On 2020-06-11 01:16, Christian Weisgerber wrote:
Paul Irofti:
This iteration of the diff adds bounds checking for tk_user and moves
the usertc.c stub to every arch in libc as recommanded by deraadt@.
It also fixes a gettimeofday issue reported by cheloha@ and tb@.
Forgot to add armv7 tk_nclock entries. Noticed by benno@, thanks!
One blemish I see is that tk_user is a magic number.
For example, sparc64 will have two timecounters: tick and stick.
They will be assigned magic numbers 1 and 2...
struct timecounter tick_timecounter = {
tick_get_timecount, NULL, ~0u, 0, "tick", 0, NULL, 1
};
struct timecounter stick_timecounter = {
stick_get_timecount, NULL, ~0u, 0, "stick", 1000, NULL, 2
};
... and sparc64 usertc.c will need the corresponding magic array order:
static uint64_t (*get_tc[])(void) =
{
rdtick,
rdstick,
};
I don't know if we want to go through the effort to make this
prettier. We would need an MD header, say, <machine/timetc.h>
that gets picked up by <sys/timetc.h>, with something like
#define TC_TICK 1
#define TC_STICK 2
The symbolic values could then be used in the kernel timecounter
definitions...
struct timecounter tick_timecounter = {
tick_get_timecount, NULL, ~0u, 0, "tick", 0, NULL, TC_TICK
};
struct timecounter stick_timecounter = {
stick_get_timecount, NULL, ~0u, 0, "stick", 1000, NULL, TC_STICK
};
... and in libc usertc.c:
static uint64_t (*get_tc[])(void) =
{
[TC_TICK] = rdtick,
[TC_STICK] = rdstick,
};
...
*tc = (*get_tc[tk_user])();
The cost would be yet another header file per arch.
Thoughts?
I think this not a functional change and it can also be done afterwards
by you or somebody else that has the energy to go through the
bikesheding. I will definitely need a break after this.
