On Tue, 2018-10-16 at 09:54 +-0200, Rafael J. Wysocki wrote: +AD4 drivers/acpi/osl.c +AHw 3 +-+-- +AD4 1 file changed, 2 insertions(+-), 1 deletion(-) +AD4 +AD4 Index: linux-pm/drivers/acpi/osl.c +AD4 +AD0APQA9AD0APQA9AD0APQA9AD0APQA9AD0APQA9AD0APQA9AD0APQA9AD0APQA9AD0APQA9AD0APQA9AD0APQA9AD0APQA9AD0APQA9AD0APQA9AD0APQA9AD0APQA9AD0APQA9AD0APQA9AD0APQA9AD0APQA9AD0APQA9AD0APQA9AD0 +AD4 --- linux-pm.orig/drivers/acpi/osl.c +AD4 +-+-+- linux-pm/drivers/acpi/osl.c +AD4 +AEAAQA -623,7 +-623,8 +AEAAQA void acpi+AF8-os+AF8-stall(u32 us) +AD4 +ACo-/ +AD4 u64 acpi+AF8-os+AF8-get+AF8-timer(void) +AD4 +AHs +AD4 - u64 time+AF8-ns +AD0 ktime+AF8-to+AF8-ns(ktime+AF8-get())+ADs +AD4 +- u64 time+AF8-ns +AD0 ktime+AF8-to+AF8-ns(unlikely(system+AF8-state +AD0APQ SYSTEM+AF8-SUSPEND) ? +AD4 +- ktime+AF8-get+AF8-mono+AF8-fast+AF8-ns() : ktime+AF8-get())+ADs +AD4 do+AF8-div(time+AF8-ns, 100)+ADs +AD4 return time+AF8-ns+ADs +AD4 +AH0 >From the Documentation/core-api/timekeeping.rst section about ktime+AF8-get+AF8-mono+AF8-fast+AF8-ns(): +ACI-most drivers should never call them, since the time is allowed to jump under certain conditions+ACI. How about the following, also untested patch? diff --git a/drivers/acpi/osl.c b/drivers/acpi/osl.c index 8df9abfa947b..2fb7911b3a28 100644 --- a/drivers/acpi/osl.c +-+-+- b/drivers/acpi/osl.c +AEAAQA -623,9 +-623,8 +AEAAQA void acpi+AF8-os+AF8-stall(u32 us) +ACo-/ u64 acpi+AF8-os+AF8-get+AF8-timer(void) +AHs - u64 time+AF8-ns +AD0 ktime+AF8-to+AF8-ns(ktime+AF8-get())+ADs - do+AF8-div(time+AF8-ns, 100)+ADs - return time+AF8-ns+ADs +- return (get+AF8-jiffies+AF8-64() - INITIAL+AF8-JIFFIES) +ACo +- (ACPI+AF8-100NSEC+AF8-PER+AF8-SEC / HZ)+ADs +AH0 acpi+AF8-status acpi+AF8-os+AF8-read+AF8-port(acpi+AF8-io+AF8-address port, u32 +ACo value, u32 width) I think that there are already multiple functions in the kernel tree that subtract the initial value of the jiffies variable from that variable: +ACQ git grep -nH ' - INITIAL+AF8-JIFFIES' arch/x86/kernel/tsc.c:228: return (jiffies+AF8-64 - INITIAL+AF8-JIFFIES) +ACo (1000000000 / HZ)+ADs drivers/net/wan/hdlc+AF8-cisco.c:119: data-+AD4-time +AD0 htonl((jiffies - INITIAL+AF8-JIFFIES) +ACo (1000 / HZ))+ADs kernel/sched/clock.c:65: return (unsigned long long)(jiffies - INITIAL+AF8-JIFFIES) kernel/time/sched+AF8-clock.c:83: return (u64)(jiffies - INITIAL+AF8-JIFFIES)+ADs kernel/trace/trace+AF8-clock.c:72: return jiffies+AF8-64+AF8-to+AF8-clock+AF8-t(jiffies+AF8-64 - INITIAL+AF8-JIFFIES)+ADs net/ipv4/devinet.c:1570: return (cstamp - INITIAL+AF8-JIFFIES) +ACo 100UL / HZ+ADs net/ipv6/addrconf.c:104: return (cstamp - INITIAL+AF8-JIFFIES) +ACo 100UL / HZ+ADs Thanks, Bart.
