linux/arch/powerpc/kernel/vdso64/gettimeofday.S

290 lines
6.7 KiB
ArmAsm
Raw Normal View History

/* SPDX-License-Identifier: GPL-2.0-or-later */
/*
* Userland implementation of gettimeofday() for 64 bits processes in a
* ppc64 kernel for use in the vDSO
*
* Copyright (C) 2004 Benjamin Herrenschmuidt (benh@kernel.crashing.org),
* IBM Corp.
*/
#include <asm/processor.h>
#include <asm/ppc_asm.h>
#include <asm/vdso.h>
#include <asm/asm-offsets.h>
#include <asm/unistd.h>
.text
/*
* Exact prototype of gettimeofday
*
* int __kernel_gettimeofday(struct timeval *tv, struct timezone *tz);
*
*/
V_FUNCTION_BEGIN(__kernel_gettimeofday)
.cfi_startproc
mflr r12
.cfi_register lr,r12
mr r11,r3 /* r11 holds tv */
mr r10,r4 /* r10 holds tz */
bl V_LOCAL_FUNC(__get_datapage) /* get data page */
cmpldi r11,0 /* check if tv is NULL */
beq 2f
powerpc: Rework VDSO gettimeofday to prevent time going backwards 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>
2010-06-20 19:03:08 +00:00
lis r7,1000000@ha /* load up USEC_PER_SEC */
addi r7,r7,1000000@l
bl V_LOCAL_FUNC(__do_get_tspec) /* get sec/us from tb & kernel */
std r4,TVAL64_TV_SEC(r11) /* store sec in tv */
std r5,TVAL64_TV_USEC(r11) /* store usec in tv */
2: cmpldi r10,0 /* check if tz is NULL */
beq 1f
lwz r4,CFG_TZ_MINUTEWEST(r3)/* fill tz */
lwz r5,CFG_TZ_DSTTIME(r3)
stw r4,TZONE_TZ_MINWEST(r10)
stw r5,TZONE_TZ_DSTTIME(r10)
1: mtlr r12
crclr cr0*4+so
li r3,0 /* always success */
blr
.cfi_endproc
V_FUNCTION_END(__kernel_gettimeofday)
/*
* Exact prototype of clock_gettime()
*
* int __kernel_clock_gettime(clockid_t clock_id, struct timespec *tp);
*
*/
V_FUNCTION_BEGIN(__kernel_clock_gettime)
.cfi_startproc
/* Check for supported clock IDs */
cmpwi cr0,r3,CLOCK_REALTIME
cmpwi cr1,r3,CLOCK_MONOTONIC
cror cr0*4+eq,cr0*4+eq,cr1*4+eq
cmpwi cr5,r3,CLOCK_REALTIME_COARSE
cmpwi cr6,r3,CLOCK_MONOTONIC_COARSE
cror cr5*4+eq,cr5*4+eq,cr6*4+eq
cror cr0*4+eq,cr0*4+eq,cr5*4+eq
bne cr0,99f
mflr r12 /* r12 saves lr */
.cfi_register lr,r12
mr r11,r4 /* r11 saves tp */
bl V_LOCAL_FUNC(__get_datapage) /* get data page */
powerpc: Rework VDSO gettimeofday to prevent time going backwards 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>
2010-06-20 19:03:08 +00:00
lis r7,NSEC_PER_SEC@h /* want nanoseconds */
ori r7,r7,NSEC_PER_SEC@l
beq cr5,70f
50: bl V_LOCAL_FUNC(__do_get_tspec) /* get time from tb & kernel */
bne cr1,80f /* if not monotonic, all done */
/*
* CLOCK_MONOTONIC
*/
/* now we must fixup using wall to monotonic. We need to snapshot
* that value and do the counter trick again. Fortunately, we still
* have the counter value in r8 that was returned by __do_get_tspec.
* At this point, r4,r5 contain our sec/nsec values.
*/
powerpc/vdso64: Fix CLOCK_MONOTONIC inconsistencies across Y2038 Jakub Drnec reported: Setting the realtime clock can sometimes make the monotonic clock go back by over a hundred years. Decreasing the realtime clock across the y2k38 threshold is one reliable way to reproduce. Allegedly this can also happen just by running ntpd, I have not managed to reproduce that other than booting with rtc at >2038 and then running ntp. When this happens, anything with timers (e.g. openjdk) breaks rather badly. And included a test case (slightly edited for brevity): #define _POSIX_C_SOURCE 199309L #include <stdio.h> #include <time.h> #include <stdlib.h> #include <unistd.h> long get_time(void) { struct timespec tp; clock_gettime(CLOCK_MONOTONIC, &tp); return tp.tv_sec + tp.tv_nsec / 1000000000; } int main(void) { long last = get_time(); while(1) { long now = get_time(); if (now < last) { printf("clock went backwards by %ld seconds!\n", last - now); } last = now; sleep(1); } return 0; } Which when run concurrently with: # date -s 2040-1-1 # date -s 2037-1-1 Will detect the clock going backward. The root cause is that wtom_clock_sec in struct vdso_data is only a 32-bit signed value, even though we set its value to be equal to tk->wall_to_monotonic.tv_sec which is 64-bits. Because the monotonic clock starts at zero when the system boots the wall_to_montonic.tv_sec offset is negative for current and future dates. Currently on a freshly booted system the offset will be in the vicinity of negative 1.5 billion seconds. However if the wall clock is set past the Y2038 boundary, the offset from wall to monotonic becomes less than negative 2^31, and no longer fits in 32-bits. When that value is assigned to wtom_clock_sec it is truncated and becomes positive, causing the VDSO assembly code to calculate CLOCK_MONOTONIC incorrectly. That causes CLOCK_MONOTONIC to jump ahead by ~4 billion seconds which it is not meant to do. Worse, if the time is then set back before the Y2038 boundary CLOCK_MONOTONIC will jump backward. We can fix it simply by storing the full 64-bit offset in the vdso_data, and using that in the VDSO assembly code. We also shuffle some of the fields in vdso_data to avoid creating a hole. The original commit that added the CLOCK_MONOTONIC support to the VDSO did actually use a 64-bit value for wtom_clock_sec, see commit a7f290dad32e ("[PATCH] powerpc: Merge vdso's and add vdso support to 32 bits kernel") (Nov 2005). However just 3 days later it was converted to 32-bits in commit 0c37ec2aa88b ("[PATCH] powerpc: vdso fixes (take #2)"), and the bug has existed since then AFAICS. Fixes: 0c37ec2aa88b ("[PATCH] powerpc: vdso fixes (take #2)") Cc: stable@vger.kernel.org # v2.6.15+ Link: http://lkml.kernel.org/r/HaC.ZfES.62bwlnvAvMP.1STMMj@seznam.cz Reported-by: Jakub Drnec <jaydee@email.cz> Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
2019-03-13 13:14:38 +00:00
ld r6,WTOM_CLOCK_SEC(r3)
lwa r9,WTOM_CLOCK_NSEC(r3)
/* We now have our result in r6,r9. We create a fake dependency
* on that result and re-check the counter
*/
or r0,r6,r9
xor r0,r0,r0
add r3,r3,r0
ld r0,CFG_TB_UPDATE_COUNT(r3)
cmpld cr0,r0,r8 /* check if updated */
bne- 50b
b 78f
/*
* For coarse clocks we get data directly from the vdso data page, so
* we don't need to call __do_get_tspec, but we still need to do the
* counter trick.
*/
70: ld r8,CFG_TB_UPDATE_COUNT(r3)
andi. r0,r8,1 /* pending update ? loop */
bne- 70b
add r3,r3,r0 /* r0 is already 0 */
/*
* CLOCK_REALTIME_COARSE, below values are needed for MONOTONIC_COARSE
* too
*/
ld r4,STAMP_XTIME_SEC(r3)
ld r5,STAMP_XTIME_NSEC(r3)
bne cr6,75f
/* CLOCK_MONOTONIC_COARSE */
powerpc/vdso64: Fix CLOCK_MONOTONIC inconsistencies across Y2038 Jakub Drnec reported: Setting the realtime clock can sometimes make the monotonic clock go back by over a hundred years. Decreasing the realtime clock across the y2k38 threshold is one reliable way to reproduce. Allegedly this can also happen just by running ntpd, I have not managed to reproduce that other than booting with rtc at >2038 and then running ntp. When this happens, anything with timers (e.g. openjdk) breaks rather badly. And included a test case (slightly edited for brevity): #define _POSIX_C_SOURCE 199309L #include <stdio.h> #include <time.h> #include <stdlib.h> #include <unistd.h> long get_time(void) { struct timespec tp; clock_gettime(CLOCK_MONOTONIC, &tp); return tp.tv_sec + tp.tv_nsec / 1000000000; } int main(void) { long last = get_time(); while(1) { long now = get_time(); if (now < last) { printf("clock went backwards by %ld seconds!\n", last - now); } last = now; sleep(1); } return 0; } Which when run concurrently with: # date -s 2040-1-1 # date -s 2037-1-1 Will detect the clock going backward. The root cause is that wtom_clock_sec in struct vdso_data is only a 32-bit signed value, even though we set its value to be equal to tk->wall_to_monotonic.tv_sec which is 64-bits. Because the monotonic clock starts at zero when the system boots the wall_to_montonic.tv_sec offset is negative for current and future dates. Currently on a freshly booted system the offset will be in the vicinity of negative 1.5 billion seconds. However if the wall clock is set past the Y2038 boundary, the offset from wall to monotonic becomes less than negative 2^31, and no longer fits in 32-bits. When that value is assigned to wtom_clock_sec it is truncated and becomes positive, causing the VDSO assembly code to calculate CLOCK_MONOTONIC incorrectly. That causes CLOCK_MONOTONIC to jump ahead by ~4 billion seconds which it is not meant to do. Worse, if the time is then set back before the Y2038 boundary CLOCK_MONOTONIC will jump backward. We can fix it simply by storing the full 64-bit offset in the vdso_data, and using that in the VDSO assembly code. We also shuffle some of the fields in vdso_data to avoid creating a hole. The original commit that added the CLOCK_MONOTONIC support to the VDSO did actually use a 64-bit value for wtom_clock_sec, see commit a7f290dad32e ("[PATCH] powerpc: Merge vdso's and add vdso support to 32 bits kernel") (Nov 2005). However just 3 days later it was converted to 32-bits in commit 0c37ec2aa88b ("[PATCH] powerpc: vdso fixes (take #2)"), and the bug has existed since then AFAICS. Fixes: 0c37ec2aa88b ("[PATCH] powerpc: vdso fixes (take #2)") Cc: stable@vger.kernel.org # v2.6.15+ Link: http://lkml.kernel.org/r/HaC.ZfES.62bwlnvAvMP.1STMMj@seznam.cz Reported-by: Jakub Drnec <jaydee@email.cz> Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
2019-03-13 13:14:38 +00:00
ld r6,WTOM_CLOCK_SEC(r3)
lwa r9,WTOM_CLOCK_NSEC(r3)
/* check if counter has updated */
or r0,r6,r9
75: or r0,r0,r4
or r0,r0,r5
xor r0,r0,r0
add r3,r3,r0
ld r0,CFG_TB_UPDATE_COUNT(r3)
cmpld cr0,r0,r8 /* check if updated */
bne- 70b
/* Counter has not updated, so continue calculating proper values for
* sec and nsec if monotonic coarse, or just return with the proper
* values for realtime.
*/
bne cr6,80f
/* Add wall->monotonic offset and check for overflow or underflow */
78: add r4,r4,r6
add r5,r5,r9
cmpd cr0,r5,r7
cmpdi cr1,r5,0
blt 79f
subf r5,r7,r5
addi r4,r4,1
79: bge cr1,80f
addi r4,r4,-1
add r5,r5,r7
80: std r4,TSPC64_TV_SEC(r11)
std r5,TSPC64_TV_NSEC(r11)
mtlr r12
crclr cr0*4+so
li r3,0
blr
/*
* syscall fallback
*/
99:
li r0,__NR_clock_gettime
powerpc/vdso: Correct call frame information Call Frame Information is used by gdb for back-traces and inserting breakpoints on function return for the "finish" command. This failed when inside __kernel_clock_gettime. More concerning than difficulty debugging is that CFI is also used by stack frame unwinding code to implement exceptions. If you have an app that needs to handle asynchronous exceptions for some reason, and you are unlucky enough to get one inside the VDSO time functions, your app will crash. What's wrong: There is control flow in __kernel_clock_gettime that reaches label 99 without saving lr in r12. CFI info however is interpreted by the unwinder without reference to control flow: It's a simple matter of "Execute all the CFI opcodes up to the current address". That means the unwinder thinks r12 contains the return address at label 99. Disabuse it of that notion by resetting CFI for the return address at label 99. Note that the ".cfi_restore lr" could have gone anywhere from the "mtlr r12" a few instructions earlier to the instruction at label 99. I put the CFI as late as possible, because in general that's best practice (and if possible grouped with other CFI in order to reduce the number of CFI opcodes executed when unwinding). Using r12 as the return address is perfectly fine after the "mtlr r12" since r12 on that code path still contains the return address. __get_datapage also has a CFI error. That function temporarily saves lr in r0, and reflects that fact with ".cfi_register lr,r0". A later use of r0 means the CFI at that point isn't correct, as r0 no longer contains the return address. Fix that too. Signed-off-by: Alan Modra <amodra@gmail.com> Tested-by: Reza Arbab <arbab@linux.ibm.com> Signed-off-by: Paul Mackerras <paulus@ozlabs.org>
2018-09-14 03:40:04 +00:00
.cfi_restore lr
sc
blr
.cfi_endproc
V_FUNCTION_END(__kernel_clock_gettime)
/*
* Exact prototype of clock_getres()
*
* int __kernel_clock_getres(clockid_t clock_id, struct timespec *res);
*
*/
V_FUNCTION_BEGIN(__kernel_clock_getres)
.cfi_startproc
/* Check for supported clock IDs */
cmpwi cr0,r3,CLOCK_REALTIME
cmpwi cr1,r3,CLOCK_MONOTONIC
cror cr0*4+eq,cr0*4+eq,cr1*4+eq
bne cr0,99f
mflr r12
.cfi_register lr,r12
bl V_LOCAL_FUNC(__get_datapage)
lwz r5, CLOCK_HRTIMER_RES(r3)
mtlr r12
li r3,0
cmpldi cr0,r4,0
crclr cr0*4+so
beqlr
std r3,TSPC64_TV_SEC(r4)
std r5,TSPC64_TV_NSEC(r4)
blr
/*
* syscall fallback
*/
99:
li r0,__NR_clock_getres
sc
blr
.cfi_endproc
V_FUNCTION_END(__kernel_clock_getres)
/*
* Exact prototype of time()
*
* time_t time(time *t);
*
*/
V_FUNCTION_BEGIN(__kernel_time)
.cfi_startproc
mflr r12
.cfi_register lr,r12
mr r11,r3 /* r11 holds t */
bl V_LOCAL_FUNC(__get_datapage)
ld r4,STAMP_XTIME_SEC(r3)
cmpldi r11,0 /* check if t is NULL */
beq 2f
std r4,0(r11) /* store result at *t */
2: mtlr r12
crclr cr0*4+so
mr r3,r4
blr
.cfi_endproc
V_FUNCTION_END(__kernel_time)
/*
powerpc: Rework VDSO gettimeofday to prevent time going backwards 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>
2010-06-20 19:03:08 +00:00
* This is the core of clock_gettime() and gettimeofday(),
* it returns the current time in r4 (seconds) and r5.
* On entry, r7 gives the resolution of r5, either USEC_PER_SEC
* or NSEC_PER_SEC, giving r5 in microseconds or nanoseconds.
* It expects the datapage ptr in r3 and doesn't clobber it.
powerpc: Rework VDSO gettimeofday to prevent time going backwards 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>
2010-06-20 19:03:08 +00:00
* It clobbers r0, r6 and r9.
* On return, r8 contains the counter value that can be reused.
* This clobbers cr0 but not any other cr field.
*/
V_FUNCTION_BEGIN(__do_get_tspec)
.cfi_startproc
/* check for update count & load values */
1: ld r8,CFG_TB_UPDATE_COUNT(r3)
andi. r0,r8,1 /* pending update ? loop */
bne- 1b
xor r0,r8,r8 /* create dependency */
add r3,r3,r0
/* Get TB & offset it. We use the MFTB macro which will generate
* workaround code for Cell.
*/
powerpc: Rework VDSO gettimeofday to prevent time going backwards 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>
2010-06-20 19:03:08 +00:00
MFTB(r6)
ld r9,CFG_TB_ORIG_STAMP(r3)
powerpc: Rework VDSO gettimeofday to prevent time going backwards 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>
2010-06-20 19:03:08 +00:00
subf r6,r9,r6
/* Scale result */
ld r5,CFG_TB_TO_XS(r3)
powerpc: Rework VDSO gettimeofday to prevent time going backwards 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>
2010-06-20 19:03:08 +00:00
sldi r6,r6,12 /* compute time since stamp_xtime */
mulhdu r6,r6,r5 /* in units of 2^-32 seconds */
/* Add stamp since epoch */
ld r4,STAMP_XTIME_SEC(r3)
powerpc: Rework VDSO gettimeofday to prevent time going backwards 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>
2010-06-20 19:03:08 +00:00
lwz r5,STAMP_SEC_FRAC(r3)
or r0,r4,r5
or r0,r0,r6
xor r0,r0,r0
add r3,r3,r0
ld r0,CFG_TB_UPDATE_COUNT(r3)
cmpld r0,r8 /* check if updated */
bne- 1b /* reload if so */
/* convert to seconds & nanoseconds and add to stamp */
powerpc: Rework VDSO gettimeofday to prevent time going backwards 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>
2010-06-20 19:03:08 +00:00
add r6,r6,r5 /* add on fractional seconds of xtime */
mulhwu r5,r6,r7 /* compute micro or nanoseconds and */
srdi r6,r6,32 /* seconds since stamp_xtime */
powerpc: Rework VDSO gettimeofday to prevent time going backwards 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>
2010-06-20 19:03:08 +00:00
clrldi r5,r5,32
add r4,r4,r6
blr
.cfi_endproc
V_FUNCTION_END(__do_get_tspec)