pwm: core: unsigned or signed ints for pwm_config
Hey Thierry, list
I'm going over the pwm core and notice that in the pwm header, duty_ns and period_ns is internally stored as an unsigned int.
struct pwm_device {
const char *label;
unsigned long flags;
unsigned int hwpwm;
unsigned int pwm;
struct pwm_chip *chip;
void *chip_data;
unsigned int period;
unsigned int duty_cycle;
enum pwm_polarity polarity;
};However, pwm_config takes signed ints
int pwm_config(struct pwm_device *pwm, int duty_ns, int period_ns);
So digging a little deeper in the PWM core, I see that pwm_config dissallows negative ints, so having them unsigned has no benefit (and technically is illegal)
if (!pwm || duty_ns < 0|| period_ns= 0 || duty_ns > period_ns)
return -EINVAL;and because (after the check) we cram the signed int into an unsigned one:
pwm->duty_cycle = duty_ns;
pwm->period = period_ns;This could end up badly when using any unsigned int larger then INT_MAX and thus ending up with a negative duty/period. I haven't checked deeper if this is accounted for later, but would it be worth my time to convert all ints to unsigned ints? Since negative period and duty cycles are really not possible anyway?
Olliver
Quoted 27 lines · Olliver Schinagl
Hey Thierry, list
I'm going over the pwm core and notice that in the pwm header,
duty_nsandperiod_nsis internally stored as an unsigned int.struct pwm_device { const char *label; unsigned long flags; unsigned int hwpwm; unsigned int pwm; struct pwm_chip *chip; void *chip_data; unsigned int period; unsigned int duty_cycle; enum pwm_polarity polarity; };However,
pwm_configtakes signed ints
intpwm_config(struct pwm_device *pwm, int duty_ns, int period_ns);So digging a little deeper in the PWM core, I see that
pwm_configdissallows negative ints, so having them unsigned has no benefit (and technically is illegal)if (!pwm || duty_ns < 0|| period_ns= 0 || duty_ns > period_ns) return -EINVAL;
Quoted 7 lines
and because (after the check) we cram the signed int into an unsigned one:
pwm->duty_cycle = duty_ns; pwm->period = period_ns;This could end up badly when using any unsigned int larger then
INT_MAXand thus ending up with a negative duty/period.
I don't think this is problematic because we're rejecting negative input values and store the non-negative ones in an unsigned int, so we can never store anything that would overflow the internal representation.
Quoted 4 lines
I haven't checked deeper if this
is accounted for later, but would it be worth my time to convert all ints to unsigned ints? Since negative period and duty cycles are really not possible anyway?
The reason for storing them as unsigned internally is precisely because they can never be negative. The reason why pwm_config() has plain ints is historic. It's always been on my TODO list to convert them over to a unsigned variant, but never high priority enough. It's also problematic because doing so needs to modify a public API and hence requires auditing all consumers and providers to make sure nothing breaks.
I'm not sure if it's worth spending this effort now. Boris Brezillon posted patches a few weeks ago to introduce an "atomic" API and that's going to require updating all users anyway. The new API also uses the correct types, so any effort should probably go into testing and migrating to the new API.
Thierry
Hey Thierry,
Quoted 50 lines · Thierry Reding
Quoted 27 lines · Olliver Schinagl
Hey Thierry, list
I'm going over the pwm core and notice that in the pwm header,
duty_nsandperiod_nsis internally stored as an unsigned int.struct pwm_device { const char *label; unsigned long flags; unsigned int hwpwm; unsigned int pwm; struct pwm_chip *chip; void *chip_data; unsigned int period; unsigned int duty_cycle; enum pwm_polarity polarity; };However,
pwm_configtakes signed ints
intpwm_config(struct pwm_device *pwm, int duty_ns, int period_ns);So digging a little deeper in the PWM core, I see that
pwm_configdissallows negative ints, so having them unsigned has no benefit (and technically is illegal)if (!pwm || duty_ns < 0|| period_ns= 0 || duty_ns > period_ns) return -EINVAL;Quoted 7 lines
and because (after the check) we cram the signed int into an unsigned one:
pwm->duty_cycle = duty_ns; pwm->period = period_ns;This could end up badly when using any unsigned int larger then
INT_MAXand thus ending up with a negative duty/period.I don't think this is problematic because we're rejecting negative input values and store the non-negative ones in an unsigned int, so we can never store anything that would overflow the internal representation.
Quoted 4 lines
I haven't checked deeper if this
is accounted for later, but would it be worth my time to convert all ints to unsigned ints? Since negative period and duty cycles are really not possible anyway?The reason for storing them as unsigned internally is precisely because they can never be negative. The reason why
pwm_config()has plain ints is historic. It's always been on my TODO list to convert them over to a unsigned variant, but never high priority enough. It's also problematic because doing so needs to modify a public API and hence requires auditing all consumers and providers to make sure nothing breaks.
Quoted 5 lines
I'm not sure if it's worth spending this effort now. Boris Brezillon posted patches a few weeks ago to introduce an "atomic" API and that's going to require updating all users anyway. The new API also uses the correct types, so any effort should probably go into testing and migrating to the new API.
Thanks for saving me from doing alot of work herin ;)
Are Boris his patches merged in some dev tree of yours? I'm working on some pwm stuff too and would love to work 'with'.
Olliver
Quoted 2 lines
Thierry
Quoted 58 lines · Olliver Schinagl
Hey Thierry,
Quoted 50 lines · Thierry Reding
Quoted 27 lines · Olliver Schinagl
Hey Thierry, list
I'm going over the pwm core and notice that in the pwm header,
duty_nsandperiod_nsis internally stored as an unsigned int.struct pwm_device { const char *label; unsigned long flags; unsigned int hwpwm; unsigned int pwm; struct pwm_chip *chip; void *chip_data; unsigned int period; unsigned int duty_cycle; enum pwm_polarity polarity; };However,
pwm_configtakes signed ints
intpwm_config(struct pwm_device *pwm, int duty_ns, int period_ns);So digging a little deeper in the PWM core, I see that
pwm_configdissallows negative ints, so having them unsigned has no benefit (and technically is illegal)if (!pwm || duty_ns < 0|| period_ns= 0 || duty_ns > period_ns) return -EINVAL;Quoted 7 lines
and because (after the check) we cram the signed int into an unsigned one:
pwm->duty_cycle = duty_ns; pwm->period = period_ns;This could end up badly when using any unsigned int larger then
INT_MAXand thus ending up with a negative duty/period.I don't think this is problematic because we're rejecting negative input values and store the non-negative ones in an unsigned int, so we can never store anything that would overflow the internal representation.
Quoted 4 lines
I haven't checked deeper if this
is accounted for later, but would it be worth my time to convert all ints to unsigned ints? Since negative period and duty cycles are really not possible anyway?The reason for storing them as unsigned internally is precisely because they can never be negative. The reason why
pwm_config()has plain ints is historic. It's always been on my TODO list to convert them over to a unsigned variant, but never high priority enough. It's also problematic because doing so needs to modify a public API and hence requires auditing all consumers and providers to make sure nothing breaks.Quoted 5 lines
I'm not sure if it's worth spending this effort now. Boris Brezillon posted patches a few weeks ago to introduce an "atomic" API and that's going to require updating all users anyway. The new API also uses the correct types, so any effort should probably go into testing and migrating to the new API.
Quoted 4 lines
Thanks for saving me from doing alot of work herin ;)
Are Boris his patches merged in some dev tree of yours? I'm working on some pwm stuff too and would love to work 'with'.
I'm hoping to get around to applying Boris' patches to my for-next branch this week.
Thierry