2023-08-04 22:03:02

by Rob Clark

Subject: [RFC] PM / QoS: Decouple request alloc from dev_pm_qos_mtx (alternative solution)

From: Rob Clark <[email protected]>

Similar to the previous patch, move the allocation out from under
dev_pm_qos_mtx, by speculatively doing the allocation and handle
any race after acquiring dev_pm_qos_mtx by freeing the redundant
allocation.

Suggested-by: Rafael J. Wysocki <[email protected]>
Signed-off-by: Rob Clark <[email protected]>
---
This is an alternative to https://patchwork.freedesktop.org/patch/551417/?series=115028&rev=4

So, this does _slightly_ change error paths, for ex
dev_pm_qos_update_user_latency_tolerance() will now allocate
dev->power.qos in some error cases. But this seems harmless?
A slightly more complicated version of this could conserve the
previous error path behavior, but I figured I'd try the simpler
thing first.

drivers/base/power/qos.c | 13 +++++++------
1 file changed, 7 insertions(+), 6 deletions(-)

diff --git a/drivers/base/power/qos.c b/drivers/base/power/qos.c
index 1b73a704aac1..c7ba85e89c42 100644
--- a/drivers/base/power/qos.c
+++ b/drivers/base/power/qos.c
@@ -920,8 +920,12 @@ s32 dev_pm_qos_get_user_latency_tolerance(struct device *dev)
int dev_pm_qos_update_user_latency_tolerance(struct device *dev, s32 val)
{
struct dev_pm_qos *qos = dev_pm_qos_constraints_allocate();
+ struct dev_pm_qos_request *req = NULL;
int ret = 0;

+ if (!dev->power.qos->latency_tolerance_req)
+ req = kzalloc(sizeof(*req), GFP_KERNEL);
+
mutex_lock(&dev_pm_qos_mtx);

dev_pm_qos_constraints_set(dev, qos);
@@ -935,8 +939,6 @@ int dev_pm_qos_update_user_latency_tolerance(struct device *dev, s32 val)
goto out;

if (!dev->power.qos->latency_tolerance_req) {
- struct dev_pm_qos_request *req;
-
if (val < 0) {
if (val == PM_QOS_LATENCY_TOLERANCE_NO_CONSTRAINT)
ret = 0;
@@ -944,17 +946,15 @@ int dev_pm_qos_update_user_latency_tolerance(struct device *dev, s32 val)
ret = -EINVAL;
goto out;
}
- req = kzalloc(sizeof(*req), GFP_KERNEL);
if (!req) {
ret = -ENOMEM;
goto out;
}
ret = __dev_pm_qos_add_request(dev, req, DEV_PM_QOS_LATENCY_TOLERANCE, val);
- if (ret < 0) {
- kfree(req);
+ if (ret < 0)
goto out;
- }
dev->power.qos->latency_tolerance_req = req;
+ req = NULL;
} else {
if (val < 0) {
__dev_pm_qos_drop_user_request(dev, DEV_PM_QOS_LATENCY_TOLERANCE);
@@ -966,6 +966,7 @@ int dev_pm_qos_update_user_latency_tolerance(struct device *dev, s32 val)

out:
mutex_unlock(&dev_pm_qos_mtx);
+ kfree(req);
return ret;
}
EXPORT_SYMBOL_GPL(dev_pm_qos_update_user_latency_tolerance);
--
2.41.0

2023-08-07 10:49:08

by Rafael J. Wysocki

Subject: Re: [RFC] PM / QoS: Decouple request alloc from dev_pm_qos_mtx (alternative solution)

On Fri, Aug 4, 2023 at 11:41 PM Rob Clark <[email protected]> wrote:
>
> From: Rob Clark <[email protected]>
>
> Similar to the previous patch, move the allocation out from under
> dev_pm_qos_mtx, by speculatively doing the allocation and handle
> any race after acquiring dev_pm_qos_mtx by freeing the redundant
> allocation.
>
> Suggested-by: Rafael J. Wysocki <[email protected]>
> Signed-off-by: Rob Clark <[email protected]>
> ---
> This is an alternative to https://patchwork.freedesktop.org/patch/551417/?series=115028&rev=4
>
> So, this does _slightly_ change error paths, for ex
> dev_pm_qos_update_user_latency_tolerance() will now allocate
> dev->power.qos in some error cases. But this seems harmless?

It is harmless AFAICS.

> A slightly more complicated version of this could conserve the
> previous error path behavior, but I figured I'd try the simpler
> thing first.

Good choice!

> drivers/base/power/qos.c | 13 +++++++------
> 1 file changed, 7 insertions(+), 6 deletions(-)
>
> diff --git a/drivers/base/power/qos.c b/drivers/base/power/qos.c
> index 1b73a704aac1..c7ba85e89c42 100644
> --- a/drivers/base/power/qos.c
> +++ b/drivers/base/power/qos.c
> @@ -920,8 +920,12 @@ s32 dev_pm_qos_get_user_latency_tolerance(struct device *dev)
> int dev_pm_qos_update_user_latency_tolerance(struct device *dev, s32 val)
> {
> struct dev_pm_qos *qos = dev_pm_qos_constraints_allocate();
> + struct dev_pm_qos_request *req = NULL;
> int ret = 0;
>
> + if (!dev->power.qos->latency_tolerance_req)
> + req = kzalloc(sizeof(*req), GFP_KERNEL);
> +
> mutex_lock(&dev_pm_qos_mtx);
>
> dev_pm_qos_constraints_set(dev, qos);
> @@ -935,8 +939,6 @@ int dev_pm_qos_update_user_latency_tolerance(struct device *dev, s32 val)
> goto out;
>
> if (!dev->power.qos->latency_tolerance_req) {
> - struct dev_pm_qos_request *req;
> -
> if (val < 0) {
> if (val == PM_QOS_LATENCY_TOLERANCE_NO_CONSTRAINT)
> ret = 0;
> @@ -944,17 +946,15 @@ int dev_pm_qos_update_user_latency_tolerance(struct device *dev, s32 val)
> ret = -EINVAL;
> goto out;
> }
> - req = kzalloc(sizeof(*req), GFP_KERNEL);
> if (!req) {
> ret = -ENOMEM;
> goto out;
> }
> ret = __dev_pm_qos_add_request(dev, req, DEV_PM_QOS_LATENCY_TOLERANCE, val);
> - if (ret < 0) {
> - kfree(req);
> + if (ret < 0)
> goto out;
> - }
> dev->power.qos->latency_tolerance_req = req;
> + req = NULL;
> } else {
> if (val < 0) {
> __dev_pm_qos_drop_user_request(dev, DEV_PM_QOS_LATENCY_TOLERANCE);
> @@ -966,6 +966,7 @@ int dev_pm_qos_update_user_latency_tolerance(struct device *dev, s32 val)
>
> out:
> mutex_unlock(&dev_pm_qos_mtx);
> + kfree(req);
> return ret;
> }
> EXPORT_SYMBOL_GPL(dev_pm_qos_update_user_latency_tolerance);
> --

Yes, something like this, please!