Commit 7930d855 authored by Dave Jiang's avatar Dave Jiang Committed by Vinod Koul

dmaengine: idxd: add knob for enqcmds retries

Add a sysfs knob to allow tuning of retries for the kernel ENQCMDS
descriptor submission. While on host, it is not as likely that ENQCMDS
return busy during normal operations due to the driver controlling the
number of descriptors allocated for submission. However, when the driver is
operating as a guest driver, the chance of retry goes up significantly due
to sharing a wq with multiple VMs. A default value is provided with the
system admin being able to tune the value on a per WQ basis.
Suggested-by: default avatarSanjay Kumar <sanjay.k.kumar@intel.com>
Signed-off-by: default avatarDave Jiang <dave.jiang@intel.com>
Link: https://lore.kernel.org/r/163820629464.2702134.7577370098568297574.stgit@djiang5-desk3.ch.intel.comSigned-off-by: default avatarVinod Koul <vkoul@kernel.org>
parent 92452a72
...@@ -220,6 +220,13 @@ Contact: dmaengine@vger.kernel.org ...@@ -220,6 +220,13 @@ Contact: dmaengine@vger.kernel.org
Description: Show the current number of entries in this WQ if WQ Occupancy Description: Show the current number of entries in this WQ if WQ Occupancy
Support bit WQ capabilities is 1. Support bit WQ capabilities is 1.
What: /sys/bus/dsa/devices/wq<m>.<n>/enqcmds_retries
Date Oct 29, 2021
KernelVersion: 5.17.0
Contact: dmaengine@vger.kernel.org
Description: Indicate the number of retires for an enqcmds submission on a shared wq.
A max value to set attribute is capped at 64.
What: /sys/bus/dsa/devices/engine<m>.<n>/group_id What: /sys/bus/dsa/devices/engine<m>.<n>/group_id
Date: Oct 25, 2019 Date: Oct 25, 2019
KernelVersion: 5.6.0 KernelVersion: 5.6.0
......
...@@ -387,6 +387,7 @@ static void idxd_wq_disable_cleanup(struct idxd_wq *wq) ...@@ -387,6 +387,7 @@ static void idxd_wq_disable_cleanup(struct idxd_wq *wq)
wq->threshold = 0; wq->threshold = 0;
wq->priority = 0; wq->priority = 0;
wq->ats_dis = 0; wq->ats_dis = 0;
wq->enqcmds_retries = IDXD_ENQCMDS_RETRIES;
clear_bit(WQ_FLAG_DEDICATED, &wq->flags); clear_bit(WQ_FLAG_DEDICATED, &wq->flags);
clear_bit(WQ_FLAG_BLOCK_ON_FAULT, &wq->flags); clear_bit(WQ_FLAG_BLOCK_ON_FAULT, &wq->flags);
memset(wq->name, 0, WQ_NAME_SIZE); memset(wq->name, 0, WQ_NAME_SIZE);
......
...@@ -52,6 +52,9 @@ enum idxd_type { ...@@ -52,6 +52,9 @@ enum idxd_type {
#define IDXD_NAME_SIZE 128 #define IDXD_NAME_SIZE 128
#define IDXD_PMU_EVENT_MAX 64 #define IDXD_PMU_EVENT_MAX 64
#define IDXD_ENQCMDS_RETRIES 32
#define IDXD_ENQCMDS_MAX_RETRIES 64
struct idxd_device_driver { struct idxd_device_driver {
const char *name; const char *name;
enum idxd_dev_type *type; enum idxd_dev_type *type;
...@@ -173,6 +176,7 @@ struct idxd_dma_chan { ...@@ -173,6 +176,7 @@ struct idxd_dma_chan {
struct idxd_wq { struct idxd_wq {
void __iomem *portal; void __iomem *portal;
u32 portal_offset; u32 portal_offset;
unsigned int enqcmds_retries;
struct percpu_ref wq_active; struct percpu_ref wq_active;
struct completion wq_dead; struct completion wq_dead;
struct completion wq_resurrect; struct completion wq_resurrect;
...@@ -584,6 +588,7 @@ int idxd_wq_init_percpu_ref(struct idxd_wq *wq); ...@@ -584,6 +588,7 @@ int idxd_wq_init_percpu_ref(struct idxd_wq *wq);
int idxd_submit_desc(struct idxd_wq *wq, struct idxd_desc *desc); int idxd_submit_desc(struct idxd_wq *wq, struct idxd_desc *desc);
struct idxd_desc *idxd_alloc_desc(struct idxd_wq *wq, enum idxd_op_type optype); struct idxd_desc *idxd_alloc_desc(struct idxd_wq *wq, enum idxd_op_type optype);
void idxd_free_desc(struct idxd_wq *wq, struct idxd_desc *desc); void idxd_free_desc(struct idxd_wq *wq, struct idxd_desc *desc);
int idxd_enqcmds(struct idxd_wq *wq, void __iomem *portal, const void *desc);
/* dmaengine */ /* dmaengine */
int idxd_register_dma_device(struct idxd_device *idxd); int idxd_register_dma_device(struct idxd_device *idxd);
......
...@@ -248,6 +248,7 @@ static int idxd_setup_wqs(struct idxd_device *idxd) ...@@ -248,6 +248,7 @@ static int idxd_setup_wqs(struct idxd_device *idxd)
init_completion(&wq->wq_resurrect); init_completion(&wq->wq_resurrect);
wq->max_xfer_bytes = WQ_DEFAULT_MAX_XFER; wq->max_xfer_bytes = WQ_DEFAULT_MAX_XFER;
wq->max_batch_size = WQ_DEFAULT_MAX_BATCH; wq->max_batch_size = WQ_DEFAULT_MAX_BATCH;
wq->enqcmds_retries = IDXD_ENQCMDS_RETRIES;
wq->wqcfg = kzalloc_node(idxd->wqcfg_size, GFP_KERNEL, dev_to_node(dev)); wq->wqcfg = kzalloc_node(idxd->wqcfg_size, GFP_KERNEL, dev_to_node(dev));
if (!wq->wqcfg) { if (!wq->wqcfg) {
put_device(conf_dev); put_device(conf_dev);
......
...@@ -98,7 +98,7 @@ static void idxd_int_handle_revoke_drain(struct idxd_irq_entry *ie) ...@@ -98,7 +98,7 @@ static void idxd_int_handle_revoke_drain(struct idxd_irq_entry *ie)
if (wq_dedicated(wq)) { if (wq_dedicated(wq)) {
iosubmit_cmds512(portal, &desc, 1); iosubmit_cmds512(portal, &desc, 1);
} else { } else {
rc = enqcmds(portal, &desc); rc = idxd_enqcmds(wq, portal, &desc);
/* This should not fail unless hardware failed. */ /* This should not fail unless hardware failed. */
if (rc < 0) if (rc < 0)
dev_warn(dev, "Failed to submit drain desc on wq %d\n", wq->id); dev_warn(dev, "Failed to submit drain desc on wq %d\n", wq->id);
......
...@@ -123,6 +123,29 @@ static void llist_abort_desc(struct idxd_wq *wq, struct idxd_irq_entry *ie, ...@@ -123,6 +123,29 @@ static void llist_abort_desc(struct idxd_wq *wq, struct idxd_irq_entry *ie,
idxd_dma_complete_txd(found, IDXD_COMPLETE_ABORT, false); idxd_dma_complete_txd(found, IDXD_COMPLETE_ABORT, false);
} }
/*
* ENQCMDS typically fail when the WQ is inactive or busy. On host submission, the driver
* has better control of number of descriptors being submitted to a shared wq by limiting
* the number of driver allocated descriptors to the wq size. However, when the swq is
* exported to a guest kernel, it may be shared with multiple guest kernels. This means
* the likelihood of getting busy returned on the swq when submitting goes significantly up.
* Having a tunable retry mechanism allows the driver to keep trying for a bit before giving
* up. The sysfs knob can be tuned by the system administrator.
*/
int idxd_enqcmds(struct idxd_wq *wq, void __iomem *portal, const void *desc)
{
int rc, retries = 0;
do {
rc = enqcmds(portal, desc);
if (rc == 0)
break;
cpu_relax();
} while (retries++ < wq->enqcmds_retries);
return rc;
}
int idxd_submit_desc(struct idxd_wq *wq, struct idxd_desc *desc) int idxd_submit_desc(struct idxd_wq *wq, struct idxd_desc *desc)
{ {
struct idxd_device *idxd = wq->idxd; struct idxd_device *idxd = wq->idxd;
...@@ -166,13 +189,7 @@ int idxd_submit_desc(struct idxd_wq *wq, struct idxd_desc *desc) ...@@ -166,13 +189,7 @@ int idxd_submit_desc(struct idxd_wq *wq, struct idxd_desc *desc)
if (wq_dedicated(wq)) { if (wq_dedicated(wq)) {
iosubmit_cmds512(portal, desc->hw, 1); iosubmit_cmds512(portal, desc->hw, 1);
} else { } else {
/* rc = idxd_enqcmds(wq, portal, desc->hw);
* It's not likely that we would receive queue full rejection
* since the descriptor allocation gates at wq size. If we
* receive a -EAGAIN, that means something went wrong such as the
* device is not accepting descriptor at all.
*/
rc = enqcmds(portal, desc->hw);
if (rc < 0) { if (rc < 0) {
percpu_ref_put(&wq->wq_active); percpu_ref_put(&wq->wq_active);
/* abort operation frees the descriptor */ /* abort operation frees the descriptor */
......
...@@ -945,6 +945,41 @@ static ssize_t wq_occupancy_show(struct device *dev, struct device_attribute *at ...@@ -945,6 +945,41 @@ static ssize_t wq_occupancy_show(struct device *dev, struct device_attribute *at
static struct device_attribute dev_attr_wq_occupancy = static struct device_attribute dev_attr_wq_occupancy =
__ATTR(occupancy, 0444, wq_occupancy_show, NULL); __ATTR(occupancy, 0444, wq_occupancy_show, NULL);
static ssize_t wq_enqcmds_retries_show(struct device *dev,
struct device_attribute *attr, char *buf)
{
struct idxd_wq *wq = confdev_to_wq(dev);
if (wq_dedicated(wq))
return -EOPNOTSUPP;
return sysfs_emit(buf, "%u\n", wq->enqcmds_retries);
}
static ssize_t wq_enqcmds_retries_store(struct device *dev, struct device_attribute *attr,
const char *buf, size_t count)
{
struct idxd_wq *wq = confdev_to_wq(dev);
int rc;
unsigned int retries;
if (wq_dedicated(wq))
return -EOPNOTSUPP;
rc = kstrtouint(buf, 10, &retries);
if (rc < 0)
return rc;
if (retries > IDXD_ENQCMDS_MAX_RETRIES)
retries = IDXD_ENQCMDS_MAX_RETRIES;
wq->enqcmds_retries = retries;
return count;
}
static struct device_attribute dev_attr_wq_enqcmds_retries =
__ATTR(enqcmds_retries, 0644, wq_enqcmds_retries_show, wq_enqcmds_retries_store);
static struct attribute *idxd_wq_attributes[] = { static struct attribute *idxd_wq_attributes[] = {
&dev_attr_wq_clients.attr, &dev_attr_wq_clients.attr,
&dev_attr_wq_state.attr, &dev_attr_wq_state.attr,
...@@ -961,6 +996,7 @@ static struct attribute *idxd_wq_attributes[] = { ...@@ -961,6 +996,7 @@ static struct attribute *idxd_wq_attributes[] = {
&dev_attr_wq_max_batch_size.attr, &dev_attr_wq_max_batch_size.attr,
&dev_attr_wq_ats_disable.attr, &dev_attr_wq_ats_disable.attr,
&dev_attr_wq_occupancy.attr, &dev_attr_wq_occupancy.attr,
&dev_attr_wq_enqcmds_retries.attr,
NULL, NULL,
}; };
......
Markdown is supported
0%
or
You are about to add 0 people to the discussion. Proceed with caution.
Finish editing this message first!
Please register or to comment