Work around lack of TX IRQs in iflib for netmap

When poll() is called via netmap, txsync is initially called,
and if there are no available buffers to reclaim, it waits for the driver
to notify of new buffers. Since the TX IRQ is generally not used in iflib
drivers, this ends up causing a timeout.

Work around this by having the reclaim DELAY(1) if it's initially unable
to reclaim anything, then schedule the tx task, which will spin by
continuously rescheduling itself until some buffers are reclaimed. In
general, the delay is enough to allow some buffers to be reclaimed, so
spinning is minimized.

Reported by:	Johannes Lundberg <johalun0@gmail.com>
Reviewed by:	sbruno
Sponsored by:	Limelight Networks
Differential Revision:	https://reviews.freebsd.org/D15455
This commit is contained in:
Stephen Hurd 2018-05-16 21:03:22 +00:00
parent 57b1f8183c
commit 5ee36c68e1
Notes: svn2git 2020-12-20 02:59:44 +00:00
svn path=/head/; revision=333686
2 changed files with 35 additions and 7 deletions

View File

@ -960,10 +960,10 @@ iflib_netmap_txsync(struct netmap_kring *kring, int flags)
*/
nm_i = netmap_idx_n2k(kring, kring->nr_hwcur);
pkt_info_zero(&pi);
pi.ipi_segs = txq->ift_segs;
pi.ipi_qsidx = kring->ring_id;
if (nm_i != head) { /* we have new packets to send */
pkt_info_zero(&pi);
pi.ipi_segs = txq->ift_segs;
pi.ipi_qsidx = kring->ring_id;
nic_i = netmap_idx_k2n(kring, nm_i);
__builtin_prefetch(&ring->slot[nm_i]);
@ -1025,11 +1025,24 @@ iflib_netmap_txsync(struct netmap_kring *kring, int flags)
/*
* Second part: reclaim buffers for completed transmissions.
*
* If there are unclaimed buffers, attempt to reclaim them.
* If none are reclaimed, and TX IRQs are not in use, do an initial
* minimal delay, then trigger the tx handler which will spin in the
* group task queue.
*/
if (iflib_tx_credits_update(ctx, txq)) {
/* some tx completed, increment avail */
nic_i = txq->ift_cidx_processed;
kring->nr_hwtail = nm_prev(netmap_idx_n2k(kring, nic_i), lim);
if (kring->nr_hwtail != nm_prev(head, lim)) {
if (iflib_tx_credits_update(ctx, txq)) {
/* some tx completed, increment avail */
nic_i = txq->ift_cidx_processed;
kring->nr_hwtail = nm_prev(netmap_idx_n2k(kring, nic_i), lim);
}
else {
if (!(ctx->ifc_flags & IFC_NETMAP_TX_IRQ)) {
DELAY(1);
GROUPTASK_ENQUEUE(&ctx->ifc_txqs[txq->ift_id].ift_task);
}
}
}
return (0);
}
@ -3702,8 +3715,20 @@ _task_fn_tx(void *context)
if (!(if_getdrvflags(ctx->ifc_ifp) & IFF_DRV_RUNNING))
return;
if (if_getcapenable(ifp) & IFCAP_NETMAP) {
/*
* If there are no available credits, and TX IRQs are not in use,
* re-schedule the task immediately.
*/
if (ctx->isc_txd_credits_update(ctx->ifc_softc, txq->ift_id, false))
netmap_tx_irq(ifp, txq->ift_id);
else {
if (!(ctx->ifc_flags & IFC_NETMAP_TX_IRQ)) {
struct netmap_kring *kring = NA(ctx->ifc_ifp)->tx_rings[txq->ift_id];
if (kring->nr_hwtail != nm_prev(kring->rhead, kring->nkr_num_slots - 1))
GROUPTASK_ENQUEUE(&txq->ift_task);
}
}
IFDI_TX_QUEUE_INTR_ENABLE(ctx, txq->ift_id);
return;
}
@ -5548,6 +5573,7 @@ iflib_irq_alloc_generic(if_ctx_t ctx, if_irq_t irq, int rid,
fn = _task_fn_tx;
intr_fast = iflib_fast_intr;
GROUPTASK_INIT(gtask, 0, fn, q);
ctx->ifc_flags |= IFC_NETMAP_TX_IRQ;
break;
case IFLIB_INTR_RX:
q = &ctx->ifc_rxqs[qid];

View File

@ -43,6 +43,8 @@
#define IFC_CHECK_HUNG 0x200
#define IFC_PSEUDO 0x400
#define IFC_NETMAP_TX_IRQ 0x80000000
MALLOC_DECLARE(M_IFLIB);
#define IFLIB_MAX_TX_BYTES (2*1024*1024)