linux-kernel.vger.kernel.org archive mirror
 help / color / mirror / Atom feed
* [RFC PATCH v4 00/11] powerpc: switch VDSO to C implementation.
@ 2020-01-16 17:58 Christophe Leroy
  2020-01-16 17:58 ` [RFC PATCH v4 01/11] powerpc/64: Don't provide time functions in compat VDSO32 Christophe Leroy
                   ` (11 more replies)
  0 siblings, 12 replies; 26+ messages in thread
From: Christophe Leroy @ 2020-01-16 17:58 UTC (permalink / raw)
  To: Benjamin Herrenschmidt, Paul Mackerras, Michael Ellerman,
	nathanl, arnd, tglx, vincenzo.frascino, luto
  Cc: linux-kernel, linuxppc-dev, linux-arm-kernel, linux-mips, x86

This is a fourth tentative to switch powerpc VDSO to generic C implementation.

This version should work on PPC64 (untested). VDSO32 for PPC64 is
impossible to build and has been de-activated, because the powerpc
ASM header files for C are not prepared to build 32 bits code with CONFIG_PPC64.

powerpc is a bit special for VDSO as well as system calls in the
way that it requires setting CR SO bit which cannot be done in C.
Therefore, entry/exit and fallback need to be performed in ASM.

Note that on previous patches, a buggy version of vdsotest was used which was
underestimating the time in gettimeofday compared to clock-get... functions.

This series applies on a merge of powerpc/merge and tip/timers/core branches,
series "lib/vdso: Bugfix and consolidation"
(https://lore.kernel.org/patchwork/project/lkml/list/?series=425784)
applied after the above merge.

On a powerpc8xx, with current powerpc/32 ASM VDSO:

gettimeofday:    vdso: 907 nsec/call
clock-getres-realtime:    vdso: 484 nsec/call
clock-gettime-realtime:    vdso: 899 nsec/call

The first patch adds VDSO generic C support without any changes to common code.
Performance is as follows:

gettimeofday:    vdso: 1211 nsec/call
clock-getres-realtime:    vdso: 722 nsec/call
clock-gettime-realtime:    vdso: 1216 nsec/call

Then a few changes in the common code have allowed performance improvement. At
the end of the series we have:

gettimeofday:    vdso: 974 nsec/call
clock-getres-realtime:    vdso: 545 nsec/call
clock-gettime-realtime:    vdso: 941 nsec/call

The final result is rather close to pure ASM VDSO:
* 7% more on gettimeofday (9 cycles)
* 5% more on clock-gettime-realtime (6 cycles)
* 12% more on clock-getres-realtime (8 cycles)

Due to the unavoidable ASM trampoline, we won't get much closer but that should be
acceptable for a port from ASM to a generic C code (here, 1 cycle is about 7,5 ns)

Christophe Leroy (11):
  powerpc/64: Don't provide time functions in compat VDSO32
  powerpc/vdso: Switch VDSO to generic C implementation.
  lib: vdso: only read hrtimer_res when needed in __cvdso_clock_getres()
  powerpc/vdso: simplify __get_datapage()
  lib: vdso: allow arches to provide vdso data pointer
  powerpc/vdso: provide inline alternative to __get_datapage()
  powerpc/vdso: provide vdso data pointer from the ASM caller.
  lib: vdso: allow fixed clock mode
  powerpc/vdso: override __arch_vdso_capable()
  lib: vdso: Allow arches to override the ns shift operation
  powerpc/32: provide vdso_shift_ns()

 arch/powerpc/Kconfig                         |   2 +
 arch/powerpc/include/asm/clocksource.h       |   6 +
 arch/powerpc/include/asm/vdso/gettimeofday.h | 117 ++++++++++++
 arch/powerpc/include/asm/vdso/vsyscall.h     |  25 +++
 arch/powerpc/include/asm/vdso_datapage.h     |  52 +++---
 arch/powerpc/kernel/asm-offsets.c            |  46 +----
 arch/powerpc/kernel/time.c                   |  91 +---------
 arch/powerpc/kernel/vdso.c                   |  58 ++----
 arch/powerpc/kernel/vdso32/Makefile          |  27 ++-
 arch/powerpc/kernel/vdso32/datapage.S        |  10 +-
 arch/powerpc/kernel/vdso32/gettimeofday.S    | 258 ++++-----------------------
 arch/powerpc/kernel/vdso32/vdso32.lds.S      |   9 +-
 arch/powerpc/kernel/vdso32/vgettimeofday.c   |  29 +++
 arch/powerpc/kernel/vdso64/Makefile          |  23 ++-
 arch/powerpc/kernel/vdso64/datapage.S        |  13 +-
 arch/powerpc/kernel/vdso64/gettimeofday.S    | 257 ++++----------------------
 arch/powerpc/kernel/vdso64/vdso64.lds.S      |   7 +-
 arch/powerpc/kernel/vdso64/vgettimeofday.c   |  29 +++
 lib/vdso/gettimeofday.c                      | 107 ++++++++---
 19 files changed, 457 insertions(+), 709 deletions(-)
 create mode 100644 arch/powerpc/include/asm/clocksource.h
 create mode 100644 arch/powerpc/include/asm/vdso/gettimeofday.h
 create mode 100644 arch/powerpc/include/asm/vdso/vsyscall.h
 create mode 100644 arch/powerpc/kernel/vdso32/vgettimeofday.c
 create mode 100644 arch/powerpc/kernel/vdso64/vgettimeofday.c

-- 
2.13.3


^ permalink raw reply	[flat|nested] 26+ messages in thread

* [RFC PATCH v4 01/11] powerpc/64: Don't provide time functions in compat VDSO32
  2020-01-16 17:58 [RFC PATCH v4 00/11] powerpc: switch VDSO to C implementation Christophe Leroy
@ 2020-01-16 17:58 ` Christophe Leroy
  2020-01-16 17:58 ` [RFC PATCH v4 02/11] powerpc/vdso: Switch VDSO to generic C implementation Christophe Leroy
                   ` (10 subsequent siblings)
  11 siblings, 0 replies; 26+ messages in thread
From: Christophe Leroy @ 2020-01-16 17:58 UTC (permalink / raw)
  To: Benjamin Herrenschmidt, Paul Mackerras, Michael Ellerman,
	nathanl, arnd, tglx, vincenzo.frascino, luto
  Cc: linux-kernel, linuxppc-dev, linux-arm-kernel, linux-mips, x86

In order to allow use of generic C VDSO, remove time functions
from the 32 bits VDSO on PPC64.

This (temporary) removal is needed because powerpc kernel C headers
are not prepared for building 32 bits code on PPC64.

Signed-off-by: Christophe Leroy <christophe.leroy@c-s.fr>
---
 arch/powerpc/kernel/vdso32/Makefile     | 3 ++-
 arch/powerpc/kernel/vdso32/vdso32.lds.S | 2 +-
 2 files changed, 3 insertions(+), 2 deletions(-)

diff --git a/arch/powerpc/kernel/vdso32/Makefile b/arch/powerpc/kernel/vdso32/Makefile
index 06f54d947057..738d52105392 100644
--- a/arch/powerpc/kernel/vdso32/Makefile
+++ b/arch/powerpc/kernel/vdso32/Makefile
@@ -3,7 +3,8 @@
 # List of files in the vdso, has to be asm only for now
 
 obj-vdso32-$(CONFIG_PPC64) = getcpu.o
-obj-vdso32 = sigtramp.o gettimeofday.o datapage.o cacheflush.o note.o \
+obj-vdso32-$(CONFIG_PPC32) = gettimeofday.o
+obj-vdso32 = sigtramp.o datapage.o cacheflush.o note.o \
 		$(obj-vdso32-y)
 
 # Build rules
diff --git a/arch/powerpc/kernel/vdso32/vdso32.lds.S b/arch/powerpc/kernel/vdso32/vdso32.lds.S
index 00c025ba4a92..9400b182e163 100644
--- a/arch/powerpc/kernel/vdso32/vdso32.lds.S
+++ b/arch/powerpc/kernel/vdso32/vdso32.lds.S
@@ -144,7 +144,7 @@ VERSION
 		__kernel_datapage_offset;
 
 		__kernel_get_syscall_map;
-#ifndef CONFIG_PPC_BOOK3S_601
+#if defined(CONFIG_PPC32) && !defined(CONFIG_PPC_BOOK3S_601)
 		__kernel_gettimeofday;
 		__kernel_clock_gettime;
 		__kernel_clock_getres;
-- 
2.13.3


^ permalink raw reply related	[flat|nested] 26+ messages in thread

* [RFC PATCH v4 02/11] powerpc/vdso: Switch VDSO to generic C implementation.
  2020-01-16 17:58 [RFC PATCH v4 00/11] powerpc: switch VDSO to C implementation Christophe Leroy
  2020-01-16 17:58 ` [RFC PATCH v4 01/11] powerpc/64: Don't provide time functions in compat VDSO32 Christophe Leroy
@ 2020-01-16 17:58 ` Christophe Leroy
  2020-01-16 17:58 ` [RFC PATCH v4 03/11] lib: vdso: only read hrtimer_res when needed in __cvdso_clock_getres() Christophe Leroy
                   ` (9 subsequent siblings)
  11 siblings, 0 replies; 26+ messages in thread
From: Christophe Leroy @ 2020-01-16 17:58 UTC (permalink / raw)
  To: Benjamin Herrenschmidt, Paul Mackerras, Michael Ellerman,
	nathanl, arnd, tglx, vincenzo.frascino, luto
  Cc: linux-kernel, linuxppc-dev, linux-arm-kernel, linux-mips, x86

This is a minimalistic conversion to VDSO generic C implementation.

On powerpc 8xx, performance is degraded by 40-45% for gettime and
by 50% for getres.

Optimisations will follow in next patches, some in the core
VDSO functions, some in the powerpc arch.

powerpc is a bit special for VDSO as well as system calls in the
way that it requires setting CR SO bit which cannot be done in C.
Therefore, entry/exit needs to be performed in ASM.

On a powerpc885 at 132MHz:
With current powerpc/32 ASM VDSO:

gettimeofday:    vdso: 907 nsec/call
clock-getres-realtime-coarse:    vdso: 3053 nsec/call
clock-gettime-realtime-coarse:    vdso: 2823 nsec/call
clock-getres-realtime:    vdso: 484 nsec/call
clock-gettime-realtime:    vdso: 899 nsec/call
clock-getres-boottime:    vdso: 2586 nsec/call
clock-gettime-boottime:    vdso: 3820 nsec/call
clock-getres-tai:    vdso: 2587 nsec/call
clock-gettime-tai:    vdso: 3819 nsec/call
clock-getres-monotonic-raw:    vdso: 2587 nsec/call
clock-gettime-monotonic-raw:    vdso: 3378 nsec/call
clock-getres-monotonic-coarse:    vdso: 3054 nsec/call
clock-gettime-monotonic-coarse:    vdso: 3376 nsec/call
clock-getres-monotonic:    vdso: 484 nsec/call
clock-gettime-monotonic:    vdso: 1026 nsec/call

Once switched to C implementation:

gettimeofday:    vdso: 1211 nsec/call
clock-getres-realtime-coarse:    vdso: 722 nsec/call
clock-gettime-realtime-coarse:    vdso: 784 nsec/call
clock-getres-realtime:    vdso: 722 nsec/call
clock-gettime-realtime:    vdso: 1216 nsec/call
clock-getres-boottime:    vdso: 722 nsec/call
clock-gettime-boottime:    vdso: 1216 nsec/call
clock-getres-tai:    vdso: 722 nsec/call
clock-gettime-tai:    vdso: 1216 nsec/call
clock-getres-monotonic-raw:    vdso: 722 nsec/call
clock-gettime-monotonic-raw:    vdso: 1277 nsec/call
clock-getres-monotonic-coarse:    vdso: 722 nsec/call
clock-gettime-monotonic-coarse:    vdso: 783 nsec/call
clock-getres-monotonic:    vdso: 722 nsec/call
clock-gettime-monotonic:    vdso: 1217 nsec/call
Signed-off-by: Christophe Leroy <christophe.leroy@c-s.fr>
---
 arch/powerpc/Kconfig                         |   2 +
 arch/powerpc/include/asm/clocksource.h       |   6 +
 arch/powerpc/include/asm/vdso/gettimeofday.h | 102 +++++++++++
 arch/powerpc/include/asm/vdso/vsyscall.h     |  25 +++
 arch/powerpc/include/asm/vdso_datapage.h     |  41 ++---
 arch/powerpc/kernel/asm-offsets.c            |  46 +----
 arch/powerpc/kernel/time.c                   |  91 +---------
 arch/powerpc/kernel/vdso.c                   |   5 +-
 arch/powerpc/kernel/vdso32/Makefile          |  24 +++
 arch/powerpc/kernel/vdso32/gettimeofday.S    | 255 +++------------------------
 arch/powerpc/kernel/vdso32/vgettimeofday.c   |  26 +++
 arch/powerpc/kernel/vdso64/Makefile          |  23 ++-
 arch/powerpc/kernel/vdso64/datapage.S        |   3 +-
 arch/powerpc/kernel/vdso64/gettimeofday.S    | 254 +++-----------------------
 arch/powerpc/kernel/vdso64/vgettimeofday.c   |  26 +++
 15 files changed, 312 insertions(+), 617 deletions(-)
 create mode 100644 arch/powerpc/include/asm/clocksource.h
 create mode 100644 arch/powerpc/include/asm/vdso/gettimeofday.h
 create mode 100644 arch/powerpc/include/asm/vdso/vsyscall.h
 create mode 100644 arch/powerpc/kernel/vdso32/vgettimeofday.c
 create mode 100644 arch/powerpc/kernel/vdso64/vgettimeofday.c

diff --git a/arch/powerpc/Kconfig b/arch/powerpc/Kconfig
index 1ec34e16ed65..bd04c68baf91 100644
--- a/arch/powerpc/Kconfig
+++ b/arch/powerpc/Kconfig
@@ -169,6 +169,7 @@ config PPC
 	select GENERIC_STRNCPY_FROM_USER
 	select GENERIC_STRNLEN_USER
 	select GENERIC_TIME_VSYSCALL
+	select GENERIC_GETTIMEOFDAY
 	select HAVE_ARCH_AUDITSYSCALL
 	select HAVE_ARCH_HUGE_VMAP		if PPC_BOOK3S_64 && PPC_RADIX_MMU
 	select HAVE_ARCH_JUMP_LABEL
@@ -198,6 +199,7 @@ config PPC
 	select HAVE_FUNCTION_GRAPH_TRACER
 	select HAVE_FUNCTION_TRACER
 	select HAVE_GCC_PLUGINS			if GCC_VERSION >= 50200   # plugin support on gcc <= 5.1 is buggy on PPC
+	select HAVE_GENERIC_VDSO
 	select HAVE_HW_BREAKPOINT		if PERF_EVENTS && (PPC_BOOK3S || PPC_8xx)
 	select HAVE_IDE
 	select HAVE_IOREMAP_PROT
diff --git a/arch/powerpc/include/asm/clocksource.h b/arch/powerpc/include/asm/clocksource.h
new file mode 100644
index 000000000000..37423d17af30
--- /dev/null
+++ b/arch/powerpc/include/asm/clocksource.h
@@ -0,0 +1,6 @@
+#ifndef _ASM_POWERPC_CLOCKSOURCE_H
+#define _ASM_POWERPC_CLOCKSOURCE_H
+
+#define VDSO_ARCH_CLOCKMODES	VDSO_CLOCKMODE_ARCHTIMER
+
+#endif
diff --git a/arch/powerpc/include/asm/vdso/gettimeofday.h b/arch/powerpc/include/asm/vdso/gettimeofday.h
new file mode 100644
index 000000000000..c2cafd85d3cb
--- /dev/null
+++ b/arch/powerpc/include/asm/vdso/gettimeofday.h
@@ -0,0 +1,102 @@
+/* SPDX-License-Identifier: GPL-2.0 */
+#ifndef __ASM_VDSO_GETTIMEOFDAY_H
+#define __ASM_VDSO_GETTIMEOFDAY_H
+
+#ifndef __ASSEMBLY__
+
+#include <asm/time.h>
+#include <asm/unistd.h>
+#include <asm/vdso_datapage.h>
+#include <uapi/linux/time.h>
+
+#define VDSO_HAS_CLOCK_GETRES		1
+
+#define VDSO_HAS_TIME			1
+
+static __always_inline int do_syscall_2(const unsigned long _r0, const unsigned long _r3,
+					const unsigned long _r4)
+{
+	register long r0 asm("r0") = _r0;
+	register unsigned long r3 asm("r3") = _r3;
+	register unsigned long r4 asm("r4") = _r4;
+	register int ret asm ("r3");
+
+	asm volatile(
+		"       sc\n"
+		"	bns+	1f\n"
+		"	neg	%0, %0\n"
+		"1:\n"
+	: "+r" (ret), "+r" (r3), "+r" (r4), "+r" (r0)
+	:
+	: "memory", "r5", "r6", "r7", "r8", "r9", "r10", "r11", "r12", "cr0", "ctr");
+
+	return ret;
+}
+
+static __always_inline
+int gettimeofday_fallback(struct __kernel_old_timeval *_tv, struct timezone *_tz)
+{
+	return do_syscall_2(__NR_gettimeofday, (unsigned long)_tv, (unsigned long)_tz);
+}
+
+static __always_inline
+int clock_gettime_fallback(clockid_t _clkid, struct __kernel_timespec *_ts)
+{
+	return do_syscall_2(__NR_clock_gettime, _clkid, (unsigned long)_ts);
+}
+
+static __always_inline
+int clock_getres_fallback(clockid_t _clkid, struct __kernel_timespec *_ts)
+{
+	return do_syscall_2(__NR_clock_getres, _clkid, (unsigned long)_ts);
+}
+
+#ifdef CONFIG_VDSO32
+
+#define BUILD_VDSO32		1
+
+static __always_inline
+int clock_gettime32_fallback(clockid_t _clkid, struct old_timespec32 *_ts)
+{
+	return do_syscall_2(__NR_clock_gettime, _clkid, (unsigned long)_ts);
+}
+
+static __always_inline
+int clock_getres32_fallback(clockid_t _clkid, struct old_timespec32 *_ts)
+{
+	return do_syscall_2(__NR_clock_getres, _clkid, (unsigned long)_ts);
+}
+#endif
+
+static __always_inline u64 __arch_get_hw_counter(s32 clock_mode)
+{
+	return get_tb();
+}
+
+void *__get_datapage(void);
+
+static __always_inline const struct vdso_data *__arch_get_vdso_data(void)
+{
+	struct vdso_arch_data *vdso_data = __get_datapage();
+
+	return vdso_data->data;
+}
+
+/*
+ * powerpc specific delta calculation.
+ *
+ * This variant removes the masking of the subtraction because the
+ * clocksource mask of all VDSO capable clocksources on powerpc is U64_MAX
+ * which would result in a pointless operation. The compiler cannot
+ * optimize it away as the mask comes from the vdso data and is not compile
+ * time constant.
+ */
+static __always_inline u64 vdso_calc_delta(u64 cycles, u64 last, u64 mask, u32 mult)
+{
+	return (cycles - last) * mult;
+}
+#define vdso_calc_delta vdso_calc_delta
+
+#endif /* !__ASSEMBLY__ */
+
+#endif /* __ASM_VDSO_GETTIMEOFDAY_H */
diff --git a/arch/powerpc/include/asm/vdso/vsyscall.h b/arch/powerpc/include/asm/vdso/vsyscall.h
new file mode 100644
index 000000000000..c56a030c0623
--- /dev/null
+++ b/arch/powerpc/include/asm/vdso/vsyscall.h
@@ -0,0 +1,25 @@
+/* SPDX-License-Identifier: GPL-2.0 */
+#ifndef __ASM_VDSO_VSYSCALL_H
+#define __ASM_VDSO_VSYSCALL_H
+
+#ifndef __ASSEMBLY__
+
+#include <linux/timekeeper_internal.h>
+#include <asm/vdso_datapage.h>
+
+/*
+ * Update the vDSO data page to keep in sync with kernel timekeeping.
+ */
+static __always_inline
+struct vdso_data *__arch_get_k_vdso_data(void)
+{
+	return vdso_data->data;
+}
+#define __arch_get_k_vdso_data __arch_get_k_vdso_data
+
+/* The asm-generic header needs to be included after the definitions above */
+#include <asm-generic/vdso/vsyscall.h>
+
+#endif /* !__ASSEMBLY__ */
+
+#endif /* __ASM_VDSO_VSYSCALL_H */
diff --git a/arch/powerpc/include/asm/vdso_datapage.h b/arch/powerpc/include/asm/vdso_datapage.h
index 40f13f3626d3..4d7965bf369e 100644
--- a/arch/powerpc/include/asm/vdso_datapage.h
+++ b/arch/powerpc/include/asm/vdso_datapage.h
@@ -36,6 +36,7 @@
 
 #include <linux/unistd.h>
 #include <linux/time.h>
+#include <vdso/datapage.h>
 
 #define SYSCALL_MAP_SIZE      ((NR_syscalls + 31) / 32)
 
@@ -45,7 +46,7 @@
 
 #ifdef CONFIG_PPC64
 
-struct vdso_data {
+struct vdso_arch_data {
 	__u8  eye_catcher[16];		/* Eyecatcher: SYSTEMCFG:PPC64	0x00 */
 	struct {			/* Systemcfg version numbers	     */
 		__u32 major;		/* Major number			0x10 */
@@ -59,13 +60,13 @@ struct vdso_data {
 	__u32 processor;		/* Processor type		0x1C */
 	__u64 processorCount;		/* # of physical processors	0x20 */
 	__u64 physicalMemorySize;	/* Size of real memory(B)	0x28 */
-	__u64 tb_orig_stamp;		/* Timebase at boot		0x30 */
+	__u64 tb_orig_stamp;		/* (NU) Timebase at boot	0x30 */
 	__u64 tb_ticks_per_sec;		/* Timebase tics / sec		0x38 */
-	__u64 tb_to_xs;			/* Inverse of TB to 2^20	0x40 */
-	__u64 stamp_xsec;		/*				0x48 */
-	__u64 tb_update_count;		/* Timebase atomicity ctr	0x50 */
-	__u32 tz_minuteswest;		/* Minutes west of Greenwich	0x58 */
-	__u32 tz_dsttime;		/* Type of dst correction	0x5C */
+	__u64 tb_to_xs;			/* (NU) Inverse of TB to 2^20	0x40 */
+	__u64 stamp_xsec;		/* (NU) 			0x48 */
+	__u64 tb_update_count;		/* (NU) Timebase atomicity ctr	0x50 */
+	__u32 tz_minuteswest;		/* (NU) Min. west of Greenwich	0x58 */
+	__u32 tz_dsttime;		/* (NU) Type of dst correction	0x5C */
 	__u32 dcache_size;		/* L1 d-cache size		0x60 */
 	__u32 dcache_line_size;		/* L1 d-cache line size		0x64 */
 	__u32 icache_size;		/* L1 i-cache size		0x68 */
@@ -78,14 +79,10 @@ struct vdso_data {
 	__u32 icache_block_size;		/* L1 i-cache block size     */
 	__u32 dcache_log_block_size;		/* L1 d-cache log block size */
 	__u32 icache_log_block_size;		/* L1 i-cache log block size */
-	__u32 stamp_sec_fraction;		/* fractional seconds of stamp_xtime */
-	__s32 wtom_clock_nsec;			/* Wall to monotonic clock nsec */
-	__s64 wtom_clock_sec;			/* Wall to monotonic clock sec */
-	__s64 stamp_xtime_sec;			/* xtime secs as at tb_orig_stamp */
-	__s64 stamp_xtime_nsec;			/* xtime nsecs as at tb_orig_stamp */
-	__u32 hrtimer_res;			/* hrtimer resolution */
    	__u32 syscall_map_64[SYSCALL_MAP_SIZE]; /* map of syscalls  */
    	__u32 syscall_map_32[SYSCALL_MAP_SIZE]; /* map of syscalls */
+
+	struct vdso_data data[CS_BASES];
 };
 
 #else /* CONFIG_PPC64 */
@@ -93,30 +90,20 @@ struct vdso_data {
 /*
  * And here is the simpler 32 bits version
  */
-struct vdso_data {
-	__u64 tb_orig_stamp;		/* Timebase at boot		0x30 */
+struct vdso_arch_data {
 	__u64 tb_ticks_per_sec;		/* Timebase tics / sec		0x38 */
-	__u64 tb_to_xs;			/* Inverse of TB to 2^20	0x40 */
-	__u64 stamp_xsec;		/*				0x48 */
-	__u32 tb_update_count;		/* Timebase atomicity ctr	0x50 */
-	__u32 tz_minuteswest;		/* Minutes west of Greenwich	0x58 */
-	__u32 tz_dsttime;		/* Type of dst correction	0x5C */
-	__s32 wtom_clock_sec;			/* Wall to monotonic clock */
-	__s32 wtom_clock_nsec;
-	__s32 stamp_xtime_sec;		/* xtime seconds as at tb_orig_stamp */
-	__s32 stamp_xtime_nsec;		/* xtime nsecs as at tb_orig_stamp */
-	__u32 stamp_sec_fraction;	/* fractional seconds of stamp_xtime */
-	__u32 hrtimer_res;		/* hrtimer resolution */
    	__u32 syscall_map_32[SYSCALL_MAP_SIZE]; /* map of syscalls */
 	__u32 dcache_block_size;	/* L1 d-cache block size     */
 	__u32 icache_block_size;	/* L1 i-cache block size     */
 	__u32 dcache_log_block_size;	/* L1 d-cache log block size */
 	__u32 icache_log_block_size;	/* L1 i-cache log block size */
+
+	struct vdso_data data[CS_BASES];
 };
 
 #endif /* CONFIG_PPC64 */
 
-extern struct vdso_data *vdso_data;
+extern struct vdso_arch_data *vdso_data;
 
 #endif /* __ASSEMBLY__ */
 
diff --git a/arch/powerpc/kernel/asm-offsets.c b/arch/powerpc/kernel/asm-offsets.c
index 3d47aec7becf..14ff4d159bc5 100644
--- a/arch/powerpc/kernel/asm-offsets.c
+++ b/arch/powerpc/kernel/asm-offsets.c
@@ -376,44 +376,16 @@ int main(void)
 #endif /* ! CONFIG_PPC64 */
 
 	/* datapage offsets for use by vdso */
-	OFFSET(CFG_TB_ORIG_STAMP, vdso_data, tb_orig_stamp);
-	OFFSET(CFG_TB_TICKS_PER_SEC, vdso_data, tb_ticks_per_sec);
-	OFFSET(CFG_TB_TO_XS, vdso_data, tb_to_xs);
-	OFFSET(CFG_TB_UPDATE_COUNT, vdso_data, tb_update_count);
-	OFFSET(CFG_TZ_MINUTEWEST, vdso_data, tz_minuteswest);
-	OFFSET(CFG_TZ_DSTTIME, vdso_data, tz_dsttime);
-	OFFSET(CFG_SYSCALL_MAP32, vdso_data, syscall_map_32);
-	OFFSET(WTOM_CLOCK_SEC, vdso_data, wtom_clock_sec);
-	OFFSET(WTOM_CLOCK_NSEC, vdso_data, wtom_clock_nsec);
-	OFFSET(STAMP_XTIME_SEC, vdso_data, stamp_xtime_sec);
-	OFFSET(STAMP_XTIME_NSEC, vdso_data, stamp_xtime_nsec);
-	OFFSET(STAMP_SEC_FRAC, vdso_data, stamp_sec_fraction);
-	OFFSET(CLOCK_HRTIMER_RES, vdso_data, hrtimer_res);
-	OFFSET(CFG_ICACHE_BLOCKSZ, vdso_data, icache_block_size);
-	OFFSET(CFG_DCACHE_BLOCKSZ, vdso_data, dcache_block_size);
-	OFFSET(CFG_ICACHE_LOGBLOCKSZ, vdso_data, icache_log_block_size);
-	OFFSET(CFG_DCACHE_LOGBLOCKSZ, vdso_data, dcache_log_block_size);
+	OFFSET(VDSO_DATA_OFFSET, vdso_arch_data, data);
+	OFFSET(CFG_TB_TICKS_PER_SEC, vdso_arch_data, tb_ticks_per_sec);
+	OFFSET(CFG_SYSCALL_MAP32, vdso_arch_data, syscall_map_32);
+	OFFSET(CFG_ICACHE_BLOCKSZ, vdso_arch_data, icache_block_size);
+	OFFSET(CFG_DCACHE_BLOCKSZ, vdso_arch_data, dcache_block_size);
+	OFFSET(CFG_ICACHE_LOGBLOCKSZ, vdso_arch_data, icache_log_block_size);
+	OFFSET(CFG_DCACHE_LOGBLOCKSZ, vdso_arch_data, dcache_log_block_size);
 #ifdef CONFIG_PPC64
-	OFFSET(CFG_SYSCALL_MAP64, vdso_data, syscall_map_64);
-	OFFSET(TVAL64_TV_SEC, __kernel_old_timeval, tv_sec);
-	OFFSET(TVAL64_TV_USEC, __kernel_old_timeval, tv_usec);
-#endif
-	OFFSET(TSPC64_TV_SEC, __kernel_timespec, tv_sec);
-	OFFSET(TSPC64_TV_NSEC, __kernel_timespec, tv_nsec);
-	OFFSET(TVAL32_TV_SEC, old_timeval32, tv_sec);
-	OFFSET(TVAL32_TV_USEC, old_timeval32, tv_usec);
-	OFFSET(TSPC32_TV_SEC, old_timespec32, tv_sec);
-	OFFSET(TSPC32_TV_NSEC, old_timespec32, tv_nsec);
-	/* timeval/timezone offsets for use by vdso */
-	OFFSET(TZONE_TZ_MINWEST, timezone, tz_minuteswest);
-	OFFSET(TZONE_TZ_DSTTIME, timezone, tz_dsttime);
-
-	/* Other bits used by the vdso */
-	DEFINE(CLOCK_REALTIME, CLOCK_REALTIME);
-	DEFINE(CLOCK_MONOTONIC, CLOCK_MONOTONIC);
-	DEFINE(CLOCK_REALTIME_COARSE, CLOCK_REALTIME_COARSE);
-	DEFINE(CLOCK_MONOTONIC_COARSE, CLOCK_MONOTONIC_COARSE);
-	DEFINE(NSEC_PER_SEC, NSEC_PER_SEC);
+	OFFSET(CFG_SYSCALL_MAP64, vdso_arch_data, syscall_map_64);
+#endif
 
 #ifdef CONFIG_BUG
 	DEFINE(BUG_ENTRY_SIZE, sizeof(struct bug_entry));
diff --git a/arch/powerpc/kernel/time.c b/arch/powerpc/kernel/time.c
index 1168e8b37e30..783b219caf4d 100644
--- a/arch/powerpc/kernel/time.c
+++ b/arch/powerpc/kernel/time.c
@@ -91,6 +91,7 @@ static struct clocksource clocksource_timebase = {
 	.flags        = CLOCK_SOURCE_IS_CONTINUOUS,
 	.mask         = CLOCKSOURCE_MASK(64),
 	.read         = timebase_read,
+	.vdso_clock_mode = VDSO_CLOCKMODE_ARCHTIMER,
 };
 
 #define DECREMENTER_DEFAULT_MAX 0x7FFFFFFF
@@ -882,95 +883,6 @@ static notrace u64 timebase_read(struct clocksource *cs)
 	return (u64)get_tb();
 }
 
-
-void update_vsyscall(struct timekeeper *tk)
-{
-	struct timespec64 xt;
-	struct clocksource *clock = tk->tkr_mono.clock;
-	u32 mult = tk->tkr_mono.mult;
-	u32 shift = tk->tkr_mono.shift;
-	u64 cycle_last = tk->tkr_mono.cycle_last;
-	u64 new_tb_to_xs, new_stamp_xsec;
-	u64 frac_sec;
-
-	if (clock != &clocksource_timebase)
-		return;
-
-	xt.tv_sec = tk->xtime_sec;
-	xt.tv_nsec = (long)(tk->tkr_mono.xtime_nsec >> tk->tkr_mono.shift);
-
-	/* Make userspace gettimeofday spin until we're done. */
-	++vdso_data->tb_update_count;
-	smp_mb();
-
-	/*
-	 * This computes ((2^20 / 1e9) * mult) >> shift as a
-	 * 0.64 fixed-point fraction.
-	 * The computation in the else clause below won't overflow
-	 * (as long as the timebase frequency is >= 1.049 MHz)
-	 * but loses precision because we lose the low bits of the constant
-	 * in the shift.  Note that 19342813113834067 ~= 2^(20+64) / 1e9.
-	 * For a shift of 24 the error is about 0.5e-9, or about 0.5ns
-	 * over a second.  (Shift values are usually 22, 23 or 24.)
-	 * For high frequency clocks such as the 512MHz timebase clock
-	 * on POWER[6789], the mult value is small (e.g. 32768000)
-	 * and so we can shift the constant by 16 initially
-	 * (295147905179 ~= 2^(20+64-16) / 1e9) and then do the
-	 * remaining shifts after the multiplication, which gives a
-	 * more accurate result (e.g. with mult = 32768000, shift = 24,
-	 * the error is only about 1.2e-12, or 0.7ns over 10 minutes).
-	 */
-	if (mult <= 62500000 && clock->shift >= 16)
-		new_tb_to_xs = ((u64) mult * 295147905179ULL) >> (clock->shift - 16);
-	else
-		new_tb_to_xs = (u64) mult * (19342813113834067ULL >> clock->shift);
-
-	/*
-	 * Compute the fractional second in units of 2^-32 seconds.
-	 * The fractional second is tk->tkr_mono.xtime_nsec >> tk->tkr_mono.shift
-	 * in nanoseconds, so multiplying that by 2^32 / 1e9 gives
-	 * it in units of 2^-32 seconds.
-	 * We assume shift <= 32 because clocks_calc_mult_shift()
-	 * generates shift values in the range 0 - 32.
-	 */
-	frac_sec = tk->tkr_mono.xtime_nsec << (32 - shift);
-	do_div(frac_sec, NSEC_PER_SEC);
-
-	/*
-	 * Work out new stamp_xsec value for any legacy users of systemcfg.
-	 * stamp_xsec is in units of 2^-20 seconds.
-	 */
-	new_stamp_xsec = frac_sec >> 12;
-	new_stamp_xsec += tk->xtime_sec * XSEC_PER_SEC;
-
-	/*
-	 * tb_update_count is used to allow the userspace gettimeofday code
-	 * to assure itself that it sees a consistent view of the tb_to_xs and
-	 * stamp_xsec variables.  It reads the tb_update_count, then reads
-	 * tb_to_xs and stamp_xsec and then reads tb_update_count again.  If
-	 * the two values of tb_update_count match and are even then the
-	 * tb_to_xs and stamp_xsec values are consistent.  If not, then it
-	 * loops back and reads them again until this criteria is met.
-	 */
-	vdso_data->tb_orig_stamp = cycle_last;
-	vdso_data->stamp_xsec = new_stamp_xsec;
-	vdso_data->tb_to_xs = new_tb_to_xs;
-	vdso_data->wtom_clock_sec = tk->wall_to_monotonic.tv_sec;
-	vdso_data->wtom_clock_nsec = tk->wall_to_monotonic.tv_nsec;
-	vdso_data->stamp_xtime_sec = xt.tv_sec;
-	vdso_data->stamp_xtime_nsec = xt.tv_nsec;
-	vdso_data->stamp_sec_fraction = frac_sec;
-	vdso_data->hrtimer_res = hrtimer_resolution;
-	smp_wmb();
-	++(vdso_data->tb_update_count);
-}
-
-void update_vsyscall_tz(void)
-{
-	vdso_data->tz_minuteswest = sys_tz.tz_minuteswest;
-	vdso_data->tz_dsttime = sys_tz.tz_dsttime;
-}
-
 static void __init clocksource_init(void)
 {
 	struct clocksource *clock;
@@ -1140,7 +1052,6 @@ void __init time_init(void)
 		sys_tz.tz_dsttime = 0;
 	}
 
-	vdso_data->tb_update_count = 0;
 	vdso_data->tb_ticks_per_sec = tb_ticks_per_sec;
 
 	/* initialise and enable the large decrementer (if we have one) */
diff --git a/arch/powerpc/kernel/vdso.c b/arch/powerpc/kernel/vdso.c
index eae9ddaecbcf..16a44bffe698 100644
--- a/arch/powerpc/kernel/vdso.c
+++ b/arch/powerpc/kernel/vdso.c
@@ -17,6 +17,7 @@
 #include <linux/elf.h>
 #include <linux/security.h>
 #include <linux/memblock.h>
+#include <vdso/datapage.h>
 
 #include <asm/pgtable.h>
 #include <asm/processor.h>
@@ -71,10 +72,10 @@ static int vdso_ready;
  * with it, it will become dynamically allocated
  */
 static union {
-	struct vdso_data	data;
+	struct vdso_arch_data	data;
 	u8			page[PAGE_SIZE];
 } vdso_data_store __page_aligned_data;
-struct vdso_data *vdso_data = &vdso_data_store.data;
+struct vdso_arch_data *vdso_data = &vdso_data_store.data;
 
 /* Format of the patch table */
 struct vdso_patch_def
diff --git a/arch/powerpc/kernel/vdso32/Makefile b/arch/powerpc/kernel/vdso32/Makefile
index 738d52105392..1adbcd74b3ad 100644
--- a/arch/powerpc/kernel/vdso32/Makefile
+++ b/arch/powerpc/kernel/vdso32/Makefile
@@ -2,11 +2,23 @@
 
 # List of files in the vdso, has to be asm only for now
 
+ARCH_REL_TYPE_ABS := R_PPC_JUMP_SLOT|R_PPC_GLOB_DAT|R_PPC_ADDR32|R_PPC_ADDR24|R_PPC_ADDR16|R_PPC_ADDR16_LO|R_PPC_ADDR16_HI|R_PPC_ADDR16_HA|R_PPC_ADDR14|R_PPC_ADDR14_BRTAKEN|R_PPC_ADDR14_BRNTAKEN
+include $(srctree)/lib/vdso/Makefile
+
 obj-vdso32-$(CONFIG_PPC64) = getcpu.o
 obj-vdso32-$(CONFIG_PPC32) = gettimeofday.o
 obj-vdso32 = sigtramp.o datapage.o cacheflush.o note.o \
 		$(obj-vdso32-y)
 
+ifneq ($(c-gettimeofday-y),)
+  CFLAGS_vgettimeofday.o += -include $(c-gettimeofday-y)
+  CFLAGS_vgettimeofday.o += $(DISABLE_LATENT_ENTROPY_PLUGIN)
+  CFLAGS_vgettimeofday.o += $(call cc-option, -fno-stack-protector)
+  CFLAGS_vgettimeofday.o += -DDISABLE_BRANCH_PROFILING
+  CFLAGS_vgettimeofday.o += -ffreestanding
+  CFLAGS_REMOVE_vgettimeofday.o = $(CC_FLAGS_FTRACE)
+endif
+
 # Build rules
 
 ifdef CROSS32_COMPILE
@@ -26,6 +38,7 @@ obj-vdso32 := $(addprefix $(obj)/, $(obj-vdso32))
 GCOV_PROFILE := n
 KCOV_INSTRUMENT := n
 UBSAN_SANITIZE := n
+KASAN_SANITIZE := n
 
 ccflags-y := -shared -fno-common -fno-builtin -nostdlib \
 	-Wl,-soname=linux-vdso32.so.1 -Wl,--hash-style=both
@@ -39,8 +52,13 @@ CPPFLAGS_vdso32.lds += -P -C -Upowerpc
 $(obj)/vdso32_wrapper.o : $(obj)/vdso32.so
 
 # link rule for the .so file, .lds has to be first
+ifdef CONFIG_PPC32
+$(obj)/vdso32.so.dbg: $(src)/vdso32.lds $(obj-vdso32) $(obj)/vgettimeofday.o FORCE
+	$(call if_changed,vdso32ld_and_check)
+else
 $(obj)/vdso32.so.dbg: $(src)/vdso32.lds $(obj-vdso32) FORCE
 	$(call if_changed,vdso32ld)
+endif
 
 # strip rule for the .so file
 $(obj)/%.so: OBJCOPYFLAGS := -S
@@ -50,12 +68,18 @@ $(obj)/%.so: $(obj)/%.so.dbg FORCE
 # assembly rules for the .S files
 $(obj-vdso32): %.o: %.S FORCE
 	$(call if_changed_dep,vdso32as)
+$(obj)/vgettimeofday.o: %.o: %.c FORCE
+	$(call if_changed_dep,vdso32cc)
 
 # actual build commands
 quiet_cmd_vdso32ld = VDSO32L $@
       cmd_vdso32ld = $(VDSOCC) $(c_flags) $(CC32FLAGS) -o $@ -Wl,-T$(filter %.lds,$^) $(filter %.o,$^)
+quiet_cmd_vdso32ld_and_check = VDSO32L $@
+      cmd_vdso32ld_and_check = $(VDSOCC) $(c_flags) $(CC32FLAGS) -o $@ -Wl,-T$(filter %.lds,$^) $(filter %.o,$^) ; $(cmd_vdso_check)
 quiet_cmd_vdso32as = VDSO32A $@
       cmd_vdso32as = $(VDSOCC) $(a_flags) $(CC32FLAGS) -c -o $@ $<
+quiet_cmd_vdso32cc = VDSO32C $@
+      cmd_vdso32cc = $(VDSOCC) $(c_flags) $(CC32FLAGS) -c -o $@ $<
 
 # install commands for the unstripped file
 quiet_cmd_vdso_install = INSTALL $@
diff --git a/arch/powerpc/kernel/vdso32/gettimeofday.S b/arch/powerpc/kernel/vdso32/gettimeofday.S
index 3306672f57a9..ba0bd64b3da3 100644
--- a/arch/powerpc/kernel/vdso32/gettimeofday.S
+++ b/arch/powerpc/kernel/vdso32/gettimeofday.S
@@ -12,14 +12,24 @@
 #include <asm/asm-offsets.h>
 #include <asm/unistd.h>
 
-/* Offset for the low 32-bit part of a field of long type */
-#ifdef CONFIG_PPC64
-#define LOPART	4
-#else
-#define LOPART	0
-#endif
-
 	.text
+
+.macro cvdso_call funct
+	stwu	r1, -16(r1)
+	mflr	r0
+	stw	r0, 20(r1)
+	bl	\funct
+	lwz	r0, 20(r1)
+	cmpwi	r3, 0
+	mtlr	r0
+	addi	r1, r1, 16
+	crclr	so
+	beqlr+
+	crset	so
+	neg	r3, r3
+	blr
+.endm
+
 /*
  * Exact prototype of gettimeofday
  *
@@ -28,32 +38,7 @@
  */
 V_FUNCTION_BEGIN(__kernel_gettimeofday)
   .cfi_startproc
-	mflr	r12
-  .cfi_register lr,r12
-
-	mr	r10,r3			/* r10 saves tv */
-	mr	r11,r4			/* r11 saves tz */
-	bl	__get_datapage@local	/* get data page */
-	mr	r9, r3			/* datapage ptr in r9 */
-	cmplwi	r10,0			/* check if tv is NULL */
-	beq	3f
-	lis	r7,1000000@ha		/* load up USEC_PER_SEC */
-	addi	r7,r7,1000000@l		/* so we get microseconds in r4 */
-	bl	__do_get_tspec@local	/* get sec/usec from tb & kernel */
-	stw	r3,TVAL32_TV_SEC(r10)
-	stw	r4,TVAL32_TV_USEC(r10)
-
-3:	cmplwi	r11,0			/* check if tz is NULL */
-	beq	1f
-	lwz	r4,CFG_TZ_MINUTEWEST(r9)/* fill tz */
-	lwz	r5,CFG_TZ_DSTTIME(r9)
-	stw	r4,TZONE_TZ_MINWEST(r11)
-	stw	r5,TZONE_TZ_DSTTIME(r11)
-
-1:	mtlr	r12
-	crclr	cr0*4+so
-	li	r3,0
-	blr
+	cvdso_call __c_kernel_gettimeofday
   .cfi_endproc
 V_FUNCTION_END(__kernel_gettimeofday)
 
@@ -65,77 +50,7 @@ V_FUNCTION_END(__kernel_gettimeofday)
  */
 V_FUNCTION_BEGIN(__kernel_clock_gettime)
   .cfi_startproc
-	/* Check for supported clock IDs */
-	cmpli	cr0,r3,CLOCK_REALTIME
-	cmpli	cr1,r3,CLOCK_MONOTONIC
-	cror	cr0*4+eq,cr0*4+eq,cr1*4+eq
-	bne	cr0,99f
-
-	mflr	r12			/* r12 saves lr */
-  .cfi_register lr,r12
-	mr	r11,r4			/* r11 saves tp */
-	bl	__get_datapage@local	/* get data page */
-	mr	r9,r3			/* datapage ptr in r9 */
-	lis	r7,NSEC_PER_SEC@h	/* want nanoseconds */
-	ori	r7,r7,NSEC_PER_SEC@l
-50:	bl	__do_get_tspec@local	/* get sec/nsec from tb & kernel */
-	bne	cr1,80f			/* 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_xsec.
-	 * At this point, r3,r4 contain our sec/nsec values, r5 and r6
-	 * can be used, r7 contains NSEC_PER_SEC.
-	 */
-
-	lwz	r5,(WTOM_CLOCK_SEC+LOPART)(r9)
-	lwz	r6,WTOM_CLOCK_NSEC(r9)
-
-	/* We now have our offset in r5,r6. We create a fake dependency
-	 * on that value and re-check the counter
-	 */
-	or	r0,r6,r5
-	xor	r0,r0,r0
-	add	r9,r9,r0
-	lwz	r0,(CFG_TB_UPDATE_COUNT+LOPART)(r9)
-        cmpl    cr0,r8,r0		/* check if updated */
-	bne-	50b
-
-	/* Calculate and store result. Note that this mimics the C code,
-	 * which may cause funny results if nsec goes negative... is that
-	 * possible at all ?
-	 */
-	add	r3,r3,r5
-	add	r4,r4,r6
-	cmpw	cr0,r4,r7
-	cmpwi	cr1,r4,0
-	blt	1f
-	subf	r4,r7,r4
-	addi	r3,r3,1
-1:	bge	cr1,80f
-	addi	r3,r3,-1
-	add	r4,r4,r7
-
-80:	stw	r3,TSPC32_TV_SEC(r11)
-	stw	r4,TSPC32_TV_NSEC(r11)
-
-	mtlr	r12
-	crclr	cr0*4+so
-	li	r3,0
-	blr
-
-	/*
-	 * syscall fallback
-	 */
-99:
-	li	r0,__NR_clock_gettime
-  .cfi_restore lr
-	sc
-	blr
+	cvdso_call __c_kernel_clock_gettime
   .cfi_endproc
 V_FUNCTION_END(__kernel_clock_gettime)
 
@@ -148,32 +63,7 @@ V_FUNCTION_END(__kernel_clock_gettime)
  */
 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	__get_datapage@local	/* get data page */
-	lwz	r5, CLOCK_HRTIMER_RES(r3)
-	mtlr	r12
-	li	r3,0
-	cmpli	cr0,r4,0
-	crclr	cr0*4+so
-	beqlr
-	stw	r3,TSPC32_TV_SEC(r4)
-	stw	r5,TSPC32_TV_NSEC(r4)
-	blr
-
-	/*
-	 * syscall fallback
-	 */
-99:
-	li	r0,__NR_clock_getres
-	sc
-	blr
+	cvdso_call __c_kernel_clock_getres
   .cfi_endproc
 V_FUNCTION_END(__kernel_clock_getres)
 
@@ -186,105 +76,14 @@ V_FUNCTION_END(__kernel_clock_getres)
  */
 V_FUNCTION_BEGIN(__kernel_time)
   .cfi_startproc
-	mflr	r12
-  .cfi_register lr,r12
-
-	mr	r11,r3			/* r11 holds t */
-	bl	__get_datapage@local
-	mr	r9, r3			/* datapage ptr in r9 */
-
-	lwz	r3,STAMP_XTIME_SEC+LOPART(r9)
-
-	cmplwi	r11,0			/* check if t is NULL */
-	beq	2f
-	stw	r3,0(r11)		/* store result at *t */
-2:	mtlr	r12
+	stwu	r1, -16(r1)
+	mflr	r0
+	stw	r0, 20(r1)
+	bl	__c_kernel_time
+	lwz	r0, 20(r1)
 	crclr	cr0*4+so
+	mtlr	r0
+	addi	r1, r1, 16
 	blr
   .cfi_endproc
 V_FUNCTION_END(__kernel_time)
-
-/*
- * This is the core of clock_gettime() and gettimeofday(),
- * it returns the current time in r3 (seconds) and r4.
- * On entry, r7 gives the resolution of r4, either USEC_PER_SEC
- * or NSEC_PER_SEC, giving r4 in microseconds or nanoseconds.
- * It expects the datapage ptr in r9 and doesn't clobber it.
- * It clobbers r0, r5 and r6.
- * On return, r8 contains the counter value that can be reused.
- * This clobbers cr0 but not any other cr field.
- */
-__do_get_tspec:
-  .cfi_startproc
-	/* Check for update count & load values. We use the low
-	 * order 32 bits of the update count
-	 */
-1:	lwz	r8,(CFG_TB_UPDATE_COUNT+LOPART)(r9)
-	andi.	r0,r8,1			/* pending update ? loop */
-	bne-	1b
-	xor	r0,r8,r8		/* create dependency */
-	add	r9,r9,r0
-
-	/* Load orig stamp (offset to TB) */
-	lwz	r5,CFG_TB_ORIG_STAMP(r9)
-	lwz	r6,(CFG_TB_ORIG_STAMP+4)(r9)
-
-	/* Get a stable TB value */
-2:	MFTBU(r3)
-	MFTBL(r4)
-	MFTBU(r0)
-	cmplw	cr0,r3,r0
-	bne-	2b
-
-	/* Subtract tb orig stamp and shift left 12 bits.
-	 */
-	subfc	r4,r6,r4
-	subfe	r0,r5,r3
-	slwi	r0,r0,12
-	rlwimi.	r0,r4,12,20,31
-	slwi	r4,r4,12
-
-	/*
-	 * Load scale factor & do multiplication.
-	 * We only use the high 32 bits of the tb_to_xs value.
-	 * Even with a 1GHz timebase clock, the high 32 bits of
-	 * tb_to_xs will be at least 4 million, so the error from
-	 * ignoring the low 32 bits will be no more than 0.25ppm.
-	 * The error will just make the clock run very very slightly
-	 * slow until the next time the kernel updates the VDSO data,
-	 * at which point the clock will catch up to the kernel's value,
-	 * so there is no long-term error accumulation.
-	 */
-	lwz	r5,CFG_TB_TO_XS(r9)	/* load values */
-	mulhwu	r4,r4,r5
-	li	r3,0
-
-	beq+	4f			/* skip high part computation if 0 */
-	mulhwu	r3,r0,r5
-	mullw	r5,r0,r5
-	addc	r4,r4,r5
-	addze	r3,r3
-4:
-	/* At this point, we have seconds since the xtime stamp
-	 * as a 32.32 fixed-point number in r3 and r4.
-	 * Load & add the xtime stamp.
-	 */
-	lwz	r5,STAMP_XTIME_SEC+LOPART(r9)
-	lwz	r6,STAMP_SEC_FRAC(r9)
-	addc	r4,r4,r6
-	adde	r3,r3,r5
-
-	/* We create a fake dependency on the result in r3/r4
-	 * and re-check the counter
-	 */
-	or	r6,r4,r3
-	xor	r0,r6,r6
-	add	r9,r9,r0
-	lwz	r0,(CFG_TB_UPDATE_COUNT+LOPART)(r9)
-        cmplw	cr0,r8,r0		/* check if updated */
-	bne-	1b
-
-	mulhwu	r4,r4,r7		/* convert to micro or nanoseconds */
-
-	blr
-  .cfi_endproc
diff --git a/arch/powerpc/kernel/vdso32/vgettimeofday.c b/arch/powerpc/kernel/vdso32/vgettimeofday.c
new file mode 100644
index 000000000000..4ed1bf2ae30e
--- /dev/null
+++ b/arch/powerpc/kernel/vdso32/vgettimeofday.c
@@ -0,0 +1,26 @@
+// SPDX-License-Identifier: GPL-2.0
+/*
+ * Powerpc userspace implementations of gettimeofday() and similar.
+ */
+#include <linux/time.h>
+#include <linux/types.h>
+
+int __c_kernel_clock_gettime(clockid_t clock, struct old_timespec32 *ts)
+{
+	return __cvdso_clock_gettime32(clock, ts);
+}
+
+int __c_kernel_gettimeofday(struct __kernel_old_timeval *tv, struct timezone *tz)
+{
+	return __cvdso_gettimeofday(tv, tz);
+}
+
+int __c_kernel_clock_getres(clockid_t clock_id, struct old_timespec32 *res)
+{
+	return __cvdso_clock_getres_time32(clock_id, res);
+}
+
+time_t __c_kernel_time(time_t *time)
+{
+	return __cvdso_time(time);
+}
diff --git a/arch/powerpc/kernel/vdso64/Makefile b/arch/powerpc/kernel/vdso64/Makefile
index 32ebb3522ea1..0d80069233e0 100644
--- a/arch/powerpc/kernel/vdso64/Makefile
+++ b/arch/powerpc/kernel/vdso64/Makefile
@@ -1,8 +1,20 @@
 # SPDX-License-Identifier: GPL-2.0
 # List of files in the vdso, has to be asm only for now
 
+ARCH_REL_TYPE_ABS := R_PPC_JUMP_SLOT|R_PPC_GLOB_DAT|R_PPC_ADDR32|R_PPC_ADDR24|R_PPC_ADDR16|R_PPC_ADDR16_LO|R_PPC_ADDR16_HI|R_PPC_ADDR16_HA|R_PPC_ADDR14|R_PPC_ADDR14_BRTAKEN|R_PPC_ADDR14_BRNTAKEN
+include $(srctree)/lib/vdso/Makefile
+
 obj-vdso64 = sigtramp.o gettimeofday.o datapage.o cacheflush.o note.o getcpu.o
 
+ifneq ($(c-gettimeofday-y),)
+  CFLAGS_vgettimeofday.o += -include $(c-gettimeofday-y)
+  CFLAGS_vgettimeofday.o += $(DISABLE_LATENT_ENTROPY_PLUGIN)
+  CFLAGS_vgettimeofday.o += $(call cc-option, -fno-stack-protector)
+  CFLAGS_vgettimeofday.o += -DDISABLE_BRANCH_PROFILING
+  CFLAGS_vgettimeofday.o += -ffreestanding
+  CFLAGS_REMOVE_vgettimeofday.o = $(CC_FLAGS_FTRACE)
+endif
+
 # Build rules
 
 targets := $(obj-vdso64) vdso64.so vdso64.so.dbg
@@ -11,6 +23,7 @@ obj-vdso64 := $(addprefix $(obj)/, $(obj-vdso64))
 GCOV_PROFILE := n
 KCOV_INSTRUMENT := n
 UBSAN_SANITIZE := n
+KASAN_SANITIZE := n
 
 ccflags-y := -shared -fno-common -fno-builtin -nostdlib \
 	-Wl,-soname=linux-vdso64.so.1 -Wl,--hash-style=both
@@ -20,12 +33,14 @@ obj-y += vdso64_wrapper.o
 extra-y += vdso64.lds
 CPPFLAGS_vdso64.lds += -P -C -U$(ARCH)
 
+$(obj)/vgettimeofday.o: %.o: %.c FORCE
+
 # Force dependency (incbin is bad)
 $(obj)/vdso64_wrapper.o : $(obj)/vdso64.so
 
 # link rule for the .so file, .lds has to be first
-$(obj)/vdso64.so.dbg: $(src)/vdso64.lds $(obj-vdso64) FORCE
-	$(call if_changed,vdso64ld)
+$(obj)/vdso64.so.dbg: $(src)/vdso64.lds $(obj-vdso64) $(obj)/vgettimeofday.o FORCE
+	$(call if_changed,vdso64ld_and_check)
 
 # strip rule for the .so file
 $(obj)/%.so: OBJCOPYFLAGS := -S
@@ -33,8 +48,8 @@ $(obj)/%.so: $(obj)/%.so.dbg FORCE
 	$(call if_changed,objcopy)
 
 # actual build commands
-quiet_cmd_vdso64ld = VDSO64L $@
-      cmd_vdso64ld = $(CC) $(c_flags) -o $@ -Wl,-T$(filter %.lds,$^) $(filter %.o,$^)
+quiet_cmd_vdso64ld_and_check = VDSO64L $@
+      cmd_vdso64ld_and_check = $(CC) $(c_flags) -o $@ -Wl,-T$(filter %.lds,$^) $(filter %.o,$^) ; $(cmd_vdso_check)
 
 # install commands for the unstripped file
 quiet_cmd_vdso_install = INSTALL $@
diff --git a/arch/powerpc/kernel/vdso64/datapage.S b/arch/powerpc/kernel/vdso64/datapage.S
index dc84f5ae3802..4808ce9eeb3d 100644
--- a/arch/powerpc/kernel/vdso64/datapage.S
+++ b/arch/powerpc/kernel/vdso64/datapage.S
@@ -16,7 +16,7 @@
 __kernel_datapage_offset:
 	.long	0
 
-V_FUNCTION_BEGIN(__get_datapage)
+_GLOBAL(__get_datapage)
   .cfi_startproc
 	/* We don't want that exposed or overridable as we want other objects
 	 * to be able to bl directly to here
@@ -37,7 +37,6 @@ data_page_branch:
 	add	r3,r0,r3
 	blr
   .cfi_endproc
-V_FUNCTION_END(__get_datapage)
 
 /*
  * void *__kernel_get_syscall_map(unsigned int *syscall_count) ;
diff --git a/arch/powerpc/kernel/vdso64/gettimeofday.S b/arch/powerpc/kernel/vdso64/gettimeofday.S
index 1c9a04703250..22f4f1f73bbc 100644
--- a/arch/powerpc/kernel/vdso64/gettimeofday.S
+++ b/arch/powerpc/kernel/vdso64/gettimeofday.S
@@ -13,6 +13,23 @@
 #include <asm/unistd.h>
 
 	.text
+
+.macro cvdso_call funct
+	mflr	r0
+	std	r0, 16(r1)
+	stdu	r1, -128(r1)
+	bl	\funct
+	addi	r1, r1, 128
+	ld	r0, 16(r1)
+	cmpwi	r3, 0
+	mtlr	r0
+	crclr	so
+	beqlr+
+	crset	so
+	neg	r3, r3
+	blr
+.endm
+
 /*
  * Exact prototype of gettimeofday
  *
@@ -21,29 +38,7 @@
  */
 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
-	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
+	cvdso_call __c_kernel_gettimeofday
   .cfi_endproc
 V_FUNCTION_END(__kernel_gettimeofday)
 
@@ -56,118 +51,7 @@ V_FUNCTION_END(__kernel_gettimeofday)
  */
 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 */
-	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.
-	 */
-
-	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 */
-	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
-  .cfi_restore lr
-	sc
-	blr
+	cvdso_call __c_kernel_clock_gettime
   .cfi_endproc
 V_FUNCTION_END(__kernel_clock_gettime)
 
@@ -180,32 +64,7 @@ V_FUNCTION_END(__kernel_clock_gettime)
  */
 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
+	cvdso_call __c_kernel_clock_getres
   .cfi_endproc
 V_FUNCTION_END(__kernel_clock_getres)
 
@@ -217,73 +76,14 @@ V_FUNCTION_END(__kernel_clock_getres)
  */
 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
+	mflr	r0
+	std	r0, 16(r1)
+	stdu	r1, -128(r1)
+	bl	__c_kernel_time
+	addi	r1, r1, 128
+	ld	r0, 16(r1)
 	crclr	cr0*4+so
-	mr	r3,r4
+	mtlr	r0
 	blr
   .cfi_endproc
 V_FUNCTION_END(__kernel_time)
-
-
-/*
- * 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.
- * 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.
-	 */
-	MFTB(r6)
-	ld	r9,CFG_TB_ORIG_STAMP(r3)
-	subf	r6,r9,r6
-
-	/* Scale result */
-	ld	r5,CFG_TB_TO_XS(r3)
-	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)
-	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 */
-	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 */
-	clrldi	r5,r5,32
-	add	r4,r4,r6
-	blr
-  .cfi_endproc
-V_FUNCTION_END(__do_get_tspec)
diff --git a/arch/powerpc/kernel/vdso64/vgettimeofday.c b/arch/powerpc/kernel/vdso64/vgettimeofday.c
new file mode 100644
index 000000000000..407c6a7ed4e2
--- /dev/null
+++ b/arch/powerpc/kernel/vdso64/vgettimeofday.c
@@ -0,0 +1,26 @@
+// SPDX-License-Identifier: GPL-2.0
+/*
+ * Powerpc userspace implementations of gettimeofday() and similar.
+ */
+#include <linux/time.h>
+#include <linux/types.h>
+
+int __c_kernel_clock_gettime(clockid_t clock, struct __kernel_timespec *ts)
+{
+	return __cvdso_clock_gettime(clock, ts);
+}
+
+int __c_kernel_gettimeofday(struct __kernel_old_timeval *tv, struct timezone *tz)
+{
+	return __cvdso_gettimeofday(tv, tz);
+}
+
+int __c_kernel_clock_getres(clockid_t clock_id, struct __kernel_timespec *res)
+{
+	return __cvdso_clock_getres(clock_id, res);
+}
+
+time_t __c_kernel_time(time_t *time)
+{
+	return __cvdso_time(time);
+}
-- 
2.13.3




^ permalink raw reply related	[flat|nested] 26+ messages in thread

* [RFC PATCH v4 03/11] lib: vdso: only read hrtimer_res when needed in __cvdso_clock_getres()
  2020-01-16 17:58 [RFC PATCH v4 00/11] powerpc: switch VDSO to C implementation Christophe Leroy
  2020-01-16 17:58 ` [RFC PATCH v4 01/11] powerpc/64: Don't provide time functions in compat VDSO32 Christophe Leroy
  2020-01-16 17:58 ` [RFC PATCH v4 02/11] powerpc/vdso: Switch VDSO to generic C implementation Christophe Leroy
@ 2020-01-16 17:58 ` Christophe Leroy
  2020-01-16 17:58 ` [RFC PATCH v4 04/11] powerpc/vdso: simplify __get_datapage() Christophe Leroy
                   ` (8 subsequent siblings)
  11 siblings, 0 replies; 26+ messages in thread
From: Christophe Leroy @ 2020-01-16 17:58 UTC (permalink / raw)
  To: Benjamin Herrenschmidt, Paul Mackerras, Michael Ellerman,
	nathanl, arnd, tglx, vincenzo.frascino, luto
  Cc: linux-kernel, linuxppc-dev, linux-arm-kernel, linux-mips, x86

Only perform READ_ONCE(vd[CS_HRES_COARSE].hrtimer_res) for
HRES and RAW clocks.

Signed-off-by: Christophe Leroy <christophe.leroy@c-s.fr>
---
 lib/vdso/gettimeofday.c | 4 +---
 1 file changed, 1 insertion(+), 3 deletions(-)

diff --git a/lib/vdso/gettimeofday.c b/lib/vdso/gettimeofday.c
index 9563be3cb5fa..8b3084d9a3ec 100644
--- a/lib/vdso/gettimeofday.c
+++ b/lib/vdso/gettimeofday.c
@@ -326,7 +326,6 @@ static __maybe_unused
 int __cvdso_clock_getres_common(clockid_t clock, struct __kernel_timespec *res)
 {
 	const struct vdso_data *vd = __arch_get_vdso_data();
-	u64 hrtimer_res;
 	u32 msk;
 	u64 ns;
 
@@ -338,7 +337,6 @@ int __cvdso_clock_getres_common(clockid_t clock, struct __kernel_timespec *res)
 	    vd->clock_mode == VDSO_CLOCKMODE_TIMENS)
 		vd = __arch_get_timens_vdso_data();
 
-	hrtimer_res = READ_ONCE(vd[CS_HRES_COARSE].hrtimer_res);
 	/*
 	 * Convert the clockid to a bitmask and use it to check which
 	 * clocks are handled in the VDSO directly.
@@ -348,7 +346,7 @@ int __cvdso_clock_getres_common(clockid_t clock, struct __kernel_timespec *res)
 		/*
 		 * Preserves the behaviour of posix_get_hrtimer_res().
 		 */
-		ns = hrtimer_res;
+		ns = READ_ONCE(vd[CS_HRES_COARSE].hrtimer_res);
 	} else if (msk & VDSO_COARSE) {
 		/*
 		 * Preserves the behaviour of posix_get_coarse_res().
-- 
2.13.3


^ permalink raw reply related	[flat|nested] 26+ messages in thread

* [RFC PATCH v4 04/11] powerpc/vdso: simplify __get_datapage()
  2020-01-16 17:58 [RFC PATCH v4 00/11] powerpc: switch VDSO to C implementation Christophe Leroy
                   ` (2 preceding siblings ...)
  2020-01-16 17:58 ` [RFC PATCH v4 03/11] lib: vdso: only read hrtimer_res when needed in __cvdso_clock_getres() Christophe Leroy
@ 2020-01-16 17:58 ` Christophe Leroy
  2020-01-16 17:58 ` [RFC PATCH v4 05/11] lib: vdso: allow arches to provide vdso data pointer Christophe Leroy
                   ` (7 subsequent siblings)
  11 siblings, 0 replies; 26+ messages in thread
From: Christophe Leroy @ 2020-01-16 17:58 UTC (permalink / raw)
  To: Benjamin Herrenschmidt, Paul Mackerras, Michael Ellerman,
	nathanl, arnd, tglx, vincenzo.frascino, luto
  Cc: linux-kernel, linuxppc-dev, linux-arm-kernel, linux-mips, x86

The VDSO datapage and the text pages are always located immediately
next to each other, so it can be hardcoded without an indirection
through __kernel_datapage_offset

In order to ease things, move the data page in front like other
arches, that way there is no need to know the size of the library
to locate the data page.

Before:
clock-getres-realtime-coarse:    vdso: 714 nsec/call
clock-gettime-realtime-coarse:    vdso: 792 nsec/call
clock-gettime-realtime:    vdso: 1243 nsec/call

After:
clock-getres-realtime-coarse:    vdso: 699 nsec/call
clock-gettime-realtime-coarse:    vdso: 784 nsec/call
clock-gettime-realtime:    vdso: 1231 nsec/call
Signed-off-by: Christophe Leroy <christophe.leroy@c-s.fr>
---
 arch/powerpc/kernel/vdso.c              | 53 +++++----------------------------
 arch/powerpc/kernel/vdso32/datapage.S   | 10 +++----
 arch/powerpc/kernel/vdso32/vdso32.lds.S |  7 ++---
 arch/powerpc/kernel/vdso64/datapage.S   | 10 +++----
 arch/powerpc/kernel/vdso64/vdso64.lds.S |  7 ++---
 5 files changed, 19 insertions(+), 68 deletions(-)

diff --git a/arch/powerpc/kernel/vdso.c b/arch/powerpc/kernel/vdso.c
index 16a44bffe698..c093d90a222a 100644
--- a/arch/powerpc/kernel/vdso.c
+++ b/arch/powerpc/kernel/vdso.c
@@ -191,7 +191,7 @@ int arch_setup_additional_pages(struct linux_binprm *bprm, int uses_interp)
 	 * install_special_mapping or the perf counter mmap tracking code
 	 * will fail to recognise it as a vDSO (since arch_vma_name fails).
 	 */
-	current->mm->context.vdso_base = vdso_base;
+	current->mm->context.vdso_base = vdso_base + PAGE_SIZE;
 
 	/*
 	 * our vma flags don't have VM_WRITE so by default, the process isn't
@@ -488,42 +488,6 @@ static __init void vdso_setup_trampolines(struct lib32_elfinfo *v32,
 	vdso32_rt_sigtramp = find_function32(v32, "__kernel_sigtramp_rt32");
 }
 
-static __init int vdso_fixup_datapage(struct lib32_elfinfo *v32,
-				       struct lib64_elfinfo *v64)
-{
-#ifdef CONFIG_VDSO32
-	Elf32_Sym *sym32;
-#endif
-#ifdef CONFIG_PPC64
-	Elf64_Sym *sym64;
-
-       	sym64 = find_symbol64(v64, "__kernel_datapage_offset");
-	if (sym64 == NULL) {
-		printk(KERN_ERR "vDSO64: Can't find symbol "
-		       "__kernel_datapage_offset !\n");
-		return -1;
-	}
-	*((int *)(vdso64_kbase + sym64->st_value - VDSO64_LBASE)) =
-		(vdso64_pages << PAGE_SHIFT) -
-		(sym64->st_value - VDSO64_LBASE);
-#endif /* CONFIG_PPC64 */
-
-#ifdef CONFIG_VDSO32
-	sym32 = find_symbol32(v32, "__kernel_datapage_offset");
-	if (sym32 == NULL) {
-		printk(KERN_ERR "vDSO32: Can't find symbol "
-		       "__kernel_datapage_offset !\n");
-		return -1;
-	}
-	*((int *)(vdso32_kbase + (sym32->st_value - VDSO32_LBASE))) =
-		(vdso32_pages << PAGE_SHIFT) -
-		(sym32->st_value - VDSO32_LBASE);
-#endif
-
-	return 0;
-}
-
-
 static __init int vdso_fixup_features(struct lib32_elfinfo *v32,
 				      struct lib64_elfinfo *v64)
 {
@@ -624,9 +588,6 @@ static __init int vdso_setup(void)
 	if (vdso_do_find_sections(&v32, &v64))
 		return -1;
 
-	if (vdso_fixup_datapage(&v32, &v64))
-		return -1;
-
 	if (vdso_fixup_features(&v32, &v64))
 		return -1;
 
@@ -771,26 +732,26 @@ static int __init vdso_init(void)
 	vdso32_pagelist = kcalloc(vdso32_pages + 2, sizeof(struct page *),
 				  GFP_KERNEL);
 	BUG_ON(vdso32_pagelist == NULL);
+	vdso32_pagelist[0] = virt_to_page(vdso_data);
 	for (i = 0; i < vdso32_pages; i++) {
 		struct page *pg = virt_to_page(vdso32_kbase + i*PAGE_SIZE);
 		get_page(pg);
-		vdso32_pagelist[i] = pg;
+		vdso32_pagelist[i + 1] = pg;
 	}
-	vdso32_pagelist[i++] = virt_to_page(vdso_data);
-	vdso32_pagelist[i] = NULL;
+	vdso32_pagelist[i + 1] = NULL;
 #endif
 
 #ifdef CONFIG_PPC64
 	vdso64_pagelist = kcalloc(vdso64_pages + 2, sizeof(struct page *),
 				  GFP_KERNEL);
 	BUG_ON(vdso64_pagelist == NULL);
+	vdso64_pagelist[0] = virt_to_page(vdso_data);
 	for (i = 0; i < vdso64_pages; i++) {
 		struct page *pg = virt_to_page(vdso64_kbase + i*PAGE_SIZE);
 		get_page(pg);
-		vdso64_pagelist[i] = pg;
+		vdso64_pagelist[i + 1] = pg;
 	}
-	vdso64_pagelist[i++] = virt_to_page(vdso_data);
-	vdso64_pagelist[i] = NULL;
+	vdso64_pagelist[i + 1] = NULL;
 #endif /* CONFIG_PPC64 */
 
 	get_page(virt_to_page(vdso_data));
diff --git a/arch/powerpc/kernel/vdso32/datapage.S b/arch/powerpc/kernel/vdso32/datapage.S
index 6c7401bd284e..d839aa1a4f01 100644
--- a/arch/powerpc/kernel/vdso32/datapage.S
+++ b/arch/powerpc/kernel/vdso32/datapage.S
@@ -12,9 +12,6 @@
 #include <asm/vdso.h>
 
 	.text
-	.global	__kernel_datapage_offset;
-__kernel_datapage_offset:
-	.long	0
 
 V_FUNCTION_BEGIN(__get_datapage)
   .cfi_startproc
@@ -31,10 +28,11 @@ V_FUNCTION_BEGIN(__get_datapage)
 data_page_branch:
 	mflr	r3
 	mtlr	r0
-	addi	r3, r3, __kernel_datapage_offset-data_page_branch
-	lwz	r0,0(r3)
+#if CONFIG_PPC_PAGE_SHIFT > 14
+	addis	r3, r3, (_vdso_datapage - data_page_branch)@ha
+#endif
+	addi	r3, r3, (_vdso_datapage - data_page_branch)@l
   .cfi_restore lr
-	add	r3,r0,r3
 	blr
   .cfi_endproc
 V_FUNCTION_END(__get_datapage)
diff --git a/arch/powerpc/kernel/vdso32/vdso32.lds.S b/arch/powerpc/kernel/vdso32/vdso32.lds.S
index 9400b182e163..f09c2354bd21 100644
--- a/arch/powerpc/kernel/vdso32/vdso32.lds.S
+++ b/arch/powerpc/kernel/vdso32/vdso32.lds.S
@@ -4,6 +4,7 @@
  * library
  */
 #include <asm/vdso.h>
+#include <asm/page.h>
 
 #ifdef __LITTLE_ENDIAN__
 OUTPUT_FORMAT("elf32-powerpcle", "elf32-powerpcle", "elf32-powerpcle")
@@ -15,6 +16,7 @@ ENTRY(_start)
 
 SECTIONS
 {
+	PROVIDE(_vdso_datapage = . - PAGE_SIZE);
 	. = VDSO32_LBASE + SIZEOF_HEADERS;
 
 	.hash          	: { *(.hash) }			:text
@@ -138,11 +140,6 @@ VERSION
 {
 	VDSO_VERSION_STRING {
 	global:
-		/*
-		 * Has to be there for the kernel to find
-		 */
-		__kernel_datapage_offset;
-
 		__kernel_get_syscall_map;
 #if defined(CONFIG_PPC32) && !defined(CONFIG_PPC_BOOK3S_601)
 		__kernel_gettimeofday;
diff --git a/arch/powerpc/kernel/vdso64/datapage.S b/arch/powerpc/kernel/vdso64/datapage.S
index 4808ce9eeb3d..8c41e1a93980 100644
--- a/arch/powerpc/kernel/vdso64/datapage.S
+++ b/arch/powerpc/kernel/vdso64/datapage.S
@@ -12,9 +12,6 @@
 #include <asm/vdso.h>
 
 	.text
-.global	__kernel_datapage_offset;
-__kernel_datapage_offset:
-	.long	0
 
 _GLOBAL(__get_datapage)
   .cfi_startproc
@@ -31,10 +28,11 @@ _GLOBAL(__get_datapage)
 data_page_branch:
 	mflr	r3
 	mtlr	r0
-	addi	r3, r3, __kernel_datapage_offset-data_page_branch
-	lwz	r0,0(r3)
+#if CONFIG_PPC_PAGE_SHIFT > 14
+	addis	r3, r3, (_vdso_datapage - data_page_branch)@ha
+#endif
+	addi	r3, r3, (_vdso_datapage - data_page_branch)@l
   .cfi_restore lr
-	add	r3,r0,r3
 	blr
   .cfi_endproc
 
diff --git a/arch/powerpc/kernel/vdso64/vdso64.lds.S b/arch/powerpc/kernel/vdso64/vdso64.lds.S
index 256fb9720298..f58c7e2e9cbd 100644
--- a/arch/powerpc/kernel/vdso64/vdso64.lds.S
+++ b/arch/powerpc/kernel/vdso64/vdso64.lds.S
@@ -4,6 +4,7 @@
  * library
  */
 #include <asm/vdso.h>
+#include <asm/page.h>
 
 #ifdef __LITTLE_ENDIAN__
 OUTPUT_FORMAT("elf64-powerpcle", "elf64-powerpcle", "elf64-powerpcle")
@@ -15,6 +16,7 @@ ENTRY(_start)
 
 SECTIONS
 {
+	PROVIDE(_vdso_datapage = . - PAGE_SIZE);
 	. = VDSO64_LBASE + SIZEOF_HEADERS;
 
 	.hash		: { *(.hash) }			:text
@@ -138,11 +140,6 @@ VERSION
 {
 	VDSO_VERSION_STRING {
 	global:
-		/*
-		 * Has to be there for the kernel to find
-		 */
-		__kernel_datapage_offset;
-
 		__kernel_get_syscall_map;
 		__kernel_gettimeofday;
 		__kernel_clock_gettime;
-- 
2.13.3


^ permalink raw reply related	[flat|nested] 26+ messages in thread

* [RFC PATCH v4 05/11] lib: vdso: allow arches to provide vdso data pointer
  2020-01-16 17:58 [RFC PATCH v4 00/11] powerpc: switch VDSO to C implementation Christophe Leroy
                   ` (3 preceding siblings ...)
  2020-01-16 17:58 ` [RFC PATCH v4 04/11] powerpc/vdso: simplify __get_datapage() Christophe Leroy
@ 2020-01-16 17:58 ` Christophe Leroy
  2020-01-16 17:58 ` [RFC PATCH v4 06/11] powerpc/vdso: provide inline alternative to __get_datapage() Christophe Leroy
                   ` (6 subsequent siblings)
  11 siblings, 0 replies; 26+ messages in thread
From: Christophe Leroy @ 2020-01-16 17:58 UTC (permalink / raw)
  To: Benjamin Herrenschmidt, Paul Mackerras, Michael Ellerman,
	nathanl, arnd, tglx, vincenzo.frascino, luto
  Cc: linux-kernel, linuxppc-dev, linux-arm-kernel, linux-mips, x86

On powerpc, __arch_get_vdso_data() clobbers the link register,
requiring the caller to save it.

As the parent function already has to set a stack frame and saves
the link register before calling the C vdso function, retriving the
vdso data pointer there is lighter.

Give arches the opportunity to hand the vdso data pointer
to C vdso functions.

Signed-off-by: Christophe Leroy <christophe.leroy@c-s.fr>
---
 lib/vdso/gettimeofday.c | 83 +++++++++++++++++++++++++++++++++++++++----------
 1 file changed, 67 insertions(+), 16 deletions(-)

diff --git a/lib/vdso/gettimeofday.c b/lib/vdso/gettimeofday.c
index 8b3084d9a3ec..9fa249809399 100644
--- a/lib/vdso/gettimeofday.c
+++ b/lib/vdso/gettimeofday.c
@@ -219,9 +219,9 @@ static __always_inline int do_coarse(const struct vdso_data *vd, clockid_t clk,
 }
 
 static __maybe_unused int
-__cvdso_clock_gettime_common(clockid_t clock, struct __kernel_timespec *ts)
+__cvdso_clock_gettime_common(const struct vdso_data *vd, clockid_t clock,
+			     struct __kernel_timespec *ts)
 {
-	const struct vdso_data *vd = __arch_get_vdso_data();
 	u32 msk;
 
 	/* Check for negative values or invalid clocks */
@@ -246,23 +246,33 @@ __cvdso_clock_gettime_common(clockid_t clock, struct __kernel_timespec *ts)
 }
 
 static __maybe_unused int
-__cvdso_clock_gettime(clockid_t clock, struct __kernel_timespec *ts)
+__cvdso_clock_gettime_data(const struct vdso_data *vd, clockid_t clock,
+			   struct __kernel_timespec *ts)
 {
-	int ret = __cvdso_clock_gettime_common(clock, ts);
+	int ret = __cvdso_clock_gettime_common(vd, clock, ts);
 
 	if (unlikely(ret))
 		return clock_gettime_fallback(clock, ts);
 	return 0;
 }
 
+static __maybe_unused int
+__cvdso_clock_gettime(clockid_t clock, struct __kernel_timespec *ts)
+{
+	const struct vdso_data *vd = __arch_get_vdso_data();
+
+	return __cvdso_clock_gettime_data(vd, clock, ts);
+}
+
 #ifdef BUILD_VDSO32
 static __maybe_unused int
-__cvdso_clock_gettime32(clockid_t clock, struct old_timespec32 *res)
+__cvdso_clock_gettime32_data(const struct vdso_data *vd, clockid_t clock,
+			     struct old_timespec32 *res)
 {
 	struct __kernel_timespec ts;
 	int ret;
 
-	ret = __cvdso_clock_gettime_common(clock, &ts);
+	ret = __cvdso_clock_gettime_common(vd, clock, &ts);
 
 	if (unlikely(ret))
 		return clock_gettime32_fallback(clock, res);
@@ -273,13 +283,21 @@ __cvdso_clock_gettime32(clockid_t clock, struct old_timespec32 *res)
 
 	return ret;
 }
-#endif /* BUILD_VDSO32 */
 
 static __maybe_unused int
-__cvdso_gettimeofday(struct __kernel_old_timeval *tv, struct timezone *tz)
+__cvdso_clock_gettime32(clockid_t clock, struct old_timespec32 *res)
 {
 	const struct vdso_data *vd = __arch_get_vdso_data();
 
+	return __cvdso_clock_gettime32_data(vd, clock, res);
+}
+#endif /* BUILD_VDSO32 */
+
+static __maybe_unused int
+__cvdso_gettimeofday_data(const struct vdso_data *vd,
+			  struct __kernel_old_timeval *tv, struct timezone *tz)
+{
+
 	if (likely(tv != NULL)) {
 		struct __kernel_timespec ts;
 
@@ -302,10 +320,18 @@ __cvdso_gettimeofday(struct __kernel_old_timeval *tv, struct timezone *tz)
 	return 0;
 }
 
-#ifdef VDSO_HAS_TIME
-static __maybe_unused __kernel_old_time_t __cvdso_time(__kernel_old_time_t *time)
+static __maybe_unused int
+__cvdso_gettimeofday(struct __kernel_old_timeval *tv, struct timezone *tz)
 {
 	const struct vdso_data *vd = __arch_get_vdso_data();
+
+	return __cvdso_gettimeofday_data(vd, tv, tz);
+}
+
+#ifdef VDSO_HAS_TIME
+static __maybe_unused __kernel_old_time_t
+__cvdso_time_data(const struct vdso_data *vd, __kernel_old_time_t *time)
+{
 	__kernel_old_time_t t;
 
 	if (IS_ENABLED(CONFIG_TIME_NS) &&
@@ -319,13 +345,20 @@ static __maybe_unused __kernel_old_time_t __cvdso_time(__kernel_old_time_t *time
 
 	return t;
 }
+
+static __maybe_unused __kernel_old_time_t __cvdso_time(__kernel_old_time_t *time)
+{
+	const struct vdso_data *vd = __arch_get_vdso_data();
+
+	return __cvdso_time_data(vd, time);
+}
 #endif /* VDSO_HAS_TIME */
 
 #ifdef VDSO_HAS_CLOCK_GETRES
 static __maybe_unused
-int __cvdso_clock_getres_common(clockid_t clock, struct __kernel_timespec *res)
+int __cvdso_clock_getres_common(const struct vdso_data *vd, clockid_t clock,
+				struct __kernel_timespec *res)
 {
-	const struct vdso_data *vd = __arch_get_vdso_data();
 	u32 msk;
 	u64 ns;
 
@@ -364,23 +397,33 @@ int __cvdso_clock_getres_common(clockid_t clock, struct __kernel_timespec *res)
 }
 
 static __maybe_unused
-int __cvdso_clock_getres(clockid_t clock, struct __kernel_timespec *res)
+int __cvdso_clock_getres_data(const struct vdso_data *vd, clockid_t clock,
+			      struct __kernel_timespec *res)
 {
-	int ret = __cvdso_clock_getres_common(clock, res);
+	int ret = __cvdso_clock_getres_common(vd, clock, res);
 
 	if (unlikely(ret))
 		return clock_getres_fallback(clock, res);
 	return 0;
 }
 
+static __maybe_unused
+int __cvdso_clock_getres(clockid_t clock, struct __kernel_timespec *res)
+{
+	const struct vdso_data *vd = __arch_get_vdso_data();
+
+	return __cvdso_clock_getres_data(vd, clock, res);
+}
+
 #ifdef BUILD_VDSO32
 static __maybe_unused int
-__cvdso_clock_getres_time32(clockid_t clock, struct old_timespec32 *res)
+__cvdso_clock_getres_time32_data(const struct vdso_data *vd, clockid_t clock,
+				 struct old_timespec32 *res)
 {
 	struct __kernel_timespec ts;
 	int ret;
 
-	ret = __cvdso_clock_getres_common(clock, &ts);
+	ret = __cvdso_clock_getres_common(vd, clock, &ts);
 
 	if (unlikely(ret))
 		return clock_getres32_fallback(clock, res);
@@ -391,5 +434,13 @@ __cvdso_clock_getres_time32(clockid_t clock, struct old_timespec32 *res)
 	}
 	return ret;
 }
+
+static __maybe_unused int
+__cvdso_clock_getres_time32(clockid_t clock, struct old_timespec32 *res)
+{
+	const struct vdso_data *vd = __arch_get_vdso_data();
+
+	return __cvdso_clock_getres_time32_data(vd, clock, res);
+}
 #endif /* BUILD_VDSO32 */
 #endif /* VDSO_HAS_CLOCK_GETRES */
-- 
2.13.3


^ permalink raw reply related	[flat|nested] 26+ messages in thread

* [RFC PATCH v4 06/11] powerpc/vdso: provide inline alternative to __get_datapage()
  2020-01-16 17:58 [RFC PATCH v4 00/11] powerpc: switch VDSO to C implementation Christophe Leroy
                   ` (4 preceding siblings ...)
  2020-01-16 17:58 ` [RFC PATCH v4 05/11] lib: vdso: allow arches to provide vdso data pointer Christophe Leroy
@ 2020-01-16 17:58 ` Christophe Leroy
  2020-01-16 17:58 ` [RFC PATCH v4 07/11] powerpc/vdso: provide vdso data pointer from the ASM caller Christophe Leroy
                   ` (5 subsequent siblings)
  11 siblings, 0 replies; 26+ messages in thread
From: Christophe Leroy @ 2020-01-16 17:58 UTC (permalink / raw)
  To: Benjamin Herrenschmidt, Paul Mackerras, Michael Ellerman,
	nathanl, arnd, tglx, vincenzo.frascino, luto
  Cc: linux-kernel, linuxppc-dev, linux-arm-kernel, linux-mips, x86

__get_datapage() is only a few instructions to retrieve the
address of the page where the kernel stores data to the VDSO.

By inlining this function into its users, a bl/blr pair and
a mflr/mtlr pair is avoided, plus a few reg moves.

The improvement is noticeable (about 55 nsec/call on an 8xx)

With current __get_datapage() function:
gettimeofday:    vdso: 731 nsec/call
clock-gettime-realtime-coarse:    vdso: 668 nsec/call
clock-gettime-monotonic-coarse:    vdso: 745 nsec/call

Using the __get_datapage macro provided by this patch:
gettimeofday:    vdso: 677 nsec/call
clock-gettime-realtime-coarse:    vdso: 613 nsec/call
clock-gettime-monotonic-coarse:    vdso: 690 nsec/call

Signed-off-by: Christophe Leroy <christophe.leroy@c-s.fr>
---
 arch/powerpc/include/asm/vdso_datapage.h | 11 +++++++++++
 1 file changed, 11 insertions(+)

diff --git a/arch/powerpc/include/asm/vdso_datapage.h b/arch/powerpc/include/asm/vdso_datapage.h
index 4d7965bf369e..7342cc0c1ae4 100644
--- a/arch/powerpc/include/asm/vdso_datapage.h
+++ b/arch/powerpc/include/asm/vdso_datapage.h
@@ -105,6 +105,17 @@ struct vdso_arch_data {
 
 extern struct vdso_arch_data *vdso_data;
 
+#else /* __ASSEMBLY__ */
+
+.macro get_datapage ptr, offset=0
+	bcl	20, 31, .+4
+	mflr	\ptr
+#if CONFIG_PPC_PAGE_SHIFT > 14
+	addis	\ptr, \ptr, (_vdso_datapage + \offset - (.-4))@ha
+#endif
+	addi	\ptr, \ptr, (_vdso_datapage + \offset - (.-4))@l
+.endm
+
 #endif /* __ASSEMBLY__ */
 
 #endif /* __KERNEL__ */
-- 
2.13.3


^ permalink raw reply related	[flat|nested] 26+ messages in thread

* [RFC PATCH v4 07/11] powerpc/vdso: provide vdso data pointer from the ASM caller.
  2020-01-16 17:58 [RFC PATCH v4 00/11] powerpc: switch VDSO to C implementation Christophe Leroy
                   ` (5 preceding siblings ...)
  2020-01-16 17:58 ` [RFC PATCH v4 06/11] powerpc/vdso: provide inline alternative to __get_datapage() Christophe Leroy
@ 2020-01-16 17:58 ` Christophe Leroy
  2020-01-16 17:58 ` [RFC PATCH v4 08/11] lib: vdso: allow fixed clock mode Christophe Leroy
                   ` (4 subsequent siblings)
  11 siblings, 0 replies; 26+ messages in thread
From: Christophe Leroy @ 2020-01-16 17:58 UTC (permalink / raw)
  To: Benjamin Herrenschmidt, Paul Mackerras, Michael Ellerman,
	nathanl, arnd, tglx, vincenzo.frascino, luto
  Cc: linux-kernel, linuxppc-dev, linux-arm-kernel, linux-mips, x86

__arch_get_vdso_data() clobbers the link register, requiring
the caller to save it.

As the ASM calling function already has to set a stack frame and
saves the link register before calling the C vdso function,
retriving the vdso data pointer there is lighter.

The improvement is significant:

Before:
gettimeofday:    vdso: 1212 nsec/call
clock-getres-realtime-coarse:    vdso: 714 nsec/call
clock-gettime-realtime-coarse:    vdso: 784 nsec/call
clock-getres-realtime:    vdso: 714 nsec/call

After:
gettimeofday:    vdso: 1094 nsec/call
getcpu:    vdso: not tested
clock-getres-realtime-coarse:    vdso: 545 nsec/call
clock-gettime-realtime-coarse:    vdso: 584 nsec/call
clock-getres-realtime:    vdso: 545 nsec/call

Signed-off-by: Christophe Leroy <christophe.leroy@c-s.fr>
---
 arch/powerpc/include/asm/vdso/gettimeofday.h | 10 +---------
 arch/powerpc/kernel/vdso32/gettimeofday.S    |  3 +++
 arch/powerpc/kernel/vdso32/vgettimeofday.c   | 19 +++++++++++--------
 arch/powerpc/kernel/vdso64/gettimeofday.S    |  3 +++
 arch/powerpc/kernel/vdso64/vgettimeofday.c   | 19 +++++++++++--------
 5 files changed, 29 insertions(+), 25 deletions(-)

diff --git a/arch/powerpc/include/asm/vdso/gettimeofday.h b/arch/powerpc/include/asm/vdso/gettimeofday.h
index c2cafd85d3cb..0a247f9b9af9 100644
--- a/arch/powerpc/include/asm/vdso/gettimeofday.h
+++ b/arch/powerpc/include/asm/vdso/gettimeofday.h
@@ -6,7 +6,6 @@
 
 #include <asm/time.h>
 #include <asm/unistd.h>
-#include <asm/vdso_datapage.h>
 #include <uapi/linux/time.h>
 
 #define VDSO_HAS_CLOCK_GETRES		1
@@ -73,14 +72,7 @@ static __always_inline u64 __arch_get_hw_counter(s32 clock_mode)
 	return get_tb();
 }
 
-void *__get_datapage(void);
-
-static __always_inline const struct vdso_data *__arch_get_vdso_data(void)
-{
-	struct vdso_arch_data *vdso_data = __get_datapage();
-
-	return vdso_data->data;
-}
+const struct vdso_data *__arch_get_vdso_data(void);
 
 /*
  * powerpc specific delta calculation.
diff --git a/arch/powerpc/kernel/vdso32/gettimeofday.S b/arch/powerpc/kernel/vdso32/gettimeofday.S
index ba0bd64b3da3..0d43878e462c 100644
--- a/arch/powerpc/kernel/vdso32/gettimeofday.S
+++ b/arch/powerpc/kernel/vdso32/gettimeofday.S
@@ -9,6 +9,7 @@
 #include <asm/processor.h>
 #include <asm/ppc_asm.h>
 #include <asm/vdso.h>
+#include <asm/vdso_datapage.h>
 #include <asm/asm-offsets.h>
 #include <asm/unistd.h>
 
@@ -18,6 +19,7 @@
 	stwu	r1, -16(r1)
 	mflr	r0
 	stw	r0, 20(r1)
+	get_datapage	r5, VDSO_DATA_OFFSET
 	bl	\funct
 	lwz	r0, 20(r1)
 	cmpwi	r3, 0
@@ -79,6 +81,7 @@ V_FUNCTION_BEGIN(__kernel_time)
 	stwu	r1, -16(r1)
 	mflr	r0
 	stw	r0, 20(r1)
+	get_datapage	r4, VDSO_DATA_OFFSET
 	bl	__c_kernel_time
 	lwz	r0, 20(r1)
 	crclr	cr0*4+so
diff --git a/arch/powerpc/kernel/vdso32/vgettimeofday.c b/arch/powerpc/kernel/vdso32/vgettimeofday.c
index 4ed1bf2ae30e..9bdc3c435846 100644
--- a/arch/powerpc/kernel/vdso32/vgettimeofday.c
+++ b/arch/powerpc/kernel/vdso32/vgettimeofday.c
@@ -5,22 +5,25 @@
 #include <linux/time.h>
 #include <linux/types.h>
 
-int __c_kernel_clock_gettime(clockid_t clock, struct old_timespec32 *ts)
+int __c_kernel_clock_gettime(clockid_t clock, struct old_timespec32 *ts,
+			     const struct vdso_data *vd)
 {
-	return __cvdso_clock_gettime32(clock, ts);
+	return __cvdso_clock_gettime32_data(vd, clock, ts);
 }
 
-int __c_kernel_gettimeofday(struct __kernel_old_timeval *tv, struct timezone *tz)
+int __c_kernel_gettimeofday(struct __kernel_old_timeval *tv, struct timezone *tz,
+			    const struct vdso_data *vd)
 {
-	return __cvdso_gettimeofday(tv, tz);
+	return __cvdso_gettimeofday_data(vd, tv, tz);
 }
 
-int __c_kernel_clock_getres(clockid_t clock_id, struct old_timespec32 *res)
+int __c_kernel_clock_getres(clockid_t clock_id, struct old_timespec32 *res,
+			    const struct vdso_data *vd)
 {
-	return __cvdso_clock_getres_time32(clock_id, res);
+	return __cvdso_clock_getres_time32_data(vd, clock_id, res);
 }
 
-time_t __c_kernel_time(time_t *time)
+time_t __c_kernel_time(time_t *time, const struct vdso_data *vd)
 {
-	return __cvdso_time(time);
+	return __cvdso_time_data(vd, time);
 }
diff --git a/arch/powerpc/kernel/vdso64/gettimeofday.S b/arch/powerpc/kernel/vdso64/gettimeofday.S
index 22f4f1f73bbc..f61c53eb6600 100644
--- a/arch/powerpc/kernel/vdso64/gettimeofday.S
+++ b/arch/powerpc/kernel/vdso64/gettimeofday.S
@@ -9,6 +9,7 @@
 #include <asm/processor.h>
 #include <asm/ppc_asm.h>
 #include <asm/vdso.h>
+#include <asm/vdso_datapage.h>
 #include <asm/asm-offsets.h>
 #include <asm/unistd.h>
 
@@ -18,6 +19,7 @@
 	mflr	r0
 	std	r0, 16(r1)
 	stdu	r1, -128(r1)
+	get_datapage	r5, VDSO_DATA_OFFSET
 	bl	\funct
 	addi	r1, r1, 128
 	ld	r0, 16(r1)
@@ -79,6 +81,7 @@ V_FUNCTION_BEGIN(__kernel_time)
 	mflr	r0
 	std	r0, 16(r1)
 	stdu	r1, -128(r1)
+	get_datapage	r4, VDSO_DATA_OFFSET
 	bl	__c_kernel_time
 	addi	r1, r1, 128
 	ld	r0, 16(r1)
diff --git a/arch/powerpc/kernel/vdso64/vgettimeofday.c b/arch/powerpc/kernel/vdso64/vgettimeofday.c
index 407c6a7ed4e2..cd80867c501c 100644
--- a/arch/powerpc/kernel/vdso64/vgettimeofday.c
+++ b/arch/powerpc/kernel/vdso64/vgettimeofday.c
@@ -5,22 +5,25 @@
 #include <linux/time.h>
 #include <linux/types.h>
 
-int __c_kernel_clock_gettime(clockid_t clock, struct __kernel_timespec *ts)
+int __c_kernel_clock_gettime(clockid_t clock, struct __kernel_timespec *ts,
+			     const struct vdso_data *vd)
 {
-	return __cvdso_clock_gettime(clock, ts);
+	return __cvdso_clock_gettime_data(vd, clock, ts);
 }
 
-int __c_kernel_gettimeofday(struct __kernel_old_timeval *tv, struct timezone *tz)
+int __c_kernel_gettimeofday(struct __kernel_old_timeval *tv, struct timezone *tz,
+			    const struct vdso_data *vd)
 {
-	return __cvdso_gettimeofday(tv, tz);
+	return __cvdso_gettimeofday_data(vd, tv, tz);
 }
 
-int __c_kernel_clock_getres(clockid_t clock_id, struct __kernel_timespec *res)
+int __c_kernel_clock_getres(clockid_t clock_id, struct __kernel_timespec *res,
+			    const struct vdso_data *vd)
 {
-	return __cvdso_clock_getres(clock_id, res);
+	return __cvdso_clock_getres_data(vd, clock_id, res);
 }
 
-time_t __c_kernel_time(time_t *time)
+time_t __c_kernel_time(time_t *time, const struct vdso_data *vd)
 {
-	return __cvdso_time(time);
+	return __cvdso_time_data(vd, time);
 }
-- 
2.13.3


^ permalink raw reply related	[flat|nested] 26+ messages in thread

* [RFC PATCH v4 08/11] lib: vdso: allow fixed clock mode
  2020-01-16 17:58 [RFC PATCH v4 00/11] powerpc: switch VDSO to C implementation Christophe Leroy
                   ` (6 preceding siblings ...)
  2020-01-16 17:58 ` [RFC PATCH v4 07/11] powerpc/vdso: provide vdso data pointer from the ASM caller Christophe Leroy
@ 2020-01-16 17:58 ` Christophe Leroy
  2020-01-16 20:13   ` Thomas Gleixner
  2020-01-16 17:58 ` [RFC PATCH v4 09/11] powerpc/vdso: override __arch_vdso_capable() Christophe Leroy
                   ` (3 subsequent siblings)
  11 siblings, 1 reply; 26+ messages in thread
From: Christophe Leroy @ 2020-01-16 17:58 UTC (permalink / raw)
  To: Benjamin Herrenschmidt, Paul Mackerras, Michael Ellerman,
	nathanl, arnd, tglx, vincenzo.frascino, luto
  Cc: linux-kernel, linuxppc-dev, linux-arm-kernel, linux-mips, x86

On arches like POWERPC, the clock is always the timebase, it
cannot be changed on the fly and it is always VDSO capable.

Therefore, give arches the opportunity to redefine the way
clock_mode is checked by moving the check into an overridable
__arch_vdso_capable() macro.

Signed-off-by: Christophe Leroy <christophe.leroy@c-s.fr>
---
 lib/vdso/gettimeofday.c | 11 +++++++++--
 1 file changed, 9 insertions(+), 2 deletions(-)

diff --git a/lib/vdso/gettimeofday.c b/lib/vdso/gettimeofday.c
index 9fa249809399..724b45c3e8ac 100644
--- a/lib/vdso/gettimeofday.c
+++ b/lib/vdso/gettimeofday.c
@@ -46,6 +46,13 @@ static inline bool __arch_vdso_hres_capable(void)
 }
 #endif
 
+#ifndef __arch_vdso_capable
+static inline bool __arch_vdso_capable(const struct vdso_data *vd)
+{
+	return vd->clock_mode != VDSO_CLOCKMODE_NONE;
+}
+#endif
+
 #ifdef CONFIG_TIME_NS
 static int do_hres_timens(const struct vdso_data *vdns, clockid_t clk,
 			  struct __kernel_timespec *ts)
@@ -66,7 +73,7 @@ static int do_hres_timens(const struct vdso_data *vdns, clockid_t clk,
 	do {
 		seq = vdso_read_begin(vd);
 
-		if (unlikely(vd->clock_mode == VDSO_CLOCKMODE_NONE))
+		if (unlikely(!__arch_vdso_capable(vd)))
 			return -1;
 
 		cycles = __arch_get_hw_counter(vd->clock_mode);
@@ -134,7 +141,7 @@ static __always_inline int do_hres(const struct vdso_data *vd, clockid_t clk,
 		}
 		smp_rmb();
 
-		if (unlikely(vd->clock_mode == VDSO_CLOCKMODE_NONE))
+		if (unlikely(!__arch_vdso_capable(vd)))
 			return -1;
 
 		cycles = __arch_get_hw_counter(vd->clock_mode);
-- 
2.13.3


^ permalink raw reply related	[flat|nested] 26+ messages in thread

* [RFC PATCH v4 09/11] powerpc/vdso: override __arch_vdso_capable()
  2020-01-16 17:58 [RFC PATCH v4 00/11] powerpc: switch VDSO to C implementation Christophe Leroy
                   ` (7 preceding siblings ...)
  2020-01-16 17:58 ` [RFC PATCH v4 08/11] lib: vdso: allow fixed clock mode Christophe Leroy
@ 2020-01-16 17:58 ` Christophe Leroy
  2020-01-16 17:58 ` [RFC PATCH v4 10/11] lib: vdso: Allow arches to override the ns shift operation Christophe Leroy
                   ` (2 subsequent siblings)
  11 siblings, 0 replies; 26+ messages in thread
From: Christophe Leroy @ 2020-01-16 17:58 UTC (permalink / raw)
  To: Benjamin Herrenschmidt, Paul Mackerras, Michael Ellerman,
	nathanl, arnd, tglx, vincenzo.frascino, luto
  Cc: linux-kernel, linuxppc-dev, linux-arm-kernel, linux-mips, x86

When the timebase is used, make __arch_vdso_capable() always
return true. When the RTC clock is used, make __arch_vdso_capable()
always return false.

Before the patch:
clock-gettime-realtime:    vdso: 1086 nsec/call

After the patch:
clock-gettime-realtime:    vdso: 1033 nsec/call

Signed-off-by: Christophe Leroy <christophe.leroy@c-s.fr>
---
 arch/powerpc/include/asm/vdso/gettimeofday.h | 6 ++++++
 1 file changed, 6 insertions(+)

diff --git a/arch/powerpc/include/asm/vdso/gettimeofday.h b/arch/powerpc/include/asm/vdso/gettimeofday.h
index 0a247f9b9af9..74b6eef8fbe9 100644
--- a/arch/powerpc/include/asm/vdso/gettimeofday.h
+++ b/arch/powerpc/include/asm/vdso/gettimeofday.h
@@ -74,6 +74,12 @@ static __always_inline u64 __arch_get_hw_counter(s32 clock_mode)
 
 const struct vdso_data *__arch_get_vdso_data(void);
 
+static inline bool __arch_vdso_capable(const struct vdso_data *vd)
+{
+	return !__USE_RTC();
+}
+#define __arch_vdso_capable __arch_vdso_capable
+
 /*
  * powerpc specific delta calculation.
  *
-- 
2.13.3


^ permalink raw reply related	[flat|nested] 26+ messages in thread

* [RFC PATCH v4 10/11] lib: vdso: Allow arches to override the ns shift operation
  2020-01-16 17:58 [RFC PATCH v4 00/11] powerpc: switch VDSO to C implementation Christophe Leroy
                   ` (8 preceding siblings ...)
  2020-01-16 17:58 ` [RFC PATCH v4 09/11] powerpc/vdso: override __arch_vdso_capable() Christophe Leroy
@ 2020-01-16 17:58 ` Christophe Leroy
  2020-01-16 19:47   ` Andy Lutomirski
  2020-01-16 17:58 ` [RFC PATCH v4 11/11] powerpc/32: provide vdso_shift_ns() Christophe Leroy
  2020-01-17  8:58 ` [RFC PATCH v4 00/11] powerpc: switch VDSO to C implementation Segher Boessenkool
  11 siblings, 1 reply; 26+ messages in thread
From: Christophe Leroy @ 2020-01-16 17:58 UTC (permalink / raw)
  To: Benjamin Herrenschmidt, Paul Mackerras, Michael Ellerman,
	nathanl, arnd, tglx, vincenzo.frascino, luto
  Cc: linux-kernel, linuxppc-dev, linux-arm-kernel, linux-mips, x86

On powerpc/32, GCC (8.1) generates pretty bad code for the
ns >>= vd->shift operation taking into account that the
shift is always < 32 and the upper part of the result is
likely to be nul. GCC makes reversed assumptions considering
the shift to be likely >= 32 and the upper part to be like not nul.

unsigned long long shift(unsigned long long x, unsigned char s)
{
	return x >> s;
}

results in:

00000018 <shift>:
  18:	35 25 ff e0 	addic.  r9,r5,-32
  1c:	41 80 00 10 	blt     2c <shift+0x14>
  20:	7c 64 4c 30 	srw     r4,r3,r9
  24:	38 60 00 00 	li      r3,0
  28:	4e 80 00 20 	blr
  2c:	54 69 08 3c 	rlwinm  r9,r3,1,0,30
  30:	21 45 00 1f 	subfic  r10,r5,31
  34:	7c 84 2c 30 	srw     r4,r4,r5
  38:	7d 29 50 30 	slw     r9,r9,r10
  3c:	7c 63 2c 30 	srw     r3,r3,r5
  40:	7d 24 23 78 	or      r4,r9,r4
  44:	4e 80 00 20 	blr

Even when forcing the shift with an &= 31, it still considers
the shift as likely >= 32.

Define a vdso_shift_ns() macro that can be overriden by
arches.

Signed-off-by: Christophe Leroy <christophe.leroy@c-s.fr>
---
 lib/vdso/gettimeofday.c | 9 ++++++++-
 1 file changed, 8 insertions(+), 1 deletion(-)

diff --git a/lib/vdso/gettimeofday.c b/lib/vdso/gettimeofday.c
index 724b45c3e8ac..9ba92058cfd7 100644
--- a/lib/vdso/gettimeofday.c
+++ b/lib/vdso/gettimeofday.c
@@ -39,6 +39,13 @@ u64 vdso_calc_delta(u64 cycles, u64 last, u64 mask, u32 mult)
 }
 #endif
 
+#ifndef vdso_shift_ns
+static __always_inline u64 vdso_shift_ns(u64 ns, unsigned long shift)
+{
+	return ns >> shift;
+}
+#endif
+
 #ifndef __arch_vdso_hres_capable
 static inline bool __arch_vdso_hres_capable(void)
 {
@@ -148,7 +155,7 @@ static __always_inline int do_hres(const struct vdso_data *vd, clockid_t clk,
 		ns = vdso_ts->nsec;
 		last = vd->cycle_last;
 		ns += vdso_calc_delta(cycles, last, vd->mask, vd->mult);
-		ns >>= vd->shift;
+		ns = vdso_shift_ns(ns, vd->shift);
 		sec = vdso_ts->sec;
 	} while (unlikely(vdso_read_retry(vd, seq)));
 
-- 
2.13.3


^ permalink raw reply related	[flat|nested] 26+ messages in thread

* [RFC PATCH v4 11/11] powerpc/32: provide vdso_shift_ns()
  2020-01-16 17:58 [RFC PATCH v4 00/11] powerpc: switch VDSO to C implementation Christophe Leroy
                   ` (9 preceding siblings ...)
  2020-01-16 17:58 ` [RFC PATCH v4 10/11] lib: vdso: Allow arches to override the ns shift operation Christophe Leroy
@ 2020-01-16 17:58 ` Christophe Leroy
  2020-01-17  8:58 ` [RFC PATCH v4 00/11] powerpc: switch VDSO to C implementation Segher Boessenkool
  11 siblings, 0 replies; 26+ messages in thread
From: Christophe Leroy @ 2020-01-16 17:58 UTC (permalink / raw)
  To: Benjamin Herrenschmidt, Paul Mackerras, Michael Ellerman,
	nathanl, arnd, tglx, vincenzo.frascino, luto
  Cc: linux-kernel, linuxppc-dev, linux-arm-kernel, linux-mips, x86

The generic x >> s gives the following result:

  18:	35 25 ff e0 	addic.  r9,r5,-32
  1c:	41 80 00 10 	blt     2c <shift+0x14>
  20:	7c 64 4c 30 	srw     r4,r3,r9
  24:	38 60 00 00 	li      r3,0
...
  2c:	54 69 08 3c 	rlwinm  r9,r3,1,0,30
  30:	21 45 00 1f 	subfic  r10,r5,31
  34:	7c 84 2c 30 	srw     r4,r4,r5
  38:	7d 29 50 30 	slw     r9,r9,r10
  3c:	7c 63 2c 30 	srw     r3,r3,r5
  40:	7d 24 23 78 	or      r4,r9,r4

In our case the shift is always < 32. In addition,  the upper 32 bits
of the result are likely nul. Lets GCC know it, it also optimises the
following calculations.

With the patch, we get:
   0:	21 25 00 20 	subfic  r9,r5,32
   4:	7c 69 48 30 	slw     r9,r3,r9
   8:	7c 84 2c 30 	srw     r4,r4,r5
   c:	7d 24 23 78 	or      r4,r9,r4
  10:	7c 63 2c 30 	srw     r3,r3,r5

Performance before the patch:
clock-gettime-realtime:    vdso: 1033 nsec/call

After the patch:
clock-gettime-realtime:    vdso: 941 nsec/call

Signed-off-by: Christophe Leroy <christophe.leroy@c-s.fr>
---
 arch/powerpc/include/asm/vdso/gettimeofday.h | 17 +++++++++++++++++
 1 file changed, 17 insertions(+)

diff --git a/arch/powerpc/include/asm/vdso/gettimeofday.h b/arch/powerpc/include/asm/vdso/gettimeofday.h
index 74b6eef8fbe9..716a137ab166 100644
--- a/arch/powerpc/include/asm/vdso/gettimeofday.h
+++ b/arch/powerpc/include/asm/vdso/gettimeofday.h
@@ -95,6 +95,23 @@ static __always_inline u64 vdso_calc_delta(u64 cycles, u64 last, u64 mask, u32 m
 }
 #define vdso_calc_delta vdso_calc_delta
 
+#ifndef __powerpc64__
+static __always_inline u64 vdso_shift_ns(u64 ns, unsigned long shift)
+{
+	u32 hi = ns >> 32;
+	u32 lo = ns;
+
+	lo = (lo >> shift) | (hi << (32 - shift));
+	hi >>= shift;
+
+	if (likely(hi == 0))
+		return lo;
+
+	return ((u64)hi << 32) | lo;
+}
+#define vdso_shift_ns vdso_shift_ns
+#endif
+
 #endif /* !__ASSEMBLY__ */
 
 #endif /* __ASM_VDSO_GETTIMEOFDAY_H */
-- 
2.13.3


^ permalink raw reply related	[flat|nested] 26+ messages in thread

* Re: [RFC PATCH v4 10/11] lib: vdso: Allow arches to override the ns shift operation
  2020-01-16 17:58 ` [RFC PATCH v4 10/11] lib: vdso: Allow arches to override the ns shift operation Christophe Leroy
@ 2020-01-16 19:47   ` Andy Lutomirski
  2020-01-16 19:57     ` Thomas Gleixner
  0 siblings, 1 reply; 26+ messages in thread
From: Andy Lutomirski @ 2020-01-16 19:47 UTC (permalink / raw)
  To: Christophe Leroy
  Cc: Benjamin Herrenschmidt, Paul Mackerras, Michael Ellerman,
	nathanl, Arnd Bergmann, Thomas Gleixner, Vincenzo Frascino,
	Andrew Lutomirski, LKML, linuxppc-dev, linux-arm-kernel,
	open list:MIPS, X86 ML

On Thu, Jan 16, 2020 at 9:58 AM Christophe Leroy
<christophe.leroy@c-s.fr> wrote:
>
> On powerpc/32, GCC (8.1) generates pretty bad code for the
> ns >>= vd->shift operation taking into account that the
> shift is always < 32 and the upper part of the result is
> likely to be nul. GCC makes reversed assumptions considering
> the shift to be likely >= 32 and the upper part to be like not nul.
>
> unsigned long long shift(unsigned long long x, unsigned char s)
> {
>         return x >> s;
> }
>
> results in:
>
> 00000018 <shift>:
>   18:   35 25 ff e0     addic.  r9,r5,-32
>   1c:   41 80 00 10     blt     2c <shift+0x14>
>   20:   7c 64 4c 30     srw     r4,r3,r9
>   24:   38 60 00 00     li      r3,0
>   28:   4e 80 00 20     blr
>   2c:   54 69 08 3c     rlwinm  r9,r3,1,0,30
>   30:   21 45 00 1f     subfic  r10,r5,31
>   34:   7c 84 2c 30     srw     r4,r4,r5
>   38:   7d 29 50 30     slw     r9,r9,r10
>   3c:   7c 63 2c 30     srw     r3,r3,r5
>   40:   7d 24 23 78     or      r4,r9,r4
>   44:   4e 80 00 20     blr
>
> Even when forcing the shift with an &= 31, it still considers
> the shift as likely >= 32.
>
> Define a vdso_shift_ns() macro that can be overriden by
> arches.

Would mul_u64_u64_shr() be a good alternative?  Could we adjust it to
assume the shift is less than 32?  That function exists to benefit
32-bit arches.

--Andy

^ permalink raw reply	[flat|nested] 26+ messages in thread

* Re: [RFC PATCH v4 10/11] lib: vdso: Allow arches to override the ns shift operation
  2020-01-16 19:47   ` Andy Lutomirski
@ 2020-01-16 19:57     ` Thomas Gleixner
  2020-01-16 20:20       ` Andy Lutomirski
  0 siblings, 1 reply; 26+ messages in thread
From: Thomas Gleixner @ 2020-01-16 19:57 UTC (permalink / raw)
  To: Andy Lutomirski, Christophe Leroy
  Cc: Benjamin Herrenschmidt, Paul Mackerras, Michael Ellerman,
	nathanl, Arnd Bergmann, Vincenzo Frascino, Andrew Lutomirski,
	LKML, linuxppc-dev, linux-arm-kernel, open list:MIPS, X86 ML

Andy Lutomirski <luto@kernel.org> writes:
> On Thu, Jan 16, 2020 at 9:58 AM Christophe Leroy
>
> Would mul_u64_u64_shr() be a good alternative?  Could we adjust it to
> assume the shift is less than 32?  That function exists to benefit
> 32-bit arches.

We'd want mul_u64_u32_shr() for this. The rules for mult and shift are:

     1 <= mult <= U32_MAX

     1 <= shift <= 32

If we want to enforce a shift < 32 we need to limit that conditionally
in the calculation/registration function.

Thanks,

        tglx


^ permalink raw reply	[flat|nested] 26+ messages in thread

* Re: [RFC PATCH v4 08/11] lib: vdso: allow fixed clock mode
  2020-01-16 17:58 ` [RFC PATCH v4 08/11] lib: vdso: allow fixed clock mode Christophe Leroy
@ 2020-01-16 20:13   ` Thomas Gleixner
  2020-01-16 20:19     ` Andy Lutomirski
  0 siblings, 1 reply; 26+ messages in thread
From: Thomas Gleixner @ 2020-01-16 20:13 UTC (permalink / raw)
  To: Christophe Leroy, Benjamin Herrenschmidt, Paul Mackerras,
	Michael Ellerman, nathanl, arnd, vincenzo.frascino, luto
  Cc: linux-kernel, linuxppc-dev, linux-arm-kernel, linux-mips, x86

Christophe Leroy <christophe.leroy@c-s.fr> writes:

Can you please adjust the prefix for future patches to lib/vdso: and
start the sentence after the colon with an uppercase letter?

> On arches like POWERPC, the clock is always the timebase, it

Please spell out architectures. Changelogs are not space constraint.

> cannot be changed on the fly and it is always VDSO capable.

Also this sentence does not make sense as it might suggests that
architectures with a fixed compile time known clocksource have something
named timebase. Something like this is more clear:

Some architectures have a fixed clocksource which is known at compile
time and cannot be replaced or disabled at runtime, e.g. timebase on
PowerPC. For such cases the clock mode check in the VDSO code is
pointless.

Hmm?

Thanks,

        tglx

^ permalink raw reply	[flat|nested] 26+ messages in thread

* Re: [RFC PATCH v4 08/11] lib: vdso: allow fixed clock mode
  2020-01-16 20:13   ` Thomas Gleixner
@ 2020-01-16 20:19     ` Andy Lutomirski
  2020-01-16 21:07       ` Thomas Gleixner
  0 siblings, 1 reply; 26+ messages in thread
From: Andy Lutomirski @ 2020-01-16 20:19 UTC (permalink / raw)
  To: Thomas Gleixner
  Cc: Christophe Leroy, Benjamin Herrenschmidt, Paul Mackerras,
	Michael Ellerman, nathanl, Arnd Bergmann, Vincenzo Frascino,
	Andrew Lutomirski, LKML, linuxppc-dev, linux-arm-kernel,
	open list:MIPS, X86 ML

On Thu, Jan 16, 2020 at 12:14 PM Thomas Gleixner <tglx@linutronix.de> wrote:
>
> Christophe Leroy <christophe.leroy@c-s.fr> writes:
>
> Can you please adjust the prefix for future patches to lib/vdso: and
> start the sentence after the colon with an uppercase letter?
>
> > On arches like POWERPC, the clock is always the timebase, it
>
> Please spell out architectures. Changelogs are not space constraint.
>
> > cannot be changed on the fly and it is always VDSO capable.
>
> Also this sentence does not make sense as it might suggests that
> architectures with a fixed compile time known clocksource have something
> named timebase. Something like this is more clear:
>
> Some architectures have a fixed clocksource which is known at compile
> time and cannot be replaced or disabled at runtime, e.g. timebase on
> PowerPC. For such cases the clock mode check in the VDSO code is
> pointless.
>

I wonder if we should use this on x86 bare-metal if we have
sufficiently invariant TSC.  (Via static_cpu_has(), not compiled in.)
Maybe there is no such x86 machine.

I really really want Intel or AMD to introduce machines where the TSC
pinky-swears to count in actual nanoseconds.

--Andy

^ permalink raw reply	[flat|nested] 26+ messages in thread

* Re: [RFC PATCH v4 10/11] lib: vdso: Allow arches to override the ns shift operation
  2020-01-16 19:57     ` Thomas Gleixner
@ 2020-01-16 20:20       ` Andy Lutomirski
  2020-01-29  7:14         ` Thomas Gleixner
  0 siblings, 1 reply; 26+ messages in thread
From: Andy Lutomirski @ 2020-01-16 20:20 UTC (permalink / raw)
  To: Thomas Gleixner
  Cc: Christophe Leroy, Benjamin Herrenschmidt, Paul Mackerras,
	Michael Ellerman, nathanl, Arnd Bergmann, Vincenzo Frascino,
	Andrew Lutomirski, LKML, linuxppc-dev, linux-arm-kernel,
	open list:MIPS, X86 ML

On Thu, Jan 16, 2020 at 11:57 AM Thomas Gleixner <tglx@linutronix.de> wrote:
>
> Andy Lutomirski <luto@kernel.org> writes:
> > On Thu, Jan 16, 2020 at 9:58 AM Christophe Leroy
> >
> > Would mul_u64_u64_shr() be a good alternative?  Could we adjust it to
> > assume the shift is less than 32?  That function exists to benefit
> > 32-bit arches.
>
> We'd want mul_u64_u32_shr() for this. The rules for mult and shift are:
>

That's what I meant to type...

>      1 <= mult <= U32_MAX
>
>      1 <= shift <= 32
>
> If we want to enforce a shift < 32 we need to limit that conditionally
> in the calculation/registration function.
>
> Thanks,
>
>         tglx
>

^ permalink raw reply	[flat|nested] 26+ messages in thread

* Re: [RFC PATCH v4 08/11] lib: vdso: allow fixed clock mode
  2020-01-16 20:19     ` Andy Lutomirski
@ 2020-01-16 21:07       ` Thomas Gleixner
  0 siblings, 0 replies; 26+ messages in thread
From: Thomas Gleixner @ 2020-01-16 21:07 UTC (permalink / raw)
  To: Andy Lutomirski
  Cc: Christophe Leroy, Benjamin Herrenschmidt, Paul Mackerras,
	Michael Ellerman, nathanl, Arnd Bergmann, Vincenzo Frascino,
	Andrew Lutomirski, LKML, linuxppc-dev, linux-arm-kernel,
	open list:MIPS, X86 ML

Andy Lutomirski <luto@kernel.org> writes:
> On Thu, Jan 16, 2020 at 12:14 PM Thomas Gleixner <tglx@linutronix.de> wrote:
>> Some architectures have a fixed clocksource which is known at compile
>> time and cannot be replaced or disabled at runtime, e.g. timebase on
>> PowerPC. For such cases the clock mode check in the VDSO code is
>> pointless.
>>
> I wonder if we should use this on x86 bare-metal if we have
> sufficiently invariant TSC.  (Via static_cpu_has(), not compiled in.)
>
> Maybe there is no such x86 machine.

There might be some, but every time I started to trust the TSC a bit
more someone reported the next variant of brokenness.

Admittedly it has become better at least up to two sockets.

For a start we could do that when the TSC is considered reliable, which
is the case when:

  - The TSC is the only available clocksource

  - tsc=reliable is on the kernel command line

> I really really want Intel or AMD to introduce machines where the TSC
> pinky-swears to count in actual nanoseconds.

and is guaranteed to be synchronized across any number of sockets/cpus
and has an enforcable protection against BIOS writers.

Ideally it'd have a writeable MSR attached which allows us to tweak the
frequency in the PPM range via NTP/PTP.

Guess how long quite some people including Linus and myself are asking
for this?

I know that Linus started bitching about the TSC before me, but it's
already a bit over 20 years on my side when I first talked to Intel and
AMD about the requirements for a reliable clocksource.

Just to set the time lines straight.

Constant frequency TSC surfaced on Intel in 2006 with the Core brand and
on AMD in 2007 with Barcelona (Fam 10h).

In 2008 the first TSC surfaced which was not affected by C-States and 5
years later in 2013 some Atoms came out where TSC even worked accross
S3.

The > 2 socket issue is still not resolved AFAICT, but we got at least
the TSC ADJUST MSR around 2012 which allowed us for the first time to
reliably detect and mitigate BIOS wreckage.

All the years I was envy on architectures which had simple designed and
just reliably working timers forever.

So now you can extrapolate how long it will take until you get your
pinky-swearing pony :)

Thanks,

        tglx

^ permalink raw reply	[flat|nested] 26+ messages in thread

* Re: [RFC PATCH v4 00/11] powerpc: switch VDSO to C implementation.
  2020-01-16 17:58 [RFC PATCH v4 00/11] powerpc: switch VDSO to C implementation Christophe Leroy
                   ` (10 preceding siblings ...)
  2020-01-16 17:58 ` [RFC PATCH v4 11/11] powerpc/32: provide vdso_shift_ns() Christophe Leroy
@ 2020-01-17  8:58 ` Segher Boessenkool
  2020-01-17  9:26   ` Christophe Leroy
  2020-01-20 14:56   ` Christophe Leroy
  11 siblings, 2 replies; 26+ messages in thread
From: Segher Boessenkool @ 2020-01-17  8:58 UTC (permalink / raw)
  To: Christophe Leroy
  Cc: Benjamin Herrenschmidt, Paul Mackerras, Michael Ellerman,
	nathanl, arnd, tglx, vincenzo.frascino, luto, x86, linuxppc-dev,
	linux-kernel, linux-arm-kernel, linux-mips

Hi!

On Thu, Jan 16, 2020 at 05:58:24PM +0000, Christophe Leroy wrote:
> On a powerpc8xx, with current powerpc/32 ASM VDSO:
> 
> gettimeofday:    vdso: 907 nsec/call
> clock-getres-realtime:    vdso: 484 nsec/call
> clock-gettime-realtime:    vdso: 899 nsec/call
> 
> The first patch adds VDSO generic C support without any changes to common code.
> Performance is as follows:
> 
> gettimeofday:    vdso: 1211 nsec/call
> clock-getres-realtime:    vdso: 722 nsec/call
> clock-gettime-realtime:    vdso: 1216 nsec/call
> 
> Then a few changes in the common code have allowed performance improvement. At
> the end of the series we have:
> 
> gettimeofday:    vdso: 974 nsec/call
> clock-getres-realtime:    vdso: 545 nsec/call
> clock-gettime-realtime:    vdso: 941 nsec/call
> 
> The final result is rather close to pure ASM VDSO:
> * 7% more on gettimeofday (9 cycles)
> * 5% more on clock-gettime-realtime (6 cycles)
> * 12% more on clock-getres-realtime (8 cycles)

Nice!  Much better.

It should be tested on more representative hardware, too, but this looks
promising alright :-)


Segher

^ permalink raw reply	[flat|nested] 26+ messages in thread

* Re: [RFC PATCH v4 00/11] powerpc: switch VDSO to C implementation.
  2020-01-17  8:58 ` [RFC PATCH v4 00/11] powerpc: switch VDSO to C implementation Segher Boessenkool
@ 2020-01-17  9:26   ` Christophe Leroy
  2020-01-20 14:56   ` Christophe Leroy
  1 sibling, 0 replies; 26+ messages in thread
From: Christophe Leroy @ 2020-01-17  9:26 UTC (permalink / raw)
  To: Segher Boessenkool
  Cc: Benjamin Herrenschmidt, Paul Mackerras, Michael Ellerman,
	nathanl, arnd, tglx, vincenzo.frascino, luto, x86, linuxppc-dev,
	linux-kernel, linux-arm-kernel, linux-mips



Le 17/01/2020 à 09:58, Segher Boessenkool a écrit :
> Hi!
> 
> On Thu, Jan 16, 2020 at 05:58:24PM +0000, Christophe Leroy wrote:
>> On a powerpc8xx, with current powerpc/32 ASM VDSO:
>>
>> gettimeofday:    vdso: 907 nsec/call
>> clock-getres-realtime:    vdso: 484 nsec/call
>> clock-gettime-realtime:    vdso: 899 nsec/call
>>
>> The first patch adds VDSO generic C support without any changes to common code.
>> Performance is as follows:
>>
>> gettimeofday:    vdso: 1211 nsec/call
>> clock-getres-realtime:    vdso: 722 nsec/call
>> clock-gettime-realtime:    vdso: 1216 nsec/call
>>
>> Then a few changes in the common code have allowed performance improvement. At
>> the end of the series we have:
>>
>> gettimeofday:    vdso: 974 nsec/call
>> clock-getres-realtime:    vdso: 545 nsec/call
>> clock-gettime-realtime:    vdso: 941 nsec/call
>>
>> The final result is rather close to pure ASM VDSO:
>> * 7% more on gettimeofday (9 cycles)
>> * 5% more on clock-gettime-realtime (6 cycles)
>> * 12% more on clock-getres-realtime (8 cycles)
> 
> Nice!  Much better.
> 
> It should be tested on more representative hardware, too, but this looks
> promising alright :-)
> 

Yes.

Now the challenge is to get VDSO32 buildable on PPC64. The big issue is 
that in most powerpc/include/asm/*.h , CONFIG_PPC64 is used to know if 
the build is a 64 bits build or a 32 bits build, so VDSO32 build fails.

I don't know how this could be easily fixed.

Christophe

^ permalink raw reply	[flat|nested] 26+ messages in thread

* Re: [RFC PATCH v4 00/11] powerpc: switch VDSO to C implementation.
  2020-01-17  8:58 ` [RFC PATCH v4 00/11] powerpc: switch VDSO to C implementation Segher Boessenkool
  2020-01-17  9:26   ` Christophe Leroy
@ 2020-01-20 14:56   ` Christophe Leroy
  2020-01-20 15:19     ` Segher Boessenkool
  1 sibling, 1 reply; 26+ messages in thread
From: Christophe Leroy @ 2020-01-20 14:56 UTC (permalink / raw)
  To: Segher Boessenkool
  Cc: Benjamin Herrenschmidt, Paul Mackerras, Michael Ellerman,
	nathanl, arnd, tglx, vincenzo.frascino, luto, x86, linuxppc-dev,
	linux-kernel, linux-arm-kernel, linux-mips

Hi

On 01/17/2020 08:58 AM, Segher Boessenkool wrote:
> Hi!
> 
> On Thu, Jan 16, 2020 at 05:58:24PM +0000, Christophe Leroy wrote:
>> On a powerpc8xx, with current powerpc/32 ASM VDSO:
>>
>> gettimeofday:    vdso: 907 nsec/call
>> clock-getres-realtime:    vdso: 484 nsec/call
>> clock-gettime-realtime:    vdso: 899 nsec/call
>>
>> The first patch adds VDSO generic C support without any changes to common code.
>> Performance is as follows:
>>
>> gettimeofday:    vdso: 1211 nsec/call
>> clock-getres-realtime:    vdso: 722 nsec/call
>> clock-gettime-realtime:    vdso: 1216 nsec/call
>>
>> Then a few changes in the common code have allowed performance improvement. At
>> the end of the series we have:
>>
>> gettimeofday:    vdso: 974 nsec/call
>> clock-getres-realtime:    vdso: 545 nsec/call
>> clock-gettime-realtime:    vdso: 941 nsec/call
>>
>> The final result is rather close to pure ASM VDSO:
>> * 7% more on gettimeofday (9 cycles)
>> * 5% more on clock-gettime-realtime (6 cycles)
>> * 12% more on clock-getres-realtime (8 cycles)
> 
> Nice!  Much better.
> 
> It should be tested on more representative hardware, too, but this looks
> promising alright :-)
> 

mpc832x (e300c2 core) at 333 MHz:

Before:

gettimeofday:    vdso: 235 nsec/call
clock-getres-realtime-coarse:    vdso: 1668 nsec/call
clock-gettime-realtime-coarse:    vdso: 1338 nsec/call
clock-getres-realtime:    vdso: 135 nsec/call
clock-gettime-realtime:    vdso: 244 nsec/call
clock-getres-boottime:    vdso: 1232 nsec/call
clock-gettime-boottime:    vdso: 1935 nsec/call
clock-getres-tai:    vdso: 1257 nsec/call
clock-gettime-tai:    vdso: 1898 nsec/call
clock-getres-monotonic-raw:    vdso: 1229 nsec/call
clock-gettime-monotonic-raw:    vdso: 1541 nsec/call
clock-getres-monotonic-coarse:    vdso: 1699 nsec/call
clock-gettime-monotonic-coarse:    vdso: 1477 nsec/call
clock-getres-monotonic:    vdso: 135 nsec/call
clock-gettime-monotonic:    vdso: 283 nsec/call

With the series:

gettimeofday:    vdso: 271 nsec/call
clock-getres-realtime-coarse:    vdso: 159 nsec/call
clock-gettime-realtime-coarse:    vdso: 184 nsec/call
clock-getres-realtime:    vdso: 163 nsec/call
clock-gettime-realtime:    vdso: 281 nsec/call
clock-getres-boottime:    vdso: 169 nsec/call
clock-gettime-boottime:    vdso: 274 nsec/call
clock-getres-tai:    vdso: 163 nsec/call
clock-gettime-tai:    vdso: 277 nsec/call
clock-getres-monotonic-raw:    vdso: 166 nsec/call
clock-gettime-monotonic-raw:    vdso: 302 nsec/call
clock-getres-monotonic-coarse:    vdso: 159 nsec/call
clock-gettime-monotonic-coarse:    vdso: 184 nsec/call
clock-getres-monotonic:    vdso: 166 nsec/call
clock-gettime-monotonic:    vdso: 274 nsec/call

Christophe

^ permalink raw reply	[flat|nested] 26+ messages in thread

* Re: [RFC PATCH v4 00/11] powerpc: switch VDSO to C implementation.
  2020-01-20 14:56   ` Christophe Leroy
@ 2020-01-20 15:19     ` Segher Boessenkool
  2020-01-20 17:08       ` Christophe Leroy
  0 siblings, 1 reply; 26+ messages in thread
From: Segher Boessenkool @ 2020-01-20 15:19 UTC (permalink / raw)
  To: Christophe Leroy
  Cc: Benjamin Herrenschmidt, Paul Mackerras, Michael Ellerman,
	nathanl, arnd, tglx, vincenzo.frascino, luto, x86, linuxppc-dev,
	linux-kernel, linux-arm-kernel, linux-mips

On Mon, Jan 20, 2020 at 02:56:00PM +0000, Christophe Leroy wrote:
> >Nice!  Much better.
> >
> >It should be tested on more representative hardware, too, but this looks
> >promising alright :-)
> 
> mpc832x (e300c2 core) at 333 MHz:
> 
> Before:
> 
> gettimeofday:    vdso: 235 nsec/call
> clock-gettime-realtime:    vdso: 244 nsec/call
> 
> With the series:
> 
> gettimeofday:    vdso: 271 nsec/call
> clock-gettime-realtime:    vdso: 281 nsec/call

Those are important, and degrade ~15%.  That is acceptable IMO, but do
you see a way to optimise this (later)?

Anyway, excellent results, thanks for your persistence!


Segher

^ permalink raw reply	[flat|nested] 26+ messages in thread

* Re: [RFC PATCH v4 00/11] powerpc: switch VDSO to C implementation.
  2020-01-20 15:19     ` Segher Boessenkool
@ 2020-01-20 17:08       ` Christophe Leroy
  2020-01-20 17:27         ` Segher Boessenkool
  0 siblings, 1 reply; 26+ messages in thread
From: Christophe Leroy @ 2020-01-20 17:08 UTC (permalink / raw)
  To: Segher Boessenkool
  Cc: Benjamin Herrenschmidt, Paul Mackerras, Michael Ellerman,
	nathanl, arnd, tglx, vincenzo.frascino, luto, x86, linuxppc-dev,
	linux-kernel, linux-arm-kernel, linux-mips



Le 20/01/2020 à 16:19, Segher Boessenkool a écrit :
> On Mon, Jan 20, 2020 at 02:56:00PM +0000, Christophe Leroy wrote:
>>> Nice!  Much better.
>>>
>>> It should be tested on more representative hardware, too, but this looks
>>> promising alright :-)
>>
>> mpc832x (e300c2 core) at 333 MHz:
>>
>> Before:
>>
>> gettimeofday:    vdso: 235 nsec/call
>> clock-gettime-realtime:    vdso: 244 nsec/call
>>
>> With the series:
>>
>> gettimeofday:    vdso: 271 nsec/call
>> clock-gettime-realtime:    vdso: 281 nsec/call
> 
> Those are important, and degrade ~15%.  That is acceptable IMO, but do
> you see a way to optimise this (later)?

Not easy I think.

First we have the unavoidable ASM entry function that can't be dropped 
because of the CR[SO] bit the set on error or clear on no error and that 
can't be done in C.

In our ASM VDSO, fixed shifts are used, while in generic C VDSO, shifts 
are generic and read from the VDSO data.

And there is still some funny code generated by GCC (8.1), like:

  620:	7d 29 3c 30 	srw     r9,r9,r7
  624:	21 87 00 20 	subfic  r12,r7,32
  628:	7d 07 3c 31 	srw.    r7,r8,r7
  62c:	7d 08 60 30 	slw     r8,r8,r12
  630:	7d 0b 4b 78 	or      r11,r8,r9
  634:	39 40 00 00 	li      r10,0
  638:	40 82 00 84 	bne     6bc <__c_kernel_clock_gettime+0x114>
  63c:	81 23 00 24 	lwz     r9,36(r3)
  640:	81 05 00 00 	lwz     r8,0(r5)
...
  6bc:	7d 69 5b 78 	mr      r9,r11
  6c0:	7c ea 3b 78 	mr      r10,r7
  6c4:	7d 2b 4b 78 	mr      r11,r9
  6c8:	4b ff ff 74 	b       63c <__c_kernel_clock_gettime+0x94>

This branch to 6bc is totally useless:
- copying r11 into r9 is pointless as r9 is overwritten in 63c
- copying back r9 into r11 is pointless as r11 has not been modified 
inbetween.
- loading r10 with 0 then overwritting r10 with r7 when r7 is not 0 is 
pointless as well, could have directly put the result of srw. in r10.

Christophe

^ permalink raw reply	[flat|nested] 26+ messages in thread

* Re: [RFC PATCH v4 00/11] powerpc: switch VDSO to C implementation.
  2020-01-20 17:08       ` Christophe Leroy
@ 2020-01-20 17:27         ` Segher Boessenkool
  0 siblings, 0 replies; 26+ messages in thread
From: Segher Boessenkool @ 2020-01-20 17:27 UTC (permalink / raw)
  To: Christophe Leroy
  Cc: Benjamin Herrenschmidt, Paul Mackerras, Michael Ellerman,
	nathanl, arnd, tglx, vincenzo.frascino, luto, x86, linuxppc-dev,
	linux-kernel, linux-arm-kernel, linux-mips

On Mon, Jan 20, 2020 at 06:08:23PM +0100, Christophe Leroy wrote:
> Not easy I think.
> 
> First we have the unavoidable ASM entry function that can't be dropped 
> because of the CR[SO] bit the set on error or clear on no error and that 
> can't be done in C.

Yup.

> In our ASM VDSO, fixed shifts are used, while in generic C VDSO, shifts 
> are generic and read from the VDSO data.

Does that cost more than just a few cycles?

> And there is still some funny code generated by GCC (8.1), like:
> 
>  620:	7d 29 3c 30 	srw     r9,r9,r7
>  624:	21 87 00 20 	subfic  r12,r7,32
>  628:	7d 07 3c 31 	srw.    r7,r8,r7
>  62c:	7d 08 60 30 	slw     r8,r8,r12
>  630:	7d 0b 4b 78 	or      r11,r8,r9

(This can be done cheaper for fixed shifts, you can use rlwimi then).

>  634:	39 40 00 00 	li      r10,0
>  638:	40 82 00 84 	bne     6bc <__c_kernel_clock_gettime+0x114>
>  63c:	81 23 00 24 	lwz     r9,36(r3)
>  640:	81 05 00 00 	lwz     r8,0(r5)
> ...
>  6bc:	7d 69 5b 78 	mr      r9,r11
>  6c0:	7c ea 3b 78 	mr      r10,r7
>  6c4:	7d 2b 4b 78 	mr      r11,r9
>  6c8:	4b ff ff 74 	b       63c <__c_kernel_clock_gettime+0x94>
> 
> This branch to 6bc is totally useless:
> - copying r11 into r9 is pointless as r9 is overwritten in 63c
> - copying back r9 into r11 is pointless as r11 has not been modified 
> inbetween.

Yeah, huh, how did that happen.

> - loading r10 with 0 then overwritting r10 with r7 when r7 is not 0 is 
> pointless as well, could have directly put the result of srw. in r10.

This may be harder to make the compiler do.

But the r9/r11 thing suggests you are preventing optimisation somewhere,
maybe with some asm?  Do you have some small testcase I can compile?


Segher

^ permalink raw reply	[flat|nested] 26+ messages in thread

* Re: [RFC PATCH v4 10/11] lib: vdso: Allow arches to override the ns shift operation
  2020-01-16 20:20       ` Andy Lutomirski
@ 2020-01-29  7:14         ` Thomas Gleixner
  2020-01-29  7:26           ` Christophe Leroy
  0 siblings, 1 reply; 26+ messages in thread
From: Thomas Gleixner @ 2020-01-29  7:14 UTC (permalink / raw)
  To: Andy Lutomirski
  Cc: Christophe Leroy, Benjamin Herrenschmidt, Paul Mackerras,
	Michael Ellerman, nathanl, Arnd Bergmann, Vincenzo Frascino,
	Andrew Lutomirski, LKML, linuxppc-dev, linux-arm-kernel,
	open list:MIPS, X86 ML

Andy Lutomirski <luto@kernel.org> writes:

> On Thu, Jan 16, 2020 at 11:57 AM Thomas Gleixner <tglx@linutronix.de> wrote:
>>
>> Andy Lutomirski <luto@kernel.org> writes:
>> > On Thu, Jan 16, 2020 at 9:58 AM Christophe Leroy
>> >
>> > Would mul_u64_u64_shr() be a good alternative?  Could we adjust it to
>> > assume the shift is less than 32?  That function exists to benefit
>> > 32-bit arches.
>>
>> We'd want mul_u64_u32_shr() for this. The rules for mult and shift are:
>>
>
> That's what I meant to type...

Just that it does not work. The math is:

     ns = d->nsecs;   // That's the nsec value shifted left by d->shift

     ns += ((cur - d->last) & d->mask) * mult;

     ns >>= d->shift;

So we cannot use mul_u64_u32_shr() because we need the addition there
before shifting. And no, we can't drop the fractional part of
d->nsecs. Been there, done that, got sporadic time going backwards
problems as a reward. Need to look at that again as stuff has changed
over time.

On x86 we enforce that mask is 64bit, so the & operation is not there,
but due to the nasties of TSC we have that conditional

    if (cur > last)
       return (cur - last) * mult;
    return 0;

Christophe, on PPC the decrementer/RTC clocksource masks are 64bit as
well, so you can spare that & operation there too.

Thanks,

        tglx




^ permalink raw reply	[flat|nested] 26+ messages in thread

* Re: [RFC PATCH v4 10/11] lib: vdso: Allow arches to override the ns shift operation
  2020-01-29  7:14         ` Thomas Gleixner
@ 2020-01-29  7:26           ` Christophe Leroy
  0 siblings, 0 replies; 26+ messages in thread
From: Christophe Leroy @ 2020-01-29  7:26 UTC (permalink / raw)
  To: Thomas Gleixner, Andy Lutomirski
  Cc: Benjamin Herrenschmidt, Paul Mackerras, Michael Ellerman,
	nathanl, Arnd Bergmann, Vincenzo Frascino, LKML, linuxppc-dev,
	linux-arm-kernel, open list:MIPS, X86 ML



Le 29/01/2020 à 08:14, Thomas Gleixner a écrit :
> Andy Lutomirski <luto@kernel.org> writes:
> 
>> On Thu, Jan 16, 2020 at 11:57 AM Thomas Gleixner <tglx@linutronix.de> wrote:
>>>
>>> Andy Lutomirski <luto@kernel.org> writes:
>>>> On Thu, Jan 16, 2020 at 9:58 AM Christophe Leroy
>>>>
>>>> Would mul_u64_u64_shr() be a good alternative?  Could we adjust it to
>>>> assume the shift is less than 32?  That function exists to benefit
>>>> 32-bit arches.
>>>
>>> We'd want mul_u64_u32_shr() for this. The rules for mult and shift are:
>>>
>>
>> That's what I meant to type...
> 
> Just that it does not work. The math is:
> 
>       ns = d->nsecs;   // That's the nsec value shifted left by d->shift
> 
>       ns += ((cur - d->last) & d->mask) * mult;
> 
>       ns >>= d->shift;
> 
> So we cannot use mul_u64_u32_shr() because we need the addition there
> before shifting. And no, we can't drop the fractional part of
> d->nsecs. Been there, done that, got sporadic time going backwards
> problems as a reward. Need to look at that again as stuff has changed
> over time.
> 
> On x86 we enforce that mask is 64bit, so the & operation is not there,
> but due to the nasties of TSC we have that conditional
> 
>      if (cur > last)
>         return (cur - last) * mult;
>      return 0;
> 
> Christophe, on PPC the decrementer/RTC clocksource masks are 64bit as
> well, so you can spare that & operation there too.
> 

Yes, I did it already. It spares reading d->mast, that the main advantage.

Christophe

^ permalink raw reply	[flat|nested] 26+ messages in thread

end of thread, other threads:[~2020-01-29  7:26 UTC | newest]

Thread overview: 26+ messages (download: mbox.gz / follow: Atom feed)
-- links below jump to the message on this page --
2020-01-16 17:58 [RFC PATCH v4 00/11] powerpc: switch VDSO to C implementation Christophe Leroy
2020-01-16 17:58 ` [RFC PATCH v4 01/11] powerpc/64: Don't provide time functions in compat VDSO32 Christophe Leroy
2020-01-16 17:58 ` [RFC PATCH v4 02/11] powerpc/vdso: Switch VDSO to generic C implementation Christophe Leroy
2020-01-16 17:58 ` [RFC PATCH v4 03/11] lib: vdso: only read hrtimer_res when needed in __cvdso_clock_getres() Christophe Leroy
2020-01-16 17:58 ` [RFC PATCH v4 04/11] powerpc/vdso: simplify __get_datapage() Christophe Leroy
2020-01-16 17:58 ` [RFC PATCH v4 05/11] lib: vdso: allow arches to provide vdso data pointer Christophe Leroy
2020-01-16 17:58 ` [RFC PATCH v4 06/11] powerpc/vdso: provide inline alternative to __get_datapage() Christophe Leroy
2020-01-16 17:58 ` [RFC PATCH v4 07/11] powerpc/vdso: provide vdso data pointer from the ASM caller Christophe Leroy
2020-01-16 17:58 ` [RFC PATCH v4 08/11] lib: vdso: allow fixed clock mode Christophe Leroy
2020-01-16 20:13   ` Thomas Gleixner
2020-01-16 20:19     ` Andy Lutomirski
2020-01-16 21:07       ` Thomas Gleixner
2020-01-16 17:58 ` [RFC PATCH v4 09/11] powerpc/vdso: override __arch_vdso_capable() Christophe Leroy
2020-01-16 17:58 ` [RFC PATCH v4 10/11] lib: vdso: Allow arches to override the ns shift operation Christophe Leroy
2020-01-16 19:47   ` Andy Lutomirski
2020-01-16 19:57     ` Thomas Gleixner
2020-01-16 20:20       ` Andy Lutomirski
2020-01-29  7:14         ` Thomas Gleixner
2020-01-29  7:26           ` Christophe Leroy
2020-01-16 17:58 ` [RFC PATCH v4 11/11] powerpc/32: provide vdso_shift_ns() Christophe Leroy
2020-01-17  8:58 ` [RFC PATCH v4 00/11] powerpc: switch VDSO to C implementation Segher Boessenkool
2020-01-17  9:26   ` Christophe Leroy
2020-01-20 14:56   ` Christophe Leroy
2020-01-20 15:19     ` Segher Boessenkool
2020-01-20 17:08       ` Christophe Leroy
2020-01-20 17:27         ` Segher Boessenkool

This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox;
as well as URLs for NNTP newsgroup(s).