| Message ID | 20220505234050.3378-1-jonathan.lemon@gmail.com (mailing list archive) |
|---|---|
| State | Superseded |
| Delegated to: | Netdev Maintainers |
| Headers | show |
| Series | [net] ptp: ocp: have adjtime handle negative delta_ns correctly | expand |
On 06.05.2022 00:40, Jonathan Lemon wrote: > delta_ns is a s64, but it was being passed ptp_ocp_adjtime_coarse > as an u64. Also, it turns out that timespec64_add_ns() only handles > positive values, so the math needs to be updated. > > Fix by passing in the correct signed value, then adding to a > nanosecond version of the timespec. > > Fixes: '90f8f4c0e3ce ("ptp: ocp: Add ptp_ocp_adjtime_coarse for large adjustments")' > Signed-off-by: Jonathan Lemon <jonathan.lemon@gmail.com> > --- > drivers/ptp/ptp_ocp.c | 6 ++++-- > 1 file changed, 4 insertions(+), 2 deletions(-) > > diff --git a/drivers/ptp/ptp_ocp.c b/drivers/ptp/ptp_ocp.c > index dd45471f6780..65e592ec272e 100644 > --- a/drivers/ptp/ptp_ocp.c > +++ b/drivers/ptp/ptp_ocp.c > @@ -841,16 +841,18 @@ __ptp_ocp_adjtime_locked(struct ptp_ocp *bp, u32 adj_val) > } > > static void > -ptp_ocp_adjtime_coarse(struct ptp_ocp *bp, u64 delta_ns) > +ptp_ocp_adjtime_coarse(struct ptp_ocp *bp, s64 delta_ns) > { > struct timespec64 ts; > unsigned long flags; > int err; > + s64 ns; > > spin_lock_irqsave(&bp->lock, flags); > err = __ptp_ocp_gettime_locked(bp, &ts, NULL); > if (likely(!err)) { > - timespec64_add_ns(&ts, delta_ns); > + ns = timespec64_to_ns(&ts) + delta_ns; > + ts = ns_to_timespec64(ns); Maybe use set_normalized_timespec64 instead of this ugly transformations and additional variable? > __ptp_ocp_settime_locked(bp, &ts); > } > spin_unlock_irqrestore(&bp->lock, flags);
On Sat, May 07, 2022 at 01:19:54AM +0100, Vadim Fedorenko wrote: > On 06.05.2022 00:40, Jonathan Lemon wrote: > > delta_ns is a s64, but it was being passed ptp_ocp_adjtime_coarse > > as an u64. Also, it turns out that timespec64_add_ns() only handles > > positive values, so the math needs to be updated. > > > > Fix by passing in the correct signed value, then adding to a > > nanosecond version of the timespec. > > > > Fixes: '90f8f4c0e3ce ("ptp: ocp: Add ptp_ocp_adjtime_coarse for large adjustments")' > > Signed-off-by: Jonathan Lemon <jonathan.lemon@gmail.com> > > --- > > drivers/ptp/ptp_ocp.c | 6 ++++-- > > 1 file changed, 4 insertions(+), 2 deletions(-) > > > > diff --git a/drivers/ptp/ptp_ocp.c b/drivers/ptp/ptp_ocp.c > > index dd45471f6780..65e592ec272e 100644 > > --- a/drivers/ptp/ptp_ocp.c > > +++ b/drivers/ptp/ptp_ocp.c > > @@ -841,16 +841,18 @@ __ptp_ocp_adjtime_locked(struct ptp_ocp *bp, u32 adj_val) > > } > > static void > > -ptp_ocp_adjtime_coarse(struct ptp_ocp *bp, u64 delta_ns) > > +ptp_ocp_adjtime_coarse(struct ptp_ocp *bp, s64 delta_ns) > > { > > struct timespec64 ts; > > unsigned long flags; > > int err; > > + s64 ns; > > spin_lock_irqsave(&bp->lock, flags); > > err = __ptp_ocp_gettime_locked(bp, &ts, NULL); > > if (likely(!err)) { > > - timespec64_add_ns(&ts, delta_ns); > > + ns = timespec64_to_ns(&ts) + delta_ns; > > + ts = ns_to_timespec64(ns); > > Maybe use set_normalized_timespec64 instead of this ugly transformations and > additional variable? I don't see how that would work - set_normalized_timespec64 just sets the ts from a <sec>.<nsec> value. In this case, delta_ns need to be added in to the ts value, but timespec64_add_ns() doeesn't handle negative values, hence the conversion / add / reconversion.
On 08.05.2022 05:55, Jonathan Lemon wrote: > On Sat, May 07, 2022 at 01:19:54AM +0100, Vadim Fedorenko wrote: >> On 06.05.2022 00:40, Jonathan Lemon wrote: >>> delta_ns is a s64, but it was being passed ptp_ocp_adjtime_coarse >>> as an u64. Also, it turns out that timespec64_add_ns() only handles >>> positive values, so the math needs to be updated. >>> >>> Fix by passing in the correct signed value, then adding to a >>> nanosecond version of the timespec. >>> >>> Fixes: '90f8f4c0e3ce ("ptp: ocp: Add ptp_ocp_adjtime_coarse for large adjustments")' >>> Signed-off-by: Jonathan Lemon <jonathan.lemon@gmail.com> >>> --- >>> drivers/ptp/ptp_ocp.c | 6 ++++-- >>> 1 file changed, 4 insertions(+), 2 deletions(-) >>> >>> diff --git a/drivers/ptp/ptp_ocp.c b/drivers/ptp/ptp_ocp.c >>> index dd45471f6780..65e592ec272e 100644 >>> --- a/drivers/ptp/ptp_ocp.c >>> +++ b/drivers/ptp/ptp_ocp.c >>> @@ -841,16 +841,18 @@ __ptp_ocp_adjtime_locked(struct ptp_ocp *bp, u32 adj_val) >>> } >>> static void >>> -ptp_ocp_adjtime_coarse(struct ptp_ocp *bp, u64 delta_ns) >>> +ptp_ocp_adjtime_coarse(struct ptp_ocp *bp, s64 delta_ns) >>> { >>> struct timespec64 ts; >>> unsigned long flags; >>> int err; >>> + s64 ns; >>> spin_lock_irqsave(&bp->lock, flags); >>> err = __ptp_ocp_gettime_locked(bp, &ts, NULL); >>> if (likely(!err)) { >>> - timespec64_add_ns(&ts, delta_ns); >>> + ns = timespec64_to_ns(&ts) + delta_ns; >>> + ts = ns_to_timespec64(ns); >> >> Maybe use set_normalized_timespec64 instead of this ugly transformations and >> additional variable? > > I don't see how that would work - set_normalized_timespec64 just sets > the ts from a <sec>.<nsec> value. In this case, delta_ns need to be > added in to the ts value, but timespec64_add_ns() doeesn't handle > negative values, hence the conversion / add / reconversion. Could be like: - timespec64_add_ns(&ts, delta_ns); + set_normalized_timespec64(&ts, ts.tv_sec, ts.tv_nsec + delta_ns); That's actually without multiplication and division caused by two conversions.
diff --git a/drivers/ptp/ptp_ocp.c b/drivers/ptp/ptp_ocp.c index dd45471f6780..65e592ec272e 100644 --- a/drivers/ptp/ptp_ocp.c +++ b/drivers/ptp/ptp_ocp.c @@ -841,16 +841,18 @@ __ptp_ocp_adjtime_locked(struct ptp_ocp *bp, u32 adj_val) } static void -ptp_ocp_adjtime_coarse(struct ptp_ocp *bp, u64 delta_ns) +ptp_ocp_adjtime_coarse(struct ptp_ocp *bp, s64 delta_ns) { struct timespec64 ts; unsigned long flags; int err; + s64 ns; spin_lock_irqsave(&bp->lock, flags); err = __ptp_ocp_gettime_locked(bp, &ts, NULL); if (likely(!err)) { - timespec64_add_ns(&ts, delta_ns); + ns = timespec64_to_ns(&ts) + delta_ns; + ts = ns_to_timespec64(ns); __ptp_ocp_settime_locked(bp, &ts); } spin_unlock_irqrestore(&bp->lock, flags);
delta_ns is a s64, but it was being passed ptp_ocp_adjtime_coarse as an u64. Also, it turns out that timespec64_add_ns() only handles positive values, so the math needs to be updated. Fix by passing in the correct signed value, then adding to a nanosecond version of the timespec. Fixes: '90f8f4c0e3ce ("ptp: ocp: Add ptp_ocp_adjtime_coarse for large adjustments")' Signed-off-by: Jonathan Lemon <jonathan.lemon@gmail.com> --- drivers/ptp/ptp_ocp.c | 6 ++++-- 1 file changed, 4 insertions(+), 2 deletions(-)