From d2ccde1c8e90d38cee87a1b0309ad2827f3fd30d Mon Sep 17 00:00:00 2001
From: hc <hc@nodka.com>
Date: Mon, 11 Dec 2023 02:45:28 +0000
Subject: [PATCH] add boot partition size
---
kernel/drivers/infiniband/hw/hfi1/qp.c | 234 ++++++++++++++++++++++++++++++++++++++++++---------------
1 files changed, 171 insertions(+), 63 deletions(-)
diff --git a/kernel/drivers/infiniband/hw/hfi1/qp.c b/kernel/drivers/infiniband/hw/hfi1/qp.c
index 63c5ba6..356518e 100644
--- a/kernel/drivers/infiniband/hw/hfi1/qp.c
+++ b/kernel/drivers/infiniband/hw/hfi1/qp.c
@@ -1,5 +1,5 @@
/*
- * Copyright(c) 2015 - 2018 Intel Corporation.
+ * Copyright(c) 2015 - 2020 Intel Corporation.
*
* This file is provided under a dual BSD/GPLv2 license. When using or
* redistributing this file, you may do so under either license.
@@ -66,7 +66,7 @@
static void flush_tx_list(struct rvt_qp *qp);
static int iowait_sleep(
struct sdma_engine *sde,
- struct iowait *wait,
+ struct iowait_work *wait,
struct sdma_txreq *stx,
unsigned int seq,
bool pkts_sent);
@@ -132,23 +132,41 @@
.qpt_support = BIT(IB_QPT_RC),
},
+[IB_WR_OPFN] = {
+ .length = sizeof(struct ib_atomic_wr),
+ .qpt_support = BIT(IB_QPT_RC),
+ .flags = RVT_OPERATION_USE_RESERVE,
+},
+
+[IB_WR_TID_RDMA_WRITE] = {
+ .length = sizeof(struct ib_rdma_wr),
+ .qpt_support = BIT(IB_QPT_RC),
+ .flags = RVT_OPERATION_IGN_RNR_CNT,
+},
+
};
-static void flush_tx_list(struct rvt_qp *qp)
+static void flush_list_head(struct list_head *l)
{
- struct hfi1_qp_priv *priv = qp->priv;
-
- while (!list_empty(&priv->s_iowait.tx_head)) {
+ while (!list_empty(l)) {
struct sdma_txreq *tx;
tx = list_first_entry(
- &priv->s_iowait.tx_head,
+ l,
struct sdma_txreq,
list);
list_del_init(&tx->list);
hfi1_put_txreq(
container_of(tx, struct verbs_txreq, txreq));
}
+}
+
+static void flush_tx_list(struct rvt_qp *qp)
+{
+ struct hfi1_qp_priv *priv = qp->priv;
+
+ flush_list_head(&iowait_get_ib_work(&priv->s_iowait)->tx_head);
+ flush_list_head(&iowait_get_tid_work(&priv->s_iowait)->tx_head);
}
static void flush_iowait(struct rvt_qp *qp)
@@ -168,15 +186,6 @@
write_sequnlock_irqrestore(lock, flags);
}
-static inline int opa_mtu_enum_to_int(int mtu)
-{
- switch (mtu) {
- case OPA_MTU_8192: return 8192;
- case OPA_MTU_10240: return 10240;
- default: return -1;
- }
-}
-
/**
* This function is what we would push to the core layer if we wanted to be a
* "first class citizen". Instead we hide this here and rely on Verbs ULPs
@@ -184,15 +193,10 @@
*/
static inline int verbs_mtu_enum_to_int(struct ib_device *dev, enum ib_mtu mtu)
{
- int val;
-
/* Constraining 10KB packets to 8KB packets */
if (mtu == (enum ib_mtu)OPA_MTU_10240)
- mtu = OPA_MTU_8192;
- val = opa_mtu_enum_to_int((int)mtu);
- if (val > 0)
- return val;
- return ib_mtu_enum_to_int(mtu);
+ mtu = (enum ib_mtu)OPA_MTU_8192;
+ return opa_mtu_enum_to_int((enum opa_mtu)mtu);
}
int hfi1_check_modify_qp(struct rvt_qp *qp, struct ib_qp_attr *attr,
@@ -279,41 +283,58 @@
priv->s_sendcontext = qp_to_send_context(qp, priv->s_sc);
qp_set_16b(qp);
}
+
+ opfn_qp_init(qp, attr, attr_mask);
}
/**
- * hfi1_check_send_wqe - validate wqe
+ * hfi1_setup_wqe - set up the wqe
* @qp - The qp
* @wqe - The built wqe
+ * @call_send - Determine if the send should be posted or scheduled.
*
- * validate wqe. This is called
- * prior to inserting the wqe into
- * the ring but after the wqe has been
- * setup.
+ * Perform setup of the wqe. This is called
+ * prior to inserting the wqe into the ring but after
+ * the wqe has been setup by RDMAVT. This function
+ * allows the driver the opportunity to perform
+ * validation and additional setup of the wqe.
*
* Returns 0 on success, -EINVAL on failure
*
*/
-int hfi1_check_send_wqe(struct rvt_qp *qp,
- struct rvt_swqe *wqe)
+int hfi1_setup_wqe(struct rvt_qp *qp, struct rvt_swqe *wqe, bool *call_send)
{
struct hfi1_ibport *ibp = to_iport(qp->ibqp.device, qp->port_num);
struct rvt_ah *ah;
+ struct hfi1_pportdata *ppd;
+ struct hfi1_devdata *dd;
switch (qp->ibqp.qp_type) {
case IB_QPT_RC:
+ hfi1_setup_tid_rdma_wqe(qp, wqe);
+ fallthrough;
case IB_QPT_UC:
if (wqe->length > 0x80000000U)
return -EINVAL;
+ if (wqe->length > qp->pmtu)
+ *call_send = false;
break;
case IB_QPT_SMI:
- ah = ibah_to_rvtah(wqe->ud_wr.ah);
- if (wqe->length > (1 << ah->log_pmtu))
+ /*
+ * SM packets should exclusively use VL15 and their SL is
+ * ignored (IBTA v1.3, Section 3.5.8.2). Therefore, when ah
+ * is created, SL is 0 in most cases and as a result some
+ * fields (vl and pmtu) in ah may not be set correctly,
+ * depending on the SL2SC and SC2VL tables at the time.
+ */
+ ppd = ppd_from_ibp(ibp);
+ dd = dd_from_ppd(ppd);
+ if (wqe->length > dd->vld[15].mtu)
return -EINVAL;
break;
case IB_QPT_GSI:
case IB_QPT_UD:
- ah = ibah_to_rvtah(wqe->ud_wr.ah);
+ ah = rvt_get_swqe_ah(wqe);
if (wqe->length > (1 << ah->log_pmtu))
return -EINVAL;
if (ibp->sl_to_sc[rdma_ah_get_sl(&ah->attr)] == 0xf)
@@ -321,7 +342,14 @@
default:
break;
}
- return wqe->length <= piothreshold;
+
+ /*
+ * System latency between send and schedule is large enough that
+ * forcing call_send to true for piothreshold packets is necessary.
+ */
+ if (wqe->length <= piothreshold)
+ *call_send = true;
+ return 0;
}
/**
@@ -333,36 +361,37 @@
* It is only used in the post send, which doesn't hold
* the s_lock.
*/
-void _hfi1_schedule_send(struct rvt_qp *qp)
+bool _hfi1_schedule_send(struct rvt_qp *qp)
{
struct hfi1_qp_priv *priv = qp->priv;
struct hfi1_ibport *ibp =
to_iport(qp->ibqp.device, qp->port_num);
struct hfi1_pportdata *ppd = ppd_from_ibp(ibp);
- struct hfi1_devdata *dd = dd_from_ibdev(qp->ibqp.device);
+ struct hfi1_devdata *dd = ppd->dd;
- iowait_schedule(&priv->s_iowait, ppd->hfi1_wq,
- priv->s_sde ?
- priv->s_sde->cpu :
- cpumask_first(cpumask_of_node(dd->node)));
+ if (dd->flags & HFI1_SHUTDOWN)
+ return true;
+
+ return iowait_schedule(&priv->s_iowait, ppd->hfi1_wq,
+ priv->s_sde ?
+ priv->s_sde->cpu :
+ cpumask_first(cpumask_of_node(dd->node)));
}
static void qp_pio_drain(struct rvt_qp *qp)
{
- struct hfi1_ibdev *dev;
struct hfi1_qp_priv *priv = qp->priv;
if (!priv->s_sendcontext)
return;
- dev = to_idev(qp->ibqp.device);
while (iowait_pio_pending(&priv->s_iowait)) {
- write_seqlock_irq(&dev->iowait_lock);
+ write_seqlock_irq(&priv->s_sendcontext->waitlock);
hfi1_sc_wantpiobuf_intr(priv->s_sendcontext, 1);
- write_sequnlock_irq(&dev->iowait_lock);
+ write_sequnlock_irq(&priv->s_sendcontext->waitlock);
iowait_pio_drain(&priv->s_iowait);
- write_seqlock_irq(&dev->iowait_lock);
+ write_seqlock_irq(&priv->s_sendcontext->waitlock);
hfi1_sc_wantpiobuf_intr(priv->s_sendcontext, 0);
- write_sequnlock_irq(&dev->iowait_lock);
+ write_sequnlock_irq(&priv->s_sendcontext->waitlock);
}
}
@@ -372,12 +401,37 @@
*
* This schedules qp progress and caller should hold
* the s_lock.
+ * @return true if the first leg is scheduled;
+ * false if the first leg is not scheduled.
*/
-void hfi1_schedule_send(struct rvt_qp *qp)
+bool hfi1_schedule_send(struct rvt_qp *qp)
{
lockdep_assert_held(&qp->s_lock);
- if (hfi1_send_ok(qp))
+ if (hfi1_send_ok(qp)) {
_hfi1_schedule_send(qp);
+ return true;
+ }
+ if (qp->s_flags & HFI1_S_ANY_WAIT_IO)
+ iowait_set_flag(&((struct hfi1_qp_priv *)qp->priv)->s_iowait,
+ IOWAIT_PENDING_IB);
+ return false;
+}
+
+static void hfi1_qp_schedule(struct rvt_qp *qp)
+{
+ struct hfi1_qp_priv *priv = qp->priv;
+ bool ret;
+
+ if (iowait_flag_set(&priv->s_iowait, IOWAIT_PENDING_IB)) {
+ ret = hfi1_schedule_send(qp);
+ if (ret)
+ iowait_clear_flag(&priv->s_iowait, IOWAIT_PENDING_IB);
+ }
+ if (iowait_flag_set(&priv->s_iowait, IOWAIT_PENDING_TID)) {
+ ret = hfi1_schedule_tid_send(qp);
+ if (ret)
+ iowait_clear_flag(&priv->s_iowait, IOWAIT_PENDING_TID);
+ }
}
void hfi1_qp_wakeup(struct rvt_qp *qp, u32 flag)
@@ -388,16 +442,41 @@
if (qp->s_flags & flag) {
qp->s_flags &= ~flag;
trace_hfi1_qpwakeup(qp, flag);
- hfi1_schedule_send(qp);
+ hfi1_qp_schedule(qp);
}
spin_unlock_irqrestore(&qp->s_lock, flags);
/* Notify hfi1_destroy_qp() if it is waiting. */
rvt_put_qp(qp);
}
+void hfi1_qp_unbusy(struct rvt_qp *qp, struct iowait_work *wait)
+{
+ struct hfi1_qp_priv *priv = qp->priv;
+
+ if (iowait_set_work_flag(wait) == IOWAIT_IB_SE) {
+ qp->s_flags &= ~RVT_S_BUSY;
+ /*
+ * If we are sending a first-leg packet from the second leg,
+ * we need to clear the busy flag from priv->s_flags to
+ * avoid a race condition when the qp wakes up before
+ * the call to hfi1_verbs_send() returns to the second
+ * leg. In that case, the second leg will terminate without
+ * being re-scheduled, resulting in failure to send TID RDMA
+ * WRITE DATA and TID RDMA ACK packets.
+ */
+ if (priv->s_flags & HFI1_S_TID_BUSY_SET) {
+ priv->s_flags &= ~(HFI1_S_TID_BUSY_SET |
+ RVT_S_BUSY);
+ iowait_set_flag(&priv->s_iowait, IOWAIT_PENDING_TID);
+ }
+ } else {
+ priv->s_flags &= ~RVT_S_BUSY;
+ }
+}
+
static int iowait_sleep(
struct sdma_engine *sde,
- struct iowait *wait,
+ struct iowait_work *wait,
struct sdma_txreq *stx,
uint seq,
bool pkts_sent)
@@ -407,7 +486,6 @@
struct hfi1_qp_priv *priv;
unsigned long flags;
int ret = 0;
- struct hfi1_ibdev *dev;
qp = tx->qp;
priv = qp->priv;
@@ -420,9 +498,8 @@
* buffer and undoing the side effects of the copy.
*/
/* Make a common routine? */
- dev = &sde->dd->verbs_dev;
list_add_tail(&stx->list, &wait->tx_head);
- write_seqlock(&dev->iowait_lock);
+ write_seqlock(&sde->waitlock);
if (sdma_progress(sde, seq, stx))
goto eagain;
if (list_empty(&priv->s_iowait.list)) {
@@ -431,14 +508,15 @@
ibp->rvp.n_dmawait++;
qp->s_flags |= RVT_S_WAIT_DMA_DESC;
+ iowait_get_priority(&priv->s_iowait);
iowait_queue(pkts_sent, &priv->s_iowait,
&sde->dmawait);
- priv->s_iowait.lock = &dev->iowait_lock;
+ priv->s_iowait.lock = &sde->waitlock;
trace_hfi1_qpsleep(qp, RVT_S_WAIT_DMA_DESC);
rvt_get_qp(qp);
}
- write_sequnlock(&dev->iowait_lock);
- qp->s_flags &= ~RVT_S_BUSY;
+ write_sequnlock(&sde->waitlock);
+ hfi1_qp_unbusy(qp, wait);
spin_unlock_irqrestore(&qp->s_lock, flags);
ret = -EBUSY;
} else {
@@ -447,7 +525,7 @@
}
return ret;
eagain:
- write_sequnlock(&dev->iowait_lock);
+ write_sequnlock(&sde->waitlock);
spin_unlock_irqrestore(&qp->s_lock, flags);
list_del_init(&stx->list);
return -EAGAIN;
@@ -478,6 +556,17 @@
hfi1_schedule_send(qp);
}
spin_unlock_irqrestore(&qp->s_lock, flags);
+}
+
+static void hfi1_init_priority(struct iowait *w)
+{
+ struct rvt_qp *qp = iowait_to_qp(w);
+ struct hfi1_qp_priv *priv = qp->priv;
+
+ if (qp->s_flags & RVT_S_ACK_PENDING)
+ w->priority++;
+ if (priv->s_flags & RVT_S_ACK_PENDING)
+ w->priority++;
}
/**
@@ -602,8 +691,8 @@
sde ? sde->this_idx : 0,
send_context,
send_context ? send_context->sw_index : 0,
- ibcq_to_rvtcq(qp->ibqp.send_cq)->queue->head,
- ibcq_to_rvtcq(qp->ibqp.send_cq)->queue->tail,
+ ib_cq_head(qp->ibqp.send_cq),
+ ib_cq_tail(qp->ibqp.send_cq),
qp->pid,
qp->s_state,
qp->s_ack_state,
@@ -637,9 +726,13 @@
&priv->s_iowait,
1,
_hfi1_do_send,
+ _hfi1_do_tid_send,
iowait_sleep,
iowait_wakeup,
- iowait_sdma_drained);
+ iowait_sdma_drained,
+ hfi1_init_priority);
+ /* Init to a value to start the running average correctly */
+ priv->s_running_pkt_size = piothreshold / 2;
return priv;
}
@@ -647,6 +740,7 @@
{
struct hfi1_qp_priv *priv = qp->priv;
+ hfi1_qp_priv_tid_free(rdi, qp);
kfree(priv->s_ahg);
kfree(priv);
}
@@ -680,19 +774,24 @@
{
lockdep_assert_held(&qp->s_lock);
flush_iowait(qp);
+ hfi1_tid_rdma_flush_wait(qp);
}
void stop_send_queue(struct rvt_qp *qp)
{
struct hfi1_qp_priv *priv = qp->priv;
- cancel_work_sync(&priv->s_iowait.iowork);
+ iowait_cancel_work(&priv->s_iowait);
+ if (cancel_work_sync(&priv->tid_rdma.trigger_work))
+ rvt_put_qp(qp);
}
void quiesce_qp(struct rvt_qp *qp)
{
struct hfi1_qp_priv *priv = qp->priv;
+ hfi1_del_tid_reap_timer(qp);
+ hfi1_del_tid_retry_timer(qp);
iowait_sdma_drain(&priv->s_iowait);
qp_pio_drain(qp);
flush_tx_list(qp);
@@ -700,8 +799,13 @@
void notify_qp_reset(struct rvt_qp *qp)
{
+ hfi1_qp_kern_exp_rcv_clear_all(qp);
qp->r_adefered = 0;
clear_ahg(qp);
+
+ /* Clear any OPFN state */
+ if (qp->ibqp.qp_type == IB_QPT_RC)
+ opfn_conn_error(qp);
}
/*
@@ -783,8 +887,11 @@
if (lock) {
write_seqlock(lock);
if (!list_empty(&priv->s_iowait.list) &&
- !(qp->s_flags & RVT_S_BUSY)) {
+ !(qp->s_flags & RVT_S_BUSY) &&
+ !(priv->s_flags & RVT_S_BUSY)) {
qp->s_flags &= ~HFI1_S_ANY_WAIT_IO;
+ iowait_clear_flag(&priv->s_iowait, IOWAIT_PENDING_IB);
+ iowait_clear_flag(&priv->s_iowait, IOWAIT_PENDING_TID);
list_del_init(&priv->s_iowait.list);
priv->s_iowait.lock = NULL;
rvt_put_qp(qp);
@@ -792,7 +899,8 @@
write_sequnlock(lock);
}
- if (!(qp->s_flags & RVT_S_BUSY)) {
+ if (!(qp->s_flags & RVT_S_BUSY) && !(priv->s_flags & RVT_S_BUSY)) {
+ qp->s_hdrwords = 0;
if (qp->s_rdma_mr) {
rvt_put_mr(qp->s_rdma_mr);
qp->s_rdma_mr = NULL;
--
Gitblit v1.6.2