Currently it is possible for userspace to see the result of
gettimeofday() going backwards by 1 microsecond, assuming that
userspace is using the gettimeofday() in the VDSO. The VDSO
gettimeofday() algorithm computes the time in "xsecs", which are
units of 2^-20 seconds, or approximately 0.954 microseconds,
using the algorithm
now = (timebase - tb_orig_stamp) * tb_to_xs + stamp_xsec
and then converts the time in xsecs to seconds and microseconds.
The kernel updates the tb_orig_stamp and stamp_xsec values every
tick in update_vsyscall(). If the length of the tick is not an
integer number of xsecs, then some precision is lost in converting
the current time to xsecs. For example, with CONFIG_HZ=1000, the
tick is 1ms long, which is 1048.576 xsecs. That means that
stamp_xsec will advance by either 1048 or 1049 on each tick.
With the right conditions, it is possible for userspace to get
(timebase - tb_orig_stamp) * tb_to_xs being 1049 if the kernel is
slightly late in updating the vdso_datapage, and then for stamp_xsec
to advance by 1048 when the kernel does update it, and for userspace
to then see (timebase - tb_orig_stamp) * tb_to_xs being zero due to
integer truncation. The result is that time appears to go backwards
by 1 microsecond.
To fix this we change the VDSO gettimeofday to use a new field in the
VDSO datapage which stores the nanoseconds part of the time as a
fractional number of seconds in a 0.32 binary fraction format.
(Or put another way, as a 32-bit number in units of 0.23283 ns.)
This is convenient because we can use the mulhwu instruction to
convert it to either microseconds or nanoseconds.
Since it turns out that computing the time of day using this new field
is simpler than either using stamp_xsec (as gettimeofday does) or
stamp_xtime.tv_nsec (as clock_gettime does), this converts both
gettimeofday and clock_gettime to use the new field. The existing
__do_get_tspec function is converted to use the new field and take
a parameter in r7 that indicates the desired resolution, 1,000,000
for microseconds or 1,000,000,000 for nanoseconds. The __do_get_xsec
function is then unused and is deleted.
The new algorithm is
now = ((timebase - tb_orig_stamp) << 12) * tb_to_xs
+ (stamp_xtime_seconds << 32) + stamp_sec_fraction
with 'now' in units of 2^-32 seconds. That is then converted to
seconds and either microseconds or nanoseconds with
seconds = now >> 32
partseconds = ((now & 0xffffffff) * resolution) >> 32
The 32-bit VDSO code also makes a further simplification: it ignores
the bottom 32 bits of the tb_to_xs value, which is a 0.64 format binary
fraction. Doing so gets rid of 4 multiply instructions. Assuming
a timebase frequency of 1GHz or less and an update interval of no
more than 10ms, the upper 32 bits of tb_to_xs will be at least
4503599, so the error from ignoring the low 32 bits will be at most
2.2ns, which is more than an order of magnitude less than the time
taken to do gettimeofday or clock_gettime on our fastest processors,
so there is no possibility of seeing inconsistent values due to this.
This also moves update_gtod() down next to its only caller, and makes
update_vsyscall use the time passed in via the wall_time argument rather
than accessing xtime directly. At present, wall_time always points to
xtime, but that could change in future.
Signed-off-by: Paul Mackerras <paulus@samba.org>
Signed-off-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
Currently the clock_gettime implementation in the VDSO produces a
result with microsecond resolution for the cases that are handled
without a system call, i.e. CLOCK_REALTIME and CLOCK_MONOTONIC. The
nanoseconds field of the result is obtained by computing a
microseconds value and multiplying by 1000.
This changes the code in the VDSO to do the computation for
clock_gettime with nanosecond resolution. That means that the
resolution of the result will ultimately depend on the timebase
frequency.
Because the timestamp in the VDSO datapage (stamp_xsec, the real time
corresponding to the timebase count in tb_orig_stamp) is in units of
2^-20 seconds, it doesn't have sufficient resolution for computing a
result with nanosecond resolution. Therefore this adds a copy of
xtime to the VDSO datapage and updates it in update_gtod() along with
the other time-related fields.
Signed-off-by: Paul Mackerras <paulus@samba.org>
The vdso64 portion of patch 74609f4536 for
fixing problems with NULL gettimeofday input mistakenly checks for a
null tz field twice, when it should be checking for null tz once, and
null tv once; by way of a r10/r11 typo.
Any application calling gettimeofday(&tv,NULL) will "fail".
This corrects that typo, and makes my G5 happy.
Tested on G5.
Signed-off-by: Will Schmidt <will_schmidt@vnet.ibm.com>
Cc: Tony Breeds <tony@bakeyournoodle.com>
Forwarded-by: Ben Herrenschmidt <benh@kernel.crashing.org>
[ Ben says: "I checked the 32 bits part of the change is correct. You
can probably blame me for originally writing the 2 versions with
inversed usage of r10 and r11, thus confusing Tony :-)"
Ben duly blamed. - Linus ]
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
Consider the prototype for gettimeofday():
int gettimofday(struct timeval *tv, struct timezone *tz);
Although it is valid to call with /either/ tv or tz being NULL, and
the C version of sys_gettimeofday() supports this, the current version
of gettimeofday() in the VDSO will SEGV if called with a NULL tv.
This adds a check for tv being NULL so that it doesn't SEGV.
Signed-off-by: Tony Breeds <tony@bakeyournoodle.com>
Acked-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
Signed-off-by: Paul Mackerras <paulus@samba.org>
The Cell CPU timebase has an erratum. When reading the entire 64 bits
of the timebase with one mftb instruction, there is a handful of cycles
window during which one might read a value with the low order 32 bits
already reset to 0x00000000 but the high order bits not yet incremeted
by one. This fixes it by reading the timebase again until the low order
32 bits is no longer 0. That might introduce occasional latencies if
hitting mftb just at the wrong time, but no more than 70ns on a cell
blade, and that was considered acceptable.
Signed-off-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
Acked-by: Olof Johansson <olof@lixom.net>
Signed-off-by: Paul Mackerras <paulus@samba.org>
A bug in the assembly code of the vdso can cause gettimeofday() to hang
or to return incorrect results. The wrong register was used to test for
pending updates of the calibration variables and to create a dependency
for subsequent loads. This fixes it.
Signed-off-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
Signed-off-by: Paul Mackerras <paulus@samba.org>
The vDSO functions should have the same calling convention as a syscall.
Unfortunately, they currently don't set the cr0.so bit which is used to
indicate an error. This patch makes them clear this bit unconditionally
since all functions currently succeed. The syscall fallback done by some
of them will eventually override this if the syscall fails.
This also changes the symbol version of all vdso exports to make sure
glibc can differenciate between old and fixed calls for existing ones
like __kernel_gettimeofday.
Signed-off-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
Signed-off-by: Paul Mackerras <paulus@samba.org>
This fixes various errors in the new functions added in the vDSO's,
I've now verified all functions on both 32 and 64 bits vDSOs. It also
fix a sign extension bug getting the initial time of day at boot that
could cause the monotonic clock value to be completely on bogus for
64 bits applications (with either the vDSO or the syscall) on
powermacs.
Signed-off-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
Signed-off-by: Paul Mackerras <paulus@samba.org>
This patch moves the vdso's to arch/powerpc, adds support for the 32
bits vdso to the 32 bits kernel, rename systemcfg (finally !), and adds
some new (still untested) routines to both vdso's: clock_gettime() with
support for CLOCK_REALTIME and CLOCK_MONOTONIC, clock_getres() (same
clocks) and get_tbfreq() for glibc to retreive the timebase frequency.
Tom,Steve: The implementation of get_tbfreq() I've done for 32 bits
returns a long long (r3, r4) not a long. This is such that if we ever
add support for >4Ghz timebases on ppc32, the userland interface won't
have to change.
I have tested gettimeofday() using some glibc patches in both ppc32 and
ppc64 kernels using 32 bits userland (I haven't had a chance to test a
64 bits userland yet, but the implementation didn't change and was
tested earlier). I haven't tested yet the new functions.
Signed-off-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
Signed-off-by: Paul Mackerras <paulus@samba.org>
Paul Mackerras
Will Schmidt
Tony Breeds
Benjamin Herrenschmidt
Jörn Engel