Support systems with time64-only syscalls
Subsystems (2)
PATCH 1/1 Support systems with time64-only syscalls
From: Palmer Dabbelt <[email protected]>
Some systems (like 32-bit RISC-V) only have the 64-bit time_t versions of syscalls. So handle the 32-bit time_t version of those being undefined.
---
This seems a little ugly, as it'll blow up when neither is defined. Some #if/#error type stuff seemed uglier, though, and that's the best I could come up with. I figure anyone without either flavor of futex call is probably deep enough in the weeds to just figure what blows up here...
--- kernel/trace/rv/monitors/sleep/sleep.c | 4 ++++ 1 file changed, 4 insertions(+)
kernel/trace/rv/monitors/sleep/sleep.c +4 -0
@@ -127,7 +127,9 @@ static void handle_sys_enter(void *data, struct pt_regs *regs, long id) mon =ltl_get_monitor (current);switch (id) { +#ifdef __NR_clock_nanosleepcase __NR_clock_nanosleep: +#endif #ifdef __NR_clock_nanosleep_time64case __NR_clock_nanosleep_time64:#endif
@@ -138,7 +140,9 @@ static void handle_sys_enter(void *data, struct pt_regs *regs, long id)ltl_atom_update (current,LTL_CLOCK_NANOSLEEP , true);break ; +#ifdef __NR_futexcase __NR_futex: +#endif #ifdef __NR_futex_time64case __NR_futex_time64:#endif
Signature · 1 line
2.50.1
Gabriele Monaco
maintainerQuoted 6 lines · Palmer Dabbelt
From: Palmer Dabbelt <[email protected]>
Some systems (like 32-bit RISC-V) only have the 64-bit
time_tversions of syscalls. So handle the 32-bittime_tversion of those being undefined.
Quoted 8 lines
Fixes f74f8bb246cf ("rv: Add rtapp_sleep monitor")Signed-off-by Palmer Dabbelt <[email protected]>---
This seems a little ugly, as it'll blow up when neither is defined. Some #if/#error type stuff seemed uglier, though, and that's the best I could come up with. I figure anyone without either flavor of futex call is probably deep enough in the weeds to just figure what blows up here...
Yeah, this is getting ugly.. I wasn't fun of this ifdeffery already but a few of them seemed acceptable, if we are really expecting any single one of them to potentially not be available, it isn't looking good.
What about doing in the beginning of the file something like:
/*
* Define dummy syscall numbers for systems not supporting them */
#ifndef __NR_whatever
#define __NR_whatever -1
#endif
#ifndef __NR_some_exotic_syscall
#define __NR_some_exotic_syscall -2
#endif
The negative number would never match, we may add a mostly insignificant overhead checking for it but we keep the function readable. What do you think?
I'm not sure if we can get the compiler rid of it completely, but it's probably not worth it.
What do you think?
Thanks,
Gabriele
Quoted 31 lines
---
kernel/trace/rv/monitors/sleep/sleep.c | 4 ++++1 file changed, 4 insertions(+)kernel/trace/rv/monitors/sleep/sleep.c +0
-0@@ -127,7 +127,9 @@ static void handle_sys_enter(void *data, struct pt_regs *regs, long id) mon = ltl_get_monitor(current);
switch (id) {
+#ifdef__NR_clock_nanosleep
case__NR_clock_nanosleep:
+#endif
#ifdef__NR_clock_nanosleep_time64
case__NR_clock_nanosleep_time64:
#endifkernel/trace/rv/monitors/sleep/sleep.c +0
-0@@ -138,7 +140,9 @@ static void handle_sys_enter(void *data, struct pt_regs *regs, long id) ltl_atom_update(current, LTL_CLOCK_NANOSLEEP, true); break;
+#ifdef__NR_futex
case__NR_futex:
+#endif
#ifdef__NR_futex_time64
case__NR_futex_time64:
#endif
Nam Cao
Acked-byQuoted 30 lines · Gabriele Monaco
Quoted 6 lines · Palmer Dabbelt
From: Palmer Dabbelt <[email protected]>
Some systems (like 32-bit RISC-V) only have the 64-bit
time_tversions of syscalls. So handle the 32-bittime_tversion of those being undefined.Quoted 8 lines
Fixes f74f8bb246cf ("rv: Add rtapp_sleep monitor")Signed-off-by Palmer Dabbelt <[email protected]>---
This seems a little ugly, as it'll blow up when neither is defined. Some #if/#error type stuff seemed uglier, though, and that's the best I could come up with. I figure anyone without either flavor of futex call is probably deep enough in the weeds to just figure what blows up here...Yeah, this is getting ugly.. I wasn't fun of this ifdeffery already but a few of them seemed acceptable, if we are really expecting any single one of them to potentially not be available, it isn't looking good.
What about doing in the beginning of the file something like:
/*
* Define dummy syscall numbers for systems not supporting them */#ifndef
__NR_whatever
#define__NR_whatever-1
#endif
Quoted 7 lines
#ifndef
__NR_some_exotic_syscall
#define__NR_some_exotic_syscall-2
#endifThe negative number would never match, we may add a mostly insignificant overhead checking for it but we keep the function readable. What do you think?
That would be much better, but the patch is fine as is, so I'm okay with merging it:
The patch is just matching the existing ugly #ifdef style (I wonder who even wrote that). So cleaning it up can be done in a follow-up patch.
Nam