[RFC PATCH] drivers: hv: Convert open-coded timeouts to msecs_to_jiffies()

Easwar Hariharan posted 1 patch 1 month, 1 week ago
drivers/hv/hv_balloon.c  | 9 +++++----
drivers/hv/hv_kvp.c      | 4 ++--
drivers/hv/hv_snapshot.c | 6 ++++--
drivers/hv/vmbus_drv.c   | 2 +-
4 files changed, 12 insertions(+), 9 deletions(-)
[RFC PATCH] drivers: hv: Convert open-coded timeouts to msecs_to_jiffies()
Posted by Easwar Hariharan 1 month, 1 week ago
We have several places where timeouts are open-coded as N (seconds) * HZ,
but best practice is to use msecs_to_jiffies(). Convert the timeouts to
make them HZ invariant.

Signed-off-by: Easwar Hariharan <eahariha@linux.microsoft.com>
---
 drivers/hv/hv_balloon.c  | 9 +++++----
 drivers/hv/hv_kvp.c      | 4 ++--
 drivers/hv/hv_snapshot.c | 6 ++++--
 drivers/hv/vmbus_drv.c   | 2 +-
 4 files changed, 12 insertions(+), 9 deletions(-)

diff --git a/drivers/hv/hv_balloon.c b/drivers/hv/hv_balloon.c
index c38dcdfcb914d..3017d41f12681 100644
--- a/drivers/hv/hv_balloon.c
+++ b/drivers/hv/hv_balloon.c
@@ -756,7 +756,7 @@ static void hv_mem_hot_add(unsigned long start, unsigned long size,
 		 * adding succeeded, it is ok to proceed even if the memory was
 		 * not onlined in time.
 		 */
-		wait_for_completion_timeout(&dm_device.ol_waitevent, 5 * HZ);
+		wait_for_completion_timeout(&dm_device.ol_waitevent, msecs_to_jiffies(5 * 1000));
 		post_status(&dm_device);
 	}
 }
@@ -1373,7 +1373,8 @@ static int dm_thread_func(void *dm_dev)
 	struct hv_dynmem_device *dm = dm_dev;
 
 	while (!kthread_should_stop()) {
-		wait_for_completion_interruptible_timeout(&dm_device.config_event, 1 * HZ);
+		wait_for_completion_interruptible_timeout(&dm_device.config_event,
+							  msecs_to_jiffies(1 * 1000));
 		/*
 		 * The host expects us to post information on the memory
 		 * pressure every second.
@@ -1748,7 +1749,7 @@ static int balloon_connect_vsp(struct hv_device *dev)
 	if (ret)
 		goto out;
 
-	t = wait_for_completion_timeout(&dm_device.host_event, 5 * HZ);
+	t = wait_for_completion_timeout(&dm_device.host_event, msecs_to_jiffies(5 * 1000));
 	if (t == 0) {
 		ret = -ETIMEDOUT;
 		goto out;
@@ -1806,7 +1807,7 @@ static int balloon_connect_vsp(struct hv_device *dev)
 	if (ret)
 		goto out;
 
-	t = wait_for_completion_timeout(&dm_device.host_event, 5 * HZ);
+	t = wait_for_completion_timeout(&dm_device.host_event, msecs_to_jiffies(5 * 1000));
 	if (t == 0) {
 		ret = -ETIMEDOUT;
 		goto out;
diff --git a/drivers/hv/hv_kvp.c b/drivers/hv/hv_kvp.c
index d35b60c061148..8d9ae1b0e8788 100644
--- a/drivers/hv/hv_kvp.c
+++ b/drivers/hv/hv_kvp.c
@@ -655,7 +655,7 @@ void hv_kvp_onchannelcallback(void *context)
 		if (host_negotiatied == NEGO_NOT_STARTED) {
 			host_negotiatied = NEGO_IN_PROGRESS;
 			schedule_delayed_work(&kvp_host_handshake_work,
-				      HV_UTIL_NEGO_TIMEOUT * HZ);
+						msecs_to_jiffies(HV_UTIL_NEGO_TIMEOUT * 1000));
 		}
 		return;
 	}
@@ -724,7 +724,7 @@ void hv_kvp_onchannelcallback(void *context)
 		 */
 		schedule_work(&kvp_sendkey_work);
 		schedule_delayed_work(&kvp_timeout_work,
-					HV_UTIL_TIMEOUT * HZ);
+				      msecs_to_jiffies(HV_UTIL_TIMEOUT * 1000));
 
 		return;
 
diff --git a/drivers/hv/hv_snapshot.c b/drivers/hv/hv_snapshot.c
index 0d2184be16912..be4955613db35 100644
--- a/drivers/hv/hv_snapshot.c
+++ b/drivers/hv/hv_snapshot.c
@@ -192,8 +192,10 @@ static void vss_send_op(void)
 
 	vss_transaction.state = HVUTIL_USERSPACE_REQ;
 
-	schedule_delayed_work(&vss_timeout_work, op == VSS_OP_FREEZE ?
-			VSS_FREEZE_TIMEOUT * HZ : HV_UTIL_TIMEOUT * HZ);
+	schedule_delayed_work(&vss_timeout_work,
+				op == VSS_OP_FREEZE ?
+				msecs_to_jiffies(VSS_FREEZE_TIMEOUT * 1000) :
+				msecs_to_jiffies(HV_UTIL_TIMEOUT * 1000));
 
 	rc = hvutil_transport_send(hvt, vss_msg, sizeof(*vss_msg), NULL);
 	if (rc) {
diff --git a/drivers/hv/vmbus_drv.c b/drivers/hv/vmbus_drv.c
index 9b15f7daf5059..e35e0a615cb58 100644
--- a/drivers/hv/vmbus_drv.c
+++ b/drivers/hv/vmbus_drv.c
@@ -2507,7 +2507,7 @@ static int vmbus_bus_resume(struct device *dev)
 	vmbus_request_offers();
 
 	if (wait_for_completion_timeout(
-		&vmbus_connection.ready_for_resume_event, 10 * HZ) == 0)
+		&vmbus_connection.ready_for_resume_event, msecs_to_jiffies(10 * 1000)) == 0)
 		pr_err("Some vmbus device is missing after suspending?\n");
 
 	/* Reset the event for the next suspend. */
-- 
2.43.0
Re: [RFC PATCH] drivers: hv: Convert open-coded timeouts to msecs_to_jiffies()
Posted by Naman Jain 1 month, 1 week ago

On 10/17/2024 4:07 AM, Easwar Hariharan wrote:
> We have several places where timeouts are open-coded as N (seconds) * HZ,
> but best practice is to use msecs_to_jiffies(). Convert the timeouts to
> make them HZ invariant.
> 
> Signed-off-by: Easwar Hariharan <eahariha@linux.microsoft.com>
> ---
>   drivers/hv/hv_balloon.c  | 9 +++++----
>   drivers/hv/hv_kvp.c      | 4 ++--
>   drivers/hv/hv_snapshot.c | 6 ++++--
>   drivers/hv/vmbus_drv.c   | 2 +-
>   4 files changed, 12 insertions(+), 9 deletions(-)
> 
> diff --git a/drivers/hv/hv_balloon.c b/drivers/hv/hv_balloon.c
> index c38dcdfcb914d..3017d41f12681 100644
> --- a/drivers/hv/hv_balloon.c

<.>


>   	if (wait_for_completion_timeout(
> -		&vmbus_connection.ready_for_resume_event, 10 * HZ) == 0)
> +		&vmbus_connection.ready_for_resume_event, msecs_to_jiffies(10 * 1000)) == 0)
>   		pr_err("Some vmbus device is missing after suspending?\n");
>   
>   	/* Reset the event for the next suspend. */


Looks good to me. There can be different ways of passing arg to
msecs_to_jiffies though-

for 10 seconds
* 10000
* 10 * 1000
* 10 * MSEC_PER_SEC

I don't have any strong opinion on this, and you can probably choose
whichever feels better.

Even the current implementation with x * HZ works fine, with different
HZ values. But, yes, I agree that using msecs_to_jiffies is better.

Regards,
Naman
Re: [RFC PATCH] drivers: hv: Convert open-coded timeouts to msecs_to_jiffies()
Posted by Praveen Kumar 1 month, 1 week ago
On 17-10-2024 04:07, Easwar Hariharan wrote:
> We have several places where timeouts are open-coded as N (seconds) * HZ,
> but best practice is to use msecs_to_jiffies(). Convert the timeouts to
> make them HZ invariant.
> > Signed-off-by: Easwar Hariharan <eahariha@linux.microsoft.com>
> ---
>  drivers/hv/hv_balloon.c  | 9 +++++----
>  drivers/hv/hv_kvp.c      | 4 ++--
>  drivers/hv/hv_snapshot.c | 6 ++++--
>  drivers/hv/vmbus_drv.c   | 2 +-
>  4 files changed, 12 insertions(+), 9 deletions(-)
> 
> diff --git a/drivers/hv/hv_balloon.c b/drivers/hv/hv_balloon.c
> index c38dcdfcb914d..3017d41f12681 100644
> --- a/drivers/hv/hv_balloon.c
> +++ b/drivers/hv/hv_balloon.c
> @@ -756,7 +756,7 @@ static void hv_mem_hot_add(unsigned long start, unsigned long size,
>  		 * adding succeeded, it is ok to proceed even if the memory was
>  		 * not onlined in time.
>  		 */
> -		wait_for_completion_timeout(&dm_device.ol_waitevent, 5 * HZ);
> +		wait_for_completion_timeout(&dm_device.ol_waitevent, msecs_to_jiffies(5 * 1000));

Is it correct to convert HZ to 1000 ?
Also, how are you testing these changes ?

Regards,

~Praveen.
Re: [RFC PATCH] drivers: hv: Convert open-coded timeouts to msecs_to_jiffies()
Posted by Easwar Hariharan 1 month, 1 week ago
On 10/18/2024 12:54 AM, Praveen Kumar wrote:
> On 17-10-2024 04:07, Easwar Hariharan wrote:
>> We have several places where timeouts are open-coded as N (seconds) * HZ,
>> but best practice is to use msecs_to_jiffies(). Convert the timeouts to
>> make them HZ invariant.
>>> Signed-off-by: Easwar Hariharan <eahariha@linux.microsoft.com>
>> ---
>>  drivers/hv/hv_balloon.c  | 9 +++++----
>>  drivers/hv/hv_kvp.c      | 4 ++--
>>  drivers/hv/hv_snapshot.c | 6 ++++--
>>  drivers/hv/vmbus_drv.c   | 2 +-
>>  4 files changed, 12 insertions(+), 9 deletions(-)
>>
>> diff --git a/drivers/hv/hv_balloon.c b/drivers/hv/hv_balloon.c
>> index c38dcdfcb914d..3017d41f12681 100644
>> --- a/drivers/hv/hv_balloon.c
>> +++ b/drivers/hv/hv_balloon.c
>> @@ -756,7 +756,7 @@ static void hv_mem_hot_add(unsigned long start, unsigned long size,
>>  		 * adding succeeded, it is ok to proceed even if the memory was
>>  		 * not onlined in time.
>>  		 */
>> -		wait_for_completion_timeout(&dm_device.ol_waitevent, 5 * HZ);
>> +		wait_for_completion_timeout(&dm_device.ol_waitevent, msecs_to_jiffies(5 * 1000));
> 
> Is it correct to convert HZ to 1000 ?
> Also, how are you testing these changes ?
> 

It's a conversion of milliseconds to seconds, rather than HZ to 1000. :)
msecs_to_jiffies() handles the conversion to jiffies with HZ. As Naman
mentioned, this could be equivalently written as 5 * MSECS_PER_SEC, and
would probably be more readable. On testing, this is only
compile-tested, and that's part of the reason why it's an RFC, since I'm
not 100% sure every one of these timeouts is measured in seconds. Hoping
for folks more familiar with the code to take a look.

Thanks,
Easwar