[PATCH] media: Revert cleanup ktime_set() usage

[PATCH] media: Revert cleanup ktime_set() usage

[Jasmin J.]

From: Jasmin Jessich <jasmin@anw.at>

This reverts 8b0e195314fa, because this will not compile for Kernels
older than 4.10.

Cc: Thomas Gleixner <tglx@linutronix.de>
Signed-off-by: Jasmin Jessich <jasmin@anw.at>
---
 drivers/media/dvb-core/dmxdev.c     | 2 +-
 drivers/media/pci/cx88/cx88-input.c | 6 ++++--
 drivers/media/pci/pt3/pt3.c         | 2 +-
 3 files changed, 6 insertions(+), 4 deletions(-)

diff --git a/drivers/media/dvb-core/dmxdev.c b/drivers/media/dvb-core/dmxdev.c
index 61a750f..cb078d6 100644
--- a/drivers/media/dvb-core/dmxdev.c
+++ b/drivers/media/dvb-core/dmxdev.c
@@ -622,7 +622,7 @@ static int dvb_dmxdev_start_feed(struct dmxdev *dmxdev,
 				 struct dmxdev_filter *filter,
 				 struct dmxdev_feed *feed)
 {
-	ktime_t timeout = 0;
+	ktime_t timeout = ktime_set(0, 0);
 	struct dmx_pes_filter_params *para = &filter->params.pes;
 	enum dmx_output otype;
 	int ret;
diff --git a/drivers/media/pci/cx88/cx88-input.c b/drivers/media/pci/cx88/cx88-input.c
index 6f4e692..b13243d 100644
--- a/drivers/media/pci/cx88/cx88-input.c
+++ b/drivers/media/pci/cx88/cx88-input.c
@@ -180,7 +180,8 @@ static enum hrtimer_restart cx88_ir_work(struct hrtimer *timer)
 	struct cx88_IR *ir = container_of(timer, struct cx88_IR, timer);
 
 	cx88_ir_handle_key(ir);
-	missed = hrtimer_forward_now(&ir->timer, ir->polling * 1000000LL);
+	missed = hrtimer_forward_now(&ir->timer,
+				     ktime_set(0, ir->polling * 1000000));
 	if (missed > 1)
 		ir_dprintk("Missed ticks %ld\n", missed - 1);
 
@@ -200,7 +201,8 @@ static int __cx88_ir_start(void *priv)
 	if (ir->polling) {
 		hrtimer_init(&ir->timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL);
 		ir->timer.function = cx88_ir_work;
-		hrtimer_start(&ir->timer, ir->polling * 1000000LL,
+		hrtimer_start(&ir->timer,
+			      ktime_set(0, ir->polling * 1000000),
 			      HRTIMER_MODE_REL);
 	}
 	if (ir->sampling) {
diff --git a/drivers/media/pci/pt3/pt3.c b/drivers/media/pci/pt3/pt3.c
index da74828..8ced807 100644
--- a/drivers/media/pci/pt3/pt3.c
+++ b/drivers/media/pci/pt3/pt3.c
@@ -464,7 +464,7 @@ static int pt3_fetch_thread(void *data)
 
 		pt3_proc_dma(adap);
 
-		delay = PT3_FETCH_DELAY * NSEC_PER_MSEC;
+		delay = ktime_set(0, PT3_FETCH_DELAY * NSEC_PER_MSEC);
 		set_current_state(TASK_UNINTERRUPTIBLE);
 		freezable_schedule_hrtimeout_range(&delay,
 					PT3_FETCH_DELAY_DELTA * NSEC_PER_MSEC,
-- 
2.7.4

Re: [PATCH] media: Revert cleanup ktime_set() usage

[Mauro Carvalho Chehab]

Em Sat, 14 Apr 2018 01:31:34 +0200
"Jasmin J." <jasmin@anw.at> escreveu:

> From: Jasmin Jessich <jasmin@anw.at>
> 
> This reverts 8b0e195314fa, because this will not compile for Kernels
> older than 4.10.

This patch looks fine, but not for the above-mentioned.

The thing is that it is not consistent to have some places with
things like:
	timeout = ktime_set(1, ir->polling * 1000000);

and others with:
	timeout = ir->polling * 1000000;

We should either use ktime_set() everywhere or remove it as a whole.
My preference is to keep using it, as it makes it better documented.
In any case, gcc should be smart enough to discard multiply by a
zero constant when evaluating ktime_set() macro (if not, it is a gcc
issue).

The fact that it makes maintainership of the media_build backport
tree easier is just a plus, but it makes no sense upstream.

> 
> Cc: Thomas Gleixner <tglx@linutronix.de>
> Signed-off-by: Jasmin Jessich <jasmin@anw.at>
> ---
>  drivers/media/dvb-core/dmxdev.c     | 2 +-
>  drivers/media/pci/cx88/cx88-input.c | 6 ++++--
>  drivers/media/pci/pt3/pt3.c         | 2 +-
>  3 files changed, 6 insertions(+), 4 deletions(-)
> 
> diff --git a/drivers/media/dvb-core/dmxdev.c b/drivers/media/dvb-core/dmxdev.c
> index 61a750f..cb078d6 100644
> --- a/drivers/media/dvb-core/dmxdev.c
> +++ b/drivers/media/dvb-core/dmxdev.c
> @@ -622,7 +622,7 @@ static int dvb_dmxdev_start_feed(struct dmxdev *dmxdev,
>  				 struct dmxdev_filter *filter,
>  				 struct dmxdev_feed *feed)
>  {
> -	ktime_t timeout = 0;
> +	ktime_t timeout = ktime_set(0, 0);
>  	struct dmx_pes_filter_params *para = &filter->params.pes;
>  	enum dmx_output otype;
>  	int ret;
> diff --git a/drivers/media/pci/cx88/cx88-input.c b/drivers/media/pci/cx88/cx88-input.c
> index 6f4e692..b13243d 100644
> --- a/drivers/media/pci/cx88/cx88-input.c
> +++ b/drivers/media/pci/cx88/cx88-input.c
> @@ -180,7 +180,8 @@ static enum hrtimer_restart cx88_ir_work(struct hrtimer *timer)
>  	struct cx88_IR *ir = container_of(timer, struct cx88_IR, timer);
>  
>  	cx88_ir_handle_key(ir);
> -	missed = hrtimer_forward_now(&ir->timer, ir->polling * 1000000LL);
> +	missed = hrtimer_forward_now(&ir->timer,
> +				     ktime_set(0, ir->polling * 1000000));
>  	if (missed > 1)
>  		ir_dprintk("Missed ticks %ld\n", missed - 1);
>  
> @@ -200,7 +201,8 @@ static int __cx88_ir_start(void *priv)
>  	if (ir->polling) {
>  		hrtimer_init(&ir->timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL);
>  		ir->timer.function = cx88_ir_work;
> -		hrtimer_start(&ir->timer, ir->polling * 1000000LL,
> +		hrtimer_start(&ir->timer,
> +			      ktime_set(0, ir->polling * 1000000),
>  			      HRTIMER_MODE_REL);
>  	}
>  	if (ir->sampling) {
> diff --git a/drivers/media/pci/pt3/pt3.c b/drivers/media/pci/pt3/pt3.c
> index da74828..8ced807 100644
> --- a/drivers/media/pci/pt3/pt3.c
> +++ b/drivers/media/pci/pt3/pt3.c
> @@ -464,7 +464,7 @@ static int pt3_fetch_thread(void *data)
>  
>  		pt3_proc_dma(adap);
>  
> -		delay = PT3_FETCH_DELAY * NSEC_PER_MSEC;
> +		delay = ktime_set(0, PT3_FETCH_DELAY * NSEC_PER_MSEC);
>  		set_current_state(TASK_UNINTERRUPTIBLE);
>  		freezable_schedule_hrtimeout_range(&delay,
>  					PT3_FETCH_DELAY_DELTA * NSEC_PER_MSEC,



Thanks,
Mauro

Re: [PATCH] media: Revert cleanup ktime_set() usage

[Jasmin J.]

Hello Mauro!

> This patch looks fine, but not for the above-mentioned.
So I shall reword the commit message?

> The thing is that it is not consistent to have some places with
> things like:
> 	timeout = ktime_set(1, ir->polling * 1000000);
> 
> and others with:
> 	timeout = ir->polling * 1000000;
With my patch applied I can find ONLY two "ir->polling * 1000000" and both are used
together with ktime_set.

$ grep -r "ir->polling" *
media/i2c/ir-kbd-i2c.c:	schedule_delayed_work(&ir->work, msecs_to_jiffies(ir->polling_interval));
media/i2c/ir-kbd-i2c.c:	ir->polling_interval = DEFAULT_POLLING_INTERVAL;
media/i2c/ir-kbd-i2c.c:			ir->polling_interval = init_data->polling_interval;
media/usb/em28xx/em28xx-input.c:	schedule_delayed_work(&ir->work, msecs_to_jiffies(ir->polling));
media/usb/em28xx/em28xx-input.c:	ir->polling = 100; /* ms */
media/usb/em28xx/em28xx-input.c:		schedule_delayed_work(&ir->work, msecs_to_jiffies(ir->polling));
media/usb/tm6000/tm6000-input.c:	if (!ir->polling)
media/usb/tm6000/tm6000-input.c:	schedule_delayed_work(&ir->work, msecs_to_jiffies(ir->polling));
media/usb/tm6000/tm6000-input.c:		ir->polling = 50;
media/usb/tm6000/tm6000-input.c:	if (!ir->polling)
media/usb/au0828/au0828-input.c:	schedule_delayed_work(&ir->work, msecs_to_jiffies(ir->polling));
media/usb/au0828/au0828-input.c:	schedule_delayed_work(&ir->work, msecs_to_jiffies(ir->polling));
media/usb/au0828/au0828-input.c:	ir->polling = 100; /* ms */
media/usb/au0828/au0828-input.c:	schedule_delayed_work(&ir->work, msecs_to_jiffies(ir->polling));
media/pci/saa7134/saa7134-input.c:	if (ir->polling) {
media/pci/saa7134/saa7134-input.c:	if (ir->polling) {
media/pci/saa7134/saa7134-input.c:	if (!ir->polling && !ir->raw_decode) {
media/pci/saa7134/saa7134-input.c:	mod_timer(&ir->timer, jiffies + msecs_to_jiffies(ir->polling));
media/pci/saa7134/saa7134-input.c:	if (ir->polling) {
media/pci/saa7134/saa7134-input.c:	if (ir->polling)
media/pci/saa7134/saa7134-input.c:	ir->polling      = polling;
media/pci/cx88/cx88-input.c:	if (ir->polling) {
media/pci/cx88/cx88-input.c:		   ir->polling ? "poll" : "irq",
media/pci/cx88/cx88-input.c:				     ktime_set(0, ir->polling * 1000000));
media/pci/cx88/cx88-input.c:	if (ir->polling) {
media/pci/cx88/cx88-input.c:			      ktime_set(0, ir->polling * 1000000),
media/pci/cx88/cx88-input.c:	if (ir->polling)
media/pci/cx88/cx88-input.c:		ir->polling = 50; /* ms */
media/pci/cx88/cx88-input.c:		ir->polling = 50; /* ms */
media/pci/cx88/cx88-input.c:		ir->polling = 1; /* ms */
media/pci/cx88/cx88-input.c:		ir->polling = 5; /* ms */
media/pci/cx88/cx88-input.c:		ir->polling = 10; /* ms */
media/pci/cx88/cx88-input.c:		ir->polling = 1; /* ms */
media/pci/cx88/cx88-input.c:		ir->polling = 1; /* ms */
media/pci/cx88/cx88-input.c:		ir->polling = 50; /* ms */
media/pci/cx88/cx88-input.c:		ir->polling = 1; /* ms */
media/pci/cx88/cx88-input.c:		ir->polling      = 50; /* ms */
media/pci/cx88/cx88-input.c:		ir->polling      = 50; /* ms */
media/pci/cx88/cx88-input.c:		ir->polling      = 50; /* ms */
media/pci/cx88/cx88-input.c:		ir->polling      = 100; /* ms */
media/pci/bt8xx/bttv-input.c:	if (ir->polling) {
media/pci/bt8xx/bttv-input.c:		ir->polling               ? "poll"  : "irq",
media/pci/bt8xx/bttv-input.c:	else if (!ir->polling)
media/pci/bt8xx/bttv-input.c:	mod_timer(&ir->timer, jiffies + msecs_to_jiffies(ir->polling));
media/pci/bt8xx/bttv-input.c:	if (ir->polling) {
media/pci/bt8xx/bttv-input.c:		ir->polling      = 50; // ms
media/pci/bt8xx/bttv-input.c:		ir->polling      = 50; // ms
media/pci/bt8xx/bttv-input.c:		ir->polling      = 50; // ms
media/pci/bt8xx/bttv-input.c:		ir->polling      = 50; // ms
media/pci/bt8xx/bttv-input.c:		ir->polling      = 50; // ms
media/pci/bt8xx/bttv-input.c:		ir->polling      = 50; // ms
media/pci/bt8xx/bttv-input.c:		ir->polling      = 50; /* ms */
media/pci/bt8xx/bttv-input.c:		ir->polling      = 50; /* ms */
media/pci/bt8xx/bttv-input.c:		ir->polling      = 1; /* ms */

Currently I am using the exact same patch in media-build. media-build would
not be able to compile without error for all Kernels, if I would have
overlook a setting of ktime_t without ktime_set.
Please point me where I can find the "timeout = ir->polling * 1000000;"
you mentioned.

> My preference is to keep using it, as it makes it better documented.
This is also my preference.

> The fact that it makes maintainership of the media_build backport
> tree easier is just a plus, but it makes no sense upstream.
We fixed all those issues in the past in media-tree. All drivers which
are able (enabled) to be compiled for Kernels older than 4.10 are using
ktime_set.
When you mean we shall change all drivers to use ktime_set (even the ones
which are currently not compiled for <4.10), then I can try to find those
and add it to the patch.

Even the mentioned 8b0e195314fa commit did not fix all existing ktime_set(0, 0):
$ grep -r ktime_set | grep "0, 0"
drivers/scsi/ufs/ufshcd.c:	hba->lrb[task_tag].compl_time_stamp = ktime_set(0, 0);
drivers/scsi/ufs/ufshcd.c:	hba->ufs_stats.last_hibern8_exit_tstamp = ktime_set(0, 0);
drivers/scsi/ufs/ufshcd.c:		hba->ufs_stats.last_hibern8_exit_tstamp = ktime_set(0, 0);
drivers/scsi/ufs/ufshcd.c:	hba->ufs_stats.last_hibern8_exit_tstamp = ktime_set(0, 0);
drivers/ntb/test/ntb_perf.c:		pthr->duration = ktime_set(0, 0);
drivers/net/can/usb/peak_usb/pcan_usb_core.c:		time_ref->tv_host = ktime_set(0, 0);
drivers/thermal/thermal_sysfs.c:		stats->time_in_state[i] = ktime_set(0, 0);
drivers/dma-buf/sync_file.c:		ktime_set(0, 0);
drivers/staging/greybus/loopback.c:	gb->ts = ktime_set(0, 0);
arch/mips/kvm/emulate.c:	ktime_t now = ktime_set(0, 0); /* silence bogus GCC warning */

BR,
   Jasmin

Re: [PATCH] media: Revert cleanup ktime_set() usage

[Mauro Carvalho Chehab]

Em Fri, 27 Apr 2018 20:15:00 +0200
"Jasmin J." <jasmin@anw.at> escreveu:

> Hello Mauro!
> 
> > This patch looks fine, but not for the above-mentioned.  
> So I shall reword the commit message?

Yes.
> 
> > The thing is that it is not consistent to have some places with
> > things like:
> > 	timeout = ktime_set(1, ir->polling * 1000000);
> > 
> > and others with:
> > 	timeout = ir->polling * 1000000;  
> With my patch applied I can find ONLY two "ir->polling * 1000000" and both are used
> together with ktime_set.
> 
> $ grep -r "ir->polling" *
> media/i2c/ir-kbd-i2c.c:	schedule_delayed_work(&ir->work, msecs_to_jiffies(ir->polling_interval));
> media/i2c/ir-kbd-i2c.c:	ir->polling_interval = DEFAULT_POLLING_INTERVAL;
> media/i2c/ir-kbd-i2c.c:			ir->polling_interval = init_data->polling_interval;
> media/usb/em28xx/em28xx-input.c:	schedule_delayed_work(&ir->work, msecs_to_jiffies(ir->polling));
> media/usb/em28xx/em28xx-input.c:	ir->polling = 100; /* ms */
> media/usb/em28xx/em28xx-input.c:		schedule_delayed_work(&ir->work, msecs_to_jiffies(ir->polling));
> media/usb/tm6000/tm6000-input.c:	if (!ir->polling)
> media/usb/tm6000/tm6000-input.c:	schedule_delayed_work(&ir->work, msecs_to_jiffies(ir->polling));
> media/usb/tm6000/tm6000-input.c:		ir->polling = 50;
> media/usb/tm6000/tm6000-input.c:	if (!ir->polling)
> media/usb/au0828/au0828-input.c:	schedule_delayed_work(&ir->work, msecs_to_jiffies(ir->polling));
> media/usb/au0828/au0828-input.c:	schedule_delayed_work(&ir->work, msecs_to_jiffies(ir->polling));
> media/usb/au0828/au0828-input.c:	ir->polling = 100; /* ms */
> media/usb/au0828/au0828-input.c:	schedule_delayed_work(&ir->work, msecs_to_jiffies(ir->polling));
> media/pci/saa7134/saa7134-input.c:	if (ir->polling) {
> media/pci/saa7134/saa7134-input.c:	if (ir->polling) {
> media/pci/saa7134/saa7134-input.c:	if (!ir->polling && !ir->raw_decode) {
> media/pci/saa7134/saa7134-input.c:	mod_timer(&ir->timer, jiffies + msecs_to_jiffies(ir->polling));
> media/pci/saa7134/saa7134-input.c:	if (ir->polling) {
> media/pci/saa7134/saa7134-input.c:	if (ir->polling)
> media/pci/saa7134/saa7134-input.c:	ir->polling      = polling;
> media/pci/cx88/cx88-input.c:	if (ir->polling) {
> media/pci/cx88/cx88-input.c:		   ir->polling ? "poll" : "irq",
> media/pci/cx88/cx88-input.c:				     ktime_set(0, ir->polling * 1000000));
> media/pci/cx88/cx88-input.c:	if (ir->polling) {
> media/pci/cx88/cx88-input.c:			      ktime_set(0, ir->polling * 1000000),
> media/pci/cx88/cx88-input.c:	if (ir->polling)
> media/pci/cx88/cx88-input.c:		ir->polling = 50; /* ms */
> media/pci/cx88/cx88-input.c:		ir->polling = 50; /* ms */
> media/pci/cx88/cx88-input.c:		ir->polling = 1; /* ms */
> media/pci/cx88/cx88-input.c:		ir->polling = 5; /* ms */
> media/pci/cx88/cx88-input.c:		ir->polling = 10; /* ms */
> media/pci/cx88/cx88-input.c:		ir->polling = 1; /* ms */
> media/pci/cx88/cx88-input.c:		ir->polling = 1; /* ms */
> media/pci/cx88/cx88-input.c:		ir->polling = 50; /* ms */
> media/pci/cx88/cx88-input.c:		ir->polling = 1; /* ms */
> media/pci/cx88/cx88-input.c:		ir->polling      = 50; /* ms */
> media/pci/cx88/cx88-input.c:		ir->polling      = 50; /* ms */
> media/pci/cx88/cx88-input.c:		ir->polling      = 50; /* ms */
> media/pci/cx88/cx88-input.c:		ir->polling      = 100; /* ms */
> media/pci/bt8xx/bttv-input.c:	if (ir->polling) {
> media/pci/bt8xx/bttv-input.c:		ir->polling               ? "poll"  : "irq",
> media/pci/bt8xx/bttv-input.c:	else if (!ir->polling)
> media/pci/bt8xx/bttv-input.c:	mod_timer(&ir->timer, jiffies + msecs_to_jiffies(ir->polling));
> media/pci/bt8xx/bttv-input.c:	if (ir->polling) {
> media/pci/bt8xx/bttv-input.c:		ir->polling      = 50; // ms
> media/pci/bt8xx/bttv-input.c:		ir->polling      = 50; // ms
> media/pci/bt8xx/bttv-input.c:		ir->polling      = 50; // ms
> media/pci/bt8xx/bttv-input.c:		ir->polling      = 50; // ms
> media/pci/bt8xx/bttv-input.c:		ir->polling      = 50; // ms
> media/pci/bt8xx/bttv-input.c:		ir->polling      = 50; // ms
> media/pci/bt8xx/bttv-input.c:		ir->polling      = 50; /* ms */
> media/pci/bt8xx/bttv-input.c:		ir->polling      = 50; /* ms */
> media/pci/bt8xx/bttv-input.c:		ir->polling      = 1; /* ms */
> 
> Currently I am using the exact same patch in media-build. media-build would
> not be able to compile without error for all Kernels, if I would have
> overlook a setting of ktime_t without ktime_set.
> Please point me where I can find the "timeout = ir->polling * 1000000;"
> you mentioned.

I don't really care on how many places it is used. The point is that,
at least on media, we should either call ktime_set() everywhere or
don't call it at all. I think that using the macro makes it better 
documented.

> > My preference is to keep using it, as it makes it better documented.  
> This is also my preference.
> 
> > The fact that it makes maintainership of the media_build backport
> > tree easier is just a plus, but it makes no sense upstream.  
> We fixed all those issues in the past in media-tree. All drivers which
> are able (enabled) to be compiled for Kernels older than 4.10 are using
> ktime_set.

That is irrelevant for upstream. We should confine backport-only
fixes at media_build, as we won't be preventing kAPI changes due
to backports. I mean: if some day ktime_set() gets replaced with
something else, we'll do such change upstream no matter how much
efforts it would mean for backports.

Yet, in this specific case, ktime_set() is not being removed,
and, IMO, mixing it with direct assigns makes the API very
confusing.

> When you mean we shall change all drivers to use ktime_set (even the ones
> which are currently not compiled for <4.10), then I can try to find those
> and add it to the patch.
> 
> Even the mentioned 8b0e195314fa commit did not fix all existing ktime_set(0, 0):
> $ grep -r ktime_set | grep "0, 0"
> drivers/scsi/ufs/ufshcd.c:	hba->lrb[task_tag].compl_time_stamp = ktime_set(0, 0);
> drivers/scsi/ufs/ufshcd.c:	hba->ufs_stats.last_hibern8_exit_tstamp = ktime_set(0, 0);
> drivers/scsi/ufs/ufshcd.c:		hba->ufs_stats.last_hibern8_exit_tstamp = ktime_set(0, 0);
> drivers/scsi/ufs/ufshcd.c:	hba->ufs_stats.last_hibern8_exit_tstamp = ktime_set(0, 0);
> drivers/ntb/test/ntb_perf.c:		pthr->duration = ktime_set(0, 0);
> drivers/net/can/usb/peak_usb/pcan_usb_core.c:		time_ref->tv_host = ktime_set(0, 0);
> drivers/thermal/thermal_sysfs.c:		stats->time_in_state[i] = ktime_set(0, 0);
> drivers/dma-buf/sync_file.c:		ktime_set(0, 0);
> drivers/staging/greybus/loopback.c:	gb->ts = ktime_set(0, 0);
> arch/mips/kvm/emulate.c:	ktime_t now = ktime_set(0, 0); /* silence bogus GCC warning */

I mean all "media drivers". We should stick with either:
	ktime_t now = 10;
or
	ktime_t now = ktime_set(0, 10);
	
everywhere along the subsystem.

It is up to other subsystem maintainers to decide what works best
for them. Nothing prevents you to contact other subsystems and submit
patches to make it more consistent along the subsystem. On such
case, there's no need to c/c linux-media (or me).

> 
> BR,
>    Jasmin



Thanks,
Mauro