freebsd-nq/sys/dev/nvme/nvme_sysctl.c
Alexander Motin 1eab19cbec Make nvme(4) driver some more NUMA aware.
- For each queue pair precalculate CPU and domain it is bound to.
If queue pairs are not per-CPU, then use the domain of the device.
 - Allocate most of queue pair memory from the domain it is bound to.
 - Bind callouts to the same CPUs as queue pair to avoid migrations.
 - Do not assign queue pairs to each SMT thread.  It just wasted
resources and increased lock congestions.
 - Remove fixed multiplier of CPUs per queue pair, spread them even.
This allows to use more queue pairs in some hardware configurations.
 - If queue pair serves multiple CPUs, bind different NVMe devices to
different CPUs.

MFC after:	1 month
Sponsored by:	iXsystems, Inc.
2019-09-23 17:53:47 +00:00

367 lines
11 KiB
C

/*-
* SPDX-License-Identifier: BSD-2-Clause-FreeBSD
*
* Copyright (C) 2012-2016 Intel Corporation
* All rights reserved.
*
* Redistribution and use in source and binary forms, with or without
* modification, are permitted provided that the following conditions
* are met:
* 1. Redistributions of source code must retain the above copyright
* notice, this list of conditions and the following disclaimer.
* 2. Redistributions in binary form must reproduce the above copyright
* notice, this list of conditions and the following disclaimer in the
* documentation and/or other materials provided with the distribution.
*
* THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
* ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
* IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
* ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
* FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
* DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
* OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
* HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
* LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
* OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
* SUCH DAMAGE.
*/
#include <sys/cdefs.h>
__FBSDID("$FreeBSD$");
#include "opt_nvme.h"
#include <sys/param.h>
#include <sys/bus.h>
#include <sys/sysctl.h>
#include "nvme_private.h"
#ifndef NVME_USE_NVD
#define NVME_USE_NVD 1
#endif
int nvme_use_nvd = NVME_USE_NVD;
bool nvme_verbose_cmd_dump = false;
SYSCTL_NODE(_hw, OID_AUTO, nvme, CTLFLAG_RD, 0, "NVMe sysctl tunables");
SYSCTL_INT(_hw_nvme, OID_AUTO, use_nvd, CTLFLAG_RDTUN,
&nvme_use_nvd, 1, "1 = Create NVD devices, 0 = Create NDA devices");
SYSCTL_BOOL(_hw_nvme, OID_AUTO, verbose_cmd_dump, CTLFLAG_RWTUN,
&nvme_verbose_cmd_dump, 0,
"enable verbose command printting when a command fails");
/*
* CTLTYPE_S64 and sysctl_handle_64 were added in r217616. Define these
* explicitly here for older kernels that don't include the r217616
* changeset.
*/
#ifndef CTLTYPE_S64
#define CTLTYPE_S64 CTLTYPE_QUAD
#define sysctl_handle_64 sysctl_handle_quad
#endif
static void
nvme_dump_queue(struct nvme_qpair *qpair)
{
struct nvme_completion *cpl;
struct nvme_command *cmd;
int i;
printf("id:%04Xh phase:%d\n", qpair->id, qpair->phase);
printf("Completion queue:\n");
for (i = 0; i < qpair->num_entries; i++) {
cpl = &qpair->cpl[i];
printf("%05d: ", i);
nvme_dump_completion(cpl);
}
printf("Submission queue:\n");
for (i = 0; i < qpair->num_entries; i++) {
cmd = &qpair->cmd[i];
printf("%05d: ", i);
nvme_dump_command(cmd);
}
}
static int
nvme_sysctl_dump_debug(SYSCTL_HANDLER_ARGS)
{
struct nvme_qpair *qpair = arg1;
uint32_t val = 0;
int error = sysctl_handle_int(oidp, &val, 0, req);
if (error)
return (error);
if (val != 0)
nvme_dump_queue(qpair);
return (0);
}
static int
nvme_sysctl_int_coal_time(SYSCTL_HANDLER_ARGS)
{
struct nvme_controller *ctrlr = arg1;
uint32_t oldval = ctrlr->int_coal_time;
int error = sysctl_handle_int(oidp, &ctrlr->int_coal_time, 0,
req);
if (error)
return (error);
if (oldval != ctrlr->int_coal_time)
nvme_ctrlr_cmd_set_interrupt_coalescing(ctrlr,
ctrlr->int_coal_time, ctrlr->int_coal_threshold, NULL,
NULL);
return (0);
}
static int
nvme_sysctl_int_coal_threshold(SYSCTL_HANDLER_ARGS)
{
struct nvme_controller *ctrlr = arg1;
uint32_t oldval = ctrlr->int_coal_threshold;
int error = sysctl_handle_int(oidp, &ctrlr->int_coal_threshold, 0,
req);
if (error)
return (error);
if (oldval != ctrlr->int_coal_threshold)
nvme_ctrlr_cmd_set_interrupt_coalescing(ctrlr,
ctrlr->int_coal_time, ctrlr->int_coal_threshold, NULL,
NULL);
return (0);
}
static int
nvme_sysctl_timeout_period(SYSCTL_HANDLER_ARGS)
{
struct nvme_controller *ctrlr = arg1;
uint32_t oldval = ctrlr->timeout_period;
int error = sysctl_handle_int(oidp, &ctrlr->timeout_period, 0, req);
if (error)
return (error);
if (ctrlr->timeout_period > NVME_MAX_TIMEOUT_PERIOD ||
ctrlr->timeout_period < NVME_MIN_TIMEOUT_PERIOD) {
ctrlr->timeout_period = oldval;
return (EINVAL);
}
return (0);
}
static void
nvme_qpair_reset_stats(struct nvme_qpair *qpair)
{
qpair->num_cmds = 0;
qpair->num_intr_handler_calls = 0;
qpair->num_retries = 0;
qpair->num_failures = 0;
}
static int
nvme_sysctl_num_cmds(SYSCTL_HANDLER_ARGS)
{
struct nvme_controller *ctrlr = arg1;
int64_t num_cmds = 0;
int i;
num_cmds = ctrlr->adminq.num_cmds;
for (i = 0; i < ctrlr->num_io_queues; i++)
num_cmds += ctrlr->ioq[i].num_cmds;
return (sysctl_handle_64(oidp, &num_cmds, 0, req));
}
static int
nvme_sysctl_num_intr_handler_calls(SYSCTL_HANDLER_ARGS)
{
struct nvme_controller *ctrlr = arg1;
int64_t num_intr_handler_calls = 0;
int i;
num_intr_handler_calls = ctrlr->adminq.num_intr_handler_calls;
for (i = 0; i < ctrlr->num_io_queues; i++)
num_intr_handler_calls += ctrlr->ioq[i].num_intr_handler_calls;
return (sysctl_handle_64(oidp, &num_intr_handler_calls, 0, req));
}
static int
nvme_sysctl_num_retries(SYSCTL_HANDLER_ARGS)
{
struct nvme_controller *ctrlr = arg1;
int64_t num_retries = 0;
int i;
num_retries = ctrlr->adminq.num_retries;
for (i = 0; i < ctrlr->num_io_queues; i++)
num_retries += ctrlr->ioq[i].num_retries;
return (sysctl_handle_64(oidp, &num_retries, 0, req));
}
static int
nvme_sysctl_num_failures(SYSCTL_HANDLER_ARGS)
{
struct nvme_controller *ctrlr = arg1;
int64_t num_failures = 0;
int i;
num_failures = ctrlr->adminq.num_failures;
for (i = 0; i < ctrlr->num_io_queues; i++)
num_failures += ctrlr->ioq[i].num_failures;
return (sysctl_handle_64(oidp, &num_failures, 0, req));
}
static int
nvme_sysctl_reset_stats(SYSCTL_HANDLER_ARGS)
{
struct nvme_controller *ctrlr = arg1;
uint32_t i, val = 0;
int error = sysctl_handle_int(oidp, &val, 0, req);
if (error)
return (error);
if (val != 0) {
nvme_qpair_reset_stats(&ctrlr->adminq);
for (i = 0; i < ctrlr->num_io_queues; i++)
nvme_qpair_reset_stats(&ctrlr->ioq[i]);
}
return (0);
}
static void
nvme_sysctl_initialize_queue(struct nvme_qpair *qpair,
struct sysctl_ctx_list *ctrlr_ctx, struct sysctl_oid *que_tree)
{
struct sysctl_oid_list *que_list = SYSCTL_CHILDREN(que_tree);
SYSCTL_ADD_UINT(ctrlr_ctx, que_list, OID_AUTO, "num_entries",
CTLFLAG_RD, &qpair->num_entries, 0,
"Number of entries in hardware queue");
SYSCTL_ADD_UINT(ctrlr_ctx, que_list, OID_AUTO, "num_trackers",
CTLFLAG_RD, &qpair->num_trackers, 0,
"Number of trackers pre-allocated for this queue pair");
SYSCTL_ADD_UINT(ctrlr_ctx, que_list, OID_AUTO, "sq_head",
CTLFLAG_RD, &qpair->sq_head, 0,
"Current head of submission queue (as observed by driver)");
SYSCTL_ADD_UINT(ctrlr_ctx, que_list, OID_AUTO, "sq_tail",
CTLFLAG_RD, &qpair->sq_tail, 0,
"Current tail of submission queue (as observed by driver)");
SYSCTL_ADD_UINT(ctrlr_ctx, que_list, OID_AUTO, "cq_head",
CTLFLAG_RD, &qpair->cq_head, 0,
"Current head of completion queue (as observed by driver)");
SYSCTL_ADD_QUAD(ctrlr_ctx, que_list, OID_AUTO, "num_cmds",
CTLFLAG_RD, &qpair->num_cmds, "Number of commands submitted");
SYSCTL_ADD_QUAD(ctrlr_ctx, que_list, OID_AUTO, "num_intr_handler_calls",
CTLFLAG_RD, &qpair->num_intr_handler_calls,
"Number of times interrupt handler was invoked (will typically be "
"less than number of actual interrupts generated due to "
"coalescing)");
SYSCTL_ADD_QUAD(ctrlr_ctx, que_list, OID_AUTO, "num_retries",
CTLFLAG_RD, &qpair->num_retries, "Number of commands retried");
SYSCTL_ADD_QUAD(ctrlr_ctx, que_list, OID_AUTO, "num_failures",
CTLFLAG_RD, &qpair->num_failures,
"Number of commands ending in failure after all retries");
SYSCTL_ADD_PROC(ctrlr_ctx, que_list, OID_AUTO,
"dump_debug", CTLTYPE_UINT | CTLFLAG_RW, qpair, 0,
nvme_sysctl_dump_debug, "IU", "Dump debug data");
}
void
nvme_sysctl_initialize_ctrlr(struct nvme_controller *ctrlr)
{
struct sysctl_ctx_list *ctrlr_ctx;
struct sysctl_oid *ctrlr_tree, *que_tree;
struct sysctl_oid_list *ctrlr_list;
#define QUEUE_NAME_LENGTH 16
char queue_name[QUEUE_NAME_LENGTH];
int i;
ctrlr_ctx = device_get_sysctl_ctx(ctrlr->dev);
ctrlr_tree = device_get_sysctl_tree(ctrlr->dev);
ctrlr_list = SYSCTL_CHILDREN(ctrlr_tree);
SYSCTL_ADD_UINT(ctrlr_ctx, ctrlr_list, OID_AUTO, "num_io_queues",
CTLFLAG_RD, &ctrlr->num_io_queues, 0,
"Number of I/O queue pairs");
SYSCTL_ADD_PROC(ctrlr_ctx, ctrlr_list, OID_AUTO,
"int_coal_time", CTLTYPE_UINT | CTLFLAG_RW, ctrlr, 0,
nvme_sysctl_int_coal_time, "IU",
"Interrupt coalescing timeout (in microseconds)");
SYSCTL_ADD_PROC(ctrlr_ctx, ctrlr_list, OID_AUTO,
"int_coal_threshold", CTLTYPE_UINT | CTLFLAG_RW, ctrlr, 0,
nvme_sysctl_int_coal_threshold, "IU",
"Interrupt coalescing threshold");
SYSCTL_ADD_PROC(ctrlr_ctx, ctrlr_list, OID_AUTO,
"timeout_period", CTLTYPE_UINT | CTLFLAG_RW, ctrlr, 0,
nvme_sysctl_timeout_period, "IU",
"Timeout period (in seconds)");
SYSCTL_ADD_PROC(ctrlr_ctx, ctrlr_list, OID_AUTO,
"num_cmds", CTLTYPE_S64 | CTLFLAG_RD,
ctrlr, 0, nvme_sysctl_num_cmds, "IU",
"Number of commands submitted");
SYSCTL_ADD_PROC(ctrlr_ctx, ctrlr_list, OID_AUTO,
"num_intr_handler_calls", CTLTYPE_S64 | CTLFLAG_RD,
ctrlr, 0, nvme_sysctl_num_intr_handler_calls, "IU",
"Number of times interrupt handler was invoked (will "
"typically be less than number of actual interrupts "
"generated due to coalescing)");
SYSCTL_ADD_PROC(ctrlr_ctx, ctrlr_list, OID_AUTO,
"num_retries", CTLTYPE_S64 | CTLFLAG_RD,
ctrlr, 0, nvme_sysctl_num_retries, "IU",
"Number of commands retried");
SYSCTL_ADD_PROC(ctrlr_ctx, ctrlr_list, OID_AUTO,
"num_failures", CTLTYPE_S64 | CTLFLAG_RD,
ctrlr, 0, nvme_sysctl_num_failures, "IU",
"Number of commands ending in failure after all retries");
SYSCTL_ADD_PROC(ctrlr_ctx, ctrlr_list, OID_AUTO,
"reset_stats", CTLTYPE_UINT | CTLFLAG_RW, ctrlr, 0,
nvme_sysctl_reset_stats, "IU", "Reset statistics to zero");
que_tree = SYSCTL_ADD_NODE(ctrlr_ctx, ctrlr_list, OID_AUTO, "adminq",
CTLFLAG_RD, NULL, "Admin Queue");
nvme_sysctl_initialize_queue(&ctrlr->adminq, ctrlr_ctx, que_tree);
for (i = 0; i < ctrlr->num_io_queues; i++) {
snprintf(queue_name, QUEUE_NAME_LENGTH, "ioq%d", i);
que_tree = SYSCTL_ADD_NODE(ctrlr_ctx, ctrlr_list, OID_AUTO,
queue_name, CTLFLAG_RD, NULL, "IO Queue");
nvme_sysctl_initialize_queue(&ctrlr->ioq[i], ctrlr_ctx,
que_tree);
}
}