The ohci driver uses the get_seconds() function to implement the 32-bit
CSR_BUS_TIME register. This was added in 2010 commit a48777e03ad5
("firewire: add CSR BUS_TIME support").
As get_seconds() returns a 32-bit value (on 32-bit architectures), it
seems like a good fit for that register, but it is also deprecated because
of the y2038/y2106 overflow problem, and should be replaced throughout
the kernel with either ktime_get_real_seconds() or ktime_get_seconds().
I'm using the latter here, which uses monotonic time. This has the
advantage of behaving better during concurrent settimeofday() updates
or leap second adjustments and won't overflow a 32-bit integer, but
the downside of using CLOCK_MONOTONIC instead of CLOCK_REALTIME is
that the observed values are not related to external clocks.
If we instead need UTC but can live with clock jumps or overflows,
then we should use ktime_get_real_seconds() instead, retaining the
existing behavior.
Reviewed-by: Clemens Ladisch <[email protected]>
Signed-off-by: Arnd Bergmann <[email protected]>
---
I notice that Stefan Richter has not been active on the mailing lists
since February 2018.
Andrew, could you pick it up in the meantime?
---
drivers/firewire/ohci.c | 2 +-
1 file changed, 1 insertion(+), 1 deletion(-)
diff --git a/drivers/firewire/ohci.c b/drivers/firewire/ohci.c
index 45c048751f3b..5125841ea338 100644
--- a/drivers/firewire/ohci.c
+++ b/drivers/firewire/ohci.c
@@ -1765,7 +1765,7 @@ static u32 update_bus_time(struct fw_ohci *ohci)
if (unlikely(!ohci->bus_time_running)) {
reg_write(ohci, OHCI1394_IntMaskSet, OHCI1394_cycle64Seconds);
- ohci->bus_time = (lower_32_bits(get_seconds()) & ~0x7f) |
+ ohci->bus_time = (lower_32_bits(ktime_get_seconds()) & ~0x7f) |
(cycle_time_seconds & 0x40);
ohci->bus_time_running = true;
}
--
2.9.0
On Jul 11 Arnd Bergmann wrote:
> The ohci driver uses the get_seconds() function to implement the 32-bit
> CSR_BUS_TIME register. This was added in 2010 commit a48777e03ad5
> ("firewire: add CSR BUS_TIME support").
>
> As get_seconds() returns a 32-bit value (on 32-bit architectures), it
> seems like a good fit for that register, but it is also deprecated because
> of the y2038/y2106 overflow problem, and should be replaced throughout
> the kernel with either ktime_get_real_seconds() or ktime_get_seconds().
>
> I'm using the latter here, which uses monotonic time. This has the
> advantage of behaving better during concurrent settimeofday() updates
> or leap second adjustments and won't overflow a 32-bit integer, but
> the downside of using CLOCK_MONOTONIC instead of CLOCK_REALTIME is
> that the observed values are not related to external clocks.
>
> If we instead need UTC but can live with clock jumps or overflows,
> then we should use ktime_get_real_seconds() instead, retaining the
> existing behavior.
>
> Reviewed-by: Clemens Ladisch <[email protected]>
> Signed-off-by: Arnd Bergmann <[email protected]>
> ---
> I notice that Stefan Richter has not been active on the mailing lists
> since February 2018.
Thanks Arnd and Clemens.
I resurrected and updated one of my FireWire enabled PCs, and try to get
pack to reasonable response times to firewire driver patches.
The switch from CLOCK_REALTIME to CLOCK_MONOTONIC looks good to me, but
I'll have another look at the context.
> Andrew, could you pick it up in the meantime?
> ---
> drivers/firewire/ohci.c | 2 +-
> 1 file changed, 1 insertion(+), 1 deletion(-)
>
> diff --git a/drivers/firewire/ohci.c b/drivers/firewire/ohci.c
> index 45c048751f3b..5125841ea338 100644
> --- a/drivers/firewire/ohci.c
> +++ b/drivers/firewire/ohci.c
> @@ -1765,7 +1765,7 @@ static u32 update_bus_time(struct fw_ohci *ohci)
>
> if (unlikely(!ohci->bus_time_running)) {
> reg_write(ohci, OHCI1394_IntMaskSet, OHCI1394_cycle64Seconds);
> - ohci->bus_time = (lower_32_bits(get_seconds()) & ~0x7f) |
> + ohci->bus_time = (lower_32_bits(ktime_get_seconds()) & ~0x7f) |
> (cycle_time_seconds & 0x40);
> ohci->bus_time_running = true;
> }
--
Stefan Richter
-======---=- -=== =-===
http://arcgraph.de/sr/