提交 b7f04513 编写于 作者: S Shlomo Pongratz 提交者: Roland Dreier

IB/iser: Accept session->cmds_max from user space

Use cmds_max passed from user space to be the number of PDUs to be
supported for the session instead of hard-coded ISCSI_DEF_XMIT_CMDS_MAX.
This allow controlling the max number of SCSI commands for the session.
Also don't ignore the qdepth passed from user space.

Derive from session->cmds_max the actual number of RX buffers and FMR
pool size to allocate during the connection bind phase.

Since the iser transport connection is established before the iscsi
session/connection are created and bound, we still use one hard-coded
quantity ISER_DEF_XMIT_CMDS_MAX to compute the maximum number of
work-requests to be supported by the RC QP used for the connection.

The above quantity is made to be a power of two between ISCSI_TOTAL_CMDS_MIN
(16) and ISER_DEF_XMIT_CMDS_MAX (512) inclusive.
Signed-off-by: NShlomo Pongratz <shlomop@mellanox.com>
Signed-off-by: NOr Gerlitz <ogerlitz@mellanox.com>
Signed-off-by: NRoland Dreier <roland@purestorage.com>
上级 986db0d6
......@@ -347,6 +347,7 @@ iscsi_iser_conn_bind(struct iscsi_cls_session *cls_session,
{
struct iscsi_conn *conn = cls_conn->dd_data;
struct iscsi_iser_conn *iser_conn;
struct iscsi_session *session;
struct iser_conn *ib_conn;
struct iscsi_endpoint *ep;
int error;
......@@ -365,7 +366,8 @@ iscsi_iser_conn_bind(struct iscsi_cls_session *cls_session,
}
ib_conn = ep->dd_data;
if (iser_alloc_rx_descriptors(ib_conn))
session = conn->session;
if (iser_alloc_rx_descriptors(ib_conn, session))
return -ENOMEM;
/* binds the iSER connection retrieved from the previously
......@@ -419,12 +421,13 @@ iscsi_iser_session_create(struct iscsi_endpoint *ep,
struct iscsi_cls_session *cls_session;
struct iscsi_session *session;
struct Scsi_Host *shost;
struct iser_conn *ib_conn;
struct iser_conn *ib_conn = NULL;
shost = iscsi_host_alloc(&iscsi_iser_sht, 0, 0);
if (!shost)
return NULL;
shost->transportt = iscsi_iser_scsi_transport;
shost->cmd_per_lun = qdepth;
shost->max_lun = iscsi_max_lun;
shost->max_id = 0;
shost->max_channel = 0;
......@@ -441,12 +444,14 @@ iscsi_iser_session_create(struct iscsi_endpoint *ep,
ep ? ib_conn->device->ib_device->dma_device : NULL))
goto free_host;
/*
* we do not support setting can_queue cmd_per_lun from userspace yet
* because we preallocate so many resources
*/
if (cmds_max > ISER_DEF_XMIT_CMDS_MAX) {
iser_info("cmds_max changed from %u to %u\n",
cmds_max, ISER_DEF_XMIT_CMDS_MAX);
cmds_max = ISER_DEF_XMIT_CMDS_MAX;
}
cls_session = iscsi_session_setup(&iscsi_iser_transport, shost,
ISCSI_DEF_XMIT_CMDS_MAX, 0,
cmds_max, 0,
sizeof(struct iscsi_iser_task),
initial_cmdsn, 0);
if (!cls_session)
......
......@@ -102,7 +102,13 @@
/* support up to 512KB in one RDMA */
#define ISCSI_ISER_SG_TABLESIZE (0x80000 >> SHIFT_4K)
#define ISER_DEF_CMD_PER_LUN ISCSI_DEF_XMIT_CMDS_MAX
#define ISER_DEF_XMIT_CMDS_DEFAULT 512
#if ISCSI_DEF_XMIT_CMDS_MAX > ISER_DEF_XMIT_CMDS_DEFAULT
#define ISER_DEF_XMIT_CMDS_MAX ISCSI_DEF_XMIT_CMDS_MAX
#else
#define ISER_DEF_XMIT_CMDS_MAX ISER_DEF_XMIT_CMDS_DEFAULT
#endif
#define ISER_DEF_CMD_PER_LUN ISER_DEF_XMIT_CMDS_MAX
/* QP settings */
/* Maximal bounds on received asynchronous PDUs */
......@@ -111,9 +117,9 @@
#define ISER_MAX_TX_MISC_PDUS 6 /* NOOP_OUT(2), TEXT(1), *
* SCSI_TMFUNC(2), LOGOUT(1) */
#define ISER_QP_MAX_RECV_DTOS (ISCSI_DEF_XMIT_CMDS_MAX)
#define ISER_QP_MAX_RECV_DTOS (ISER_DEF_XMIT_CMDS_MAX)
#define ISER_MIN_POSTED_RX (ISCSI_DEF_XMIT_CMDS_MAX >> 2)
#define ISER_MIN_POSTED_RX (ISER_DEF_XMIT_CMDS_MAX >> 2)
/* the max TX (send) WR supported by the iSER QP is defined by *
* max_send_wr = T * (1 + D) + C ; D is how many inflight dataouts we expect *
......@@ -123,7 +129,7 @@
#define ISER_INFLIGHT_DATAOUTS 8
#define ISER_QP_MAX_REQ_DTOS (ISCSI_DEF_XMIT_CMDS_MAX * \
#define ISER_QP_MAX_REQ_DTOS (ISER_DEF_XMIT_CMDS_MAX * \
(1 + ISER_INFLIGHT_DATAOUTS) + \
ISER_MAX_TX_MISC_PDUS + \
ISER_MAX_RX_MISC_PDUS)
......@@ -272,6 +278,9 @@ struct iser_conn {
struct ib_qp *qp; /* QP */
struct ib_fmr_pool *fmr_pool; /* pool of IB FMRs */
wait_queue_head_t wait; /* waitq for conn/disconn */
unsigned qp_max_recv_dtos; /* num of rx buffers */
unsigned qp_max_recv_dtos_mask; /* above minus 1 */
unsigned min_posted_rx; /* qp_max_recv_dtos >> 2 */
int post_recv_buf_count; /* posted rx count */
atomic_t post_send_buf_count; /* posted tx count */
char name[ISER_OBJECT_NAME_SIZE];
......@@ -394,7 +403,7 @@ int iser_dma_map_task_data(struct iscsi_iser_task *iser_task,
void iser_dma_unmap_task_data(struct iscsi_iser_task *iser_task);
int iser_initialize_task_headers(struct iscsi_task *task,
struct iser_tx_desc *tx_desc);
int iser_alloc_rx_descriptors(struct iser_conn *ib_conn);
int iser_create_fmr_pool(struct iser_conn *ib_conn);
int iser_alloc_rx_descriptors(struct iser_conn *ib_conn, struct iscsi_session *session);
int iser_create_fmr_pool(struct iser_conn *ib_conn, unsigned cmds_max);
void iser_free_fmr_pool(struct iser_conn *ib_conn);
#endif
......@@ -241,7 +241,7 @@ static int iser_alloc_login_buf(struct iser_conn *ib_conn)
return -ENOMEM;
}
int iser_alloc_rx_descriptors(struct iser_conn *ib_conn)
int iser_alloc_rx_descriptors(struct iser_conn *ib_conn, struct iscsi_session *session)
{
int i, j;
u64 dma_addr;
......@@ -249,20 +249,24 @@ int iser_alloc_rx_descriptors(struct iser_conn *ib_conn)
struct ib_sge *rx_sg;
struct iser_device *device = ib_conn->device;
if (iser_create_fmr_pool(ib_conn))
ib_conn->qp_max_recv_dtos = session->cmds_max;
ib_conn->qp_max_recv_dtos_mask = session->cmds_max - 1; /* cmds_max is 2^N */
ib_conn->min_posted_rx = ib_conn->qp_max_recv_dtos >> 2;
if (iser_create_fmr_pool(ib_conn, session->scsi_cmds_max))
goto create_fmr_pool_failed;
if (iser_alloc_login_buf(ib_conn))
goto alloc_login_buf_fail;
ib_conn->rx_descs = kmalloc(ISER_QP_MAX_RECV_DTOS *
ib_conn->rx_descs = kmalloc(session->cmds_max *
sizeof(struct iser_rx_desc), GFP_KERNEL);
if (!ib_conn->rx_descs)
goto rx_desc_alloc_fail;
rx_desc = ib_conn->rx_descs;
for (i = 0; i < ISER_QP_MAX_RECV_DTOS; i++, rx_desc++) {
for (i = 0; i < ib_conn->qp_max_recv_dtos; i++, rx_desc++) {
dma_addr = ib_dma_map_single(device->ib_device, (void *)rx_desc,
ISER_RX_PAYLOAD_SIZE, DMA_FROM_DEVICE);
if (ib_dma_mapping_error(device->ib_device, dma_addr))
......@@ -305,7 +309,7 @@ void iser_free_rx_descriptors(struct iser_conn *ib_conn)
goto free_login_buf;
rx_desc = ib_conn->rx_descs;
for (i = 0; i < ISER_QP_MAX_RECV_DTOS; i++, rx_desc++)
for (i = 0; i < ib_conn->qp_max_recv_dtos; i++, rx_desc++)
ib_dma_unmap_single(device->ib_device, rx_desc->dma_addr,
ISER_RX_PAYLOAD_SIZE, DMA_FROM_DEVICE);
kfree(ib_conn->rx_descs);
......@@ -334,9 +338,10 @@ static int iser_post_rx_bufs(struct iscsi_conn *conn, struct iscsi_hdr *req)
WARN_ON(iser_conn->ib_conn->post_recv_buf_count != 1);
WARN_ON(atomic_read(&iser_conn->ib_conn->post_send_buf_count) != 0);
iser_dbg("Initially post: %d\n", ISER_MIN_POSTED_RX);
iser_dbg("Initially post: %d\n", iser_conn->ib_conn->min_posted_rx);
/* Initial post receive buffers */
if (iser_post_recvm(iser_conn->ib_conn, ISER_MIN_POSTED_RX))
if (iser_post_recvm(iser_conn->ib_conn,
iser_conn->ib_conn->min_posted_rx))
return -ENOMEM;
return 0;
......@@ -573,9 +578,9 @@ void iser_rcv_completion(struct iser_rx_desc *rx_desc,
return;
outstanding = ib_conn->post_recv_buf_count;
if (outstanding + ISER_MIN_POSTED_RX <= ISER_QP_MAX_RECV_DTOS) {
count = min(ISER_QP_MAX_RECV_DTOS - outstanding,
ISER_MIN_POSTED_RX);
if (outstanding + ib_conn->min_posted_rx <= ib_conn->qp_max_recv_dtos) {
count = min(ib_conn->qp_max_recv_dtos - outstanding,
ib_conn->min_posted_rx);
err = iser_post_recvm(ib_conn, count);
if (err)
iser_err("posting %d rx bufs err %d\n", count, err);
......
......@@ -182,7 +182,7 @@ static void iser_free_device_ib_res(struct iser_device *device)
*
* returns 0 on success, or errno code on failure
*/
int iser_create_fmr_pool(struct iser_conn *ib_conn)
int iser_create_fmr_pool(struct iser_conn *ib_conn, unsigned cmds_max)
{
struct iser_device *device = ib_conn->device;
struct ib_fmr_pool_param params;
......@@ -202,8 +202,8 @@ int iser_create_fmr_pool(struct iser_conn *ib_conn)
params.max_pages_per_fmr = ISCSI_ISER_SG_TABLESIZE + 1;
/* make the pool size twice the max number of SCSI commands *
* the ML is expected to queue, watermark for unmap at 50% */
params.pool_size = ISCSI_DEF_XMIT_CMDS_MAX * 2;
params.dirty_watermark = ISCSI_DEF_XMIT_CMDS_MAX;
params.pool_size = cmds_max * 2;
params.dirty_watermark = cmds_max;
params.cache = 0;
params.flush_function = NULL;
params.access = (IB_ACCESS_LOCAL_WRITE |
......@@ -771,7 +771,7 @@ int iser_post_recvm(struct iser_conn *ib_conn, int count)
rx_wr->sg_list = &rx_desc->rx_sg;
rx_wr->num_sge = 1;
rx_wr->next = rx_wr + 1;
my_rx_head = (my_rx_head + 1) & (ISER_QP_MAX_RECV_DTOS - 1);
my_rx_head = (my_rx_head + 1) & ib_conn->qp_max_recv_dtos_mask;
}
rx_wr--;
......
Markdown is supported
0% .
You are about to add 0 people to the discussion. Proceed with caution.
先完成此消息的编辑!
想要评论请 注册