* [Xenomai-core] [PATCH] Use tsc for implementation of clock_gettime.
@ 2007-05-17 11:29 Gilles Chanteperdrix
2007-05-17 12:43 ` Jan Kiszka
0 siblings, 1 reply; 9+ messages in thread
From: Gilles Chanteperdrix @ 2007-05-17 11:29 UTC (permalink / raw)
To: xenomai
[-- Attachment #1: message body and .signature --]
[-- Type: text/plain, Size: 197 bytes --]
Hi,
here comes, for review, a patch which reduces the overhead of
clock_gettime by directly reading the tsc in user-space for
architectures that support it.
--
Gilles Chanteperdrix.
[-- Attachment #2: xeno-clock_gettime-direct-tsc.diff --]
[-- Type: text/plain, Size: 37956 bytes --]
Index: include/asm-ia64/Makefile.in
===================================================================
--- include/asm-ia64/Makefile.in (revision 2454)
+++ include/asm-ia64/Makefile.in (working copy)
@@ -230,7 +230,9 @@
target_os = @target_os@
target_vendor = @target_vendor@
includesubdir = $(includedir)/asm-ia64
-includesub_HEADERS = atomic.h calibration.h features.h hal.h syscall.h system.h wrappers.h fptest.h
+includesub_HEADERS = arith.h atomic.h calibration.h features.h fptest.h \
+ hal.h syscall.h system.h wrappers.h
+
SUBDIRS = bits
all: all-recursive
Index: include/asm-ia64/Makefile.am
===================================================================
--- include/asm-ia64/Makefile.am (revision 2454)
+++ include/asm-ia64/Makefile.am (working copy)
@@ -1,5 +1,6 @@
includesubdir = $(includedir)/asm-ia64
-includesub_HEADERS = atomic.h calibration.h features.h hal.h syscall.h system.h wrappers.h fptest.h
+includesub_HEADERS = arith.h atomic.h calibration.h features.h fptest.h \
+ hal.h syscall.h system.h wrappers.h
SUBDIRS = bits
Index: include/asm-ia64/arith.h
===================================================================
--- include/asm-ia64/arith.h (revision 0)
+++ include/asm-ia64/arith.h (revision 0)
@@ -0,0 +1,6 @@
+#ifndef _XENO_ASM_IA64_ARITH_H
+#define _XENO_ASM_IA64_ARITH_H
+
+#include <asm-generic/xenomai/arith.h>
+
+#endif /* _XENO_ASM_IA64_ARITH_H */
Property changes on: include/asm-ia64/arith.h
___________________________________________________________________
Name: svn:executable
+ *
Index: include/asm-blackfin/Makefile.in
===================================================================
--- include/asm-blackfin/Makefile.in (revision 2454)
+++ include/asm-blackfin/Makefile.in (working copy)
@@ -230,8 +230,8 @@
target_os = @target_os@
target_vendor = @target_vendor@
includesubdir = $(includedir)/asm-blackfin
-includesub_HEADERS = atomic.h calibration.h features.h hal.h \
- syscall.h system.h wrappers.h fptest.h
+includesub_HEADERS = arith.h atomic.h calibration.h features.h fptest.h \
+ hal.h syscall.h system.h wrappers.h
SUBDIRS = bits
all: all-recursive
Index: include/asm-blackfin/Makefile.am
===================================================================
--- include/asm-blackfin/Makefile.am (revision 2454)
+++ include/asm-blackfin/Makefile.am (working copy)
@@ -1,6 +1,6 @@
includesubdir = $(includedir)/asm-blackfin
-includesub_HEADERS = atomic.h calibration.h features.h hal.h \
- syscall.h system.h wrappers.h fptest.h
+includesub_HEADERS = arith.h atomic.h calibration.h features.h fptest.h \
+ hal.h syscall.h system.h wrappers.h
SUBDIRS = bits
Index: include/asm-blackfin/arith.h
===================================================================
--- include/asm-blackfin/arith.h (revision 0)
+++ include/asm-blackfin/arith.h (revision 0)
@@ -0,0 +1,6 @@
+#ifndef _XENO_ASM_BLACKFIN_ARITH_H
+#define _XENO_ASM_BLACKFIN_ARITH_H
+
+#include <asm-generic/xenomai/arith.h>
+
+#endif /* _XENO_ASM_BLACKFIN_ARITH_H */
Property changes on: include/asm-blackfin/arith.h
___________________________________________________________________
Name: svn:executable
+ *
Index: include/asm-generic/syscall.h
===================================================================
--- include/asm-generic/syscall.h (revision 2454)
+++ include/asm-generic/syscall.h (working copy)
@@ -55,7 +55,6 @@
unsigned long long cpufreq; /* CPU frequency */
unsigned long tickval; /* Tick duration (ns) */
-
} xnsysinfo_t;
#define SIGHARDEN SIGWINCH
Index: include/asm-generic/bits/bind.h
===================================================================
--- include/asm-generic/bits/bind.h (revision 2454)
+++ include/asm-generic/bits/bind.h (working copy)
@@ -62,7 +62,7 @@
sa.sa_flags = 0;
sigaction(SIGXCPU, &sa, NULL);
- return __xn_mux_shifted_id(muxid);
+ return muxid;
}
static inline int
@@ -105,7 +105,7 @@
exit(1);
}
- return __xn_mux_shifted_id(muxid);
+ return muxid;
}
#endif /* _XENO_ASM_GENERIC_BITS_BIND_H */
Index: include/asm-generic/Makefile.in
===================================================================
--- include/asm-generic/Makefile.in (revision 2454)
+++ include/asm-generic/Makefile.in (working copy)
@@ -230,7 +230,7 @@
target_os = @target_os@
target_vendor = @target_vendor@
includesubdir = $(includedir)/asm-generic
-includesub_HEADERS = features.h hal.h syscall.h system.h wrappers.h
+includesub_HEADERS = arith.h features.h hal.h syscall.h system.h wrappers.h
SUBDIRS = bits
all: all-recursive
Index: include/asm-generic/Makefile.am
===================================================================
--- include/asm-generic/Makefile.am (revision 2454)
+++ include/asm-generic/Makefile.am (working copy)
@@ -1,5 +1,5 @@
includesubdir = $(includedir)/asm-generic
-includesub_HEADERS = features.h hal.h syscall.h system.h wrappers.h
+includesub_HEADERS = arith.h features.h hal.h syscall.h system.h wrappers.h
SUBDIRS = bits
Index: include/asm-generic/hal.h
===================================================================
--- include/asm-generic/hal.h (revision 2454)
+++ include/asm-generic/hal.h (working copy)
@@ -37,8 +37,8 @@
#include <linux/kallsyms.h>
#include <linux/init.h>
#include <asm/byteorder.h>
-#include <asm/div64.h>
#include <asm/xenomai/wrappers.h>
+#include <asm/xenomai/arith.h>
#define RTHAL_DOMAIN_ID 0x58454e4f
@@ -385,121 +385,6 @@
#define rthal_printk printk
-#ifdef __BIG_ENDIAN
-#define endianstruct struct { unsigned _h; unsigned _l; } _s
-#else /* __LITTLE_ENDIAN */
-#define endianstruct struct { unsigned _l; unsigned _h; } _s
-#endif
-
-#ifndef __rthal_u64tou32
-#define __rthal_u64tou32(ull, h, l) ({ \
- union { unsigned long long _ull; \
- endianstruct; \
- } _u; \
- _u._ull = (ull); \
- (h) = _u._s._h; \
- (l) = _u._s._l; \
-})
-#endif /* !__rthal_u64tou32 */
-
-#ifndef __rthal_u64fromu32
-#define __rthal_u64fromu32(h, l) ({ \
- union { unsigned long long _ull; \
- endianstruct; \
- } _u; \
- _u._s._h = (h); \
- _u._s._l = (l); \
- _u._ull; \
-})
-#endif /* !__rthal_u64fromu32 */
-
-#ifndef rthal_ullmul
-static inline __attribute_const__ unsigned long long
-__rthal_generic_ullmul(const unsigned m0, const unsigned m1)
-{
- return (unsigned long long) m0 * m1;
-}
-#define rthal_ullmul(m0,m1) __rthal_generic_ullmul((m0),(m1))
-#endif /* !rthal_ullmul */
-
-#ifndef rthal_ulldiv
-static inline unsigned long long __rthal_generic_ulldiv (unsigned long long ull,
- const unsigned uld,
- unsigned long *const rp)
-{
- const unsigned r = do_div(ull, uld);
-
- if (rp)
- *rp = r;
-
- return ull;
-}
-#define rthal_ulldiv(ull,uld,rp) __rthal_generic_ulldiv((ull),(uld),(rp))
-#endif /* !rthal_ulldiv */
-
-#ifndef rthal_uldivrem
-#define rthal_uldivrem(ull,ul,rp) ((unsigned) rthal_ulldiv((ull),(ul),(rp)))
-#endif /* !rthal_uldivrem */
-
-#ifndef rthal_imuldiv
-static inline __attribute_const__ int __rthal_generic_imuldiv (int i,
- int mult,
- int div)
-{
- /* Returns (int)i = (unsigned long long)i*(unsigned)(mult)/(unsigned)div. */
- const unsigned long long ull = rthal_ullmul(i, mult);
- return rthal_uldivrem(ull, div, NULL);
-}
-#define rthal_imuldiv(i,m,d) __rthal_generic_imuldiv((i),(m),(d))
-#endif /* !rthal_imuldiv */
-
-#ifndef rthal_llimd
-/* Division of an unsigned 96 bits ((h << 32) + l) by an unsigned 32 bits.
- Building block for llimd. Without const qualifiers, gcc reload registers
- after each call to uldivrem. */
-static inline unsigned long long
-__rthal_generic_div96by32 (const unsigned long long h,
- const unsigned l,
- const unsigned d,
- unsigned long *const rp)
-{
- unsigned long rh;
- const unsigned qh = rthal_uldivrem(h, d, &rh);
- const unsigned long long t = __rthal_u64fromu32(rh, l);
- const unsigned ql = rthal_uldivrem(t, d, rp);
-
- return __rthal_u64fromu32(qh, ql);
-}
-
-static inline __attribute_const__
-unsigned long long __rthal_generic_ullimd (const unsigned long long op,
- const unsigned m,
- const unsigned d)
-{
- unsigned oph, opl, tlh, tll;
- unsigned long long th, tl;
-
- __rthal_u64tou32(op, oph, opl);
- tl = rthal_ullmul(opl, m);
- __rthal_u64tou32(tl, tlh, tll);
- th = rthal_ullmul(oph, m);
- th += tlh;
-
- return __rthal_generic_div96by32(th, tll, d, NULL);
-}
-
-static inline __attribute_const__ long long __rthal_generic_llimd (long long op,
- unsigned m,
- unsigned d)
-{
-
- if(op < 0LL)
- return -__rthal_generic_ullimd(-op, m, d);
- return __rthal_generic_ullimd(op, m, d);
-}
-#define rthal_llimd(ll,m,d) __rthal_generic_llimd((ll),(m),(d))
-#endif /* !rthal_llimd */
-
typedef ipipe_irq_handler_t rthal_irq_handler_t;
typedef ipipe_irq_ackfn_t rthal_irq_ackfn_t;
Index: include/asm-generic/system.h
===================================================================
--- include/asm-generic/system.h (revision 2454)
+++ include/asm-generic/system.h (working copy)
@@ -133,18 +133,6 @@
#define xnarch_logerr(fmt,args...) printk(KERN_ERR XNARCH_PROMPT fmt , ##args)
#define xnarch_printf(fmt,args...) printk(KERN_INFO XNARCH_PROMPT fmt , ##args)
-#define xnarch_ullmod(ull,uld,rem) ({ xnarch_ulldiv(ull,uld,rem); (*rem); })
-#define xnarch_uldiv(ull, d) rthal_uldivrem(ull, d, NULL)
-#define xnarch_ulmod(ull, d) ({ u_long _rem; \
- rthal_uldivrem(ull,d,&_rem); _rem; })
-
-#define xnarch_ullmul rthal_ullmul
-#define xnarch_uldivrem rthal_uldivrem
-#define xnarch_ulldiv rthal_ulldiv
-#define xnarch_imuldiv rthal_imuldiv
-#define xnarch_llimd rthal_llimd
-#define xnarch_get_cpu_tsc rthal_rdtsc
-
typedef cpumask_t xnarch_cpumask_t;
#ifdef CONFIG_SMP
Index: include/asm-generic/arith.h
===================================================================
--- include/asm-generic/arith.h (revision 0)
+++ include/asm-generic/arith.h (revision 0)
@@ -0,0 +1,173 @@
+/**
+ * @ingroup hal
+ * @file
+ *
+ * Generic arithmetic/conversion routines.
+ * Copyright © 2005 Stelian Pop.
+ * Copyright © 2005 Gilles Chanteperdrix.
+ *
+ * This program is free software; you can redistribute it and/or modify
+ * it under the terms of the GNU General Public License as published by
+ * the Free Software Foundation, Inc., 675 Mass Ave, Cambridge MA 02139,
+ * USA; either version 2 of the License, or (at your option) any later
+ * version.
+ *
+ * This program is distributed in the hope that it will be useful,
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
+ * GNU General Public License for more details.
+ *
+ * You should have received a copy of the GNU General Public License
+ * along with this program; if not, write to the Free Software
+ * Foundation, Inc., 59 Temple Place - Suite 330, Boston, MA 02111-1307, USA.
+ */
+
+/**
+ * @addtogroup hal
+ *@{*/
+
+#ifndef _XENO_ASM_GENERIC_ARITH_H
+#define _XENO_ASM_GENERIC_ARITH_H
+
+#ifdef __KERNEL__
+#include <asm/div64.h>
+#else /* !__KERNEL__ */
+static inline unsigned __rthal_do_div(unsigned long long *a, unsigned d)
+{
+ unsigned r = *a % d;
+ *a /= d;
+ return r;
+}
+#define do_div(a, d) __rthal_do_div(&(a), (d))
+#endif /* !__KERNEL__ */
+
+#ifdef __BIG_ENDIAN
+#define endianstruct struct { unsigned _h; unsigned _l; } _s
+#else /* __LITTLE_ENDIAN */
+#define endianstruct struct { unsigned _l; unsigned _h; } _s
+#endif
+
+#ifndef __rthal_u64tou32
+#define __rthal_u64tou32(ull, h, l) ({ \
+ union { unsigned long long _ull; \
+ endianstruct; \
+ } _u; \
+ _u._ull = (ull); \
+ (h) = _u._s._h; \
+ (l) = _u._s._l; \
+})
+#endif /* !__rthal_u64tou32 */
+
+#ifndef __rthal_u64fromu32
+#define __rthal_u64fromu32(h, l) ({ \
+ union { unsigned long long _ull; \
+ endianstruct; \
+ } _u; \
+ _u._s._h = (h); \
+ _u._s._l = (l); \
+ _u._ull; \
+})
+#endif /* !__rthal_u64fromu32 */
+
+#ifndef rthal_ullmul
+static inline __attribute_const__ unsigned long long
+__rthal_generic_ullmul(const unsigned m0, const unsigned m1)
+{
+ return (unsigned long long) m0 * m1;
+}
+#define rthal_ullmul(m0,m1) __rthal_generic_ullmul((m0),(m1))
+#endif /* !rthal_ullmul */
+
+#ifndef rthal_ulldiv
+static inline unsigned long long __rthal_generic_ulldiv (unsigned long long ull,
+ const unsigned uld,
+ unsigned long *const rp)
+{
+ const unsigned r = do_div(ull, uld);
+
+ if (rp)
+ *rp = r;
+
+ return ull;
+}
+#define rthal_ulldiv(ull,uld,rp) __rthal_generic_ulldiv((ull),(uld),(rp))
+#endif /* !rthal_ulldiv */
+
+#ifndef rthal_uldivrem
+#define rthal_uldivrem(ull,ul,rp) ((unsigned) rthal_ulldiv((ull),(ul),(rp)))
+#endif /* !rthal_uldivrem */
+
+#ifndef rthal_imuldiv
+static inline __attribute_const__ int __rthal_generic_imuldiv (int i,
+ int mult,
+ int div)
+{
+ /* Returns (int)i = (unsigned long long)i*(unsigned)(mult)/(unsigned)div. */
+ const unsigned long long ull = rthal_ullmul(i, mult);
+ return rthal_uldivrem(ull, div, NULL);
+}
+#define rthal_imuldiv(i,m,d) __rthal_generic_imuldiv((i),(m),(d))
+#endif /* !rthal_imuldiv */
+
+#ifndef rthal_llimd
+/* Division of an unsigned 96 bits ((h << 32) + l) by an unsigned 32 bits.
+ Building block for llimd. Without const qualifiers, gcc reload registers
+ after each call to uldivrem. */
+static inline unsigned long long
+__rthal_generic_div96by32 (const unsigned long long h,
+ const unsigned l,
+ const unsigned d,
+ unsigned long *const rp)
+{
+ unsigned long rh;
+ const unsigned qh = rthal_uldivrem(h, d, &rh);
+ const unsigned long long t = __rthal_u64fromu32(rh, l);
+ const unsigned ql = rthal_uldivrem(t, d, rp);
+
+ return __rthal_u64fromu32(qh, ql);
+}
+
+static inline __attribute_const__
+unsigned long long __rthal_generic_ullimd (const unsigned long long op,
+ const unsigned m,
+ const unsigned d)
+{
+ unsigned oph, opl, tlh, tll;
+ unsigned long long th, tl;
+
+ __rthal_u64tou32(op, oph, opl);
+ tl = rthal_ullmul(opl, m);
+ __rthal_u64tou32(tl, tlh, tll);
+ th = rthal_ullmul(oph, m);
+ th += tlh;
+
+ return __rthal_generic_div96by32(th, tll, d, NULL);
+}
+
+static inline __attribute_const__ long long __rthal_generic_llimd (long long op,
+ unsigned m,
+ unsigned d)
+{
+
+ if(op < 0LL)
+ return -__rthal_generic_ullimd(-op, m, d);
+ return __rthal_generic_ullimd(op, m, d);
+}
+#define rthal_llimd(ll,m,d) __rthal_generic_llimd((ll),(m),(d))
+#endif /* !rthal_llimd */
+
+#define xnarch_ullmod(ull,uld,rem) ({ xnarch_ulldiv(ull,uld,rem); (*rem); })
+#define xnarch_uldiv(ull, d) rthal_uldivrem(ull, d, NULL)
+#define xnarch_ulmod(ull, d) ({ u_long _rem; \
+ rthal_uldivrem(ull,d,&_rem); _rem; })
+
+#define xnarch_ullmul rthal_ullmul
+#define xnarch_uldivrem rthal_uldivrem
+#define xnarch_ulldiv rthal_ulldiv
+#define xnarch_imuldiv rthal_imuldiv
+#define xnarch_llimd rthal_llimd
+#define xnarch_get_cpu_tsc rthal_rdtsc
+
+/*@}*/
+
+#endif /* _XENO_ASM_GENERIC_ARITH_H */
Property changes on: include/asm-generic/arith.h
___________________________________________________________________
Name: svn:executable
+ *
Index: include/asm-powerpc/Makefile.in
===================================================================
--- include/asm-powerpc/Makefile.in (revision 2454)
+++ include/asm-powerpc/Makefile.in (working copy)
@@ -230,8 +230,8 @@
target_os = @target_os@
target_vendor = @target_vendor@
includesubdir = $(includedir)/asm-powerpc
-includesub_HEADERS = atomic.h calibration.h features.h hal.h \
- syscall.h system.h wrappers.h fptest.h
+includesub_HEADERS = arith.h atomic.h calibration.h features.h fptest.h \
+ hal.h syscall.h system.h wrappers.h
SUBDIRS = bits
all: all-recursive
Index: include/asm-powerpc/Makefile.am
===================================================================
--- include/asm-powerpc/Makefile.am (revision 2454)
+++ include/asm-powerpc/Makefile.am (working copy)
@@ -1,6 +1,6 @@
includesubdir = $(includedir)/asm-powerpc
-includesub_HEADERS = atomic.h calibration.h features.h hal.h \
- syscall.h system.h wrappers.h fptest.h
+includesub_HEADERS = arith.h atomic.h calibration.h features.h fptest.h \
+ hal.h syscall.h system.h wrappers.h
SUBDIRS = bits
Index: include/asm-powerpc/arith.h
===================================================================
--- include/asm-powerpc/arith.h (revision 0)
+++ include/asm-powerpc/arith.h (revision 0)
@@ -0,0 +1,6 @@
+#ifndef _XENO_ASM_POWERPC_ARITH_H
+#define _XENO_ASM_POWERPC_ARITH_H
+
+#include <asm-generic/xenomai/arith.h>
+
+#endif /* _XENO_ASM_POWERPC_ARITH_H */
Property changes on: include/asm-powerpc/arith.h
___________________________________________________________________
Name: svn:executable
+ *
Index: include/asm-arm/Makefile.in
===================================================================
--- include/asm-arm/Makefile.in (revision 2454)
+++ include/asm-arm/Makefile.in (working copy)
@@ -230,8 +230,8 @@
target_os = @target_os@
target_vendor = @target_vendor@
includesubdir = $(includedir)/asm-arm
-includesub_HEADERS = atomic.h calibration.h features.h hal.h \
- syscall.h system.h wrappers.h fptest.h
+includesub_HEADERS = arith.h atomic.h calibration.h features.h fptest.h \
+ hal.h syscall.h system.h wrappers.h
SUBDIRS = bits
all: all-recursive
Index: include/asm-arm/Makefile.am
===================================================================
--- include/asm-arm/Makefile.am (revision 2454)
+++ include/asm-arm/Makefile.am (working copy)
@@ -1,6 +1,6 @@
includesubdir = $(includedir)/asm-arm
-includesub_HEADERS = atomic.h calibration.h features.h hal.h \
- syscall.h system.h wrappers.h fptest.h
+includesub_HEADERS = arith.h atomic.h calibration.h features.h fptest.h \
+ hal.h syscall.h system.h wrappers.h
SUBDIRS = bits
Index: include/asm-arm/arith.h
===================================================================
--- include/asm-arm/arith.h (revision 0)
+++ include/asm-arm/arith.h (revision 0)
@@ -0,0 +1,6 @@
+#ifndef _XENO_ASM_ARM_ARITH_H
+#define _XENO_ASM_ARM_ARITH_H
+
+#include <asm-generic/xenomai/arith.h>
+
+#endif /* _XENO_ASM_ARM_ARITH_H */
Property changes on: include/asm-arm/arith.h
___________________________________________________________________
Name: svn:executable
+ *
Index: include/posix/syscall.h
===================================================================
--- include/posix/syscall.h (revision 2454)
+++ include/posix/syscall.h (working copy)
@@ -100,6 +100,11 @@
#define __pse51_condattr_setpshared 74
#define __pse51_thread_getschedparam 75
+typedef struct pse51_sysinfo {
+ unsigned period;
+ unsigned long long wallclock_offset;
+} pse51_sysinfo_t;
+
#ifdef __KERNEL__
#ifdef __cplusplus
Index: include/asm-i386/Makefile.in
===================================================================
--- include/asm-i386/Makefile.in (revision 2454)
+++ include/asm-i386/Makefile.in (working copy)
@@ -230,8 +230,8 @@
target_os = @target_os@
target_vendor = @target_vendor@
includesubdir = $(includedir)/asm-i386
-includesub_HEADERS = atomic.h calibration.h features.h \
- hal.h smi.h switch.h syscall.h system.h wrappers.h fptest.h
+includesub_HEADERS = arith.h atomic.h calibration.h features.h fptest.h \
+ hal.h smi.h switch.h syscall.h system.h wrappers.h
SUBDIRS = bits
all: all-recursive
Index: include/asm-i386/Makefile.am
===================================================================
--- include/asm-i386/Makefile.am (revision 2454)
+++ include/asm-i386/Makefile.am (working copy)
@@ -1,6 +1,6 @@
includesubdir = $(includedir)/asm-i386
-includesub_HEADERS = atomic.h calibration.h features.h \
- hal.h smi.h switch.h syscall.h system.h wrappers.h fptest.h
+includesub_HEADERS = arith.h atomic.h calibration.h features.h fptest.h \
+ hal.h smi.h switch.h syscall.h system.h wrappers.h
SUBDIRS = bits
Index: include/asm-i386/hal.h
===================================================================
--- include/asm-i386/hal.h (revision 2454)
+++ include/asm-i386/hal.h (working copy)
@@ -39,71 +39,6 @@
#include <asm/xenomai/wrappers.h>
-#define __rthal_u64tou32(ull, h, l) ({ \
- unsigned long long _ull = (ull); \
- (l) = _ull & 0xffffffff; \
- (h) = _ull >> 32; \
-})
-
-#define __rthal_u64fromu32(h, l) ({ \
- unsigned long long _ull; \
- asm ( "": "=A"(_ull) : "d"(h), "a"(l)); \
- _ull; \
-})
-
-/* const helper for rthal_uldivrem, so that the compiler will eliminate
- multiple calls with same arguments, at no additionnal cost. */
-static inline __attribute_const__ unsigned long long
-__rthal_uldivrem(const unsigned long long ull, const unsigned long d)
-{
- unsigned long long ret;
- __asm__ ("divl %1" : "=A,A"(ret) : "r,?m"(d), "A,A"(ull));
- /* Exception if quotient does not fit on unsigned long. */
- return ret;
-}
-
-/* Fast long long division: when the quotient and remainder fit on 32 bits. */
-static inline unsigned long __rthal_i386_uldivrem(unsigned long long ull,
- const unsigned d,
- unsigned long *const rp)
-{
- unsigned long q, r;
- ull = __rthal_uldivrem(ull, d);
- __asm__ ( "": "=d"(r), "=a"(q) : "A"(ull));
- if(rp)
- *rp = r;
- return q;
-}
-#define rthal_uldivrem(ull, d, rp) __rthal_i386_uldivrem((ull),(d),(rp))
-
-/* Division of an unsigned 96 bits ((h << 32) + l) by an unsigned 32 bits.
- Building block for ulldiv. */
-static inline unsigned long long __rthal_div96by32 (const unsigned long long h,
- const unsigned long l,
- const unsigned long d,
- unsigned long *const rp)
-{
- u_long rh;
- const u_long qh = rthal_uldivrem(h, d, &rh);
- const unsigned long long t = __rthal_u64fromu32(rh, l);
- const u_long ql = rthal_uldivrem(t, d, rp);
-
- return __rthal_u64fromu32(qh, ql);
-}
-
-/* Slow long long division. Uses rthal_uldivrem, hence has the same property:
- the compiler removes redundant calls. */
-static inline unsigned long long
-__rthal_i386_ulldiv (const unsigned long long ull,
- const unsigned d,
- unsigned long *const rp)
-{
- unsigned long h, l;
- __rthal_u64tou32(ull, h, l);
- return __rthal_div96by32(h, l, d, rp);
-}
-#define rthal_ulldiv(ull,d,rp) __rthal_i386_ulldiv((ull),(d),(rp))
-
#include <asm-generic/xenomai/hal.h> /* Read the generic bits. */
#ifndef CONFIG_X86_WP_WORKS_OK
Index: include/asm-i386/arith.h
===================================================================
--- include/asm-i386/arith.h (revision 0)
+++ include/asm-i386/arith.h (revision 0)
@@ -0,0 +1,107 @@
+/**
+ * @ingroup hal
+ * @file
+ *
+ * Arithmetic/conversion routines for x86.
+ *
+ * Original RTAI/x86 HAL services from: \n
+ * Copyright © 2000 Paolo Mantegazza, \n
+ * Copyright © 2000 Steve Papacharalambous, \n
+ * Copyright © 2000 Stuart Hughes, \n
+ * and others.
+ *
+ * RTAI/x86 rewrite over Adeos: \n
+ * Copyright © 2002,2003 Philippe Gerum.
+ * Major refactoring for Xenomai: \n
+ * Copyright © 2004,2005 Philippe Gerum.
+ * Arithmetic/conversion routines: \n
+ * Copyright © 2005 Gilles Chanteperdrix.
+ *
+ * Xenomai is free software; you can redistribute it and/or
+ * modify it under the terms of the GNU General Public License as
+ * published by the Free Software Foundation, Inc., 675 Mass Ave,
+ * Cambridge MA 02139, USA; either version 2 of the License, or (at
+ * your option) any later version.
+ *
+ * Xenomai is distributed in the hope that it will be useful,
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
+ * General Public License for more details.
+ *
+ * You should have received a copy of the GNU General Public License
+ * along with Xenomai; if not, write to the Free Software
+ * Foundation, Inc., 59 Temple Place - Suite 330, Boston, MA
+ * 02111-1307, USA.
+ */
+
+#ifndef _XENO_ASM_I386_ARITH_H
+#define _XENO_ASM_I386_ARITH_H
+
+#define __rthal_u64tou32(ull, h, l) ({ \
+ unsigned long long _ull = (ull); \
+ (l) = _ull & 0xffffffff; \
+ (h) = _ull >> 32; \
+})
+
+#define __rthal_u64fromu32(h, l) ({ \
+ unsigned long long _ull; \
+ asm ( "": "=A"(_ull) : "d"(h), "a"(l)); \
+ _ull; \
+})
+
+/* const helper for rthal_uldivrem, so that the compiler will eliminate
+ multiple calls with same arguments, at no additionnal cost. */
+static inline __attribute_const__ unsigned long long
+__rthal_uldivrem(const unsigned long long ull, const unsigned long d)
+{
+ unsigned long long ret;
+ __asm__ ("divl %1" : "=A,A"(ret) : "r,?m"(d), "A,A"(ull));
+ /* Exception if quotient does not fit on unsigned long. */
+ return ret;
+}
+
+/* Fast long long division: when the quotient and remainder fit on 32 bits. */
+static inline unsigned long __rthal_i386_uldivrem(unsigned long long ull,
+ const unsigned d,
+ unsigned long *const rp)
+{
+ unsigned long q, r;
+ ull = __rthal_uldivrem(ull, d);
+ __asm__ ( "": "=d"(r), "=a"(q) : "A"(ull));
+ if(rp)
+ *rp = r;
+ return q;
+}
+#define rthal_uldivrem(ull, d, rp) __rthal_i386_uldivrem((ull),(d),(rp))
+
+/* Division of an unsigned 96 bits ((h << 32) + l) by an unsigned 32 bits.
+ Building block for ulldiv. */
+static inline unsigned long long __rthal_div96by32 (const unsigned long long h,
+ const unsigned long l,
+ const unsigned long d,
+ unsigned long *const rp)
+{
+ u_long rh;
+ const u_long qh = rthal_uldivrem(h, d, &rh);
+ const unsigned long long t = __rthal_u64fromu32(rh, l);
+ const u_long ql = rthal_uldivrem(t, d, rp);
+
+ return __rthal_u64fromu32(qh, ql);
+}
+
+/* Slow long long division. Uses rthal_uldivrem, hence has the same property:
+ the compiler removes redundant calls. */
+static inline unsigned long long
+__rthal_i386_ulldiv (const unsigned long long ull,
+ const unsigned d,
+ unsigned long *const rp)
+{
+ unsigned long h, l;
+ __rthal_u64tou32(ull, h, l);
+ return __rthal_div96by32(h, l, d, rp);
+}
+#define rthal_ulldiv(ull,d,rp) __rthal_i386_ulldiv((ull),(d),(rp))
+
+#include <asm-generic/xenomai/arith.h>
+
+#endif /* _XENO_ASM_I386_ARITH_H */
Property changes on: include/asm-i386/arith.h
___________________________________________________________________
Name: svn:executable
+ *
Index: include/asm-x86_64/Makefile.in
===================================================================
--- include/asm-x86_64/Makefile.in (revision 2454)
+++ include/asm-x86_64/Makefile.in (working copy)
@@ -230,8 +230,8 @@
target_os = @target_os@
target_vendor = @target_vendor@
includesubdir = $(includedir)/asm-x86_64
-includesub_HEADERS = atomic.h smi.h switch.h calibration.h features.h \
- hal.h syscall.h system.h wrappers.h fptest.h
+includesub_HEADERS = arith.h atomic.h calibration.h features.h fptest.h \
+ hal.h smi.h switch.h syscall.h system.h wrappers.h
SUBDIRS = bits
all: all-recursive
Index: include/asm-x86_64/Makefile.am
===================================================================
--- include/asm-x86_64/Makefile.am (revision 2454)
+++ include/asm-x86_64/Makefile.am (working copy)
@@ -1,6 +1,6 @@
includesubdir = $(includedir)/asm-x86_64
-includesub_HEADERS = atomic.h smi.h switch.h calibration.h features.h \
- hal.h syscall.h system.h wrappers.h fptest.h
+includesub_HEADERS = arith.h atomic.h calibration.h features.h fptest.h \
+ hal.h smi.h switch.h syscall.h system.h wrappers.h
SUBDIRS = bits
Index: include/asm-x86_64/arith.h
===================================================================
--- include/asm-x86_64/arith.h (revision 0)
+++ include/asm-x86_64/arith.h (revision 0)
@@ -0,0 +1,6 @@
+#ifndef _XENO_ASM_X86_64_ARITH_H
+#define _XENO_ASM_X86_64_ARITH_H
+
+#include <asm-generic/xenomai/arith.h>
+
+#endif /* _XENO_ASM_X86_64_ARITH_H */
Property changes on: include/asm-x86_64/arith.h
___________________________________________________________________
Name: svn:executable
+ *
Index: include/nucleus/shadow.h
===================================================================
--- include/nucleus/shadow.h (revision 2454)
+++ include/nucleus/shadow.h (working copy)
@@ -31,6 +31,7 @@
/* Events sent to the interface callback */
#define XNSHADOW_CLIENT_ATTACH 0
#define XNSHADOW_CLIENT_DETACH 1
+#define XNSHADOW_CLIENT_INFO 2
#ifdef __cplusplus
extern "C" {
Index: src/skins/psos+/init.c
===================================================================
--- src/skins/psos+/init.c (revision 2454)
+++ src/skins/psos+/init.c (working copy)
@@ -27,6 +27,7 @@
{
__psos_muxid =
xeno_bind_skin(PSOS_SKIN_MAGIC, "psos", "xeno_psos");
+ __psos_muxid = __xn_mux_shifted_id(__psos_muxid);
}
void k_fatal(u_long err_code, u_long flags)
Index: src/skins/rtai/init.c
===================================================================
--- src/skins/rtai/init.c (revision 2454)
+++ src/skins/rtai/init.c (working copy)
@@ -30,4 +30,5 @@
{
__rtai_muxid =
xeno_bind_skin(RTAI_SKIN_MAGIC, "RTAI", "xeno_rtai");
+ __rtai_muxid = __xn_mux_shifted_id(__rtai_muxid);
}
Index: src/skins/posix/init.c
===================================================================
--- src/skins/posix/init.c (revision 2454)
+++ src/skins/posix/init.c (working copy)
@@ -35,6 +35,7 @@
static int fork_handler_registered;
int __wrap_pthread_setschedparam(pthread_t, int, const struct sched_param *);
+void pse51_clock_init(int);
static __attribute__ ((constructor))
void __init_posix_interface(void)
@@ -42,9 +43,15 @@
struct sched_param parm;
int muxid, err;
- __pse51_muxid =
+ muxid =
xeno_bind_skin(PSE51_SKIN_MAGIC, "POSIX", "xeno_posix");
+#ifdef CONFIG_XENO_HW_DIRECT_TSC
+ pse51_clock_init(muxid);
+#endif /* CONFIG_XENO_HW_DIRECT_TSC */
+
+ __pse51_muxid = __xn_mux_shifted_id(muxid);
+
muxid = XENOMAI_SYSBIND(RTDM_SKIN_MAGIC,
XENOMAI_FEAT_DEP, XENOMAI_ABI_REV, NULL);
if (muxid > 0) {
Index: src/skins/posix/clock.c
===================================================================
--- src/skins/posix/clock.c (revision 2454)
+++ src/skins/posix/clock.c (working copy)
@@ -20,9 +20,25 @@
#include <pthread.h> /* For pthread_setcanceltype. */
#include <posix/syscall.h>
#include <time.h>
+#include <asm/xenomai/arith.h>
extern int __pse51_muxid;
+#ifdef CONFIG_XENO_HW_DIRECT_TSC
+static xnsysinfo_t sysinfo;
+static pse51_sysinfo_t skininfo;
+
+void pse51_clock_init(int muxid)
+{
+ int err = -XENOMAI_SYSCALL3(__xn_sys_info, muxid, &sysinfo, &skininfo);
+ if (err) {
+ fprintf(stderr, "Xenomai Posix skin init: "
+ "sys_info: %s\n", strerror(err));
+ exit(EXIT_FAILURE);
+ }
+}
+#endif /* CONFIG_XENO_HW_DIRECT_TSC */
+
int __wrap_clock_getres(clockid_t clock_id, struct timespec *tp)
{
int err = -XENOMAI_SKINCALL2(__pse51_muxid,
@@ -39,11 +55,39 @@
int __wrap_clock_gettime(clockid_t clock_id, struct timespec *tp)
{
- int err = -XENOMAI_SKINCALL2(__pse51_muxid,
- __pse51_clock_gettime,
- clock_id,
- tp);
+ int err;
+#ifdef CONFIG_XENO_HW_DIRECT_TSC
+ unsigned long long tsc, ns;
+ switch(clock_id) {
+ case CLOCK_MONOTONIC:
+ tsc = __xn_rdtsc();
+ ns = xnarch_llimd(tsc, 1000000000, sysinfo.cpufreq);
+ tp->tv_sec = ns / 1000000000;
+ tp->tv_nsec = ns % 1000000000;
+ return 0;
+ case CLOCK_REALTIME:
+ if (skininfo.period != 1)
+ break;
+
+ tsc = __xn_rdtsc();
+ ns = xnarch_llimd(tsc, 1000000000, sysinfo.cpufreq);
+ ns += skininfo.wallclock_offset;
+ tp->tv_sec = ns / 1000000000;
+ tp->tv_nsec = ns % 1000000000;
+ return 0;
+
+ default:
+ errno = EINVAL;
+ return -1;
+ }
+#endif /* CONFIG_XENO_HW_DIRECT_TSC */
+
+ err = -XENOMAI_SKINCALL2(__pse51_muxid,
+ __pse51_clock_gettime,
+ clock_id,
+ tp);
+
if (!err)
return 0;
Index: src/skins/vrtx/init.c
===================================================================
--- src/skins/vrtx/init.c (revision 2454)
+++ src/skins/vrtx/init.c (working copy)
@@ -43,6 +43,7 @@
__vrtx_muxid =
xeno_bind_skin(VRTX_SKIN_MAGIC, "vrtx", "xeno_vrtx");
+ __vrtx_muxid = __xn_mux_shifted_id(__vrtx_muxid);
/* Allocate a TSD key for indexing self task pointers. */
Index: src/skins/vxworks/init.c
===================================================================
--- src/skins/vxworks/init.c (revision 2454)
+++ src/skins/vxworks/init.c (working copy)
@@ -41,6 +41,7 @@
{
__vxworks_muxid = xeno_bind_skin(VXWORKS_SKIN_MAGIC,
"vxworks", "xeno_vxworks");
+ __vxworks_muxid = __xn_mux_shifted_id(__vxworks_muxid);
/* Allocate a TSD key for indexing self task pointers. */
Index: src/skins/native/init.c
===================================================================
--- src/skins/native/init.c (revision 2454)
+++ src/skins/native/init.c (working copy)
@@ -42,6 +42,7 @@
{
__native_muxid =
xeno_bind_skin(XENO_SKIN_MAGIC, "native", "xeno_native");
+ __native_muxid = __xn_mux_shifted_id(__native_muxid);
/* Allocate a TSD key for indexing self task pointers. */
Index: src/skins/rtdm/init.c
===================================================================
--- src/skins/rtdm/init.c (revision 2454)
+++ src/skins/rtdm/init.c (working copy)
@@ -34,4 +34,6 @@
__rtdm_muxid =
xeno_bind_skin_opt(RTDM_SKIN_MAGIC, "rtdm", "xeno_rtdm");
+ __rtdm_muxid = __xn_mux_shifted_id(__rtdm_muxid);
+
}
Index: ksrc/skins/posix/syscall.c
===================================================================
--- ksrc/skins/posix/syscall.c (revision 2454)
+++ ksrc/skins/posix/syscall.c (working copy)
@@ -2843,6 +2843,7 @@
static void *pse51_eventcb(int event, void *data)
{
+ pse51_sysinfo_t si;
pse51_queues_t *q;
switch(event) {
@@ -2890,8 +2891,18 @@
xnarch_sysfree(q, sizeof(*q));
return NULL;
+
+ case XNSHADOW_CLIENT_INFO:
+ if (!__xn_access_ok(current, VERIFY_WRITE, data, sizeof(si)))
+ return ERR_PTR(-EFAULT);
+
+ si.period = xntbase_get_tickval(pse51_tbase);
+ si.wallclock_offset = pse51_tbase->wallclock_offset;
+
+ __xn_copy_to_user(current, data, &si, sizeof(si));
+
+ return 0;
}
-
return ERR_PTR(-EINVAL);
}
Index: ksrc/nucleus/shadow.c
===================================================================
--- ksrc/nucleus/shadow.c (revision 2454)
+++ ksrc/nucleus/shadow.c (working copy)
@@ -1557,10 +1557,14 @@
static int xnshadow_sys_info(struct task_struct *curr, struct pt_regs *regs)
{
- /* UNUSED int muxid = __xn_reg_arg1(regs); */
+ int muxid = __xn_reg_arg1(regs);
u_long infarg = __xn_reg_arg2(regs);
xnsysinfo_t info;
+ void *err;
+ if (muxid < 0 || muxid >= XENOMAI_MUX_NR)
+ return -EINVAL;
+
if (!__xn_access_ok(curr, VERIFY_WRITE, infarg, sizeof(info)))
return -EFAULT;
@@ -1568,6 +1572,13 @@
info.tickval = xntbase_get_tickval(&nktbase);
__xn_copy_to_user(curr, (void *)infarg, &info, sizeof(info));
+ if (muxtable[muxid].systab && muxtable[muxid].eventcb) {
+ err = muxtable[muxid].eventcb(XNSHADOW_CLIENT_INFO,
+ (void *) __xn_reg_arg3(regs));
+ if (IS_ERR(err))
+ return PTR_ERR(err);
+ }
+
return 0;
}
^ permalink raw reply [flat|nested] 9+ messages in thread
* Re: [Xenomai-core] [PATCH] Use tsc for implementation of clock_gettime.
2007-05-17 11:29 [Xenomai-core] [PATCH] Use tsc for implementation of clock_gettime Gilles Chanteperdrix
@ 2007-05-17 12:43 ` Jan Kiszka
2007-05-17 12:54 ` Gilles Chanteperdrix
0 siblings, 1 reply; 9+ messages in thread
From: Jan Kiszka @ 2007-05-17 12:43 UTC (permalink / raw)
To: Gilles Chanteperdrix; +Cc: xenomai
[-- Attachment #1: Type: text/plain, Size: 569 bytes --]
Gilles Chanteperdrix wrote:
> Hi,
>
> here comes, for review, a patch which reduces the overhead of
> clock_gettime by directly reading the tsc in user-space for
> architectures that support it.
Highly welcome. But I have one concern: How and when do you propagate
wallclock_offset changes to user space?
I think we need some vsyscall-alike approach for this, some read-only
page that is mapped into every RT process, containing things like a
regularly updated offset (seqlock fashioned) or other read-only
information (shadow mode? cpu id?).
Jan
[-- Attachment #2: OpenPGP digital signature --]
[-- Type: application/pgp-signature, Size: 249 bytes --]
^ permalink raw reply [flat|nested] 9+ messages in thread
* Re: [Xenomai-core] [PATCH] Use tsc for implementation of clock_gettime.
2007-05-17 12:43 ` Jan Kiszka
@ 2007-05-17 12:54 ` Gilles Chanteperdrix
2007-05-17 13:00 ` Jan Kiszka
0 siblings, 1 reply; 9+ messages in thread
From: Gilles Chanteperdrix @ 2007-05-17 12:54 UTC (permalink / raw)
To: Jan Kiszka; +Cc: xenomai
Jan Kiszka wrote:
> Gilles Chanteperdrix wrote:
> > Hi,
> >
> > here comes, for review, a patch which reduces the overhead of
> > clock_gettime by directly reading the tsc in user-space for
> > architectures that support it.
>
> Highly welcome. But I have one concern: How and when do you propagate
> wallclock_offset changes to user space?
Since clock_settime is not implemented, never, but if clock_settime was
implemented, clock_settime would re-issue the __xn_sys_info syscall.
>
> I think we need some vsyscall-alike approach for this, some read-only
> page that is mapped into every RT process, containing things like a
> regularly updated offset (seqlock fashioned) or other read-only
> information (shadow mode? cpu id?).
As I said yesterday, having a page mapped at a different address in
kernel-space and user-space is not an option on ARM, because each
mapping of this page would have a separated cache.
--
Gilles Chanteperdrix.
^ permalink raw reply [flat|nested] 9+ messages in thread
* Re: [Xenomai-core] [PATCH] Use tsc for implementation of clock_gettime.
2007-05-17 12:54 ` Gilles Chanteperdrix
@ 2007-05-17 13:00 ` Jan Kiszka
2007-05-17 13:10 ` Gilles Chanteperdrix
0 siblings, 1 reply; 9+ messages in thread
From: Jan Kiszka @ 2007-05-17 13:00 UTC (permalink / raw)
To: Gilles Chanteperdrix; +Cc: xenomai
[-- Attachment #1: Type: text/plain, Size: 1404 bytes --]
Gilles Chanteperdrix wrote:
> Jan Kiszka wrote:
> > Gilles Chanteperdrix wrote:
> > > Hi,
> > >
> > > here comes, for review, a patch which reduces the overhead of
> > > clock_gettime by directly reading the tsc in user-space for
> > > architectures that support it.
> >
> > Highly welcome. But I have one concern: How and when do you propagate
> > wallclock_offset changes to user space?
>
> Since clock_settime is not implemented, never, but if clock_settime was
> implemented, clock_settime would re-issue the __xn_sys_info syscall.
This excludes automatic clock adjustment, something I'm convinced we
will have to provide in the future.
>
> >
> > I think we need some vsyscall-alike approach for this, some read-only
> > page that is mapped into every RT process, containing things like a
> > regularly updated offset (seqlock fashioned) or other read-only
> > information (shadow mode? cpu id?).
>
> As I said yesterday, having a page mapped at a different address in
> kernel-space and user-space is not an option on ARM, because each
> mapping of this page would have a separated cache.
>
Hmm, what about the kernel writing directly to the (current) user
address? We then only need to keep track of the per-mm address and have
not fixed kernel equivalent. At least on arm, other archs may handle
this differently at compile time.
Jan
[-- Attachment #2: OpenPGP digital signature --]
[-- Type: application/pgp-signature, Size: 249 bytes --]
^ permalink raw reply [flat|nested] 9+ messages in thread
* Re: [Xenomai-core] [PATCH] Use tsc for implementation of clock_gettime.
2007-05-17 13:00 ` Jan Kiszka
@ 2007-05-17 13:10 ` Gilles Chanteperdrix
2007-05-17 13:31 ` Jan Kiszka
0 siblings, 1 reply; 9+ messages in thread
From: Gilles Chanteperdrix @ 2007-05-17 13:10 UTC (permalink / raw)
To: Jan Kiszka; +Cc: xenomai
Jan Kiszka wrote:
> Gilles Chanteperdrix wrote:
> > Jan Kiszka wrote:
> > > Gilles Chanteperdrix wrote:
> > > > Hi,
> > > >
> > > > here comes, for review, a patch which reduces the overhead of
> > > > clock_gettime by directly reading the tsc in user-space for
> > > > architectures that support it.
> > >
> > > Highly welcome. But I have one concern: How and when do you propagate
> > > wallclock_offset changes to user space?
> >
> > Since clock_settime is not implemented, never, but if clock_settime was
> > implemented, clock_settime would re-issue the __xn_sys_info syscall.
>
> This excludes automatic clock adjustment, something I'm convinced we
> will have to provide in the future.
When we provide automatic clock adjustment, we will have to devise
something more subtle than just an offset, so we will have to redesign
posix clocks support anyway. Maybe clock_gettime(CLOCK_REALTIME) will
then always be a syscall. After all, rt_timer_read is a syscall. If you
want the fast clock, use CLOCK_MONOTONIC or rt_timer_tsc.
--
Gilles Chanteperdrix.
^ permalink raw reply [flat|nested] 9+ messages in thread
* Re: [Xenomai-core] [PATCH] Use tsc for implementation of clock_gettime.
2007-05-17 13:10 ` Gilles Chanteperdrix
@ 2007-05-17 13:31 ` Jan Kiszka
2007-05-17 15:21 ` Gilles Chanteperdrix
0 siblings, 1 reply; 9+ messages in thread
From: Jan Kiszka @ 2007-05-17 13:31 UTC (permalink / raw)
To: Gilles Chanteperdrix; +Cc: xenomai
Gilles Chanteperdrix wrote:
> Jan Kiszka wrote:
> > Gilles Chanteperdrix wrote:
> > > Jan Kiszka wrote:
> > > > Gilles Chanteperdrix wrote:
> > > > > Hi,
> > > > >
> > > > > here comes, for review, a patch which reduces the overhead of
> > > > > clock_gettime by directly reading the tsc in user-space for
> > > > > architectures that support it.
> > > >
> > > > Highly welcome. But I have one concern: How and when do you propagate
> > > > wallclock_offset changes to user space?
> > >
> > > Since clock_settime is not implemented, never, but if clock_settime was
> > > implemented, clock_settime would re-issue the __xn_sys_info syscall.
> >
> > This excludes automatic clock adjustment, something I'm convinced we
> > will have to provide in the future.
>
> When we provide automatic clock adjustment, we will have to devise
> something more subtle than just an offset, so we will have to redesign
I think offset + scaling factor.
> posix clocks support anyway. Maybe clock_gettime(CLOCK_REALTIME) will
> then always be a syscall. After all, rt_timer_read is a syscall. If you
> want the fast clock, use CLOCK_MONOTONIC or rt_timer_tsc.
Actually, the issue of the intermediate approach starts earlier: as soon
as you have clock_settime. Then you need to sync the offset across
applications in different processes.
Having clock_monotonic (and maybe also rt_timer_tsc2ns) as a fast
variant already now is not bad. But beyond that, before introducing a
new interface between kernel and user space, I would like to consider
the effort to get it future-proof immediately. That doesn't mean that we
already have to implement clock adjustment, but we may already prepare
the prerequisites.
Jan
^ permalink raw reply [flat|nested] 9+ messages in thread
* Re: [Xenomai-core] [PATCH] Use tsc for implementation of clock_gettime.
2007-05-17 13:31 ` Jan Kiszka
@ 2007-05-17 15:21 ` Gilles Chanteperdrix
2007-05-17 15:56 ` Gilles Chanteperdrix
0 siblings, 1 reply; 9+ messages in thread
From: Gilles Chanteperdrix @ 2007-05-17 15:21 UTC (permalink / raw)
To: Jan Kiszka; +Cc: xenomai
[-- Attachment #1: message body and .signature --]
[-- Type: text/plain, Size: 2095 bytes --]
Jan Kiszka wrote:
> Gilles Chanteperdrix wrote:
> > Jan Kiszka wrote:
> > > Gilles Chanteperdrix wrote:
> > > > Jan Kiszka wrote:
> > > > > Gilles Chanteperdrix wrote:
> > > > > > Hi,
> > > > > >
> > > > > > here comes, for review, a patch which reduces the overhead of
> > > > > > clock_gettime by directly reading the tsc in user-space for
> > > > > > architectures that support it.
> > > > >
> > > > > Highly welcome. But I have one concern: How and when do you propagate
> > > > > wallclock_offset changes to user space?
> > > >
> > > > Since clock_settime is not implemented, never, but if clock_settime was
> > > > implemented, clock_settime would re-issue the __xn_sys_info syscall.
> > >
> > > This excludes automatic clock adjustment, something I'm convinced we
> > > will have to provide in the future.
> >
> > When we provide automatic clock adjustment, we will have to devise
> > something more subtle than just an offset, so we will have to redesign
>
> I think offset + scaling factor.
>
> > posix clocks support anyway. Maybe clock_gettime(CLOCK_REALTIME) will
> > then always be a syscall. After all, rt_timer_read is a syscall. If you
> > want the fast clock, use CLOCK_MONOTONIC or rt_timer_tsc.
>
> Actually, the issue of the intermediate approach starts earlier: as soon
> as you have clock_settime. Then you need to sync the offset across
> applications in different processes.
>
> Having clock_monotonic (and maybe also rt_timer_tsc2ns) as a fast
> variant already now is not bad. But beyond that, before introducing a
> new interface between kernel and user space, I would like to consider
> the effort to get it future-proof immediately. That doesn't mean that we
> already have to implement clock adjustment, but we may already prepare
> the prerequisites.
Ok. Here is a new version that does not break the ABI, and where
clock_gettime(CLOCK_REALTIME) remains a syscall.
Note that it also adds user-space conversions between tsc and ns to the
native skin.
--
Gilles Chanteperdrix.
[-- Attachment #2: xeno-clock_gettime-direct-tsc.2.diff --]
[-- Type: text/plain, Size: 36393 bytes --]
Index: include/asm-ia64/Makefile.in
===================================================================
--- include/asm-ia64/Makefile.in (revision 2454)
+++ include/asm-ia64/Makefile.in (working copy)
@@ -230,7 +230,9 @@
target_os = @target_os@
target_vendor = @target_vendor@
includesubdir = $(includedir)/asm-ia64
-includesub_HEADERS = atomic.h calibration.h features.h hal.h syscall.h system.h wrappers.h fptest.h
+includesub_HEADERS = arith.h atomic.h calibration.h features.h fptest.h \
+ hal.h syscall.h system.h wrappers.h
+
SUBDIRS = bits
all: all-recursive
Index: include/asm-ia64/Makefile.am
===================================================================
--- include/asm-ia64/Makefile.am (revision 2454)
+++ include/asm-ia64/Makefile.am (working copy)
@@ -1,5 +1,6 @@
includesubdir = $(includedir)/asm-ia64
-includesub_HEADERS = atomic.h calibration.h features.h hal.h syscall.h system.h wrappers.h fptest.h
+includesub_HEADERS = arith.h atomic.h calibration.h features.h fptest.h \
+ hal.h syscall.h system.h wrappers.h
SUBDIRS = bits
Index: include/asm-ia64/arith.h
===================================================================
--- include/asm-ia64/arith.h (revision 0)
+++ include/asm-ia64/arith.h (revision 0)
@@ -0,0 +1,6 @@
+#ifndef _XENO_ASM_IA64_ARITH_H
+#define _XENO_ASM_IA64_ARITH_H
+
+#include <asm-generic/xenomai/arith.h>
+
+#endif /* _XENO_ASM_IA64_ARITH_H */
Index: include/asm-blackfin/Makefile.in
===================================================================
--- include/asm-blackfin/Makefile.in (revision 2454)
+++ include/asm-blackfin/Makefile.in (working copy)
@@ -230,8 +230,8 @@
target_os = @target_os@
target_vendor = @target_vendor@
includesubdir = $(includedir)/asm-blackfin
-includesub_HEADERS = atomic.h calibration.h features.h hal.h \
- syscall.h system.h wrappers.h fptest.h
+includesub_HEADERS = arith.h atomic.h calibration.h features.h fptest.h \
+ hal.h syscall.h system.h wrappers.h
SUBDIRS = bits
all: all-recursive
Index: include/asm-blackfin/Makefile.am
===================================================================
--- include/asm-blackfin/Makefile.am (revision 2454)
+++ include/asm-blackfin/Makefile.am (working copy)
@@ -1,6 +1,6 @@
includesubdir = $(includedir)/asm-blackfin
-includesub_HEADERS = atomic.h calibration.h features.h hal.h \
- syscall.h system.h wrappers.h fptest.h
+includesub_HEADERS = arith.h atomic.h calibration.h features.h fptest.h \
+ hal.h syscall.h system.h wrappers.h
SUBDIRS = bits
Index: include/asm-blackfin/arith.h
===================================================================
--- include/asm-blackfin/arith.h (revision 0)
+++ include/asm-blackfin/arith.h (revision 0)
@@ -0,0 +1,6 @@
+#ifndef _XENO_ASM_BLACKFIN_ARITH_H
+#define _XENO_ASM_BLACKFIN_ARITH_H
+
+#include <asm-generic/xenomai/arith.h>
+
+#endif /* _XENO_ASM_BLACKFIN_ARITH_H */
Index: include/asm-generic/syscall.h
===================================================================
--- include/asm-generic/syscall.h (revision 2454)
+++ include/asm-generic/syscall.h (working copy)
@@ -55,7 +55,6 @@
unsigned long long cpufreq; /* CPU frequency */
unsigned long tickval; /* Tick duration (ns) */
-
} xnsysinfo_t;
#define SIGHARDEN SIGWINCH
Index: include/asm-generic/bits/bind.h
===================================================================
--- include/asm-generic/bits/bind.h (revision 2454)
+++ include/asm-generic/bits/bind.h (working copy)
@@ -62,7 +62,7 @@
sa.sa_flags = 0;
sigaction(SIGXCPU, &sa, NULL);
- return __xn_mux_shifted_id(muxid);
+ return muxid;
}
static inline int
@@ -105,7 +105,7 @@
exit(1);
}
- return __xn_mux_shifted_id(muxid);
+ return muxid;
}
#endif /* _XENO_ASM_GENERIC_BITS_BIND_H */
Index: include/asm-generic/Makefile.in
===================================================================
--- include/asm-generic/Makefile.in (revision 2454)
+++ include/asm-generic/Makefile.in (working copy)
@@ -230,7 +230,7 @@
target_os = @target_os@
target_vendor = @target_vendor@
includesubdir = $(includedir)/asm-generic
-includesub_HEADERS = features.h hal.h syscall.h system.h wrappers.h
+includesub_HEADERS = arith.h features.h hal.h syscall.h system.h wrappers.h
SUBDIRS = bits
all: all-recursive
Index: include/asm-generic/Makefile.am
===================================================================
--- include/asm-generic/Makefile.am (revision 2454)
+++ include/asm-generic/Makefile.am (working copy)
@@ -1,5 +1,5 @@
includesubdir = $(includedir)/asm-generic
-includesub_HEADERS = features.h hal.h syscall.h system.h wrappers.h
+includesub_HEADERS = arith.h features.h hal.h syscall.h system.h wrappers.h
SUBDIRS = bits
Index: include/asm-generic/hal.h
===================================================================
--- include/asm-generic/hal.h (revision 2454)
+++ include/asm-generic/hal.h (working copy)
@@ -37,8 +37,8 @@
#include <linux/kallsyms.h>
#include <linux/init.h>
#include <asm/byteorder.h>
-#include <asm/div64.h>
#include <asm/xenomai/wrappers.h>
+#include <asm/xenomai/arith.h>
#define RTHAL_DOMAIN_ID 0x58454e4f
@@ -385,121 +385,6 @@
#define rthal_printk printk
-#ifdef __BIG_ENDIAN
-#define endianstruct struct { unsigned _h; unsigned _l; } _s
-#else /* __LITTLE_ENDIAN */
-#define endianstruct struct { unsigned _l; unsigned _h; } _s
-#endif
-
-#ifndef __rthal_u64tou32
-#define __rthal_u64tou32(ull, h, l) ({ \
- union { unsigned long long _ull; \
- endianstruct; \
- } _u; \
- _u._ull = (ull); \
- (h) = _u._s._h; \
- (l) = _u._s._l; \
-})
-#endif /* !__rthal_u64tou32 */
-
-#ifndef __rthal_u64fromu32
-#define __rthal_u64fromu32(h, l) ({ \
- union { unsigned long long _ull; \
- endianstruct; \
- } _u; \
- _u._s._h = (h); \
- _u._s._l = (l); \
- _u._ull; \
-})
-#endif /* !__rthal_u64fromu32 */
-
-#ifndef rthal_ullmul
-static inline __attribute_const__ unsigned long long
-__rthal_generic_ullmul(const unsigned m0, const unsigned m1)
-{
- return (unsigned long long) m0 * m1;
-}
-#define rthal_ullmul(m0,m1) __rthal_generic_ullmul((m0),(m1))
-#endif /* !rthal_ullmul */
-
-#ifndef rthal_ulldiv
-static inline unsigned long long __rthal_generic_ulldiv (unsigned long long ull,
- const unsigned uld,
- unsigned long *const rp)
-{
- const unsigned r = do_div(ull, uld);
-
- if (rp)
- *rp = r;
-
- return ull;
-}
-#define rthal_ulldiv(ull,uld,rp) __rthal_generic_ulldiv((ull),(uld),(rp))
-#endif /* !rthal_ulldiv */
-
-#ifndef rthal_uldivrem
-#define rthal_uldivrem(ull,ul,rp) ((unsigned) rthal_ulldiv((ull),(ul),(rp)))
-#endif /* !rthal_uldivrem */
-
-#ifndef rthal_imuldiv
-static inline __attribute_const__ int __rthal_generic_imuldiv (int i,
- int mult,
- int div)
-{
- /* Returns (int)i = (unsigned long long)i*(unsigned)(mult)/(unsigned)div. */
- const unsigned long long ull = rthal_ullmul(i, mult);
- return rthal_uldivrem(ull, div, NULL);
-}
-#define rthal_imuldiv(i,m,d) __rthal_generic_imuldiv((i),(m),(d))
-#endif /* !rthal_imuldiv */
-
-#ifndef rthal_llimd
-/* Division of an unsigned 96 bits ((h << 32) + l) by an unsigned 32 bits.
- Building block for llimd. Without const qualifiers, gcc reload registers
- after each call to uldivrem. */
-static inline unsigned long long
-__rthal_generic_div96by32 (const unsigned long long h,
- const unsigned l,
- const unsigned d,
- unsigned long *const rp)
-{
- unsigned long rh;
- const unsigned qh = rthal_uldivrem(h, d, &rh);
- const unsigned long long t = __rthal_u64fromu32(rh, l);
- const unsigned ql = rthal_uldivrem(t, d, rp);
-
- return __rthal_u64fromu32(qh, ql);
-}
-
-static inline __attribute_const__
-unsigned long long __rthal_generic_ullimd (const unsigned long long op,
- const unsigned m,
- const unsigned d)
-{
- unsigned oph, opl, tlh, tll;
- unsigned long long th, tl;
-
- __rthal_u64tou32(op, oph, opl);
- tl = rthal_ullmul(opl, m);
- __rthal_u64tou32(tl, tlh, tll);
- th = rthal_ullmul(oph, m);
- th += tlh;
-
- return __rthal_generic_div96by32(th, tll, d, NULL);
-}
-
-static inline __attribute_const__ long long __rthal_generic_llimd (long long op,
- unsigned m,
- unsigned d)
-{
-
- if(op < 0LL)
- return -__rthal_generic_ullimd(-op, m, d);
- return __rthal_generic_ullimd(op, m, d);
-}
-#define rthal_llimd(ll,m,d) __rthal_generic_llimd((ll),(m),(d))
-#endif /* !rthal_llimd */
-
typedef ipipe_irq_handler_t rthal_irq_handler_t;
typedef ipipe_irq_ackfn_t rthal_irq_ackfn_t;
Index: include/asm-generic/system.h
===================================================================
--- include/asm-generic/system.h (revision 2454)
+++ include/asm-generic/system.h (working copy)
@@ -133,18 +133,6 @@
#define xnarch_logerr(fmt,args...) printk(KERN_ERR XNARCH_PROMPT fmt , ##args)
#define xnarch_printf(fmt,args...) printk(KERN_INFO XNARCH_PROMPT fmt , ##args)
-#define xnarch_ullmod(ull,uld,rem) ({ xnarch_ulldiv(ull,uld,rem); (*rem); })
-#define xnarch_uldiv(ull, d) rthal_uldivrem(ull, d, NULL)
-#define xnarch_ulmod(ull, d) ({ u_long _rem; \
- rthal_uldivrem(ull,d,&_rem); _rem; })
-
-#define xnarch_ullmul rthal_ullmul
-#define xnarch_uldivrem rthal_uldivrem
-#define xnarch_ulldiv rthal_ulldiv
-#define xnarch_imuldiv rthal_imuldiv
-#define xnarch_llimd rthal_llimd
-#define xnarch_get_cpu_tsc rthal_rdtsc
-
typedef cpumask_t xnarch_cpumask_t;
#ifdef CONFIG_SMP
Index: include/asm-generic/arith.h
===================================================================
--- include/asm-generic/arith.h (revision 0)
+++ include/asm-generic/arith.h (revision 0)
@@ -0,0 +1,173 @@
+/**
+ * @ingroup hal
+ * @file
+ *
+ * Generic arithmetic/conversion routines.
+ * Copyright © 2005 Stelian Pop.
+ * Copyright © 2005 Gilles Chanteperdrix.
+ *
+ * This program is free software; you can redistribute it and/or modify
+ * it under the terms of the GNU General Public License as published by
+ * the Free Software Foundation, Inc., 675 Mass Ave, Cambridge MA 02139,
+ * USA; either version 2 of the License, or (at your option) any later
+ * version.
+ *
+ * This program is distributed in the hope that it will be useful,
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
+ * GNU General Public License for more details.
+ *
+ * You should have received a copy of the GNU General Public License
+ * along with this program; if not, write to the Free Software
+ * Foundation, Inc., 59 Temple Place - Suite 330, Boston, MA 02111-1307, USA.
+ */
+
+/**
+ * @addtogroup hal
+ *@{*/
+
+#ifndef _XENO_ASM_GENERIC_ARITH_H
+#define _XENO_ASM_GENERIC_ARITH_H
+
+#ifdef __KERNEL__
+#include <asm/div64.h>
+#else /* !__KERNEL__ */
+static inline unsigned __rthal_do_div(unsigned long long *a, unsigned d)
+{
+ unsigned r = *a % d;
+ *a /= d;
+ return r;
+}
+#define do_div(a, d) __rthal_do_div(&(a), (d))
+#endif /* !__KERNEL__ */
+
+#ifdef __BIG_ENDIAN
+#define endianstruct struct { unsigned _h; unsigned _l; } _s
+#else /* __LITTLE_ENDIAN */
+#define endianstruct struct { unsigned _l; unsigned _h; } _s
+#endif
+
+#ifndef __rthal_u64tou32
+#define __rthal_u64tou32(ull, h, l) ({ \
+ union { unsigned long long _ull; \
+ endianstruct; \
+ } _u; \
+ _u._ull = (ull); \
+ (h) = _u._s._h; \
+ (l) = _u._s._l; \
+})
+#endif /* !__rthal_u64tou32 */
+
+#ifndef __rthal_u64fromu32
+#define __rthal_u64fromu32(h, l) ({ \
+ union { unsigned long long _ull; \
+ endianstruct; \
+ } _u; \
+ _u._s._h = (h); \
+ _u._s._l = (l); \
+ _u._ull; \
+})
+#endif /* !__rthal_u64fromu32 */
+
+#ifndef rthal_ullmul
+static inline __attribute_const__ unsigned long long
+__rthal_generic_ullmul(const unsigned m0, const unsigned m1)
+{
+ return (unsigned long long) m0 * m1;
+}
+#define rthal_ullmul(m0,m1) __rthal_generic_ullmul((m0),(m1))
+#endif /* !rthal_ullmul */
+
+#ifndef rthal_ulldiv
+static inline unsigned long long __rthal_generic_ulldiv (unsigned long long ull,
+ const unsigned uld,
+ unsigned long *const rp)
+{
+ const unsigned r = do_div(ull, uld);
+
+ if (rp)
+ *rp = r;
+
+ return ull;
+}
+#define rthal_ulldiv(ull,uld,rp) __rthal_generic_ulldiv((ull),(uld),(rp))
+#endif /* !rthal_ulldiv */
+
+#ifndef rthal_uldivrem
+#define rthal_uldivrem(ull,ul,rp) ((unsigned) rthal_ulldiv((ull),(ul),(rp)))
+#endif /* !rthal_uldivrem */
+
+#ifndef rthal_imuldiv
+static inline __attribute_const__ int __rthal_generic_imuldiv (int i,
+ int mult,
+ int div)
+{
+ /* Returns (int)i = (unsigned long long)i*(unsigned)(mult)/(unsigned)div. */
+ const unsigned long long ull = rthal_ullmul(i, mult);
+ return rthal_uldivrem(ull, div, NULL);
+}
+#define rthal_imuldiv(i,m,d) __rthal_generic_imuldiv((i),(m),(d))
+#endif /* !rthal_imuldiv */
+
+#ifndef rthal_llimd
+/* Division of an unsigned 96 bits ((h << 32) + l) by an unsigned 32 bits.
+ Building block for llimd. Without const qualifiers, gcc reload registers
+ after each call to uldivrem. */
+static inline unsigned long long
+__rthal_generic_div96by32 (const unsigned long long h,
+ const unsigned l,
+ const unsigned d,
+ unsigned long *const rp)
+{
+ unsigned long rh;
+ const unsigned qh = rthal_uldivrem(h, d, &rh);
+ const unsigned long long t = __rthal_u64fromu32(rh, l);
+ const unsigned ql = rthal_uldivrem(t, d, rp);
+
+ return __rthal_u64fromu32(qh, ql);
+}
+
+static inline __attribute_const__
+unsigned long long __rthal_generic_ullimd (const unsigned long long op,
+ const unsigned m,
+ const unsigned d)
+{
+ unsigned oph, opl, tlh, tll;
+ unsigned long long th, tl;
+
+ __rthal_u64tou32(op, oph, opl);
+ tl = rthal_ullmul(opl, m);
+ __rthal_u64tou32(tl, tlh, tll);
+ th = rthal_ullmul(oph, m);
+ th += tlh;
+
+ return __rthal_generic_div96by32(th, tll, d, NULL);
+}
+
+static inline __attribute_const__ long long __rthal_generic_llimd (long long op,
+ unsigned m,
+ unsigned d)
+{
+
+ if(op < 0LL)
+ return -__rthal_generic_ullimd(-op, m, d);
+ return __rthal_generic_ullimd(op, m, d);
+}
+#define rthal_llimd(ll,m,d) __rthal_generic_llimd((ll),(m),(d))
+#endif /* !rthal_llimd */
+
+#define xnarch_ullmod(ull,uld,rem) ({ xnarch_ulldiv(ull,uld,rem); (*rem); })
+#define xnarch_uldiv(ull, d) rthal_uldivrem(ull, d, NULL)
+#define xnarch_ulmod(ull, d) ({ u_long _rem; \
+ rthal_uldivrem(ull,d,&_rem); _rem; })
+
+#define xnarch_ullmul rthal_ullmul
+#define xnarch_uldivrem rthal_uldivrem
+#define xnarch_ulldiv rthal_ulldiv
+#define xnarch_imuldiv rthal_imuldiv
+#define xnarch_llimd rthal_llimd
+#define xnarch_get_cpu_tsc rthal_rdtsc
+
+/*@}*/
+
+#endif /* _XENO_ASM_GENERIC_ARITH_H */
Index: include/asm-powerpc/Makefile.in
===================================================================
--- include/asm-powerpc/Makefile.in (revision 2454)
+++ include/asm-powerpc/Makefile.in (working copy)
@@ -230,8 +230,8 @@
target_os = @target_os@
target_vendor = @target_vendor@
includesubdir = $(includedir)/asm-powerpc
-includesub_HEADERS = atomic.h calibration.h features.h hal.h \
- syscall.h system.h wrappers.h fptest.h
+includesub_HEADERS = arith.h atomic.h calibration.h features.h fptest.h \
+ hal.h syscall.h system.h wrappers.h
SUBDIRS = bits
all: all-recursive
Index: include/asm-powerpc/Makefile.am
===================================================================
--- include/asm-powerpc/Makefile.am (revision 2454)
+++ include/asm-powerpc/Makefile.am (working copy)
@@ -1,6 +1,6 @@
includesubdir = $(includedir)/asm-powerpc
-includesub_HEADERS = atomic.h calibration.h features.h hal.h \
- syscall.h system.h wrappers.h fptest.h
+includesub_HEADERS = arith.h atomic.h calibration.h features.h fptest.h \
+ hal.h syscall.h system.h wrappers.h
SUBDIRS = bits
Index: include/asm-powerpc/arith.h
===================================================================
--- include/asm-powerpc/arith.h (revision 0)
+++ include/asm-powerpc/arith.h (revision 0)
@@ -0,0 +1,6 @@
+#ifndef _XENO_ASM_POWERPC_ARITH_H
+#define _XENO_ASM_POWERPC_ARITH_H
+
+#include <asm-generic/xenomai/arith.h>
+
+#endif /* _XENO_ASM_POWERPC_ARITH_H */
Index: include/asm-arm/Makefile.in
===================================================================
--- include/asm-arm/Makefile.in (revision 2454)
+++ include/asm-arm/Makefile.in (working copy)
@@ -230,8 +230,8 @@
target_os = @target_os@
target_vendor = @target_vendor@
includesubdir = $(includedir)/asm-arm
-includesub_HEADERS = atomic.h calibration.h features.h hal.h \
- syscall.h system.h wrappers.h fptest.h
+includesub_HEADERS = arith.h atomic.h calibration.h features.h fptest.h \
+ hal.h syscall.h system.h wrappers.h
SUBDIRS = bits
all: all-recursive
Index: include/asm-arm/Makefile.am
===================================================================
--- include/asm-arm/Makefile.am (revision 2454)
+++ include/asm-arm/Makefile.am (working copy)
@@ -1,6 +1,6 @@
includesubdir = $(includedir)/asm-arm
-includesub_HEADERS = atomic.h calibration.h features.h hal.h \
- syscall.h system.h wrappers.h fptest.h
+includesub_HEADERS = arith.h atomic.h calibration.h features.h fptest.h \
+ hal.h syscall.h system.h wrappers.h
SUBDIRS = bits
Index: include/asm-arm/arith.h
===================================================================
--- include/asm-arm/arith.h (revision 0)
+++ include/asm-arm/arith.h (revision 0)
@@ -0,0 +1,6 @@
+#ifndef _XENO_ASM_ARM_ARITH_H
+#define _XENO_ASM_ARM_ARITH_H
+
+#include <asm-generic/xenomai/arith.h>
+
+#endif /* _XENO_ASM_ARM_ARITH_H */
Index: include/posix/syscall.h
===================================================================
--- include/posix/syscall.h (revision 2454)
+++ include/posix/syscall.h (working copy)
@@ -100,6 +100,11 @@
#define __pse51_condattr_setpshared 74
#define __pse51_thread_getschedparam 75
+typedef struct pse51_sysinfo {
+ unsigned period;
+ unsigned long long wallclock_offset;
+} pse51_sysinfo_t;
+
#ifdef __KERNEL__
#ifdef __cplusplus
Index: include/asm-i386/Makefile.in
===================================================================
--- include/asm-i386/Makefile.in (revision 2454)
+++ include/asm-i386/Makefile.in (working copy)
@@ -230,8 +230,8 @@
target_os = @target_os@
target_vendor = @target_vendor@
includesubdir = $(includedir)/asm-i386
-includesub_HEADERS = atomic.h calibration.h features.h \
- hal.h smi.h switch.h syscall.h system.h wrappers.h fptest.h
+includesub_HEADERS = arith.h atomic.h calibration.h features.h fptest.h \
+ hal.h smi.h switch.h syscall.h system.h wrappers.h
SUBDIRS = bits
all: all-recursive
Index: include/asm-i386/Makefile.am
===================================================================
--- include/asm-i386/Makefile.am (revision 2454)
+++ include/asm-i386/Makefile.am (working copy)
@@ -1,6 +1,6 @@
includesubdir = $(includedir)/asm-i386
-includesub_HEADERS = atomic.h calibration.h features.h \
- hal.h smi.h switch.h syscall.h system.h wrappers.h fptest.h
+includesub_HEADERS = arith.h atomic.h calibration.h features.h fptest.h \
+ hal.h smi.h switch.h syscall.h system.h wrappers.h
SUBDIRS = bits
Index: include/asm-i386/hal.h
===================================================================
--- include/asm-i386/hal.h (revision 2454)
+++ include/asm-i386/hal.h (working copy)
@@ -39,71 +39,6 @@
#include <asm/xenomai/wrappers.h>
-#define __rthal_u64tou32(ull, h, l) ({ \
- unsigned long long _ull = (ull); \
- (l) = _ull & 0xffffffff; \
- (h) = _ull >> 32; \
-})
-
-#define __rthal_u64fromu32(h, l) ({ \
- unsigned long long _ull; \
- asm ( "": "=A"(_ull) : "d"(h), "a"(l)); \
- _ull; \
-})
-
-/* const helper for rthal_uldivrem, so that the compiler will eliminate
- multiple calls with same arguments, at no additionnal cost. */
-static inline __attribute_const__ unsigned long long
-__rthal_uldivrem(const unsigned long long ull, const unsigned long d)
-{
- unsigned long long ret;
- __asm__ ("divl %1" : "=A,A"(ret) : "r,?m"(d), "A,A"(ull));
- /* Exception if quotient does not fit on unsigned long. */
- return ret;
-}
-
-/* Fast long long division: when the quotient and remainder fit on 32 bits. */
-static inline unsigned long __rthal_i386_uldivrem(unsigned long long ull,
- const unsigned d,
- unsigned long *const rp)
-{
- unsigned long q, r;
- ull = __rthal_uldivrem(ull, d);
- __asm__ ( "": "=d"(r), "=a"(q) : "A"(ull));
- if(rp)
- *rp = r;
- return q;
-}
-#define rthal_uldivrem(ull, d, rp) __rthal_i386_uldivrem((ull),(d),(rp))
-
-/* Division of an unsigned 96 bits ((h << 32) + l) by an unsigned 32 bits.
- Building block for ulldiv. */
-static inline unsigned long long __rthal_div96by32 (const unsigned long long h,
- const unsigned long l,
- const unsigned long d,
- unsigned long *const rp)
-{
- u_long rh;
- const u_long qh = rthal_uldivrem(h, d, &rh);
- const unsigned long long t = __rthal_u64fromu32(rh, l);
- const u_long ql = rthal_uldivrem(t, d, rp);
-
- return __rthal_u64fromu32(qh, ql);
-}
-
-/* Slow long long division. Uses rthal_uldivrem, hence has the same property:
- the compiler removes redundant calls. */
-static inline unsigned long long
-__rthal_i386_ulldiv (const unsigned long long ull,
- const unsigned d,
- unsigned long *const rp)
-{
- unsigned long h, l;
- __rthal_u64tou32(ull, h, l);
- return __rthal_div96by32(h, l, d, rp);
-}
-#define rthal_ulldiv(ull,d,rp) __rthal_i386_ulldiv((ull),(d),(rp))
-
#include <asm-generic/xenomai/hal.h> /* Read the generic bits. */
#ifndef CONFIG_X86_WP_WORKS_OK
Index: include/asm-i386/arith.h
===================================================================
--- include/asm-i386/arith.h (revision 0)
+++ include/asm-i386/arith.h (revision 0)
@@ -0,0 +1,107 @@
+/**
+ * @ingroup hal
+ * @file
+ *
+ * Arithmetic/conversion routines for x86.
+ *
+ * Original RTAI/x86 HAL services from: \n
+ * Copyright © 2000 Paolo Mantegazza, \n
+ * Copyright © 2000 Steve Papacharalambous, \n
+ * Copyright © 2000 Stuart Hughes, \n
+ * and others.
+ *
+ * RTAI/x86 rewrite over Adeos: \n
+ * Copyright © 2002,2003 Philippe Gerum.
+ * Major refactoring for Xenomai: \n
+ * Copyright © 2004,2005 Philippe Gerum.
+ * Arithmetic/conversion routines: \n
+ * Copyright © 2005 Gilles Chanteperdrix.
+ *
+ * Xenomai is free software; you can redistribute it and/or
+ * modify it under the terms of the GNU General Public License as
+ * published by the Free Software Foundation, Inc., 675 Mass Ave,
+ * Cambridge MA 02139, USA; either version 2 of the License, or (at
+ * your option) any later version.
+ *
+ * Xenomai is distributed in the hope that it will be useful,
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
+ * General Public License for more details.
+ *
+ * You should have received a copy of the GNU General Public License
+ * along with Xenomai; if not, write to the Free Software
+ * Foundation, Inc., 59 Temple Place - Suite 330, Boston, MA
+ * 02111-1307, USA.
+ */
+
+#ifndef _XENO_ASM_I386_ARITH_H
+#define _XENO_ASM_I386_ARITH_H
+
+#define __rthal_u64tou32(ull, h, l) ({ \
+ unsigned long long _ull = (ull); \
+ (l) = _ull & 0xffffffff; \
+ (h) = _ull >> 32; \
+})
+
+#define __rthal_u64fromu32(h, l) ({ \
+ unsigned long long _ull; \
+ asm ( "": "=A"(_ull) : "d"(h), "a"(l)); \
+ _ull; \
+})
+
+/* const helper for rthal_uldivrem, so that the compiler will eliminate
+ multiple calls with same arguments, at no additionnal cost. */
+static inline __attribute_const__ unsigned long long
+__rthal_uldivrem(const unsigned long long ull, const unsigned long d)
+{
+ unsigned long long ret;
+ __asm__ ("divl %1" : "=A,A"(ret) : "r,?m"(d), "A,A"(ull));
+ /* Exception if quotient does not fit on unsigned long. */
+ return ret;
+}
+
+/* Fast long long division: when the quotient and remainder fit on 32 bits. */
+static inline unsigned long __rthal_i386_uldivrem(unsigned long long ull,
+ const unsigned d,
+ unsigned long *const rp)
+{
+ unsigned long q, r;
+ ull = __rthal_uldivrem(ull, d);
+ __asm__ ( "": "=d"(r), "=a"(q) : "A"(ull));
+ if(rp)
+ *rp = r;
+ return q;
+}
+#define rthal_uldivrem(ull, d, rp) __rthal_i386_uldivrem((ull),(d),(rp))
+
+/* Division of an unsigned 96 bits ((h << 32) + l) by an unsigned 32 bits.
+ Building block for ulldiv. */
+static inline unsigned long long __rthal_div96by32 (const unsigned long long h,
+ const unsigned long l,
+ const unsigned long d,
+ unsigned long *const rp)
+{
+ u_long rh;
+ const u_long qh = rthal_uldivrem(h, d, &rh);
+ const unsigned long long t = __rthal_u64fromu32(rh, l);
+ const u_long ql = rthal_uldivrem(t, d, rp);
+
+ return __rthal_u64fromu32(qh, ql);
+}
+
+/* Slow long long division. Uses rthal_uldivrem, hence has the same property:
+ the compiler removes redundant calls. */
+static inline unsigned long long
+__rthal_i386_ulldiv (const unsigned long long ull,
+ const unsigned d,
+ unsigned long *const rp)
+{
+ unsigned long h, l;
+ __rthal_u64tou32(ull, h, l);
+ return __rthal_div96by32(h, l, d, rp);
+}
+#define rthal_ulldiv(ull,d,rp) __rthal_i386_ulldiv((ull),(d),(rp))
+
+#include <asm-generic/xenomai/arith.h>
+
+#endif /* _XENO_ASM_I386_ARITH_H */
Index: include/asm-x86_64/Makefile.in
===================================================================
--- include/asm-x86_64/Makefile.in (revision 2454)
+++ include/asm-x86_64/Makefile.in (working copy)
@@ -230,8 +230,8 @@
target_os = @target_os@
target_vendor = @target_vendor@
includesubdir = $(includedir)/asm-x86_64
-includesub_HEADERS = atomic.h smi.h switch.h calibration.h features.h \
- hal.h syscall.h system.h wrappers.h fptest.h
+includesub_HEADERS = arith.h atomic.h calibration.h features.h fptest.h \
+ hal.h smi.h switch.h syscall.h system.h wrappers.h
SUBDIRS = bits
all: all-recursive
Index: include/asm-x86_64/Makefile.am
===================================================================
--- include/asm-x86_64/Makefile.am (revision 2454)
+++ include/asm-x86_64/Makefile.am (working copy)
@@ -1,6 +1,6 @@
includesubdir = $(includedir)/asm-x86_64
-includesub_HEADERS = atomic.h smi.h switch.h calibration.h features.h \
- hal.h syscall.h system.h wrappers.h fptest.h
+includesub_HEADERS = arith.h atomic.h calibration.h features.h fptest.h \
+ hal.h smi.h switch.h syscall.h system.h wrappers.h
SUBDIRS = bits
Index: include/asm-x86_64/arith.h
===================================================================
--- include/asm-x86_64/arith.h (revision 0)
+++ include/asm-x86_64/arith.h (revision 0)
@@ -0,0 +1,6 @@
+#ifndef _XENO_ASM_X86_64_ARITH_H
+#define _XENO_ASM_X86_64_ARITH_H
+
+#include <asm-generic/xenomai/arith.h>
+
+#endif /* _XENO_ASM_X86_64_ARITH_H */
Index: src/skins/psos+/init.c
===================================================================
--- src/skins/psos+/init.c (revision 2454)
+++ src/skins/psos+/init.c (working copy)
@@ -27,6 +27,7 @@
{
__psos_muxid =
xeno_bind_skin(PSOS_SKIN_MAGIC, "psos", "xeno_psos");
+ __psos_muxid = __xn_mux_shifted_id(__psos_muxid);
}
void k_fatal(u_long err_code, u_long flags)
Index: src/skins/rtai/init.c
===================================================================
--- src/skins/rtai/init.c (revision 2454)
+++ src/skins/rtai/init.c (working copy)
@@ -30,4 +30,5 @@
{
__rtai_muxid =
xeno_bind_skin(RTAI_SKIN_MAGIC, "RTAI", "xeno_rtai");
+ __rtai_muxid = __xn_mux_shifted_id(__rtai_muxid);
}
Index: src/skins/posix/init.c
===================================================================
--- src/skins/posix/init.c (revision 2454)
+++ src/skins/posix/init.c (working copy)
@@ -35,6 +35,7 @@
static int fork_handler_registered;
int __wrap_pthread_setschedparam(pthread_t, int, const struct sched_param *);
+void pse51_clock_init(int);
static __attribute__ ((constructor))
void __init_posix_interface(void)
@@ -42,9 +43,15 @@
struct sched_param parm;
int muxid, err;
- __pse51_muxid =
+ muxid =
xeno_bind_skin(PSE51_SKIN_MAGIC, "POSIX", "xeno_posix");
+#ifdef CONFIG_XENO_HW_DIRECT_TSC
+ pse51_clock_init(muxid);
+#endif /* CONFIG_XENO_HW_DIRECT_TSC */
+
+ __pse51_muxid = __xn_mux_shifted_id(muxid);
+
muxid = XENOMAI_SYSBIND(RTDM_SKIN_MAGIC,
XENOMAI_FEAT_DEP, XENOMAI_ABI_REV, NULL);
if (muxid > 0) {
Index: src/skins/posix/clock.c
===================================================================
--- src/skins/posix/clock.c (revision 2454)
+++ src/skins/posix/clock.c (working copy)
@@ -20,9 +20,24 @@
#include <pthread.h> /* For pthread_setcanceltype. */
#include <posix/syscall.h>
#include <time.h>
+#include <asm/xenomai/arith.h>
extern int __pse51_muxid;
+#ifdef CONFIG_XENO_HW_DIRECT_TSC
+static xnsysinfo_t sysinfo;
+
+void pse51_clock_init(int muxid)
+{
+ int err = -XENOMAI_SYSCALL3(__xn_sys_info, muxid, &sysinfo);
+ if (err) {
+ fprintf(stderr, "Xenomai Posix skin init: "
+ "sys_info: %s\n", strerror(err));
+ exit(EXIT_FAILURE);
+ }
+}
+#endif /* CONFIG_XENO_HW_DIRECT_TSC */
+
int __wrap_clock_getres(clockid_t clock_id, struct timespec *tp)
{
int err = -XENOMAI_SKINCALL2(__pse51_muxid,
@@ -39,11 +54,25 @@
int __wrap_clock_gettime(clockid_t clock_id, struct timespec *tp)
{
- int err = -XENOMAI_SKINCALL2(__pse51_muxid,
- __pse51_clock_gettime,
- clock_id,
- tp);
+ int err;
+#ifdef CONFIG_XENO_HW_DIRECT_TSC
+ if (clock_id == CLOCK_MONOTONIC) {
+ unsigned long long tsc, ns;
+ unsigned long rem;
+ tsc = __xn_rdtsc();
+ ns = xnarch_llimd(tsc, 1000000000, sysinfo.cpufreq);
+ tp->tv_sec = xnarch_ulldiv(ns, 1000000000, &rem);
+ tp->tv_nsec = rem;
+ return 0;
+ }
+#endif /* CONFIG_XENO_HW_DIRECT_TSC */
+
+ err = -XENOMAI_SKINCALL2(__pse51_muxid,
+ __pse51_clock_gettime,
+ clock_id,
+ tp);
+
if (!err)
return 0;
Index: src/skins/vrtx/init.c
===================================================================
--- src/skins/vrtx/init.c (revision 2454)
+++ src/skins/vrtx/init.c (working copy)
@@ -43,6 +43,7 @@
__vrtx_muxid =
xeno_bind_skin(VRTX_SKIN_MAGIC, "vrtx", "xeno_vrtx");
+ __vrtx_muxid = __xn_mux_shifted_id(__vrtx_muxid);
/* Allocate a TSD key for indexing self task pointers. */
Index: src/skins/vxworks/init.c
===================================================================
--- src/skins/vxworks/init.c (revision 2454)
+++ src/skins/vxworks/init.c (working copy)
@@ -41,6 +41,7 @@
{
__vxworks_muxid = xeno_bind_skin(VXWORKS_SKIN_MAGIC,
"vxworks", "xeno_vxworks");
+ __vxworks_muxid = __xn_mux_shifted_id(__vxworks_muxid);
/* Allocate a TSD key for indexing self task pointers. */
Index: src/skins/native/init.c
===================================================================
--- src/skins/native/init.c (revision 2454)
+++ src/skins/native/init.c (working copy)
@@ -30,6 +30,7 @@
pthread_key_t __native_tskey;
int __native_muxid = -1;
+void native_timer_init(int);
static void __flush_tsd(void *tsd)
{
@@ -43,6 +44,12 @@
__native_muxid =
xeno_bind_skin(XENO_SKIN_MAGIC, "native", "xeno_native");
+#ifdef CONFIG_XENO_HW_DIRECT_TSC
+ native_timer_init(__native_muxid);
+#endif /* CONFIG_XENO_HW_DIRECT_TSC */
+
+ __native_muxid = __xn_mux_shifted_id(__native_muxid);
+
/* Allocate a TSD key for indexing self task pointers. */
if (pthread_key_create(&__native_tskey, &__flush_tsd) != 0) {
Index: src/skins/native/timer.c
===================================================================
--- src/skins/native/timer.c (revision 2454)
+++ src/skins/native/timer.c (working copy)
@@ -21,6 +21,20 @@
extern int __native_muxid;
+#ifdef CONFIG_XENO_HW_DIRECT_TSC
+static xnsysinfo_t sysinfo;
+
+void native_timer_init(int muxid)
+{
+ int err = XENOMAI_SYSCALL3(__xn_sys_info, muxid, &sysinfo);
+ if (err) {
+ fprintf(stderr, "Native skin init: "
+ "sys_info: %s\n", strerror(-err));
+ exit(EXIT_FAILURE);
+ }
+}
+#endif /* CONFIG_XENO_HW_DIRECT_TSC */
+
int rt_timer_set_mode(RTIME tickval)
{
return XENOMAI_SKINCALL1(__native_muxid, __native_timer_set_mode,
@@ -68,7 +82,11 @@
{
RTIME ticks;
+#ifdef CONFIG_XENO_HW_DIRECT_TSC
+ ticks = xnarch_llimd(ns, sysinfo.cpufreq, 1000000000);
+#else /* !CONFIG_XENO_HW_DIRECT_TSC */
XENOMAI_SKINCALL2(__native_muxid, __native_timer_ns2tsc, &ticks, &ns);
+#endif /* !CONFIG_XENO_HW_DIRECT_TSC */
return ticks;
}
@@ -76,7 +94,11 @@
{
SRTIME ns;
+#ifdef CONFIG_XENO_HW_DIRECT_TSC
+ ticks = xnarch_llimd(ticks, 1000000000, sysinfo.cpufreq);
+#else /* !CONFIG_XENO_HW_DIRECT_TSC */
XENOMAI_SKINCALL2(__native_muxid, __native_timer_tsc2ns, &ns, &ticks);
+#endif /* !CONFIG_XENO_HW_DIRECT_TSC */
return ns;
}
Index: src/skins/rtdm/init.c
===================================================================
--- src/skins/rtdm/init.c (revision 2454)
+++ src/skins/rtdm/init.c (working copy)
@@ -34,4 +34,6 @@
__rtdm_muxid =
xeno_bind_skin_opt(RTDM_SKIN_MAGIC, "rtdm", "xeno_rtdm");
+ __rtdm_muxid = __xn_mux_shifted_id(__rtdm_muxid);
+
}
Index: ksrc/skins/posix/syscall.c
===================================================================
--- ksrc/skins/posix/syscall.c (revision 2454)
+++ ksrc/skins/posix/syscall.c (working copy)
@@ -2891,7 +2891,7 @@
return NULL;
}
-
+
return ERR_PTR(-EINVAL);
}
^ permalink raw reply [flat|nested] 9+ messages in thread
* Re: [Xenomai-core] [PATCH] Use tsc for implementation of clock_gettime.
2007-05-17 15:21 ` Gilles Chanteperdrix
@ 2007-05-17 15:56 ` Gilles Chanteperdrix
2007-05-18 5:52 ` Jan Kiszka
0 siblings, 1 reply; 9+ messages in thread
From: Gilles Chanteperdrix @ 2007-05-17 15:56 UTC (permalink / raw)
To: Jan Kiszka, xenomai
[-- Attachment #1: message body and .signature --]
[-- Type: text/plain, Size: 2291 bytes --]
Gilles Chanteperdrix wrote:
> Jan Kiszka wrote:
> > Gilles Chanteperdrix wrote:
> > > Jan Kiszka wrote:
> > > > Gilles Chanteperdrix wrote:
> > > > > Jan Kiszka wrote:
> > > > > > Gilles Chanteperdrix wrote:
> > > > > > > Hi,
> > > > > > >
> > > > > > > here comes, for review, a patch which reduces the overhead of
> > > > > > > clock_gettime by directly reading the tsc in user-space for
> > > > > > > architectures that support it.
> > > > > >
> > > > > > Highly welcome. But I have one concern: How and when do you propagate
> > > > > > wallclock_offset changes to user space?
> > > > >
> > > > > Since clock_settime is not implemented, never, but if clock_settime was
> > > > > implemented, clock_settime would re-issue the __xn_sys_info syscall.
> > > >
> > > > This excludes automatic clock adjustment, something I'm convinced we
> > > > will have to provide in the future.
> > >
> > > When we provide automatic clock adjustment, we will have to devise
> > > something more subtle than just an offset, so we will have to redesign
> >
> > I think offset + scaling factor.
> >
> > > posix clocks support anyway. Maybe clock_gettime(CLOCK_REALTIME) will
> > > then always be a syscall. After all, rt_timer_read is a syscall. If you
> > > want the fast clock, use CLOCK_MONOTONIC or rt_timer_tsc.
> >
> > Actually, the issue of the intermediate approach starts earlier: as soon
> > as you have clock_settime. Then you need to sync the offset across
> > applications in different processes.
> >
> > Having clock_monotonic (and maybe also rt_timer_tsc2ns) as a fast
> > variant already now is not bad. But beyond that, before introducing a
> > new interface between kernel and user space, I would like to consider
> > the effort to get it future-proof immediately. That doesn't mean that we
> > already have to implement clock adjustment, but we may already prepare
> > the prerequisites.
>
> Ok. Here is a new version that does not break the ABI, and where
> clock_gettime(CLOCK_REALTIME) remains a syscall.
>
> Note that it also adds user-space conversions between tsc and ns to the
> native skin.
This one is actually tested.
--
Gilles Chanteperdrix.
[-- Attachment #2: xeno-clock_gettime-direct-tsc.3.diff --]
[-- Type: text/plain, Size: 36274 bytes --]
Index: include/asm-ia64/Makefile.in
===================================================================
--- include/asm-ia64/Makefile.in (revision 2454)
+++ include/asm-ia64/Makefile.in (working copy)
@@ -230,7 +230,9 @@
target_os = @target_os@
target_vendor = @target_vendor@
includesubdir = $(includedir)/asm-ia64
-includesub_HEADERS = atomic.h calibration.h features.h hal.h syscall.h system.h wrappers.h fptest.h
+includesub_HEADERS = arith.h atomic.h calibration.h features.h fptest.h \
+ hal.h syscall.h system.h wrappers.h
+
SUBDIRS = bits
all: all-recursive
Index: include/asm-ia64/Makefile.am
===================================================================
--- include/asm-ia64/Makefile.am (revision 2454)
+++ include/asm-ia64/Makefile.am (working copy)
@@ -1,5 +1,6 @@
includesubdir = $(includedir)/asm-ia64
-includesub_HEADERS = atomic.h calibration.h features.h hal.h syscall.h system.h wrappers.h fptest.h
+includesub_HEADERS = arith.h atomic.h calibration.h features.h fptest.h \
+ hal.h syscall.h system.h wrappers.h
SUBDIRS = bits
Index: include/asm-ia64/arith.h
===================================================================
--- include/asm-ia64/arith.h (revision 0)
+++ include/asm-ia64/arith.h (revision 0)
@@ -0,0 +1,6 @@
+#ifndef _XENO_ASM_IA64_ARITH_H
+#define _XENO_ASM_IA64_ARITH_H
+
+#include <asm-generic/xenomai/arith.h>
+
+#endif /* _XENO_ASM_IA64_ARITH_H */
Index: include/asm-blackfin/Makefile.in
===================================================================
--- include/asm-blackfin/Makefile.in (revision 2454)
+++ include/asm-blackfin/Makefile.in (working copy)
@@ -230,8 +230,8 @@
target_os = @target_os@
target_vendor = @target_vendor@
includesubdir = $(includedir)/asm-blackfin
-includesub_HEADERS = atomic.h calibration.h features.h hal.h \
- syscall.h system.h wrappers.h fptest.h
+includesub_HEADERS = arith.h atomic.h calibration.h features.h fptest.h \
+ hal.h syscall.h system.h wrappers.h
SUBDIRS = bits
all: all-recursive
Index: include/asm-blackfin/Makefile.am
===================================================================
--- include/asm-blackfin/Makefile.am (revision 2454)
+++ include/asm-blackfin/Makefile.am (working copy)
@@ -1,6 +1,6 @@
includesubdir = $(includedir)/asm-blackfin
-includesub_HEADERS = atomic.h calibration.h features.h hal.h \
- syscall.h system.h wrappers.h fptest.h
+includesub_HEADERS = arith.h atomic.h calibration.h features.h fptest.h \
+ hal.h syscall.h system.h wrappers.h
SUBDIRS = bits
Index: include/asm-blackfin/arith.h
===================================================================
--- include/asm-blackfin/arith.h (revision 0)
+++ include/asm-blackfin/arith.h (revision 0)
@@ -0,0 +1,6 @@
+#ifndef _XENO_ASM_BLACKFIN_ARITH_H
+#define _XENO_ASM_BLACKFIN_ARITH_H
+
+#include <asm-generic/xenomai/arith.h>
+
+#endif /* _XENO_ASM_BLACKFIN_ARITH_H */
Index: include/asm-generic/syscall.h
===================================================================
--- include/asm-generic/syscall.h (revision 2454)
+++ include/asm-generic/syscall.h (working copy)
@@ -55,7 +55,6 @@
unsigned long long cpufreq; /* CPU frequency */
unsigned long tickval; /* Tick duration (ns) */
-
} xnsysinfo_t;
#define SIGHARDEN SIGWINCH
Index: include/asm-generic/bits/bind.h
===================================================================
--- include/asm-generic/bits/bind.h (revision 2454)
+++ include/asm-generic/bits/bind.h (working copy)
@@ -62,7 +62,7 @@
sa.sa_flags = 0;
sigaction(SIGXCPU, &sa, NULL);
- return __xn_mux_shifted_id(muxid);
+ return muxid;
}
static inline int
@@ -105,7 +105,7 @@
exit(1);
}
- return __xn_mux_shifted_id(muxid);
+ return muxid;
}
#endif /* _XENO_ASM_GENERIC_BITS_BIND_H */
Index: include/asm-generic/Makefile.in
===================================================================
--- include/asm-generic/Makefile.in (revision 2454)
+++ include/asm-generic/Makefile.in (working copy)
@@ -230,7 +230,7 @@
target_os = @target_os@
target_vendor = @target_vendor@
includesubdir = $(includedir)/asm-generic
-includesub_HEADERS = features.h hal.h syscall.h system.h wrappers.h
+includesub_HEADERS = arith.h features.h hal.h syscall.h system.h wrappers.h
SUBDIRS = bits
all: all-recursive
Index: include/asm-generic/Makefile.am
===================================================================
--- include/asm-generic/Makefile.am (revision 2454)
+++ include/asm-generic/Makefile.am (working copy)
@@ -1,5 +1,5 @@
includesubdir = $(includedir)/asm-generic
-includesub_HEADERS = features.h hal.h syscall.h system.h wrappers.h
+includesub_HEADERS = arith.h features.h hal.h syscall.h system.h wrappers.h
SUBDIRS = bits
Index: include/asm-generic/hal.h
===================================================================
--- include/asm-generic/hal.h (revision 2454)
+++ include/asm-generic/hal.h (working copy)
@@ -37,8 +37,8 @@
#include <linux/kallsyms.h>
#include <linux/init.h>
#include <asm/byteorder.h>
-#include <asm/div64.h>
#include <asm/xenomai/wrappers.h>
+#include <asm/xenomai/arith.h>
#define RTHAL_DOMAIN_ID 0x58454e4f
@@ -385,121 +385,6 @@
#define rthal_printk printk
-#ifdef __BIG_ENDIAN
-#define endianstruct struct { unsigned _h; unsigned _l; } _s
-#else /* __LITTLE_ENDIAN */
-#define endianstruct struct { unsigned _l; unsigned _h; } _s
-#endif
-
-#ifndef __rthal_u64tou32
-#define __rthal_u64tou32(ull, h, l) ({ \
- union { unsigned long long _ull; \
- endianstruct; \
- } _u; \
- _u._ull = (ull); \
- (h) = _u._s._h; \
- (l) = _u._s._l; \
-})
-#endif /* !__rthal_u64tou32 */
-
-#ifndef __rthal_u64fromu32
-#define __rthal_u64fromu32(h, l) ({ \
- union { unsigned long long _ull; \
- endianstruct; \
- } _u; \
- _u._s._h = (h); \
- _u._s._l = (l); \
- _u._ull; \
-})
-#endif /* !__rthal_u64fromu32 */
-
-#ifndef rthal_ullmul
-static inline __attribute_const__ unsigned long long
-__rthal_generic_ullmul(const unsigned m0, const unsigned m1)
-{
- return (unsigned long long) m0 * m1;
-}
-#define rthal_ullmul(m0,m1) __rthal_generic_ullmul((m0),(m1))
-#endif /* !rthal_ullmul */
-
-#ifndef rthal_ulldiv
-static inline unsigned long long __rthal_generic_ulldiv (unsigned long long ull,
- const unsigned uld,
- unsigned long *const rp)
-{
- const unsigned r = do_div(ull, uld);
-
- if (rp)
- *rp = r;
-
- return ull;
-}
-#define rthal_ulldiv(ull,uld,rp) __rthal_generic_ulldiv((ull),(uld),(rp))
-#endif /* !rthal_ulldiv */
-
-#ifndef rthal_uldivrem
-#define rthal_uldivrem(ull,ul,rp) ((unsigned) rthal_ulldiv((ull),(ul),(rp)))
-#endif /* !rthal_uldivrem */
-
-#ifndef rthal_imuldiv
-static inline __attribute_const__ int __rthal_generic_imuldiv (int i,
- int mult,
- int div)
-{
- /* Returns (int)i = (unsigned long long)i*(unsigned)(mult)/(unsigned)div. */
- const unsigned long long ull = rthal_ullmul(i, mult);
- return rthal_uldivrem(ull, div, NULL);
-}
-#define rthal_imuldiv(i,m,d) __rthal_generic_imuldiv((i),(m),(d))
-#endif /* !rthal_imuldiv */
-
-#ifndef rthal_llimd
-/* Division of an unsigned 96 bits ((h << 32) + l) by an unsigned 32 bits.
- Building block for llimd. Without const qualifiers, gcc reload registers
- after each call to uldivrem. */
-static inline unsigned long long
-__rthal_generic_div96by32 (const unsigned long long h,
- const unsigned l,
- const unsigned d,
- unsigned long *const rp)
-{
- unsigned long rh;
- const unsigned qh = rthal_uldivrem(h, d, &rh);
- const unsigned long long t = __rthal_u64fromu32(rh, l);
- const unsigned ql = rthal_uldivrem(t, d, rp);
-
- return __rthal_u64fromu32(qh, ql);
-}
-
-static inline __attribute_const__
-unsigned long long __rthal_generic_ullimd (const unsigned long long op,
- const unsigned m,
- const unsigned d)
-{
- unsigned oph, opl, tlh, tll;
- unsigned long long th, tl;
-
- __rthal_u64tou32(op, oph, opl);
- tl = rthal_ullmul(opl, m);
- __rthal_u64tou32(tl, tlh, tll);
- th = rthal_ullmul(oph, m);
- th += tlh;
-
- return __rthal_generic_div96by32(th, tll, d, NULL);
-}
-
-static inline __attribute_const__ long long __rthal_generic_llimd (long long op,
- unsigned m,
- unsigned d)
-{
-
- if(op < 0LL)
- return -__rthal_generic_ullimd(-op, m, d);
- return __rthal_generic_ullimd(op, m, d);
-}
-#define rthal_llimd(ll,m,d) __rthal_generic_llimd((ll),(m),(d))
-#endif /* !rthal_llimd */
-
typedef ipipe_irq_handler_t rthal_irq_handler_t;
typedef ipipe_irq_ackfn_t rthal_irq_ackfn_t;
Index: include/asm-generic/system.h
===================================================================
--- include/asm-generic/system.h (revision 2454)
+++ include/asm-generic/system.h (working copy)
@@ -133,18 +133,6 @@
#define xnarch_logerr(fmt,args...) printk(KERN_ERR XNARCH_PROMPT fmt , ##args)
#define xnarch_printf(fmt,args...) printk(KERN_INFO XNARCH_PROMPT fmt , ##args)
-#define xnarch_ullmod(ull,uld,rem) ({ xnarch_ulldiv(ull,uld,rem); (*rem); })
-#define xnarch_uldiv(ull, d) rthal_uldivrem(ull, d, NULL)
-#define xnarch_ulmod(ull, d) ({ u_long _rem; \
- rthal_uldivrem(ull,d,&_rem); _rem; })
-
-#define xnarch_ullmul rthal_ullmul
-#define xnarch_uldivrem rthal_uldivrem
-#define xnarch_ulldiv rthal_ulldiv
-#define xnarch_imuldiv rthal_imuldiv
-#define xnarch_llimd rthal_llimd
-#define xnarch_get_cpu_tsc rthal_rdtsc
-
typedef cpumask_t xnarch_cpumask_t;
#ifdef CONFIG_SMP
Index: include/asm-generic/arith.h
===================================================================
--- include/asm-generic/arith.h (revision 0)
+++ include/asm-generic/arith.h (revision 0)
@@ -0,0 +1,172 @@
+/**
+ * @ingroup hal
+ * @file
+ *
+ * Generic arithmetic/conversion routines.
+ * Copyright © 2005 Stelian Pop.
+ * Copyright © 2005 Gilles Chanteperdrix.
+ *
+ * This program is free software; you can redistribute it and/or modify
+ * it under the terms of the GNU General Public License as published by
+ * the Free Software Foundation, Inc., 675 Mass Ave, Cambridge MA 02139,
+ * USA; either version 2 of the License, or (at your option) any later
+ * version.
+ *
+ * This program is distributed in the hope that it will be useful,
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
+ * GNU General Public License for more details.
+ *
+ * You should have received a copy of the GNU General Public License
+ * along with this program; if not, write to the Free Software
+ * Foundation, Inc., 59 Temple Place - Suite 330, Boston, MA 02111-1307, USA.
+ */
+
+/**
+ * @addtogroup hal
+ *@{*/
+
+#ifndef _XENO_ASM_GENERIC_ARITH_H
+#define _XENO_ASM_GENERIC_ARITH_H
+
+#ifdef __KERNEL__
+#include <asm/div64.h>
+#else /* !__KERNEL__ */
+static inline unsigned __rthal_do_div(unsigned long long *a, unsigned d)
+{
+ unsigned r = *a % d;
+ *a /= d;
+ return r;
+}
+#define do_div(a, d) __rthal_do_div(&(a), (d))
+#endif /* !__KERNEL__ */
+
+#ifdef __BIG_ENDIAN
+#define endianstruct struct { unsigned _h; unsigned _l; } _s
+#else /* __LITTLE_ENDIAN */
+#define endianstruct struct { unsigned _l; unsigned _h; } _s
+#endif
+
+#ifndef __rthal_u64tou32
+#define __rthal_u64tou32(ull, h, l) ({ \
+ union { unsigned long long _ull; \
+ endianstruct; \
+ } _u; \
+ _u._ull = (ull); \
+ (h) = _u._s._h; \
+ (l) = _u._s._l; \
+})
+#endif /* !__rthal_u64tou32 */
+
+#ifndef __rthal_u64fromu32
+#define __rthal_u64fromu32(h, l) ({ \
+ union { unsigned long long _ull; \
+ endianstruct; \
+ } _u; \
+ _u._s._h = (h); \
+ _u._s._l = (l); \
+ _u._ull; \
+})
+#endif /* !__rthal_u64fromu32 */
+
+#ifndef rthal_ullmul
+static inline __attribute__((__const__)) unsigned long long
+__rthal_generic_ullmul(const unsigned m0, const unsigned m1)
+{
+ return (unsigned long long) m0 * m1;
+}
+#define rthal_ullmul(m0,m1) __rthal_generic_ullmul((m0),(m1))
+#endif /* !rthal_ullmul */
+
+#ifndef rthal_ulldiv
+static inline unsigned long long __rthal_generic_ulldiv (unsigned long long ull,
+ const unsigned uld,
+ unsigned long *const rp)
+{
+ const unsigned r = do_div(ull, uld);
+
+ if (rp)
+ *rp = r;
+
+ return ull;
+}
+#define rthal_ulldiv(ull,uld,rp) __rthal_generic_ulldiv((ull),(uld),(rp))
+#endif /* !rthal_ulldiv */
+
+#ifndef rthal_uldivrem
+#define rthal_uldivrem(ull,ul,rp) ((unsigned) rthal_ulldiv((ull),(ul),(rp)))
+#endif /* !rthal_uldivrem */
+
+#ifndef rthal_imuldiv
+static inline __attribute__((__const__)) int __rthal_generic_imuldiv (int i,
+ int mult,
+ int div)
+{
+ /* Returns (int)i = (unsigned long long)i*(unsigned)(mult)/(unsigned)div. */
+ const unsigned long long ull = rthal_ullmul(i, mult);
+ return rthal_uldivrem(ull, div, NULL);
+}
+#define rthal_imuldiv(i,m,d) __rthal_generic_imuldiv((i),(m),(d))
+#endif /* !rthal_imuldiv */
+
+#ifndef rthal_llimd
+/* Division of an unsigned 96 bits ((h << 32) + l) by an unsigned 32 bits.
+ Building block for llimd. Without const qualifiers, gcc reload registers
+ after each call to uldivrem. */
+static inline unsigned long long
+__rthal_generic_div96by32 (const unsigned long long h,
+ const unsigned l,
+ const unsigned d,
+ unsigned long *const rp)
+{
+ unsigned long rh;
+ const unsigned qh = rthal_uldivrem(h, d, &rh);
+ const unsigned long long t = __rthal_u64fromu32(rh, l);
+ const unsigned ql = rthal_uldivrem(t, d, rp);
+
+ return __rthal_u64fromu32(qh, ql);
+}
+
+static inline __attribute__((__const__))
+unsigned long long __rthal_generic_ullimd (const unsigned long long op,
+ const unsigned m,
+ const unsigned d)
+{
+ unsigned oph, opl, tlh, tll;
+ unsigned long long th, tl;
+
+ __rthal_u64tou32(op, oph, opl);
+ tl = rthal_ullmul(opl, m);
+ __rthal_u64tou32(tl, tlh, tll);
+ th = rthal_ullmul(oph, m);
+ th += tlh;
+
+ return __rthal_generic_div96by32(th, tll, d, NULL);
+}
+
+static inline __attribute__((__const__)) long long
+__rthal_generic_llimd (long long op, unsigned m, unsigned d)
+{
+
+ if(op < 0LL)
+ return -__rthal_generic_ullimd(-op, m, d);
+ return __rthal_generic_ullimd(op, m, d);
+}
+#define rthal_llimd(ll,m,d) __rthal_generic_llimd((ll),(m),(d))
+#endif /* !rthal_llimd */
+
+#define xnarch_ullmod(ull,uld,rem) ({ xnarch_ulldiv(ull,uld,rem); (*rem); })
+#define xnarch_uldiv(ull, d) rthal_uldivrem(ull, d, NULL)
+#define xnarch_ulmod(ull, d) ({ u_long _rem; \
+ rthal_uldivrem(ull,d,&_rem); _rem; })
+
+#define xnarch_ullmul rthal_ullmul
+#define xnarch_uldivrem rthal_uldivrem
+#define xnarch_ulldiv rthal_ulldiv
+#define xnarch_imuldiv rthal_imuldiv
+#define xnarch_llimd rthal_llimd
+#define xnarch_get_cpu_tsc rthal_rdtsc
+
+/*@}*/
+
+#endif /* _XENO_ASM_GENERIC_ARITH_H */
Index: include/asm-powerpc/Makefile.in
===================================================================
--- include/asm-powerpc/Makefile.in (revision 2454)
+++ include/asm-powerpc/Makefile.in (working copy)
@@ -230,8 +230,8 @@
target_os = @target_os@
target_vendor = @target_vendor@
includesubdir = $(includedir)/asm-powerpc
-includesub_HEADERS = atomic.h calibration.h features.h hal.h \
- syscall.h system.h wrappers.h fptest.h
+includesub_HEADERS = arith.h atomic.h calibration.h features.h fptest.h \
+ hal.h syscall.h system.h wrappers.h
SUBDIRS = bits
all: all-recursive
Index: include/asm-powerpc/Makefile.am
===================================================================
--- include/asm-powerpc/Makefile.am (revision 2454)
+++ include/asm-powerpc/Makefile.am (working copy)
@@ -1,6 +1,6 @@
includesubdir = $(includedir)/asm-powerpc
-includesub_HEADERS = atomic.h calibration.h features.h hal.h \
- syscall.h system.h wrappers.h fptest.h
+includesub_HEADERS = arith.h atomic.h calibration.h features.h fptest.h \
+ hal.h syscall.h system.h wrappers.h
SUBDIRS = bits
Index: include/asm-powerpc/arith.h
===================================================================
--- include/asm-powerpc/arith.h (revision 0)
+++ include/asm-powerpc/arith.h (revision 0)
@@ -0,0 +1,6 @@
+#ifndef _XENO_ASM_POWERPC_ARITH_H
+#define _XENO_ASM_POWERPC_ARITH_H
+
+#include <asm-generic/xenomai/arith.h>
+
+#endif /* _XENO_ASM_POWERPC_ARITH_H */
Index: include/asm-arm/Makefile.in
===================================================================
--- include/asm-arm/Makefile.in (revision 2454)
+++ include/asm-arm/Makefile.in (working copy)
@@ -230,8 +230,8 @@
target_os = @target_os@
target_vendor = @target_vendor@
includesubdir = $(includedir)/asm-arm
-includesub_HEADERS = atomic.h calibration.h features.h hal.h \
- syscall.h system.h wrappers.h fptest.h
+includesub_HEADERS = arith.h atomic.h calibration.h features.h fptest.h \
+ hal.h syscall.h system.h wrappers.h
SUBDIRS = bits
all: all-recursive
Index: include/asm-arm/Makefile.am
===================================================================
--- include/asm-arm/Makefile.am (revision 2454)
+++ include/asm-arm/Makefile.am (working copy)
@@ -1,6 +1,6 @@
includesubdir = $(includedir)/asm-arm
-includesub_HEADERS = atomic.h calibration.h features.h hal.h \
- syscall.h system.h wrappers.h fptest.h
+includesub_HEADERS = arith.h atomic.h calibration.h features.h fptest.h \
+ hal.h syscall.h system.h wrappers.h
SUBDIRS = bits
Index: include/asm-arm/arith.h
===================================================================
--- include/asm-arm/arith.h (revision 0)
+++ include/asm-arm/arith.h (revision 0)
@@ -0,0 +1,6 @@
+#ifndef _XENO_ASM_ARM_ARITH_H
+#define _XENO_ASM_ARM_ARITH_H
+
+#include <asm-generic/xenomai/arith.h>
+
+#endif /* _XENO_ASM_ARM_ARITH_H */
Index: include/posix/syscall.h
===================================================================
--- include/posix/syscall.h (revision 2454)
+++ include/posix/syscall.h (working copy)
@@ -100,6 +100,11 @@
#define __pse51_condattr_setpshared 74
#define __pse51_thread_getschedparam 75
+typedef struct pse51_sysinfo {
+ unsigned period;
+ unsigned long long wallclock_offset;
+} pse51_sysinfo_t;
+
#ifdef __KERNEL__
#ifdef __cplusplus
Index: include/asm-i386/Makefile.in
===================================================================
--- include/asm-i386/Makefile.in (revision 2454)
+++ include/asm-i386/Makefile.in (working copy)
@@ -230,8 +230,8 @@
target_os = @target_os@
target_vendor = @target_vendor@
includesubdir = $(includedir)/asm-i386
-includesub_HEADERS = atomic.h calibration.h features.h \
- hal.h smi.h switch.h syscall.h system.h wrappers.h fptest.h
+includesub_HEADERS = arith.h atomic.h calibration.h features.h fptest.h \
+ hal.h smi.h switch.h syscall.h system.h wrappers.h
SUBDIRS = bits
all: all-recursive
Index: include/asm-i386/Makefile.am
===================================================================
--- include/asm-i386/Makefile.am (revision 2454)
+++ include/asm-i386/Makefile.am (working copy)
@@ -1,6 +1,6 @@
includesubdir = $(includedir)/asm-i386
-includesub_HEADERS = atomic.h calibration.h features.h \
- hal.h smi.h switch.h syscall.h system.h wrappers.h fptest.h
+includesub_HEADERS = arith.h atomic.h calibration.h features.h fptest.h \
+ hal.h smi.h switch.h syscall.h system.h wrappers.h
SUBDIRS = bits
Index: include/asm-i386/hal.h
===================================================================
--- include/asm-i386/hal.h (revision 2454)
+++ include/asm-i386/hal.h (working copy)
@@ -39,71 +39,6 @@
#include <asm/xenomai/wrappers.h>
-#define __rthal_u64tou32(ull, h, l) ({ \
- unsigned long long _ull = (ull); \
- (l) = _ull & 0xffffffff; \
- (h) = _ull >> 32; \
-})
-
-#define __rthal_u64fromu32(h, l) ({ \
- unsigned long long _ull; \
- asm ( "": "=A"(_ull) : "d"(h), "a"(l)); \
- _ull; \
-})
-
-/* const helper for rthal_uldivrem, so that the compiler will eliminate
- multiple calls with same arguments, at no additionnal cost. */
-static inline __attribute_const__ unsigned long long
-__rthal_uldivrem(const unsigned long long ull, const unsigned long d)
-{
- unsigned long long ret;
- __asm__ ("divl %1" : "=A,A"(ret) : "r,?m"(d), "A,A"(ull));
- /* Exception if quotient does not fit on unsigned long. */
- return ret;
-}
-
-/* Fast long long division: when the quotient and remainder fit on 32 bits. */
-static inline unsigned long __rthal_i386_uldivrem(unsigned long long ull,
- const unsigned d,
- unsigned long *const rp)
-{
- unsigned long q, r;
- ull = __rthal_uldivrem(ull, d);
- __asm__ ( "": "=d"(r), "=a"(q) : "A"(ull));
- if(rp)
- *rp = r;
- return q;
-}
-#define rthal_uldivrem(ull, d, rp) __rthal_i386_uldivrem((ull),(d),(rp))
-
-/* Division of an unsigned 96 bits ((h << 32) + l) by an unsigned 32 bits.
- Building block for ulldiv. */
-static inline unsigned long long __rthal_div96by32 (const unsigned long long h,
- const unsigned long l,
- const unsigned long d,
- unsigned long *const rp)
-{
- u_long rh;
- const u_long qh = rthal_uldivrem(h, d, &rh);
- const unsigned long long t = __rthal_u64fromu32(rh, l);
- const u_long ql = rthal_uldivrem(t, d, rp);
-
- return __rthal_u64fromu32(qh, ql);
-}
-
-/* Slow long long division. Uses rthal_uldivrem, hence has the same property:
- the compiler removes redundant calls. */
-static inline unsigned long long
-__rthal_i386_ulldiv (const unsigned long long ull,
- const unsigned d,
- unsigned long *const rp)
-{
- unsigned long h, l;
- __rthal_u64tou32(ull, h, l);
- return __rthal_div96by32(h, l, d, rp);
-}
-#define rthal_ulldiv(ull,d,rp) __rthal_i386_ulldiv((ull),(d),(rp))
-
#include <asm-generic/xenomai/hal.h> /* Read the generic bits. */
#ifndef CONFIG_X86_WP_WORKS_OK
Index: include/asm-i386/arith.h
===================================================================
--- include/asm-i386/arith.h (revision 0)
+++ include/asm-i386/arith.h (revision 0)
@@ -0,0 +1,107 @@
+/**
+ * @ingroup hal
+ * @file
+ *
+ * Arithmetic/conversion routines for x86.
+ *
+ * Original RTAI/x86 HAL services from: \n
+ * Copyright © 2000 Paolo Mantegazza, \n
+ * Copyright © 2000 Steve Papacharalambous, \n
+ * Copyright © 2000 Stuart Hughes, \n
+ * and others.
+ *
+ * RTAI/x86 rewrite over Adeos: \n
+ * Copyright © 2002,2003 Philippe Gerum.
+ * Major refactoring for Xenomai: \n
+ * Copyright © 2004,2005 Philippe Gerum.
+ * Arithmetic/conversion routines: \n
+ * Copyright © 2005 Gilles Chanteperdrix.
+ *
+ * Xenomai is free software; you can redistribute it and/or
+ * modify it under the terms of the GNU General Public License as
+ * published by the Free Software Foundation, Inc., 675 Mass Ave,
+ * Cambridge MA 02139, USA; either version 2 of the License, or (at
+ * your option) any later version.
+ *
+ * Xenomai is distributed in the hope that it will be useful,
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
+ * General Public License for more details.
+ *
+ * You should have received a copy of the GNU General Public License
+ * along with Xenomai; if not, write to the Free Software
+ * Foundation, Inc., 59 Temple Place - Suite 330, Boston, MA
+ * 02111-1307, USA.
+ */
+
+#ifndef _XENO_ASM_I386_ARITH_H
+#define _XENO_ASM_I386_ARITH_H
+
+#define __rthal_u64tou32(ull, h, l) ({ \
+ unsigned long long _ull = (ull); \
+ (l) = _ull & 0xffffffff; \
+ (h) = _ull >> 32; \
+})
+
+#define __rthal_u64fromu32(h, l) ({ \
+ unsigned long long _ull; \
+ asm ( "": "=A"(_ull) : "d"(h), "a"(l)); \
+ _ull; \
+})
+
+/* const helper for rthal_uldivrem, so that the compiler will eliminate
+ multiple calls with same arguments, at no additionnal cost. */
+static inline __attribute__((__const__)) unsigned long long
+__rthal_uldivrem(const unsigned long long ull, const unsigned long d)
+{
+ unsigned long long ret;
+ __asm__ ("divl %1" : "=A,A"(ret) : "r,?m"(d), "A,A"(ull));
+ /* Exception if quotient does not fit on unsigned long. */
+ return ret;
+}
+
+/* Fast long long division: when the quotient and remainder fit on 32 bits. */
+static inline unsigned long __rthal_i386_uldivrem(unsigned long long ull,
+ const unsigned d,
+ unsigned long *const rp)
+{
+ unsigned long q, r;
+ ull = __rthal_uldivrem(ull, d);
+ __asm__ ( "": "=d"(r), "=a"(q) : "A"(ull));
+ if(rp)
+ *rp = r;
+ return q;
+}
+#define rthal_uldivrem(ull, d, rp) __rthal_i386_uldivrem((ull),(d),(rp))
+
+/* Division of an unsigned 96 bits ((h << 32) + l) by an unsigned 32 bits.
+ Building block for ulldiv. */
+static inline unsigned long long __rthal_div96by32 (const unsigned long long h,
+ const unsigned long l,
+ const unsigned long d,
+ unsigned long *const rp)
+{
+ u_long rh;
+ const u_long qh = rthal_uldivrem(h, d, &rh);
+ const unsigned long long t = __rthal_u64fromu32(rh, l);
+ const u_long ql = rthal_uldivrem(t, d, rp);
+
+ return __rthal_u64fromu32(qh, ql);
+}
+
+/* Slow long long division. Uses rthal_uldivrem, hence has the same property:
+ the compiler removes redundant calls. */
+static inline unsigned long long
+__rthal_i386_ulldiv (const unsigned long long ull,
+ const unsigned d,
+ unsigned long *const rp)
+{
+ unsigned long h, l;
+ __rthal_u64tou32(ull, h, l);
+ return __rthal_div96by32(h, l, d, rp);
+}
+#define rthal_ulldiv(ull,d,rp) __rthal_i386_ulldiv((ull),(d),(rp))
+
+#include <asm-generic/xenomai/arith.h>
+
+#endif /* _XENO_ASM_I386_ARITH_H */
Index: include/asm-x86_64/Makefile.in
===================================================================
--- include/asm-x86_64/Makefile.in (revision 2454)
+++ include/asm-x86_64/Makefile.in (working copy)
@@ -230,8 +230,8 @@
target_os = @target_os@
target_vendor = @target_vendor@
includesubdir = $(includedir)/asm-x86_64
-includesub_HEADERS = atomic.h smi.h switch.h calibration.h features.h \
- hal.h syscall.h system.h wrappers.h fptest.h
+includesub_HEADERS = arith.h atomic.h calibration.h features.h fptest.h \
+ hal.h smi.h switch.h syscall.h system.h wrappers.h
SUBDIRS = bits
all: all-recursive
Index: include/asm-x86_64/Makefile.am
===================================================================
--- include/asm-x86_64/Makefile.am (revision 2454)
+++ include/asm-x86_64/Makefile.am (working copy)
@@ -1,6 +1,6 @@
includesubdir = $(includedir)/asm-x86_64
-includesub_HEADERS = atomic.h smi.h switch.h calibration.h features.h \
- hal.h syscall.h system.h wrappers.h fptest.h
+includesub_HEADERS = arith.h atomic.h calibration.h features.h fptest.h \
+ hal.h smi.h switch.h syscall.h system.h wrappers.h
SUBDIRS = bits
Index: include/asm-x86_64/arith.h
===================================================================
--- include/asm-x86_64/arith.h (revision 0)
+++ include/asm-x86_64/arith.h (revision 0)
@@ -0,0 +1,6 @@
+#ifndef _XENO_ASM_X86_64_ARITH_H
+#define _XENO_ASM_X86_64_ARITH_H
+
+#include <asm-generic/xenomai/arith.h>
+
+#endif /* _XENO_ASM_X86_64_ARITH_H */
Index: src/skins/psos+/init.c
===================================================================
--- src/skins/psos+/init.c (revision 2454)
+++ src/skins/psos+/init.c (working copy)
@@ -27,6 +27,7 @@
{
__psos_muxid =
xeno_bind_skin(PSOS_SKIN_MAGIC, "psos", "xeno_psos");
+ __psos_muxid = __xn_mux_shifted_id(__psos_muxid);
}
void k_fatal(u_long err_code, u_long flags)
Index: src/skins/rtai/init.c
===================================================================
--- src/skins/rtai/init.c (revision 2454)
+++ src/skins/rtai/init.c (working copy)
@@ -30,4 +30,5 @@
{
__rtai_muxid =
xeno_bind_skin(RTAI_SKIN_MAGIC, "RTAI", "xeno_rtai");
+ __rtai_muxid = __xn_mux_shifted_id(__rtai_muxid);
}
Index: src/skins/posix/init.c
===================================================================
--- src/skins/posix/init.c (revision 2454)
+++ src/skins/posix/init.c (working copy)
@@ -35,6 +35,7 @@
static int fork_handler_registered;
int __wrap_pthread_setschedparam(pthread_t, int, const struct sched_param *);
+void pse51_clock_init(int);
static __attribute__ ((constructor))
void __init_posix_interface(void)
@@ -42,9 +43,15 @@
struct sched_param parm;
int muxid, err;
- __pse51_muxid =
+ muxid =
xeno_bind_skin(PSE51_SKIN_MAGIC, "POSIX", "xeno_posix");
+#ifdef CONFIG_XENO_HW_DIRECT_TSC
+ pse51_clock_init(muxid);
+#endif /* CONFIG_XENO_HW_DIRECT_TSC */
+
+ __pse51_muxid = __xn_mux_shifted_id(muxid);
+
muxid = XENOMAI_SYSBIND(RTDM_SKIN_MAGIC,
XENOMAI_FEAT_DEP, XENOMAI_ABI_REV, NULL);
if (muxid > 0) {
Index: src/skins/posix/clock.c
===================================================================
--- src/skins/posix/clock.c (revision 2454)
+++ src/skins/posix/clock.c (working copy)
@@ -20,9 +20,24 @@
#include <pthread.h> /* For pthread_setcanceltype. */
#include <posix/syscall.h>
#include <time.h>
+#include <asm/xenomai/arith.h>
extern int __pse51_muxid;
+#ifdef CONFIG_XENO_HW_DIRECT_TSC
+static xnsysinfo_t sysinfo;
+
+void pse51_clock_init(int muxid)
+{
+ int err = -XENOMAI_SYSCALL2(__xn_sys_info, muxid, &sysinfo);
+ if (err) {
+ fprintf(stderr, "Xenomai Posix skin init: "
+ "sys_info: %s\n", strerror(err));
+ exit(EXIT_FAILURE);
+ }
+}
+#endif /* CONFIG_XENO_HW_DIRECT_TSC */
+
int __wrap_clock_getres(clockid_t clock_id, struct timespec *tp)
{
int err = -XENOMAI_SKINCALL2(__pse51_muxid,
@@ -39,11 +54,25 @@
int __wrap_clock_gettime(clockid_t clock_id, struct timespec *tp)
{
- int err = -XENOMAI_SKINCALL2(__pse51_muxid,
- __pse51_clock_gettime,
- clock_id,
- tp);
+ int err;
+#ifdef CONFIG_XENO_HW_DIRECT_TSC
+ if (clock_id == CLOCK_MONOTONIC) {
+ unsigned long long tsc, ns;
+ unsigned long rem;
+ tsc = __xn_rdtsc();
+ ns = xnarch_llimd(tsc, 1000000000, sysinfo.cpufreq);
+ tp->tv_sec = xnarch_ulldiv(ns, 1000000000, &rem);
+ tp->tv_nsec = rem;
+ return 0;
+ }
+#endif /* CONFIG_XENO_HW_DIRECT_TSC */
+
+ err = -XENOMAI_SKINCALL2(__pse51_muxid,
+ __pse51_clock_gettime,
+ clock_id,
+ tp);
+
if (!err)
return 0;
Index: src/skins/vrtx/init.c
===================================================================
--- src/skins/vrtx/init.c (revision 2454)
+++ src/skins/vrtx/init.c (working copy)
@@ -43,6 +43,7 @@
__vrtx_muxid =
xeno_bind_skin(VRTX_SKIN_MAGIC, "vrtx", "xeno_vrtx");
+ __vrtx_muxid = __xn_mux_shifted_id(__vrtx_muxid);
/* Allocate a TSD key for indexing self task pointers. */
Index: src/skins/vxworks/init.c
===================================================================
--- src/skins/vxworks/init.c (revision 2454)
+++ src/skins/vxworks/init.c (working copy)
@@ -41,6 +41,7 @@
{
__vxworks_muxid = xeno_bind_skin(VXWORKS_SKIN_MAGIC,
"vxworks", "xeno_vxworks");
+ __vxworks_muxid = __xn_mux_shifted_id(__vxworks_muxid);
/* Allocate a TSD key for indexing self task pointers. */
Index: src/skins/native/init.c
===================================================================
--- src/skins/native/init.c (revision 2454)
+++ src/skins/native/init.c (working copy)
@@ -30,6 +30,7 @@
pthread_key_t __native_tskey;
int __native_muxid = -1;
+void native_timer_init(int);
static void __flush_tsd(void *tsd)
{
@@ -43,6 +44,12 @@
__native_muxid =
xeno_bind_skin(XENO_SKIN_MAGIC, "native", "xeno_native");
+#ifdef CONFIG_XENO_HW_DIRECT_TSC
+ native_timer_init(__native_muxid);
+#endif /* CONFIG_XENO_HW_DIRECT_TSC */
+
+ __native_muxid = __xn_mux_shifted_id(__native_muxid);
+
/* Allocate a TSD key for indexing self task pointers. */
if (pthread_key_create(&__native_tskey, &__flush_tsd) != 0) {
Index: src/skins/native/timer.c
===================================================================
--- src/skins/native/timer.c (revision 2454)
+++ src/skins/native/timer.c (working copy)
@@ -18,9 +18,24 @@
#include <native/syscall.h>
#include <native/timer.h>
+#include <asm/xenomai/arith.h>
extern int __native_muxid;
+#ifdef CONFIG_XENO_HW_DIRECT_TSC
+static xnsysinfo_t sysinfo;
+
+void native_timer_init(int muxid)
+{
+ int err = XENOMAI_SYSCALL2(__xn_sys_info, muxid, &sysinfo);
+ if (err) {
+ fprintf(stderr, "Native skin init: "
+ "sys_info: %s\n", strerror(-err));
+ exit(EXIT_FAILURE);
+ }
+}
+#endif /* CONFIG_XENO_HW_DIRECT_TSC */
+
int rt_timer_set_mode(RTIME tickval)
{
return XENOMAI_SKINCALL1(__native_muxid, __native_timer_set_mode,
@@ -68,7 +83,11 @@
{
RTIME ticks;
+#ifdef CONFIG_XENO_HW_DIRECT_TSC
+ ticks = xnarch_llimd(ns, sysinfo.cpufreq, 1000000000);
+#else /* !CONFIG_XENO_HW_DIRECT_TSC */
XENOMAI_SKINCALL2(__native_muxid, __native_timer_ns2tsc, &ticks, &ns);
+#endif /* !CONFIG_XENO_HW_DIRECT_TSC */
return ticks;
}
@@ -76,7 +95,11 @@
{
SRTIME ns;
+#ifdef CONFIG_XENO_HW_DIRECT_TSC
+ ns = xnarch_llimd(ticks, 1000000000, sysinfo.cpufreq);
+#else /* !CONFIG_XENO_HW_DIRECT_TSC */
XENOMAI_SKINCALL2(__native_muxid, __native_timer_tsc2ns, &ns, &ticks);
+#endif /* !CONFIG_XENO_HW_DIRECT_TSC */
return ns;
}
Index: src/skins/rtdm/init.c
===================================================================
--- src/skins/rtdm/init.c (revision 2454)
+++ src/skins/rtdm/init.c (working copy)
@@ -34,4 +34,6 @@
__rtdm_muxid =
xeno_bind_skin_opt(RTDM_SKIN_MAGIC, "rtdm", "xeno_rtdm");
+ __rtdm_muxid = __xn_mux_shifted_id(__rtdm_muxid);
+
}
Index: ksrc/skins/posix/syscall.c
===================================================================
--- ksrc/skins/posix/syscall.c (revision 2454)
+++ ksrc/skins/posix/syscall.c (working copy)
@@ -2891,7 +2891,7 @@
return NULL;
}
-
+
return ERR_PTR(-EINVAL);
}
^ permalink raw reply [flat|nested] 9+ messages in thread
* Re: [Xenomai-core] [PATCH] Use tsc for implementation of clock_gettime.
2007-05-17 15:56 ` Gilles Chanteperdrix
@ 2007-05-18 5:52 ` Jan Kiszka
0 siblings, 0 replies; 9+ messages in thread
From: Jan Kiszka @ 2007-05-18 5:52 UTC (permalink / raw)
To: Gilles Chanteperdrix; +Cc: xenomai
[-- Attachment #1: Type: text/plain, Size: 2490 bytes --]
Gilles Chanteperdrix wrote:
> Gilles Chanteperdrix wrote:
> > Jan Kiszka wrote:
> > > Gilles Chanteperdrix wrote:
> > > > Jan Kiszka wrote:
> > > > > Gilles Chanteperdrix wrote:
> > > > > > Jan Kiszka wrote:
> > > > > > > Gilles Chanteperdrix wrote:
> > > > > > > > Hi,
> > > > > > > >
> > > > > > > > here comes, for review, a patch which reduces the overhead of
> > > > > > > > clock_gettime by directly reading the tsc in user-space for
> > > > > > > > architectures that support it.
> > > > > > >
> > > > > > > Highly welcome. But I have one concern: How and when do you propagate
> > > > > > > wallclock_offset changes to user space?
> > > > > >
> > > > > > Since clock_settime is not implemented, never, but if clock_settime was
> > > > > > implemented, clock_settime would re-issue the __xn_sys_info syscall.
> > > > >
> > > > > This excludes automatic clock adjustment, something I'm convinced we
> > > > > will have to provide in the future.
> > > >
> > > > When we provide automatic clock adjustment, we will have to devise
> > > > something more subtle than just an offset, so we will have to redesign
> > >
> > > I think offset + scaling factor.
> > >
> > > > posix clocks support anyway. Maybe clock_gettime(CLOCK_REALTIME) will
> > > > then always be a syscall. After all, rt_timer_read is a syscall. If you
> > > > want the fast clock, use CLOCK_MONOTONIC or rt_timer_tsc.
> > >
> > > Actually, the issue of the intermediate approach starts earlier: as soon
> > > as you have clock_settime. Then you need to sync the offset across
> > > applications in different processes.
> > >
> > > Having clock_monotonic (and maybe also rt_timer_tsc2ns) as a fast
> > > variant already now is not bad. But beyond that, before introducing a
> > > new interface between kernel and user space, I would like to consider
> > > the effort to get it future-proof immediately. That doesn't mean that we
> > > already have to implement clock adjustment, but we may already prepare
> > > the prerequisites.
> >
> > Ok. Here is a new version that does not break the ABI, and where
> > clock_gettime(CLOCK_REALTIME) remains a syscall.
> >
> > Note that it also adds user-space conversions between tsc and ns to the
> > native skin.
>
> This one is actually tested.
>
Looks good to me, builds and runs fine here.
Jan
[-- Attachment #2: OpenPGP digital signature --]
[-- Type: application/pgp-signature, Size: 249 bytes --]
^ permalink raw reply [flat|nested] 9+ messages in thread
end of thread, other threads:[~2007-05-18 5:52 UTC | newest]
Thread overview: 9+ messages (download: mbox.gz follow: Atom feed
-- links below jump to the message on this page --
2007-05-17 11:29 [Xenomai-core] [PATCH] Use tsc for implementation of clock_gettime Gilles Chanteperdrix
2007-05-17 12:43 ` Jan Kiszka
2007-05-17 12:54 ` Gilles Chanteperdrix
2007-05-17 13:00 ` Jan Kiszka
2007-05-17 13:10 ` Gilles Chanteperdrix
2007-05-17 13:31 ` Jan Kiszka
2007-05-17 15:21 ` Gilles Chanteperdrix
2007-05-17 15:56 ` Gilles Chanteperdrix
2007-05-18 5:52 ` Jan Kiszka
This is an external index of several public inboxes,
see mirroring instructions on how to clone and mirror
all data and code used by this external index.