softirq_name(SCHED) \
softirq_name(HRTIMER) \
diff --git a/lib/Kconfig b/lib/Kconfig
index f0df318..e00e196 100644
--- a/lib/Kconfig
+++ b/lib/Kconfig
@@ -475,6 +475,11 @@ config DDR
information. This data is useful for drivers handling
DDR SDRAM controllers.
+config IRQ_POLL
+ bool "IRQ polling library"
+ help
+ Helper library to poll interrupt mitigation using polling.
+
config MPILIB
tristate
select CLZ_TAB
diff --git a/lib/Makefile b/lib/Makefile
index 7f1de26..1478ae2 100644
--- a/lib/Makefile
+++ b/lib/Makefile
@@ -164,6 +164,7 @@ obj-$(CONFIG_GENERIC_NET_UTILS) += net_utils.o
obj-$(CONFIG_SG_SPLIT) += sg_split.o
obj-$(CONFIG_STMP_DEVICE) += stmp_device.o
+obj-$(CONFIG_IRQ_POLL) += irq_poll.o
libfdt_files = fdt.o fdt_ro.o fdt_wip.o fdt_rw.o fdt_sw.o fdt_strerror.o \
fdt_empty_tree.o
diff --git a/lib/irq_poll.c b/lib/irq_poll.c
new file mode 100644
index 0000000..e6fd1dc
--- /dev/null
+++ b/lib/irq_poll.c
@@ -0,0 +1,221 @@
+/*
+ * Functions related to interrupt-poll handling in the block layer. This
+ * is similar to NAPI for network devices.
+ */
+#include <linux/kernel.h>
+#include <linux/module.h>
+#include <linux/init.h>
+#include <linux/bio.h>
+#include <linux/interrupt.h>
+#include <linux/cpu.h>
+#include <linux/irq_poll.h>
+#include <linux/delay.h>
+
+static unsigned int irq_poll_budget __read_mostly = 256;
+
+static DEFINE_PER_CPU(struct list_head, blk_cpu_iopoll);
+
+/**
+ * irq_poll_sched - Schedule a run of the iopoll handler
+ * @iop: The parent iopoll structure
+ *
+ * Description:
+ * Add this irq_poll structure to the pending poll list and trigger the
+ * raise of the blk iopoll softirq. The driver must already have gotten a
+ * successful return from irq_poll_sched_prep() before calling this.
+ **/
+void irq_poll_sched(struct irq_poll *iop)
+{
+ unsigned long flags;
+
+ local_irq_save(flags);
+ list_add_tail(&iop->list, this_cpu_ptr(&blk_cpu_iopoll));
+ __raise_softirq_irqoff(IRQ_POLL_SOFTIRQ);
+ local_irq_restore(flags);
+}
+EXPORT_SYMBOL(irq_poll_sched);
+
+/**
+ * __irq_poll_complete - Mark this @iop as un-polled again
+ * @iop: The parent iopoll structure
+ *
+ * Description:
+ * See irq_poll_complete(). This function must be called with interrupts
+ * disabled.
+ **/
+void __irq_poll_complete(struct irq_poll *iop)
+{
+ list_del(&iop->list);
+ smp_mb__before_atomic();
+ clear_bit_unlock(IRQ_POLL_F_SCHED, &iop->state);
+}
+EXPORT_SYMBOL(__irq_poll_complete);
+
+/**
+ * irq_poll_complete - Mark this @iop as un-polled again
+ * @iop: The parent iopoll structure
+ *
+ * Description:
+ * If a driver consumes less than the assigned budget in its run of the
+ * iopoll handler, it'll end the polled mode by calling this function. The
+ * iopoll handler will not be invoked again before irq_poll_sched_prep()
+ * is called.
+ **/
+void irq_poll_complete(struct irq_poll *iop)
+{
+ unsigned long flags;
+
+ local_irq_save(flags);
+ __irq_poll_complete(iop);
+ local_irq_restore(flags);
+}
+EXPORT_SYMBOL(irq_poll_complete);
+
+static void irq_poll_softirq(struct softirq_action *h)
+{
+ struct list_head *list = this_cpu_ptr(&blk_cpu_iopoll);
+ int rearm = 0, budget = irq_poll_budget;
+ unsigned long start_time = jiffies;
+
+ local_irq_disable();
+
+ while (!list_empty(list)) {
+ struct irq_poll *iop;
+ int work, weight;
+
+ /*
+ * If softirq window is exhausted then punt.
+ */
+ if (budget <= 0 || time_after(jiffies, start_time)) {
+ rearm = 1;
+ break;
+ }
+
+ local_irq_enable();
+
+ /* Even though interrupts have been re-enabled, this
+ * access is safe because interrupts can only add new
+ * entries to the tail of this list, and only ->poll()
+ * calls can remove this head entry from the list.
+ */
+ iop = list_entry(list->next, struct irq_poll, list);
+
+ weight = iop->weight;
+ work = 0;
+ if (test_bit(IRQ_POLL_F_SCHED, &iop->state))
+ work = iop->poll(iop, weight);
+
+ budget -= work;
+
+ local_irq_disable();
+
+ /*
+ * Drivers must not modify the iopoll state, if they
+ * consume their assigned weight (or more, some drivers can't
+ * easily just stop processing, they have to complete an
+ * entire mask of commands).In such cases this code
+ * still "owns" the iopoll instance and therefore can
+ * move the instance around on the list at-will.
+ */
+ if (work >= weight) {
+ if (irq_poll_disable_pending(iop))
+ __irq_poll_complete(iop);
+ else
+ list_move_tail(&iop->list, list);
+ }
+ }
+
+ if (rearm)
+ __raise_softirq_irqoff(IRQ_POLL_SOFTIRQ);
+
+ local_irq_enable();
+}
+
+/**
+ * irq_poll_disable - Disable iopoll on this @iop
+ * @iop: The parent iopoll structure
+ *
+ * Description:
+ * Disable io polling and wait for any pending callbacks to have completed.
+ **/
+void irq_poll_disable(struct irq_poll *iop)
+{
+ set_bit(IRQ_POLL_F_DISABLE, &iop->state);
+ while (test_and_set_bit(IRQ_POLL_F_SCHED, &iop->state))
+ msleep(1);
+ clear_bit(IRQ_POLL_F_DISABLE, &iop->state);
+}
+EXPORT_SYMBOL(irq_poll_disable);
+
+/**
+ * irq_poll_enable - Enable iopoll on this @iop
+ * @iop: The parent iopoll structure
+ *
+ * Description:
+ * Enable iopoll on this @iop. Note that the handler run will not be
+ * scheduled, it will only mark it as active.
+ **/
+void irq_poll_enable(struct irq_poll *iop)
+{
+ BUG_ON(!test_bit(IRQ_POLL_F_SCHED, &iop->state));
+ smp_mb__before_atomic();
+ clear_bit_unlock(IRQ_POLL_F_SCHED, &iop->state);
+}
+EXPORT_SYMBOL(irq_poll_enable);
+
+/**
+ * irq_poll_init - Initialize this @iop
+ * @iop: The parent iopoll structure
+ * @weight: The default weight (or command completion budget)
+ * @poll_fn: The handler to invoke
+ *
+ * Description:
+ * Initialize this irq_poll structure. Before being actively used, the
+ * driver must call irq_poll_enable().
+ **/
+void irq_poll_init(struct irq_poll *iop, int weight, irq_poll_fn *poll_fn)
+{
+ memset(iop, 0, sizeof(*iop));
+ INIT_LIST_HEAD(&iop->list);
+ iop->weight = weight;
+ iop->poll = poll_fn;
+ set_bit(IRQ_POLL_F_SCHED, &iop->state);
+}
+EXPORT_SYMBOL(irq_poll_init);
+
+static int irq_poll_cpu_notify(struct notifier_block *self,
+ unsigned long action, void *hcpu)
+{
+ /*
+ * If a CPU goes away, splice its entries to the current CPU
+ * and trigger a run of the softirq
+ */
+ if (action == CPU_DEAD || action == CPU_DEAD_FROZEN) {
+ int cpu = (unsigned long) hcpu;
+
+ local_irq_disable();
+ list_splice_init(&per_cpu(blk_cpu_iopoll, cpu),
+ this_cpu_ptr(&blk_cpu_iopoll));
+ __raise_softirq_irqoff(IRQ_POLL_SOFTIRQ);
+ local_irq_enable();
+ }
+
+ return NOTIFY_OK;
+}
+
+static struct notifier_block irq_poll_cpu_notifier = {
+ .notifier_call = irq_poll_cpu_notify,
+};
+
+static __init int irq_poll_setup(void)
+{
+ int i;
+
+ for_each_possible_cpu(i)
+ INIT_LIST_HEAD(&per_cpu(blk_cpu_iopoll, i));
+
+ open_softirq(IRQ_POLL_SOFTIRQ, irq_poll_softirq);
+ register_hotcpu_notifier(&irq_poll_cpu_notifier);
+ return 0;
+}
+subsys_initcall(irq_poll_setup);
diff --git a/tools/lib/traceevent/event-parse.c b/tools/lib/traceevent/event-parse.c
index 2a912df..af5a316 100644
--- a/tools/lib/traceevent/event-parse.c
+++ b/tools/lib/traceevent/event-parse.c
@@ -3746,7 +3746,7 @@ static const struct flag flags[] = {
{ "NET_TX_SOFTIRQ", 2 },
{ "NET_RX_SOFTIRQ", 3 },
{ "BLOCK_SOFTIRQ", 4 },
- { "BLOCK_IOPOLL_SOFTIRQ", 5 },
+ { "IRQ_POLL_SOFTIRQ", 5 },
{ "TASKLET_SOFTIRQ", 6 },
{ "SCHED_SOFTIRQ", 7 },
{ "HRTIMER_SOFTIRQ", 8 },
diff --git a/tools/perf/util/trace-event-parse.c b/tools/perf/util/trace-event-parse.c
index 8ff7d62..33b52ea 100644
--- a/tools/perf/util/trace-event-parse.c
+++ b/tools/perf/util/trace-event-parse.c
@@ -209,7 +209,7 @@ static const struct flag flags[] = {
{ "NET_TX_SOFTIRQ", 2 },
{ "NET_RX_SOFTIRQ", 3 },
{ "BLOCK_SOFTIRQ", 4 },
- { "BLOCK_IOPOLL_SOFTIRQ", 5 },
+ { "IRQ_POLL_SOFTIRQ", 5 },
{ "TASKLET_SOFTIRQ", 6 },
{ "SCHED_SOFTIRQ", 7 },
{ "HRTIMER_SOFTIRQ", 8 },
--
1.9.1
--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at http://vger.kernel.org/majordomo-info.html
Please read the FAQ at http://www.tux.org/lkml/
[toc] | [prev] | [next] | [standalone]
| From | Or Gerlitz <gerlitz.or@gmail.com> |
|---|---|
| Date | 2015-11-13 16:30 +0100 |
| Subject | Re: [PATCH 1/9] move blk_iopoll to limit and make it generally available |
| Message-ID | <quosy-88l-3@gated-at.bofh.it> |
| In reply to | #1268955 |
On Fri, Nov 13, 2015 at 3:46 PM, Christoph Hellwig <hch@lst.de> wrote: > The new name is irq_poll as iopoll is already taken. Better suggestions > welcome. Sagi (or Christoph if you can address that), @ some pointer over the last 18 months there was a port done at mellanox for iser to use blk-iopoll and AFAIR it didn't work well or didn't work at all. Can you tell now what was the problem and how did you address it at your generalization? Or. -- To unsubscribe from this list: send the line "unsubscribe linux-kernel" in the body of a message to majordomo@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html Please read the FAQ at http://www.tux.org/lkml/
[toc] | [prev] | [next] | [standalone]
| From | Christoph Hellwig <hch@lst.de> |
|---|---|
| Date | 2015-11-14 08:10 +0100 |
| Subject | Re: [PATCH 1/9] move blk_iopoll to limit and make it generally available |
| Message-ID | <quD8e-Wt-11@gated-at.bofh.it> |
| In reply to | #1269015 |
On Fri, Nov 13, 2015 at 05:23:39PM +0200, Or Gerlitz wrote: > On Fri, Nov 13, 2015 at 3:46 PM, Christoph Hellwig <hch@lst.de> wrote: > > The new name is irq_poll as iopoll is already taken. Better suggestions > > welcome. > > Sagi (or Christoph if you can address that), > > @ some pointer over the last 18 months there was a port done at > mellanox for iser to use blk-iopoll and AFAIR it didn't work well or > didn't work at all. Can you tell now what was the problem and how did > you address it at your generalization? Hi Or, Sagi mentioned last time he tried a similar approach in iSER he saw some large latency sparks. We've seen nothing worse than the original approach. The Flash memory summit slide set has some numbers: http://www.flashmemorysummit.com/English/Collaterals/Proceedings/2015/20150811_FA11_Bandic.pdf they aren't quite up to date, but the latency distribution hasn't really changed. -- To unsubscribe from this list: send the line "unsubscribe linux-kernel" in the body of a message to majordomo@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html Please read the FAQ at http://www.tux.org/lkml/
[toc] | [prev] | [next] | [standalone]
| From | Sagi Grimberg <sagig@dev.mellanox.co.il> |
|---|---|
| Date | 2015-11-15 09:50 +0100 |
| Subject | Re: [PATCH 1/9] move blk_iopoll to limit and make it generally available |
| Message-ID | <qv1ay-7t7-9@gated-at.bofh.it> |
| In reply to | #1269430 |
>> On Fri, Nov 13, 2015 at 3:46 PM, Christoph Hellwig <hch@lst.de> wrote: >>> The new name is irq_poll as iopoll is already taken. Better suggestions >>> welcome. >> >> Sagi (or Christoph if you can address that), >> >> @ some pointer over the last 18 months there was a port done at >> mellanox for iser to use blk-iopoll and AFAIR it didn't work well or >> didn't work at all. Can you tell now what was the problem and how did >> you address it at your generalization? > > Hi Or, > > Sagi mentioned last time he tried a similar approach in iSER he saw > some large latency sparks. We've seen nothing worse than the original > approach. The Flash memory summit slide set has some numbers: > > http://www.flashmemorysummit.com/English/Collaterals/Proceedings/2015/20150811_FA11_Bandic.pdf > > they aren't quite up to date, but the latency distribution hasn't > really changed. Or is correct, I have attempted to convert iser to use blk_iopoll in the past, however I've seen inconsistent performance and latency skews (comparing to tasklets iser is using today). This was manifested in IOPs test cases where I ran multiple threads with higher queue-depth and not in sanitized pure latency (QD=1) test cases. Unfortunately I didn't have the time to pick it up since. I do have every intention of testing it again with this. If it still exist we will need to find the root-cause of it before converting drivers to use it. Sagi. -- To unsubscribe from this list: send the line "unsubscribe linux-kernel" in the body of a message to majordomo@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html Please read the FAQ at http://www.tux.org/lkml/
[toc] | [prev] | [next] | [standalone]
| From | Or Gerlitz <gerlitz.or@gmail.com> |
|---|---|
| Date | 2015-11-15 10:10 +0100 |
| Subject | Re: [PATCH 1/9] move blk_iopoll to limit and make it generally available |
| Message-ID | <qv1tU-7Qe-5@gated-at.bofh.it> |
| In reply to | #1269652 |
On Sun, Nov 15, 2015 at 10:48 AM, Sagi Grimberg <sagig@dev.mellanox.co.il> wrote: > Or is correct, > > I have attempted to convert iser to use blk_iopoll in the past, however > I've seen inconsistent performance and latency skews (comparing to > tasklets iser is using today). This was manifested in IOPs test cases > where I ran multiple threads with higher queue-depth and not in > sanitized pure latency (QD=1) test cases. Unfortunately I didn't have > the time to pick it up since. > > I do have every intention of testing it again with this. If it still > exist we will need to find the root-cause of it before converting > drivers to use it. Good, this way (inconsistent performance and latency skews) or another (all shines up) -- please let us know your findings, best through commenting within V > 0 the cover letter posts of this series -- To unsubscribe from this list: send the line "unsubscribe linux-kernel" in the body of a message to majordomo@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html Please read the FAQ at http://www.tux.org/lkml/
[toc] | [prev] | [next] | [standalone]
| From | Sagi Grimberg <sagig@dev.mellanox.co.il> |
|---|---|
| Date | 2015-11-15 14:20 +0100 |
| Subject | Re: [PATCH 1/9] move blk_iopoll to limit and make it generally available |
| Message-ID | <qv5nQ-1Ta-3@gated-at.bofh.it> |
| In reply to | #1269654 |
On 15/11/2015 11:04, Or Gerlitz wrote: > On Sun, Nov 15, 2015 at 10:48 AM, Sagi Grimberg > <sagig@dev.mellanox.co.il> wrote: >> Or is correct, >> >> I have attempted to convert iser to use blk_iopoll in the past, however >> I've seen inconsistent performance and latency skews (comparing to >> tasklets iser is using today). This was manifested in IOPs test cases >> where I ran multiple threads with higher queue-depth and not in >> sanitized pure latency (QD=1) test cases. Unfortunately I didn't have >> the time to pick it up since. >> >> I do have every intention of testing it again with this. If it still >> exist we will need to find the root-cause of it before converting >> drivers to use it. > > Good, this way (inconsistent performance and latency skews) or another > (all shines up) -- please > let us know your findings, best through commenting within V > 0 the > cover letter posts of this series > Hi Or & Co, I ran some tests on the iser code with this patchset applied. I can confirm that I did not see any performance degradations. summary (on my test servers): 1 LUN: ~530K (IOPs) 2 LUNs: ~1080K (IOPs) 4 LUNs: ~1350K (IOPs) 8 LUNs: ~1930K (IOPs) 16 LUns: ~2250K (IOPs) These results are true both for tasklet and iopoll. So, I don't have anything smart to say here, the IO stack (block, scsi) has gone through major changes since the last time I looked into this, so it'll be pretty hard to figure out what was the root cause back then... Sagi. -- To unsubscribe from this list: send the line "unsubscribe linux-kernel" in the body of a message to majordomo@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html Please read the FAQ at http://www.tux.org/lkml/
[toc] | [prev] | [next] | [standalone]
| From | Christoph Hellwig <hch@lst.de> |
|---|---|
| Date | 2015-11-15 14:00 +0100 |
| Subject | Re: [PATCH 1/9] move blk_iopoll to limit and make it generally available |
| Message-ID | <qv54u-1vn-15@gated-at.bofh.it> |
| In reply to | #1269652 |
On Sun, Nov 15, 2015 at 10:48:41AM +0200, Sagi Grimberg wrote:
> I have attempted to convert iser to use blk_iopoll in the past, however
> I've seen inconsistent performance and latency skews (comparing to
> tasklets iser is using today). This was manifested in IOPs test cases
> where I ran multiple threads with higher queue-depth and not in
> sanitized pure latency (QD=1) test cases. Unfortunately I didn't have
> the time to pick it up since.
>
> I do have every intention of testing it again with this. If it still
> exist we will need to find the root-cause of it before converting
> drivers to use it.
Thanks. If you see issue like that with high iops and high loads
the next thing to check are:
a) increasing the budget (both in the main poll loop and the IB
code)
b) try without the single jiffie time limit in the main softirq handler
--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at http://vger.kernel.org/majordomo-info.html
Please read the FAQ at http://www.tux.org/lkml/
[toc] | [prev] | [next] | [standalone]
| From | Christoph Hellwig <hch@lst.de> |
|---|---|
| Date | 2015-11-14 08:10 +0100 |
| Subject | Re: [PATCH 1/9] move blk_iopoll to limit and make it generally available |
| Message-ID | <quD8e-Wt-15@gated-at.bofh.it> |
| In reply to | #1268955 |
On Fri, Nov 13, 2015 at 11:19:24AM -0800, Bart Van Assche wrote: > On 11/13/2015 05:46 AM, Christoph Hellwig wrote: >> The new name is irq_poll as iopoll is already taken. Better suggestions >> welcome. > > Hello Christoph, > > Would it be possible to provide more background information about this ? > Which other kernel subsystem is using the name iopoll ? Take a look at include/linux/iopoll.h - I can't reaplly make much sense of it to be honest, but it's used in a quite a few places. -- To unsubscribe from this list: send the line "unsubscribe linux-kernel" in the body of a message to majordomo@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html Please read the FAQ at http://www.tux.org/lkml/
[toc] | [prev] | [next] | [standalone]
| From | Christoph Hellwig <hch@lst.de> |
|---|---|
| Date | 2015-11-13 15:00 +0100 |
| Subject | [PATCH 6/9] srp: use the new CQ API |
| Message-ID | <qun3y-77o-113@gated-at.bofh.it> |
| In reply to | #1268953 |
This also moves recv completion handling from hardirq context into
softirq context.
Signed-off-by: Christoph Hellwig <hch@lst.de>
---
drivers/infiniband/ulp/srp/ib_srp.c | 198 ++++++++++++++----------------------
drivers/infiniband/ulp/srp/ib_srp.h | 7 +-
2 files changed, 76 insertions(+), 129 deletions(-)
diff --git a/drivers/infiniband/ulp/srp/ib_srp.c b/drivers/infiniband/ulp/srp/ib_srp.c
index 3027824..57237e1 100644
--- a/drivers/infiniband/ulp/srp/ib_srp.c
+++ b/drivers/infiniband/ulp/srp/ib_srp.c
@@ -132,8 +132,9 @@ MODULE_PARM_DESC(ch_count,
static void srp_add_one(struct ib_device *device);
static void srp_remove_one(struct ib_device *device, void *client_data);
-static void srp_recv_completion(struct ib_cq *cq, void *ch_ptr);
-static void srp_send_completion(struct ib_cq *cq, void *ch_ptr);
+static void srp_recv_done(struct ib_cq *cq, struct ib_wc *wc);
+static void srp_handle_qp_err(struct ib_cq *cq, struct ib_wc *wc,
+ const char *opname);
static int srp_cm_handler(struct ib_cm_id *cm_id, struct ib_cm_event *event);
static struct scsi_transport_template *ib_srp_transport_template;
@@ -445,41 +446,6 @@ static struct srp_fr_pool *srp_alloc_fr_pool(struct srp_target_port *target)
dev->max_pages_per_mr);
}
-/**
- * srp_destroy_qp() - destroy an RDMA queue pair
- * @ch: SRP RDMA channel.
- *
- * Change a queue pair into the error state and wait until all receive
- * completions have been processed before destroying it. This avoids that
- * the receive completion handler can access the queue pair while it is
- * being destroyed.
- */
-static void srp_destroy_qp(struct srp_rdma_ch *ch)
-{
- static struct ib_qp_attr attr = { .qp_state = IB_QPS_ERR };
- static struct ib_recv_wr wr = { 0 };
- struct ib_recv_wr *bad_wr;
- int ret;
-
- wr.wr_id = SRP_LAST_WR_ID;
- /* Destroying a QP and reusing ch->done is only safe if not connected */
- WARN_ON_ONCE(ch->connected);
-
- ret = ib_modify_qp(ch->qp, &attr, IB_QP_STATE);
- WARN_ONCE(ret, "ib_cm_init_qp_attr() returned %d\n", ret);
- if (ret)
- goto out;
-
- init_completion(&ch->done);
- ret = ib_post_recv(ch->qp, &wr, &bad_wr);
- WARN_ONCE(ret, "ib_post_recv() returned %d\n", ret);
- if (ret == 0)
- wait_for_completion(&ch->done);
-
-out:
- ib_destroy_qp(ch->qp);
-}
-
static int srp_create_ch_ib(struct srp_rdma_ch *ch)
{
struct srp_target_port *target = ch->target;
@@ -490,34 +456,27 @@ static int srp_create_ch_ib(struct srp_rdma_ch *ch)
struct ib_fmr_pool *fmr_pool = NULL;
struct srp_fr_pool *fr_pool = NULL;
const int m = 1 + dev->use_fast_reg;
- struct ib_cq_init_attr cq_attr = {};
int ret;
init_attr = kzalloc(sizeof *init_attr, GFP_KERNEL);
if (!init_attr)
return -ENOMEM;
- /* + 1 for SRP_LAST_WR_ID */
- cq_attr.cqe = target->queue_size + 1;
- cq_attr.comp_vector = ch->comp_vector;
- recv_cq = ib_create_cq(dev->dev, srp_recv_completion, NULL, ch,
- &cq_attr);
+ /* queue_size + 1 for ib_drain_qp */
+ recv_cq = ib_alloc_cq(dev->dev, ch, target->queue_size + 1, ch->comp_vector,
+ IB_POLL_SOFTIRQ);
if (IS_ERR(recv_cq)) {
ret = PTR_ERR(recv_cq);
goto err;
}
- cq_attr.cqe = m * target->queue_size;
- cq_attr.comp_vector = ch->comp_vector;
- send_cq = ib_create_cq(dev->dev, srp_send_completion, NULL, ch,
- &cq_attr);
+ send_cq = ib_alloc_cq(dev->dev, ch, m * target->queue_size, ch->comp_vector,
+ IB_POLL_DIRECT);
if (IS_ERR(send_cq)) {
ret = PTR_ERR(send_cq);
goto err_recv_cq;
}
- ib_req_notify_cq(recv_cq, IB_CQ_NEXT_COMP);
-
init_attr->event_handler = srp_qp_event;
init_attr->cap.max_send_wr = m * target->queue_size;
init_attr->cap.max_recv_wr = target->queue_size + 1;
@@ -557,11 +516,11 @@ static int srp_create_ch_ib(struct srp_rdma_ch *ch)
}
if (ch->qp)
- srp_destroy_qp(ch);
+ ib_destroy_qp(ch->qp);
if (ch->recv_cq)
- ib_destroy_cq(ch->recv_cq);
+ ib_free_cq(ch->recv_cq);
if (ch->send_cq)
- ib_destroy_cq(ch->send_cq);
+ ib_free_cq(ch->send_cq);
ch->qp = qp;
ch->recv_cq = recv_cq;
@@ -584,10 +543,10 @@ err_qp:
ib_destroy_qp(qp);
err_send_cq:
- ib_destroy_cq(send_cq);
+ ib_free_cq(send_cq);
err_recv_cq:
- ib_destroy_cq(recv_cq);
+ ib_free_cq(recv_cq);
err:
kfree(init_attr);
@@ -623,9 +582,11 @@ static void srp_free_ch_ib(struct srp_target_port *target,
if (ch->fmr_pool)
ib_destroy_fmr_pool(ch->fmr_pool);
}
- srp_destroy_qp(ch);
- ib_destroy_cq(ch->send_cq);
- ib_destroy_cq(ch->recv_cq);
+
+ ib_drain_qp(ch->qp);
+ ib_destroy_qp(ch->qp);
+ ib_free_cq(ch->send_cq);
+ ib_free_cq(ch->recv_cq);
/*
* Avoid that the SCSI error handler tries to use this channel after
@@ -1038,7 +999,13 @@ static int srp_connect_ch(struct srp_rdma_ch *ch, bool multich)
}
}
-static int srp_inv_rkey(struct srp_rdma_ch *ch, u32 rkey)
+static void srp_inv_rkey_done(struct ib_cq *cq, struct ib_wc *wc)
+{
+ srp_handle_qp_err(cq, wc, "INV RKEY");
+}
+
+static int srp_inv_rkey(struct srp_request *req, struct srp_rdma_ch *ch,
+ u32 rkey)
{
struct ib_send_wr *bad_wr;
struct ib_send_wr wr = {
@@ -1049,8 +1016,8 @@ static int srp_inv_rkey(struct srp_rdma_ch *ch, u32 rkey)
.ex.invalidate_rkey = rkey,
};
- wr.wr_id = LOCAL_INV_WR_ID_MASK;
-
+ wr.wr_cqe = &req->reg_cqe;
+ req->reg_cqe.done = srp_inv_rkey_done;
return ib_post_send(ch->qp, &wr, &bad_wr);
}
@@ -1072,7 +1039,7 @@ static void srp_unmap_data(struct scsi_cmnd *scmnd,
struct srp_fr_desc **pfr;
for (i = req->nmdesc, pfr = req->fr_list; i > 0; i--, pfr++) {
- res = srp_inv_rkey(ch, (*pfr)->mr->rkey);
+ res = srp_inv_rkey(req, ch, (*pfr)->mr->rkey);
if (res < 0) {
shost_printk(KERN_ERR, target->scsi_host, PFX
"Queueing INV WR for rkey %#x failed (%d)\n",
@@ -1310,7 +1277,13 @@ reset_state:
return 0;
}
+static void srp_reg_mr_done(struct ib_cq *cq, struct ib_wc *wc)
+{
+ srp_handle_qp_err(cq, wc, "FAST REG");
+}
+
static int srp_map_finish_fr(struct srp_map_state *state,
+ struct srp_request *req,
struct srp_rdma_ch *ch)
{
struct srp_target_port *target = ch->target;
@@ -1348,9 +1321,11 @@ static int srp_map_finish_fr(struct srp_map_state *state,
if (unlikely(n < 0))
return n;
+ req->reg_cqe.done = srp_reg_mr_done;
+
wr.wr.next = NULL;
wr.wr.opcode = IB_WR_REG_MR;
- wr.wr.wr_id = FAST_REG_WR_ID_MASK;
+ wr.wr.wr_cqe = &req->reg_cqe;
wr.wr.num_sge = 0;
wr.wr.send_flags = 0;
wr.mr = desc->mr;
@@ -1455,7 +1430,7 @@ static int srp_map_sg_fr(struct srp_map_state *state, struct srp_rdma_ch *ch,
while (state->sg_nents) {
int i, n;
- n = srp_map_finish_fr(state, ch);
+ n = srp_map_finish_fr(state, req, ch);
if (unlikely(n < 0))
return n;
@@ -1522,7 +1497,7 @@ static int srp_map_idb(struct srp_rdma_ch *ch, struct srp_request *req,
state.sg_nents = 1;
sg_set_buf(idb_sg, req->indirect_desc, idb_len);
idb_sg->dma_address = req->indirect_dma_addr; /* hack! */
- ret = srp_map_finish_fr(&state, ch);
+ ret = srp_map_finish_fr(&state, req, ch);
if (ret < 0)
return ret;
} else if (dev->use_fmr) {
@@ -1717,7 +1692,7 @@ static struct srp_iu *__srp_get_tx_iu(struct srp_rdma_ch *ch,
s32 rsv = (iu_type == SRP_IU_TSK_MGMT) ? 0 : SRP_TSK_MGMT_SQ_SIZE;
struct srp_iu *iu;
- srp_send_completion(ch->send_cq, ch);
+ ib_process_cq_direct(ch->send_cq);
if (list_empty(&ch->free_tx))
return NULL;
@@ -1737,6 +1712,19 @@ static struct srp_iu *__srp_get_tx_iu(struct srp_rdma_ch *ch,
return iu;
}
+static void srp_send_done(struct ib_cq *cq, struct ib_wc *wc)
+{
+ struct srp_iu *iu = container_of(wc->wr_cqe, struct srp_iu, cqe);
+ struct srp_rdma_ch *ch = cq->cq_context;
+
+ if (likely(wc->status != IB_WC_SUCCESS)) {
+ srp_handle_qp_err(cq, wc, "SEND");
+ return;
+ }
+
+ list_add(&iu->list, &ch->free_tx);
+}
+
static int srp_post_send(struct srp_rdma_ch *ch, struct srp_iu *iu, int len)
{
struct srp_target_port *target = ch->target;
@@ -1747,8 +1735,10 @@ static int srp_post_send(struct srp_rdma_ch *ch, struct srp_iu *iu, int len)
list.length = len;
list.lkey = target->lkey;
+ iu->cqe.done = srp_send_done;
+
wr.next = NULL;
- wr.wr_id = (uintptr_t) iu;
+ wr.wr_cqe = &iu->cqe;
wr.sg_list = &list;
wr.num_sge = 1;
wr.opcode = IB_WR_SEND;
@@ -1767,8 +1757,10 @@ static int srp_post_recv(struct srp_rdma_ch *ch, struct srp_iu *iu)
list.length = iu->size;
list.lkey = target->lkey;
+ iu->cqe.done = srp_recv_done;
+
wr.next = NULL;
- wr.wr_id = (uintptr_t) iu;
+ wr.wr_cqe = &iu->cqe;
wr.sg_list = &list;
wr.num_sge = 1;
@@ -1900,14 +1892,20 @@ static void srp_process_aer_req(struct srp_rdma_ch *ch,
"problems processing SRP_AER_REQ\n");
}
-static void srp_handle_recv(struct srp_rdma_ch *ch, struct ib_wc *wc)
+static void srp_recv_done(struct ib_cq *cq, struct ib_wc *wc)
{
+ struct srp_iu *iu = container_of(wc->wr_cqe, struct srp_iu, cqe);
+ struct srp_rdma_ch *ch = cq->cq_context;
struct srp_target_port *target = ch->target;
struct ib_device *dev = target->srp_host->srp_dev->dev;
- struct srp_iu *iu = (struct srp_iu *) (uintptr_t) wc->wr_id;
int res;
u8 opcode;
+ if (unlikely(wc->status != IB_WC_SUCCESS)) {
+ srp_handle_qp_err(cq, wc, "RECV");
+ return;
+ }
+
ib_dma_sync_single_for_cpu(dev, iu->dma, ch->max_ti_iu_len,
DMA_FROM_DEVICE);
@@ -1970,68 +1968,22 @@ static void srp_tl_err_work(struct work_struct *work)
srp_start_tl_fail_timers(target->rport);
}
-static void srp_handle_qp_err(u64 wr_id, enum ib_wc_status wc_status,
- bool send_err, struct srp_rdma_ch *ch)
+static void srp_handle_qp_err(struct ib_cq *cq, struct ib_wc *wc,
+ const char *opname)
{
+ struct srp_rdma_ch *ch = cq->cq_context;
struct srp_target_port *target = ch->target;
- if (wr_id == SRP_LAST_WR_ID) {
- complete(&ch->done);
- return;
- }
-
if (ch->connected && !target->qp_in_error) {
- if (wr_id & LOCAL_INV_WR_ID_MASK) {
- shost_printk(KERN_ERR, target->scsi_host, PFX
- "LOCAL_INV failed with status %s (%d)\n",
- ib_wc_status_msg(wc_status), wc_status);
- } else if (wr_id & FAST_REG_WR_ID_MASK) {
- shost_printk(KERN_ERR, target->scsi_host, PFX
- "FAST_REG_MR failed status %s (%d)\n",
- ib_wc_status_msg(wc_status), wc_status);
- } else {
- shost_printk(KERN_ERR, target->scsi_host,
- PFX "failed %s status %s (%d) for iu %p\n",
- send_err ? "send" : "receive",
- ib_wc_status_msg(wc_status), wc_status,
- (void *)(uintptr_t)wr_id);
- }
+ shost_printk(KERN_ERR, target->scsi_host,
+ PFX "failed %s status %s (%d) for CQE %p\n",
+ opname, ib_wc_status_msg(wc->status), wc->status,
+ wc->wr_cqe);
queue_work(system_long_wq, &target->tl_err_work);
}
target->qp_in_error = true;
}
-static void srp_recv_completion(struct ib_cq *cq, void *ch_ptr)
-{
- struct srp_rdma_ch *ch = ch_ptr;
- struct ib_wc wc;
-
- ib_req_notify_cq(cq, IB_CQ_NEXT_COMP);
- while (ib_poll_cq(cq, 1, &wc) > 0) {
- if (likely(wc.status == IB_WC_SUCCESS)) {
- srp_handle_recv(ch, &wc);
- } else {
- srp_handle_qp_err(wc.wr_id, wc.status, false, ch);
- }
- }
-}
-
-static void srp_send_completion(struct ib_cq *cq, void *ch_ptr)
-{
- struct srp_rdma_ch *ch = ch_ptr;
- struct ib_wc wc;
- struct srp_iu *iu;
-
- while (ib_poll_cq(cq, 1, &wc) > 0) {
- if (likely(wc.status == IB_WC_SUCCESS)) {
- iu = (struct srp_iu *) (uintptr_t) wc.wr_id;
- list_add(&iu->list, &ch->free_tx);
- } else {
- srp_handle_qp_err(wc.wr_id, wc.status, true, ch);
- }
- }
-}
-
static int srp_queuecommand(struct Scsi_Host *shost, struct scsi_cmnd *scmnd)
{
struct srp_target_port *target = host_to_target(shost);
@@ -3576,8 +3528,6 @@ static int __init srp_init_module(void)
{
int ret;
- BUILD_BUG_ON(FIELD_SIZEOF(struct ib_wc, wr_id) < sizeof(void *));
-
if (srp_sg_tablesize) {
pr_warn("srp_sg_tablesize is deprecated, please use cmd_sg_entries\n");
if (!cmd_sg_entries)
diff --git a/drivers/infiniband/ulp/srp/ib_srp.h b/drivers/infiniband/ulp/srp/ib_srp.h
index 87a2a91..7fec482 100644
--- a/drivers/infiniband/ulp/srp/ib_srp.h
+++ b/drivers/infiniband/ulp/srp/ib_srp.h
@@ -66,11 +66,6 @@ enum {
SRP_TAG_TSK_MGMT = 1U << 31,
SRP_MAX_PAGES_PER_MR = 512,
-
- LOCAL_INV_WR_ID_MASK = 1,
- FAST_REG_WR_ID_MASK = 2,
-
- SRP_LAST_WR_ID = 0xfffffffcU,
};
enum srp_target_state {
@@ -128,6 +123,7 @@ struct srp_request {
struct srp_direct_buf *indirect_desc;
dma_addr_t indirect_dma_addr;
short nmdesc;
+ struct ib_cqe reg_cqe;
};
/**
@@ -231,6 +227,7 @@ struct srp_iu {
void *buf;
size_t size;
enum dma_data_direction direction;
+ struct ib_cqe cqe;
};
/**
--
1.9.1
--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at http://vger.kernel.org/majordomo-info.html
Please read the FAQ at http://www.tux.org/lkml/
[toc] | [prev] | [next] | [standalone]
| From | Christoph Hellwig <hch@lst.de> |
|---|---|
| Date | 2015-11-13 15:00 +0100 |
| Subject | [PATCH 4/9] srpt: chain RDMA READ/WRITE requests |
| Message-ID | <qun3y-77o-125@gated-at.bofh.it> |
| In reply to | #1268953 |
Remove struct rdma_iu and instead allocate the struct ib_rdma_wr array
early and fill out directly. This allows us to chain the WRs, and thus
archive both less lock contention on the HCA workqueue as well as much
simpler error handling.
Signed-off-by: Christoph Hellwig <hch@lst.de>
---
drivers/infiniband/ulp/srpt/ib_srpt.c | 100 +++++++++++++---------------------
drivers/infiniband/ulp/srpt/ib_srpt.h | 14 +----
2 files changed, 39 insertions(+), 75 deletions(-)
diff --git a/drivers/infiniband/ulp/srpt/ib_srpt.c b/drivers/infiniband/ulp/srpt/ib_srpt.c
index 14b361a..2b6dd71 100644
--- a/drivers/infiniband/ulp/srpt/ib_srpt.c
+++ b/drivers/infiniband/ulp/srpt/ib_srpt.c
@@ -1057,7 +1057,7 @@ static void srpt_unmap_sg_to_ib_sge(struct srpt_rdma_ch *ch,
BUG_ON(ioctx->n_rdma && !ioctx->rdma_ius);
while (ioctx->n_rdma)
- kfree(ioctx->rdma_ius[--ioctx->n_rdma].sge);
+ kfree(ioctx->rdma_ius[--ioctx->n_rdma].wr.sg_list);
kfree(ioctx->rdma_ius);
ioctx->rdma_ius = NULL;
@@ -1084,7 +1084,7 @@ static int srpt_map_sg_to_ib_sge(struct srpt_rdma_ch *ch,
struct scatterlist *sg, *sg_orig;
int sg_cnt;
enum dma_data_direction dir;
- struct rdma_iu *riu;
+ struct ib_rdma_wr *riu;
struct srp_direct_buf *db;
dma_addr_t dma_addr;
struct ib_sge *sge;
@@ -1117,7 +1117,8 @@ static int srpt_map_sg_to_ib_sge(struct srpt_rdma_ch *ch,
nrdma = (count + SRPT_DEF_SG_PER_WQE - 1) / SRPT_DEF_SG_PER_WQE
+ ioctx->n_rbuf;
- ioctx->rdma_ius = kzalloc(nrdma * sizeof *riu, GFP_KERNEL);
+ ioctx->rdma_ius = kcalloc(nrdma, sizeof(*ioctx->rdma_ius),
+ GFP_KERNEL);
if (!ioctx->rdma_ius)
goto free_mem;
@@ -1141,9 +1142,9 @@ static int srpt_map_sg_to_ib_sge(struct srpt_rdma_ch *ch,
j < count && i < ioctx->n_rbuf && tsize > 0; ++i, ++riu, ++db) {
rsize = be32_to_cpu(db->len);
raddr = be64_to_cpu(db->va);
- riu->raddr = raddr;
+ riu->remote_addr = raddr;
riu->rkey = be32_to_cpu(db->key);
- riu->sge_cnt = 0;
+ riu->wr.num_sge = 0;
/* calculate how many sge required for this remote_buf */
while (rsize > 0 && tsize > 0) {
@@ -1167,27 +1168,29 @@ static int srpt_map_sg_to_ib_sge(struct srpt_rdma_ch *ch,
rsize = 0;
}
- ++riu->sge_cnt;
+ ++riu->wr.num_sge;
- if (rsize > 0 && riu->sge_cnt == SRPT_DEF_SG_PER_WQE) {
+ if (rsize > 0 &&
+ riu->wr.num_sge == SRPT_DEF_SG_PER_WQE) {
++ioctx->n_rdma;
- riu->sge =
- kmalloc(riu->sge_cnt * sizeof *riu->sge,
- GFP_KERNEL);
- if (!riu->sge)
+ riu->wr.sg_list = kmalloc_array(riu->wr.num_sge,
+ sizeof(*riu->wr.sg_list),
+ GFP_KERNEL);
+ if (!riu->wr.sg_list)
goto free_mem;
++riu;
- riu->sge_cnt = 0;
- riu->raddr = raddr;
+ riu->wr.num_sge = 0;
+ riu->remote_addr = raddr;
riu->rkey = be32_to_cpu(db->key);
}
}
++ioctx->n_rdma;
- riu->sge = kmalloc(riu->sge_cnt * sizeof *riu->sge,
- GFP_KERNEL);
- if (!riu->sge)
+ riu->wr.sg_list = kmalloc_array(riu->wr.num_sge,
+ sizeof(*riu->wr.sg_list),
+ GFP_KERNEL);
+ if (!riu->wr.sg_list)
goto free_mem;
}
@@ -1202,7 +1205,7 @@ static int srpt_map_sg_to_ib_sge(struct srpt_rdma_ch *ch,
for (i = 0, j = 0;
j < count && i < ioctx->n_rbuf && tsize > 0; ++i, ++riu, ++db) {
rsize = be32_to_cpu(db->len);
- sge = riu->sge;
+ sge = riu->wr.sg_list;
k = 0;
while (rsize > 0 && tsize > 0) {
@@ -1234,9 +1237,9 @@ static int srpt_map_sg_to_ib_sge(struct srpt_rdma_ch *ch,
}
++k;
- if (k == riu->sge_cnt && rsize > 0 && tsize > 0) {
+ if (k == riu->wr.num_sge && rsize > 0 && tsize > 0) {
++riu;
- sge = riu->sge;
+ sge = riu->wr.sg_list;
k = 0;
} else if (rsize > 0 && tsize > 0)
++sge;
@@ -1457,8 +1460,6 @@ static void srpt_handle_rdma_comp(struct srpt_rdma_ch *ch,
else
pr_err("%s[%d]: wrong state = %d\n", __func__,
__LINE__, srpt_get_cmd_state(ioctx));
- } else if (opcode == SRPT_RDMA_ABORT) {
- ioctx->rdma_aborted = true;
} else {
WARN(true, "unexpected opcode %d\n", opcode);
}
@@ -1981,8 +1982,7 @@ static void srpt_process_send_completion(struct ib_cq *cq,
if (opcode == SRPT_SEND)
srpt_handle_send_comp(ch, send_ioctx);
else {
- WARN_ON(opcode != SRPT_RDMA_ABORT &&
- wc->opcode != IB_WC_RDMA_READ);
+ WARN_ON(wc->opcode != IB_WC_RDMA_READ);
srpt_handle_rdma_comp(ch, send_ioctx, opcode);
}
} else {
@@ -2823,9 +2823,7 @@ static int srpt_cm_handler(struct ib_cm_id *cm_id, struct ib_cm_event *event)
static int srpt_perform_rdmas(struct srpt_rdma_ch *ch,
struct srpt_send_ioctx *ioctx)
{
- struct ib_rdma_wr wr;
struct ib_send_wr *bad_wr;
- struct rdma_iu *riu;
int i;
int ret;
int sq_wr_avail;
@@ -2844,59 +2842,37 @@ static int srpt_perform_rdmas(struct srpt_rdma_ch *ch,
}
}
- ioctx->rdma_aborted = false;
- ret = 0;
- riu = ioctx->rdma_ius;
- memset(&wr, 0, sizeof wr);
+ for (i = 0; i < n_rdma; i++) {
+ struct ib_rdma_wr *wr = &ioctx->rdma_ius[i];
- for (i = 0; i < n_rdma; ++i, ++riu) {
if (dir == DMA_FROM_DEVICE) {
- wr.wr.opcode = IB_WR_RDMA_WRITE;
- wr.wr.wr_id = encode_wr_id(i == n_rdma - 1 ?
+ wr->wr.opcode = IB_WR_RDMA_WRITE;
+ wr->wr.wr_id = encode_wr_id(i == n_rdma - 1 ?
SRPT_RDMA_WRITE_LAST :
SRPT_RDMA_MID,
ioctx->ioctx.index);
} else {
- wr.wr.opcode = IB_WR_RDMA_READ;
- wr.wr.wr_id = encode_wr_id(i == n_rdma - 1 ?
+ wr->wr.opcode = IB_WR_RDMA_READ;
+ wr->wr.wr_id = encode_wr_id(i == n_rdma - 1 ?
SRPT_RDMA_READ_LAST :
SRPT_RDMA_MID,
ioctx->ioctx.index);
}
- wr.wr.next = NULL;
- wr.remote_addr = riu->raddr;
- wr.rkey = riu->rkey;
- wr.wr.num_sge = riu->sge_cnt;
- wr.wr.sg_list = riu->sge;
-
- /* only get completion event for the last rdma write */
- if (i == (n_rdma - 1) && dir == DMA_TO_DEVICE)
- wr.wr.send_flags = IB_SEND_SIGNALED;
- ret = ib_post_send(ch->qp, &wr.wr, &bad_wr);
- if (ret)
- break;
+ if (i == n_rdma - 1) {
+ /* only get completion event for the last rdma read */
+ if (dir == DMA_TO_DEVICE)
+ wr->wr.send_flags = IB_SEND_SIGNALED;
+ wr->wr.next = NULL;
+ } else {
+ wr->wr.next = &ioctx->rdma_ius[i + 1].wr;
+ }
}
+ ret = ib_post_send(ch->qp, &ioctx->rdma_ius->wr, &bad_wr);
if (ret)
pr_err("%s[%d]: ib_post_send() returned %d for %d/%d\n",
__func__, __LINE__, ret, i, n_rdma);
- if (ret && i > 0) {
- wr.wr.num_sge = 0;
- wr.wr.wr_id = encode_wr_id(SRPT_RDMA_ABORT, ioctx->ioctx.index);
- wr.wr.send_flags = IB_SEND_SIGNALED;
- while (ch->state == CH_LIVE &&
- ib_post_send(ch->qp, &wr.wr, &bad_wr) != 0) {
- pr_info("Trying to abort failed RDMA transfer [%d]\n",
- ioctx->ioctx.index);
- msleep(1000);
- }
- while (ch->state != CH_RELEASING && !ioctx->rdma_aborted) {
- pr_info("Waiting until RDMA abort finished [%d]\n",
- ioctx->ioctx.index);
- msleep(1000);
- }
- }
out:
if (unlikely(dir == DMA_TO_DEVICE && ret < 0))
atomic_add(n_rdma, &ch->sq_wr_avail);
diff --git a/drivers/infiniband/ulp/srpt/ib_srpt.h b/drivers/infiniband/ulp/srpt/ib_srpt.h
index 0df7d61..fd6097e 100644
--- a/drivers/infiniband/ulp/srpt/ib_srpt.h
+++ b/drivers/infiniband/ulp/srpt/ib_srpt.h
@@ -132,7 +132,6 @@ enum srpt_opcode {
SRPT_RECV,
SRPT_SEND,
SRPT_RDMA_MID,
- SRPT_RDMA_ABORT,
SRPT_RDMA_READ_LAST,
SRPT_RDMA_WRITE_LAST,
};
@@ -150,14 +149,6 @@ static inline u32 idx_from_wr_id(u64 wr_id)
return (u32)wr_id;
}
-struct rdma_iu {
- u64 raddr;
- u32 rkey;
- struct ib_sge *sge;
- u32 sge_cnt;
- int mem_id;
-};
-
/**
* enum srpt_command_state - SCSI command state managed by SRPT.
* @SRPT_STATE_NEW: New command arrived and is being processed.
@@ -220,22 +211,19 @@ struct srpt_recv_ioctx {
* @tag: Tag of the received SRP information unit.
* @spinlock: Protects 'state'.
* @state: I/O context state.
- * @rdma_aborted: If initiating a multipart RDMA transfer failed, whether
- * the already initiated transfers have finished.
* @cmd: Target core command data structure.
* @sense_data: SCSI sense data.
*/
struct srpt_send_ioctx {
struct srpt_ioctx ioctx;
struct srpt_rdma_ch *ch;
- struct rdma_iu *rdma_ius;
+ struct ib_rdma_wr *rdma_ius;
struct srp_direct_buf *rbufs;
struct srp_direct_buf single_rbuf;
struct scatterlist *sg;
struct list_head free_list;
spinlock_t spinlock;
enum srpt_command_state state;
- bool rdma_aborted;
struct se_cmd cmd;
struct completion tx_done;
int sg_cnt;
--
1.9.1
--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at http://vger.kernel.org/majordomo-info.html
Please read the FAQ at http://www.tux.org/lkml/
[toc] | [prev] | [next] | [standalone]
| From | Christoph Hellwig <hch@lst.de> |
|---|---|
| Date | 2015-11-13 15:00 +0100 |
| Subject | [PATCH 9/9] IB/iser: Convert to CQ abstraction |
| Message-ID | <qun3z-77o-133@gated-at.bofh.it> |
| In reply to | #1268953 |
From: Sagi Grimberg <sagig@mellanox.com>
Signed-off-by: Sagi Grimberg <sagig@mellanox.com>
Signed-off-by: Christoph Hellwig <hch@lst.de>
---
drivers/infiniband/ulp/iser/iscsi_iser.h | 68 ++++---
drivers/infiniband/ulp/iser/iser_initiator.c | 142 ++++++++++-----
drivers/infiniband/ulp/iser/iser_memory.c | 21 ++-
drivers/infiniband/ulp/iser/iser_verbs.c | 258 ++++++---------------------
4 files changed, 209 insertions(+), 280 deletions(-)
diff --git a/drivers/infiniband/ulp/iser/iscsi_iser.h b/drivers/infiniband/ulp/iser/iscsi_iser.h
index cf4c4ce..1799c87 100644
--- a/drivers/infiniband/ulp/iser/iscsi_iser.h
+++ b/drivers/infiniband/ulp/iser/iscsi_iser.h
@@ -151,16 +151,12 @@
- ISER_MAX_RX_MISC_PDUS) / \
(1 + ISER_INFLIGHT_DATAOUTS))
-#define ISER_WC_BATCH_COUNT 16
#define ISER_SIGNAL_CMD_COUNT 32
#define ISER_VER 0x10
#define ISER_WSV 0x08
#define ISER_RSV 0x04
-#define ISER_FASTREG_LI_WRID 0xffffffffffffffffULL
-#define ISER_BEACON_WRID 0xfffffffffffffffeULL
-
/**
* struct iser_hdr - iSER header
*
@@ -269,7 +265,7 @@ enum iser_desc_type {
#define ISER_MAX_WRS 7
/**
- * struct iser_tx_desc - iSER TX descriptor (for send wr_id)
+ * struct iser_tx_desc - iSER TX descriptor
*
* @iser_header: iser header
* @iscsi_header: iscsi header
@@ -293,6 +289,7 @@ struct iser_tx_desc {
u64 dma_addr;
struct ib_sge tx_sg[2];
int num_sge;
+ struct ib_cqe cqe;
bool mapped;
u8 wr_idx;
union iser_wr {
@@ -306,9 +303,10 @@ struct iser_tx_desc {
};
#define ISER_RX_PAD_SIZE (256 - (ISER_RX_PAYLOAD_SIZE + \
- sizeof(u64) + sizeof(struct ib_sge)))
+ sizeof(u64) + sizeof(struct ib_sge) + \
+ sizeof(struct ib_cqe)))
/**
- * struct iser_rx_desc - iSER RX descriptor (for recv wr_id)
+ * struct iser_rx_desc - iSER RX descriptor
*
* @iser_header: iser header
* @iscsi_header: iscsi header
@@ -323,6 +321,7 @@ struct iser_rx_desc {
char data[ISER_RECV_DATA_SEG_LEN];
u64 dma_addr;
struct ib_sge rx_sg;
+ struct ib_cqe cqe;
char pad[ISER_RX_PAD_SIZE];
} __attribute__((packed));
@@ -335,6 +334,7 @@ struct iser_rx_desc {
* @req_dma: DMA address of login request buffer
* @rsp_dma: DMA address of login response buffer
* @sge: IB sge for login post recv
+ * @cqe: completion handler
*/
struct iser_login_desc {
void *req;
@@ -342,6 +342,7 @@ struct iser_login_desc {
u64 req_dma;
u64 rsp_dma;
struct ib_sge sge;
+ struct ib_cqe cqe;
} __attribute__((packed));
@@ -352,18 +353,12 @@ struct iscsi_iser_task;
/**
* struct iser_comp - iSER completion context
*
- * @device: pointer to device handle
* @cq: completion queue
- * @wcs: work completion array
- * @tasklet: Tasklet handle
* @active_qps: Number of active QPs attached
* to completion context
*/
struct iser_comp {
- struct iser_device *device;
struct ib_cq *cq;
- struct ib_wc wcs[ISER_WC_BATCH_COUNT];
- struct tasklet_struct tasklet;
int active_qps;
};
@@ -492,10 +487,11 @@ struct iser_fr_pool {
* @rx_wr: receive work request for batch posts
* @device: reference to iser device
* @comp: iser completion context
- * @pi_support: Indicate device T10-PI support
- * @beacon: beacon send wr to signal all flush errors were drained
- * @flush_comp: completes when all connection completions consumed
* @fr_pool: connection fast registration poool
+ * @pi_support: Indicate device T10-PI support
+ * @last: last send wr to signal all flush errors were drained
+ * @last_cqe: cqe handler for last wr
+ * @last_comp: completes when all connection completions consumed
*/
struct ib_conn {
struct rdma_cm_id *cma_id;
@@ -505,10 +501,12 @@ struct ib_conn {
struct ib_recv_wr rx_wr[ISER_MIN_POSTED_RX];
struct iser_device *device;
struct iser_comp *comp;
- bool pi_support;
- struct ib_send_wr beacon;
- struct completion flush_comp;
struct iser_fr_pool fr_pool;
+ bool pi_support;
+ struct ib_send_wr last;
+ struct ib_cqe last_cqe;
+ struct ib_cqe reg_cqe;
+ struct completion last_comp;
};
/**
@@ -643,12 +641,14 @@ int iser_conn_terminate(struct iser_conn *iser_conn);
void iser_release_work(struct work_struct *work);
-void iser_rcv_completion(struct iser_rx_desc *desc,
- unsigned long dto_xfer_len,
- struct ib_conn *ib_conn);
-
-void iser_snd_completion(struct iser_tx_desc *desc,
- struct ib_conn *ib_conn);
+void iser_err_comp(struct ib_wc *wc, const char *type);
+void iser_login_rsp(struct ib_cq *cq, struct ib_wc *wc);
+void iser_task_rsp(struct ib_cq *cq, struct ib_wc *wc);
+void iser_cmd_comp(struct ib_cq *cq, struct ib_wc *wc);
+void iser_ctrl_comp(struct ib_cq *cq, struct ib_wc *wc);
+void iser_dataout_comp(struct ib_cq *cq, struct ib_wc *wc);
+void iser_reg_comp(struct ib_cq *cq, struct ib_wc *wc);
+void iser_last_comp(struct ib_cq *cq, struct ib_wc *wc);
void iser_task_rdma_init(struct iscsi_iser_task *task);
@@ -735,4 +735,22 @@ to_iser_conn(struct ib_conn *ib_conn)
return container_of(ib_conn, struct iser_conn, ib_conn);
}
+static inline struct iser_rx_desc *
+iser_rx(struct ib_cqe *cqe)
+{
+ return container_of(cqe, struct iser_rx_desc, cqe);
+}
+
+static inline struct iser_tx_desc *
+iser_tx(struct ib_cqe *cqe)
+{
+ return container_of(cqe, struct iser_tx_desc, cqe);
+}
+
+static inline struct iser_login_desc *
+iser_login(struct ib_cqe *cqe)
+{
+ return container_of(cqe, struct iser_login_desc, cqe);
+}
+
#endif
diff --git a/drivers/infiniband/ulp/iser/iser_initiator.c b/drivers/infiniband/ulp/iser/iser_initiator.c
index 21148b6..44ecb89 100644
--- a/drivers/infiniband/ulp/iser/iser_initiator.c
+++ b/drivers/infiniband/ulp/iser/iser_initiator.c
@@ -270,11 +270,11 @@ int iser_alloc_rx_descriptors(struct iser_conn *iser_conn,
goto rx_desc_dma_map_failed;
rx_desc->dma_addr = dma_addr;
-
+ rx_desc->cqe.done = iser_task_rsp;
rx_sg = &rx_desc->rx_sg;
- rx_sg->addr = rx_desc->dma_addr;
+ rx_sg->addr = rx_desc->dma_addr;
rx_sg->length = ISER_RX_PAYLOAD_SIZE;
- rx_sg->lkey = device->pd->local_dma_lkey;
+ rx_sg->lkey = device->pd->local_dma_lkey;
}
iser_conn->rx_desc_head = 0;
@@ -373,6 +373,7 @@ int iser_send_command(struct iscsi_conn *conn,
/* build the tx desc regd header and add it to the tx desc dto */
tx_desc->type = ISCSI_TX_SCSI_COMMAND;
+ tx_desc->cqe.done = iser_cmd_comp;
iser_create_send_desc(iser_conn, tx_desc);
if (hdr->flags & ISCSI_FLAG_CMD_READ) {
@@ -454,6 +455,7 @@ int iser_send_data_out(struct iscsi_conn *conn,
}
tx_desc->type = ISCSI_TX_DATAOUT;
+ tx_desc->cqe.done = iser_dataout_comp;
tx_desc->iser_header.flags = ISER_VER;
memcpy(&tx_desc->iscsi_header, hdr, sizeof(struct iscsi_hdr));
@@ -503,6 +505,7 @@ int iser_send_control(struct iscsi_conn *conn,
/* build the tx desc regd header and add it to the tx desc dto */
mdesc->type = ISCSI_TX_CONTROL;
+ mdesc->cqe.done = iser_ctrl_comp;
iser_create_send_desc(iser_conn, mdesc);
device = iser_conn->ib_conn.device;
@@ -552,44 +555,69 @@ send_control_error:
return err;
}
-/**
- * iser_rcv_dto_completion - recv DTO completion
- */
-void iser_rcv_completion(struct iser_rx_desc *rx_desc,
- unsigned long rx_xfer_len,
- struct ib_conn *ib_conn)
+void iser_login_rsp(struct ib_cq *cq, struct ib_wc *wc)
{
+ struct ib_conn *ib_conn = wc->qp->qp_context;
struct iser_conn *iser_conn = to_iser_conn(ib_conn);
+ struct iser_login_desc *desc = iser_login(wc->wr_cqe);
struct iscsi_hdr *hdr;
char *data;
- u64 rx_dma;
- int rx_buflen, outstanding, count, err;
-
- /* differentiate between login to all other PDUs */
- if (rx_desc == (void *)&iser_conn->login_desc) {
- rx_dma = iser_conn->login_desc.rsp_dma;
- rx_buflen = ISER_RX_LOGIN_SIZE;
- hdr = iser_conn->login_desc.rsp + sizeof(struct iser_hdr);
- data = iser_conn->login_desc.rsp + ISER_HEADERS_LEN;
- } else {
- rx_dma = rx_desc->dma_addr;
- rx_buflen = ISER_RX_PAYLOAD_SIZE;
- hdr = &rx_desc->iscsi_header;
- data = rx_desc->data;
+ int length;
+
+ if (unlikely(wc->status != IB_WC_SUCCESS)) {
+ iser_err_comp(wc, "login_rsp");
+ return;
+ }
+
+ ib_dma_sync_single_for_cpu(ib_conn->device->ib_device,
+ desc->rsp_dma, ISER_RX_LOGIN_SIZE,
+ DMA_FROM_DEVICE);
+
+ hdr = desc->rsp + sizeof(struct iser_hdr);
+ data = desc->rsp + ISER_HEADERS_LEN;
+ length = wc->byte_len - ISER_HEADERS_LEN;
+
+ iser_dbg("op 0x%x itt 0x%x dlen %d\n", hdr->opcode,
+ hdr->itt, length);
+
+ iscsi_iser_recv(iser_conn->iscsi_conn, hdr, data, length);
+
+ ib_dma_sync_single_for_device(ib_conn->device->ib_device,
+ desc->rsp_dma, ISER_RX_LOGIN_SIZE,
+ DMA_FROM_DEVICE);
+
+ ib_conn->post_recv_buf_count--;
+}
+
+void iser_task_rsp(struct ib_cq *cq, struct ib_wc *wc)
+{
+ struct ib_conn *ib_conn = wc->qp->qp_context;
+ struct iser_conn *iser_conn = to_iser_conn(ib_conn);
+ struct iser_rx_desc *desc = iser_rx(wc->wr_cqe);
+ struct iscsi_hdr *hdr;
+ int length;
+ int outstanding, count, err;
+
+ if (unlikely(wc->status != IB_WC_SUCCESS)) {
+ iser_err_comp(wc, "task_rsp");
+ return;
}
- ib_dma_sync_single_for_cpu(ib_conn->device->ib_device, rx_dma,
- rx_buflen, DMA_FROM_DEVICE);
+ ib_dma_sync_single_for_cpu(ib_conn->device->ib_device,
+ desc->dma_addr, ISER_RX_PAYLOAD_SIZE,
+ DMA_FROM_DEVICE);
+ hdr = &desc->iscsi_header;
+ length = wc->byte_len - ISER_HEADERS_LEN;
iser_dbg("op 0x%x itt 0x%x dlen %d\n", hdr->opcode,
- hdr->itt, (int)(rx_xfer_len - ISER_HEADERS_LEN));
+ hdr->itt, length);
- iscsi_iser_recv(iser_conn->iscsi_conn, hdr, data,
- rx_xfer_len - ISER_HEADERS_LEN);
+ iscsi_iser_recv(iser_conn->iscsi_conn, hdr, desc->data, length);
- ib_dma_sync_single_for_device(ib_conn->device->ib_device, rx_dma,
- rx_buflen, DMA_FROM_DEVICE);
+ ib_dma_sync_single_for_device(ib_conn->device->ib_device,
+ desc->dma_addr, ISER_RX_PAYLOAD_SIZE,
+ DMA_FROM_DEVICE);
/* decrementing conn->post_recv_buf_count only --after-- freeing the *
* task eliminates the need to worry on tasks which are completed in *
@@ -597,9 +625,6 @@ void iser_rcv_completion(struct iser_rx_desc *rx_desc,
* for the posted rx bufs refcount to become zero handles everything */
ib_conn->post_recv_buf_count--;
- if (rx_desc == (void *)&iser_conn->login_desc)
- return;
-
outstanding = ib_conn->post_recv_buf_count;
if (outstanding + iser_conn->min_posted_rx <= iser_conn->qp_max_recv_dtos) {
count = min(iser_conn->qp_max_recv_dtos - outstanding,
@@ -610,26 +635,47 @@ void iser_rcv_completion(struct iser_rx_desc *rx_desc,
}
}
-void iser_snd_completion(struct iser_tx_desc *tx_desc,
- struct ib_conn *ib_conn)
+void iser_cmd_comp(struct ib_cq *cq, struct ib_wc *wc)
+{
+ if (unlikely(wc->status != IB_WC_SUCCESS))
+ iser_err_comp(wc, "command");
+}
+
+void iser_ctrl_comp(struct ib_cq *cq, struct ib_wc *wc)
{
+ struct iser_tx_desc *desc = iser_tx(wc->wr_cqe);
struct iscsi_task *task;
- struct iser_device *device = ib_conn->device;
- if (tx_desc->type == ISCSI_TX_DATAOUT) {
- ib_dma_unmap_single(device->ib_device, tx_desc->dma_addr,
- ISER_HEADERS_LEN, DMA_TO_DEVICE);
- kmem_cache_free(ig.desc_cache, tx_desc);
- tx_desc = NULL;
+ if (unlikely(wc->status != IB_WC_SUCCESS)) {
+ iser_err_comp(wc, "control");
+ return;
}
- if (tx_desc && tx_desc->type == ISCSI_TX_CONTROL) {
- /* this arithmetic is legal by libiscsi dd_data allocation */
- task = (void *) ((long)(void *)tx_desc -
- sizeof(struct iscsi_task));
- if (task->hdr->itt == RESERVED_ITT)
- iscsi_put_task(task);
- }
+ /* this arithmetic is legal by libiscsi dd_data allocation */
+ task = (void *)desc - sizeof(struct iscsi_task);
+ if (task->hdr->itt == RESERVED_ITT)
+ iscsi_put_task(task);
+}
+
+void iser_dataout_comp(struct ib_cq *cq, struct ib_wc *wc)
+{
+ struct iser_tx_desc *desc = iser_tx(wc->wr_cqe);
+ struct ib_conn *ib_conn = wc->qp->qp_context;
+ struct iser_device *device = ib_conn->device;
+
+ if (unlikely(wc->status != IB_WC_SUCCESS))
+ iser_err_comp(wc, "dataout");
+
+ ib_dma_unmap_single(device->ib_device, desc->dma_addr,
+ ISER_HEADERS_LEN, DMA_TO_DEVICE);
+ kmem_cache_free(ig.desc_cache, desc);
+}
+
+void iser_last_comp(struct ib_cq *cq, struct ib_wc *wc)
+{
+ struct ib_conn *ib_conn = wc->qp->qp_context;
+
+ complete(&ib_conn->last_comp);
}
void iser_task_rdma_init(struct iscsi_iser_task *iser_task)
diff --git a/drivers/infiniband/ulp/iser/iser_memory.c b/drivers/infiniband/ulp/iser/iser_memory.c
index 81ad5e9..454c8cd 100644
--- a/drivers/infiniband/ulp/iser/iser_memory.c
+++ b/drivers/infiniband/ulp/iser/iser_memory.c
@@ -67,6 +67,11 @@ static struct iser_reg_ops fmr_ops = {
.reg_desc_put = iser_reg_desc_put_fmr,
};
+void iser_reg_comp(struct ib_cq *cq, struct ib_wc *wc)
+{
+ iser_err_comp(wc, "memreg");
+}
+
int iser_assign_reg_ops(struct iser_device *device)
{
struct ib_device *ib_dev = device->ib_device;
@@ -413,12 +418,14 @@ iser_set_prot_checks(struct scsi_cmnd *sc, u8 *mask)
}
static void
-iser_inv_rkey(struct ib_send_wr *inv_wr, struct ib_mr *mr)
+iser_inv_rkey(struct ib_send_wr *inv_wr,
+ struct ib_mr *mr,
+ struct ib_cqe *cqe)
{
u32 rkey;
inv_wr->opcode = IB_WR_LOCAL_INV;
- inv_wr->wr_id = ISER_FASTREG_LI_WRID;
+ inv_wr->wr_cqe = cqe;
inv_wr->ex.invalidate_rkey = mr->rkey;
inv_wr->send_flags = 0;
inv_wr->num_sge = 0;
@@ -436,6 +443,7 @@ iser_reg_sig_mr(struct iscsi_iser_task *iser_task,
{
struct iser_tx_desc *tx_desc = &iser_task->desc;
struct ib_sig_attrs *sig_attrs = &tx_desc->sig_attrs;
+ struct ib_cqe *cqe = &iser_task->iser_conn->ib_conn.reg_cqe;
struct ib_sig_handover_wr *wr;
int ret;
@@ -447,11 +455,11 @@ iser_reg_sig_mr(struct iscsi_iser_task *iser_task,
iser_set_prot_checks(iser_task->sc, &sig_attrs->check_mask);
if (!pi_ctx->sig_mr_valid)
- iser_inv_rkey(iser_tx_next_wr(tx_desc), pi_ctx->sig_mr);
+ iser_inv_rkey(iser_tx_next_wr(tx_desc), pi_ctx->sig_mr, cqe);
wr = sig_handover_wr(iser_tx_next_wr(tx_desc));
wr->wr.opcode = IB_WR_REG_SIG_MR;
- wr->wr.wr_id = ISER_FASTREG_LI_WRID;
+ wr->wr.wr_cqe = cqe;
wr->wr.sg_list = &data_reg->sge;
wr->wr.num_sge = 1;
wr->wr.send_flags = 0;
@@ -484,12 +492,13 @@ static int iser_fast_reg_mr(struct iscsi_iser_task *iser_task,
struct iser_mem_reg *reg)
{
struct iser_tx_desc *tx_desc = &iser_task->desc;
+ struct ib_cqe *cqe = &iser_task->iser_conn->ib_conn.reg_cqe;
struct ib_mr *mr = rsc->mr;
struct ib_reg_wr *wr;
int n;
if (!rsc->mr_valid)
- iser_inv_rkey(iser_tx_next_wr(tx_desc), mr);
+ iser_inv_rkey(iser_tx_next_wr(tx_desc), mr, cqe);
n = ib_map_mr_sg(mr, mem->sg, mem->size, SIZE_4K);
if (unlikely(n != mem->size)) {
@@ -500,7 +509,7 @@ static int iser_fast_reg_mr(struct iscsi_iser_task *iser_task,
wr = reg_wr(iser_tx_next_wr(tx_desc));
wr->wr.opcode = IB_WR_REG_MR;
- wr->wr.wr_id = ISER_FASTREG_LI_WRID;
+ wr->wr.wr_cqe = cqe;
wr->wr.send_flags = 0;
wr->wr.num_sge = 0;
wr->mr = mr;
diff --git a/drivers/infiniband/ulp/iser/iser_verbs.c b/drivers/infiniband/ulp/iser/iser_verbs.c
index f75ef0c..29d9046 100644
--- a/drivers/infiniband/ulp/iser/iser_verbs.c
+++ b/drivers/infiniband/ulp/iser/iser_verbs.c
@@ -44,17 +44,6 @@
#define ISER_MAX_CQ_LEN (ISER_MAX_RX_LEN + ISER_MAX_TX_LEN + \
ISCSI_ISER_MAX_CONN)
-static int iser_cq_poll_limit = 512;
-
-static void iser_cq_tasklet_fn(unsigned long data);
-static void iser_cq_callback(struct ib_cq *cq, void *cq_context);
-
-static void iser_cq_event_callback(struct ib_event *cause, void *context)
-{
- iser_err("cq event %s (%d)\n",
- ib_event_msg(cause->event), cause->event);
-}
-
static void iser_qp_event_callback(struct ib_event *cause, void *context)
{
iser_err("qp event %s (%d)\n",
@@ -104,27 +93,14 @@ static int iser_create_device_ib_res(struct iser_device *device)
goto pd_err;
for (i = 0; i < device->comps_used; i++) {
- struct ib_cq_init_attr cq_attr = {};
struct iser_comp *comp = &device->comps[i];
- comp->device = device;
- cq_attr.cqe = max_cqe;
- cq_attr.comp_vector = i;
- comp->cq = ib_create_cq(ib_dev,
- iser_cq_callback,
- iser_cq_event_callback,
- (void *)comp,
- &cq_attr);
+ comp->cq = ib_alloc_cq(ib_dev, comp, max_cqe, i,
+ IB_POLL_SOFTIRQ);
if (IS_ERR(comp->cq)) {
comp->cq = NULL;
goto cq_err;
}
-
- if (ib_req_notify_cq(comp->cq, IB_CQ_NEXT_COMP))
- goto cq_err;
-
- tasklet_init(&comp->tasklet, iser_cq_tasklet_fn,
- (unsigned long)comp);
}
if (!iser_always_reg) {
@@ -134,7 +110,7 @@ static int iser_create_device_ib_res(struct iser_device *device)
device->mr = ib_get_dma_mr(device->pd, access);
if (IS_ERR(device->mr))
- goto dma_mr_err;
+ goto cq_err;
}
INIT_IB_EVENT_HANDLER(&device->event_handler, device->ib_device,
@@ -147,15 +123,12 @@ static int iser_create_device_ib_res(struct iser_device *device)
handler_err:
if (device->mr)
ib_dereg_mr(device->mr);
-dma_mr_err:
- for (i = 0; i < device->comps_used; i++)
- tasklet_kill(&device->comps[i].tasklet);
cq_err:
for (i = 0; i < device->comps_used; i++) {
struct iser_comp *comp = &device->comps[i];
if (comp->cq)
- ib_destroy_cq(comp->cq);
+ ib_free_cq(comp->cq);
}
ib_dealloc_pd(device->pd);
pd_err:
@@ -176,8 +149,7 @@ static void iser_free_device_ib_res(struct iser_device *device)
for (i = 0; i < device->comps_used; i++) {
struct iser_comp *comp = &device->comps[i];
- tasklet_kill(&comp->tasklet);
- ib_destroy_cq(comp->cq);
+ ib_free_cq(comp->cq);
comp->cq = NULL;
}
@@ -717,13 +689,13 @@ int iser_conn_terminate(struct iser_conn *iser_conn)
iser_conn, err);
/* post an indication that all flush errors were consumed */
- err = ib_post_send(ib_conn->qp, &ib_conn->beacon, &bad_wr);
+ err = ib_post_send(ib_conn->qp, &ib_conn->last, &bad_wr);
if (err) {
- iser_err("conn %p failed to post beacon", ib_conn);
+ iser_err("conn %p failed to post last wr", ib_conn);
return 1;
}
- wait_for_completion(&ib_conn->flush_comp);
+ wait_for_completion(&ib_conn->last_comp);
}
return 1;
@@ -960,14 +932,21 @@ static int iser_cma_handler(struct rdma_cm_id *cma_id, struct rdma_cm_event *eve
void iser_conn_init(struct iser_conn *iser_conn)
{
+ struct ib_conn *ib_conn = &iser_conn->ib_conn;
+
iser_conn->state = ISER_CONN_INIT;
- iser_conn->ib_conn.post_recv_buf_count = 0;
- init_completion(&iser_conn->ib_conn.flush_comp);
init_completion(&iser_conn->stop_completion);
init_completion(&iser_conn->ib_completion);
init_completion(&iser_conn->up_completion);
INIT_LIST_HEAD(&iser_conn->conn_list);
mutex_init(&iser_conn->state_mutex);
+
+ ib_conn->post_recv_buf_count = 0;
+ ib_conn->reg_cqe.done = iser_reg_comp;
+ ib_conn->last_cqe.done = iser_last_comp;
+ ib_conn->last.wr_cqe = &ib_conn->last_cqe;
+ ib_conn->last.opcode = IB_WR_SEND;
+ init_completion(&ib_conn->last_comp);
}
/**
@@ -993,9 +972,6 @@ int iser_connect(struct iser_conn *iser_conn,
iser_conn->state = ISER_CONN_PENDING;
- ib_conn->beacon.wr_id = ISER_BEACON_WRID;
- ib_conn->beacon.opcode = IB_WR_SEND;
-
ib_conn->cma_id = rdma_create_id(&init_net, iser_cma_handler,
(void *)iser_conn,
RDMA_PS_TCP, IB_QPT_RC);
@@ -1038,56 +1014,60 @@ connect_failure:
int iser_post_recvl(struct iser_conn *iser_conn)
{
- struct ib_recv_wr rx_wr, *rx_wr_failed;
struct ib_conn *ib_conn = &iser_conn->ib_conn;
struct iser_login_desc *desc = &iser_conn->login_desc;
+ struct ib_recv_wr wr, *wr_failed;
int ib_ret;
desc->sge.addr = desc->rsp_dma;
desc->sge.length = ISER_RX_LOGIN_SIZE;
desc->sge.lkey = ib_conn->device->pd->local_dma_lkey;
- rx_wr.wr_id = (uintptr_t)desc;
- rx_wr.sg_list = &desc->sge;
- rx_wr.num_sge = 1;
- rx_wr.next = NULL;
+ desc->cqe.done = iser_login_rsp;
+ wr.wr_cqe = &desc->cqe;
+ wr.sg_list = &desc->sge;
+ wr.num_sge = 1;
+ wr.next = NULL;
ib_conn->post_recv_buf_count++;
- ib_ret = ib_post_recv(ib_conn->qp, &rx_wr, &rx_wr_failed);
+ ib_ret = ib_post_recv(ib_conn->qp, &wr, &wr_failed);
if (ib_ret) {
iser_err("ib_post_recv failed ret=%d\n", ib_ret);
ib_conn->post_recv_buf_count--;
}
+
return ib_ret;
}
int iser_post_recvm(struct iser_conn *iser_conn, int count)
{
- struct ib_recv_wr *rx_wr, *rx_wr_failed;
- int i, ib_ret;
struct ib_conn *ib_conn = &iser_conn->ib_conn;
unsigned int my_rx_head = iser_conn->rx_desc_head;
struct iser_rx_desc *rx_desc;
+ struct ib_recv_wr *wr, *wr_failed;
+ int i, ib_ret;
- for (rx_wr = ib_conn->rx_wr, i = 0; i < count; i++, rx_wr++) {
- rx_desc = &iser_conn->rx_descs[my_rx_head];
- rx_wr->wr_id = (uintptr_t)rx_desc;
- rx_wr->sg_list = &rx_desc->rx_sg;
- rx_wr->num_sge = 1;
- rx_wr->next = rx_wr + 1;
+ for (wr = ib_conn->rx_wr, i = 0; i < count; i++, wr++) {
+ rx_desc = &iser_conn->rx_descs[my_rx_head];
+ rx_desc->cqe.done = iser_task_rsp;
+ wr->wr_cqe = &rx_desc->cqe;
+ wr->sg_list = &rx_desc->rx_sg;
+ wr->num_sge = 1;
+ wr->next = wr + 1;
my_rx_head = (my_rx_head + 1) & iser_conn->qp_max_recv_dtos_mask;
}
- rx_wr--;
- rx_wr->next = NULL; /* mark end of work requests list */
+ wr--;
+ wr->next = NULL; /* mark end of work requests list */
ib_conn->post_recv_buf_count += count;
- ib_ret = ib_post_recv(ib_conn->qp, ib_conn->rx_wr, &rx_wr_failed);
+ ib_ret = ib_post_recv(ib_conn->qp, ib_conn->rx_wr, &wr_failed);
if (ib_ret) {
iser_err("ib_post_recv failed ret=%d\n", ib_ret);
ib_conn->post_recv_buf_count -= count;
} else
iser_conn->rx_desc_head = my_rx_head;
+
return ib_ret;
}
@@ -1108,7 +1088,7 @@ int iser_post_send(struct ib_conn *ib_conn, struct iser_tx_desc *tx_desc,
DMA_TO_DEVICE);
wr->next = NULL;
- wr->wr_id = (uintptr_t)tx_desc;
+ wr->wr_cqe = &tx_desc->cqe;
wr->sg_list = tx_desc->tx_sg;
wr->num_sge = tx_desc->num_sge;
wr->opcode = IB_WR_SEND;
@@ -1122,148 +1102,6 @@ int iser_post_send(struct ib_conn *ib_conn, struct iser_tx_desc *tx_desc,
return ib_ret;
}
-/**
- * is_iser_tx_desc - Indicate if the completion wr_id
- * is a TX descriptor or not.
- * @iser_conn: iser connection
- * @wr_id: completion WR identifier
- *
- * Since we cannot rely on wc opcode in FLUSH errors
- * we must work around it by checking if the wr_id address
- * falls in the iser connection rx_descs buffer. If so
- * it is an RX descriptor, otherwize it is a TX.
- */
-static inline bool
-is_iser_tx_desc(struct iser_conn *iser_conn, void *wr_id)
-{
- void *start = iser_conn->rx_descs;
- int len = iser_conn->num_rx_descs * sizeof(*iser_conn->rx_descs);
-
- if (wr_id >= start && wr_id < start + len)
- return false;
-
- return true;
-}
-
-/**
- * iser_handle_comp_error() - Handle error completion
- * @ib_conn: connection RDMA resources
- * @wc: work completion
- *
- * Notes: We may handle a FLUSH error completion and in this case
- * we only cleanup in case TX type was DATAOUT. For non-FLUSH
- * error completion we should also notify iscsi layer that
- * connection is failed (in case we passed bind stage).
- */
-static void
-iser_handle_comp_error(struct ib_conn *ib_conn,
- struct ib_wc *wc)
-{
- struct iser_conn *iser_conn = to_iser_conn(ib_conn);
- void *wr_id = (void *)(uintptr_t)wc->wr_id;
-
- if (wc->status != IB_WC_WR_FLUSH_ERR)
- if (iser_conn->iscsi_conn)
- iscsi_conn_failure(iser_conn->iscsi_conn,
- ISCSI_ERR_CONN_FAILED);
-
- if (wc->wr_id == ISER_FASTREG_LI_WRID)
- return;
-
- if (is_iser_tx_desc(iser_conn, wr_id)) {
- struct iser_tx_desc *desc = wr_id;
-
- if (desc->type == ISCSI_TX_DATAOUT)
- kmem_cache_free(ig.desc_cache, desc);
- } else {
- ib_conn->post_recv_buf_count--;
- }
-}
-
-/**
- * iser_handle_wc - handle a single work completion
- * @wc: work completion
- *
- * Soft-IRQ context, work completion can be either
- * SEND or RECV, and can turn out successful or
- * with error (or flush error).
- */
-static void iser_handle_wc(struct ib_wc *wc)
-{
- struct ib_conn *ib_conn;
- struct iser_tx_desc *tx_desc;
- struct iser_rx_desc *rx_desc;
-
- ib_conn = wc->qp->qp_context;
- if (likely(wc->status == IB_WC_SUCCESS)) {
- if (wc->opcode == IB_WC_RECV) {
- rx_desc = (struct iser_rx_desc *)(uintptr_t)wc->wr_id;
- iser_rcv_completion(rx_desc, wc->byte_len,
- ib_conn);
- } else
- if (wc->opcode == IB_WC_SEND) {
- tx_desc = (struct iser_tx_desc *)(uintptr_t)wc->wr_id;
- iser_snd_completion(tx_desc, ib_conn);
- } else {
- iser_err("Unknown wc opcode %d\n", wc->opcode);
- }
- } else {
- if (wc->status != IB_WC_WR_FLUSH_ERR)
- iser_err("%s (%d): wr id %llx vend_err %x\n",
- ib_wc_status_msg(wc->status), wc->status,
- wc->wr_id, wc->vendor_err);
- else
- iser_dbg("%s (%d): wr id %llx\n",
- ib_wc_status_msg(wc->status), wc->status,
- wc->wr_id);
-
- if (wc->wr_id == ISER_BEACON_WRID)
- /* all flush errors were consumed */
- complete(&ib_conn->flush_comp);
- else
- iser_handle_comp_error(ib_conn, wc);
- }
-}
-
-/**
- * iser_cq_tasklet_fn - iSER completion polling loop
- * @data: iSER completion context
- *
- * Soft-IRQ context, polling connection CQ until
- * either CQ was empty or we exausted polling budget
- */
-static void iser_cq_tasklet_fn(unsigned long data)
-{
- struct iser_comp *comp = (struct iser_comp *)data;
- struct ib_cq *cq = comp->cq;
- struct ib_wc *const wcs = comp->wcs;
- int i, n, completed = 0;
-
- while ((n = ib_poll_cq(cq, ARRAY_SIZE(comp->wcs), wcs)) > 0) {
- for (i = 0; i < n; i++)
- iser_handle_wc(&wcs[i]);
-
- completed += n;
- if (completed >= iser_cq_poll_limit)
- break;
- }
-
- /*
- * It is assumed here that arming CQ only once its empty
- * would not cause interrupts to be missed.
- */
- ib_req_notify_cq(cq, IB_CQ_NEXT_COMP);
-
- iser_dbg("got %d completions\n", completed);
-}
-
-static void iser_cq_callback(struct ib_cq *cq, void *cq_context)
-{
- struct iser_comp *comp = cq_context;
-
- tasklet_schedule(&comp->tasklet);
-}
-
u8 iser_check_task_pi_status(struct iscsi_iser_task *iser_task,
enum iser_data_dir cmd_dir, sector_t *sector)
{
@@ -1311,3 +1149,21 @@ err:
/* Not alot we can do here, return ambiguous guard error */
return 0x1;
}
+
+void iser_err_comp(struct ib_wc *wc, const char *type)
+{
+ if (wc->status != IB_WC_WR_FLUSH_ERR) {
+ struct iser_conn *iser_conn = to_iser_conn(wc->qp->qp_context);
+
+ iser_err("%s failure: %s (%d) vend_err %x\n", type,
+ ib_wc_status_msg(wc->status), wc->status,
+ wc->vendor_err);
+
+ if (iser_conn->iscsi_conn)
+ iscsi_conn_failure(iser_conn->iscsi_conn,
+ ISCSI_ERR_CONN_FAILED);
+ } else {
+ iser_dbg("%s failure: %s (%d)\n", type,
+ ib_wc_status_msg(wc->status), wc->status);
+ }
+}
--
1.9.1
--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at http://vger.kernel.org/majordomo-info.html
Please read the FAQ at http://www.tux.org/lkml/
[toc] | [prev] | [next] | [standalone]
| From | Christoph Hellwig <hch@lst.de> |
|---|---|
| Date | 2015-11-13 15:00 +0100 |
| Subject | [PATCH 8/9] IB/iser: Use helper for container_of |
| Message-ID | <qun3z-77o-137@gated-at.bofh.it> |
| In reply to | #1268953 |
From: Sagi Grimberg <sagig@mellanox.com>
Nicer this way.
Signed-off-by: Sagi Grimberg <sagig@mellanox.com>
Signed-off-by: Christoph Hellwig <hch@lst.de>
---
drivers/infiniband/ulp/iser/iscsi_iser.h | 6 ++++++
drivers/infiniband/ulp/iser/iser_initiator.c | 3 +--
drivers/infiniband/ulp/iser/iser_verbs.c | 6 ++----
3 files changed, 9 insertions(+), 6 deletions(-)
diff --git a/drivers/infiniband/ulp/iser/iscsi_iser.h b/drivers/infiniband/ulp/iser/iscsi_iser.h
index 5648409..cf4c4ce 100644
--- a/drivers/infiniband/ulp/iser/iscsi_iser.h
+++ b/drivers/infiniband/ulp/iser/iscsi_iser.h
@@ -729,4 +729,10 @@ iser_tx_next_wr(struct iser_tx_desc *tx_desc)
return cur_wr;
}
+static inline struct iser_conn *
+to_iser_conn(struct ib_conn *ib_conn)
+{
+ return container_of(ib_conn, struct iser_conn, ib_conn);
+}
+
#endif
diff --git a/drivers/infiniband/ulp/iser/iser_initiator.c b/drivers/infiniband/ulp/iser/iser_initiator.c
index 21f28c8..21148b6 100644
--- a/drivers/infiniband/ulp/iser/iser_initiator.c
+++ b/drivers/infiniband/ulp/iser/iser_initiator.c
@@ -559,8 +559,7 @@ void iser_rcv_completion(struct iser_rx_desc *rx_desc,
unsigned long rx_xfer_len,
struct ib_conn *ib_conn)
{
- struct iser_conn *iser_conn = container_of(ib_conn, struct iser_conn,
- ib_conn);
+ struct iser_conn *iser_conn = to_iser_conn(ib_conn);
struct iscsi_hdr *hdr;
char *data;
u64 rx_dma;
diff --git a/drivers/infiniband/ulp/iser/iser_verbs.c b/drivers/infiniband/ulp/iser/iser_verbs.c
index ee4cebc..f75ef0c 100644
--- a/drivers/infiniband/ulp/iser/iser_verbs.c
+++ b/drivers/infiniband/ulp/iser/iser_verbs.c
@@ -455,8 +455,7 @@ void iser_free_fastreg_pool(struct ib_conn *ib_conn)
*/
static int iser_create_ib_conn_res(struct ib_conn *ib_conn)
{
- struct iser_conn *iser_conn = container_of(ib_conn, struct iser_conn,
- ib_conn);
+ struct iser_conn *iser_conn = to_iser_conn(ib_conn);
struct iser_device *device;
struct ib_device *ib_dev;
struct ib_qp_init_attr init_attr;
@@ -1160,9 +1159,8 @@ static void
iser_handle_comp_error(struct ib_conn *ib_conn,
struct ib_wc *wc)
{
+ struct iser_conn *iser_conn = to_iser_conn(ib_conn);
void *wr_id = (void *)(uintptr_t)wc->wr_id;
- struct iser_conn *iser_conn = container_of(ib_conn, struct iser_conn,
- ib_conn);
if (wc->status != IB_WC_WR_FLUSH_ERR)
if (iser_conn->iscsi_conn)
--
1.9.1
--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at http://vger.kernel.org/majordomo-info.html
Please read the FAQ at http://www.tux.org/lkml/
[toc] | [prev] | [standalone]
Back to top | Article view | linux.kernel
csiph-web