Return-Path: Received: (majordomo@vger.kernel.org) by vger.kernel.org via listexpand id S1030919AbXAZQ5a (ORCPT ); Fri, 26 Jan 2007 11:57:30 -0500 Received: (majordomo@vger.kernel.org) by vger.kernel.org id S1030879AbXAZQ5C (ORCPT ); Fri, 26 Jan 2007 11:57:02 -0500 Received: from mtagate2.de.ibm.com ([195.212.29.151]:25746 "EHLO mtagate2.de.ibm.com" rhost-flags-OK-OK-OK-OK) by vger.kernel.org with ESMTP id S1030786AbXAZQ4Z (ORCPT ); Fri, 26 Jan 2007 11:56:25 -0500 Date: Fri, 26 Jan 2007 17:56:16 +0100 From: Martin Schwidefsky To: linux-kernel@vger.kernel.org, linux-s390@vger.kernel.org Subject: [S390] Calibrate delay and bogomips. Message-ID: <20070126165616.GB11609@skybase> MIME-Version: 1.0 Content-Type: text/plain; charset=us-ascii Content-Disposition: inline User-Agent: Mutt/1.5.13 (2006-08-11) Sender: linux-kernel-owner@vger.kernel.org X-Mailing-List: linux-kernel@vger.kernel.org Content-Length: 15638 Lines: 467 From: Martin Schwidefsky [S390] Calibrate delay and bogomips. Preset the bogomips number to the cpu capacity value reported by store system information in SYSIB 1.2.2. This value is constant for a particular machine model and can be used to determine relative performance differences between machines. Signed-off-by: Martin Schwidefsky --- arch/s390/Kconfig | 4 -- arch/s390/lib/Makefile | 2 - arch/s390/lib/qrnnd.S | 77 ++++++++++++++++++++++++++++++++++++++++++ arch/s390/math-emu/Makefile | 2 - arch/s390/math-emu/math.c | 2 - arch/s390/math-emu/qrnnd.S | 77 ------------------------------------------ arch/s390/math-emu/sfp-util.h | 66 ------------------------------------ drivers/s390/Makefile | 2 + drivers/s390/sysinfo.c | 51 +++++++++++++++++++++++++++ include/asm-s390/sfp-util.h | 66 ++++++++++++++++++++++++++++++++++++ 10 files changed, 198 insertions(+), 151 deletions(-) diff -urpN linux-2.6/arch/s390/Kconfig linux-2.6-patched/arch/s390/Kconfig --- linux-2.6/arch/s390/Kconfig 2007-01-26 17:29:31.000000000 +0100 +++ linux-2.6-patched/arch/s390/Kconfig 2007-01-26 17:29:37.000000000 +0100 @@ -34,10 +34,6 @@ config GENERIC_HWEIGHT bool default y -config GENERIC_CALIBRATE_DELAY - bool - default y - config GENERIC_TIME def_bool y diff -urpN linux-2.6/arch/s390/lib/Makefile linux-2.6-patched/arch/s390/lib/Makefile --- linux-2.6/arch/s390/lib/Makefile 2007-01-26 17:27:27.000000000 +0100 +++ linux-2.6-patched/arch/s390/lib/Makefile 2007-01-26 17:29:37.000000000 +0100 @@ -4,7 +4,7 @@ EXTRA_AFLAGS := -traditional -lib-y += delay.o string.o uaccess_std.o uaccess_pt.o +lib-y += delay.o string.o uaccess_std.o uaccess_pt.o qrnnd.o lib-$(CONFIG_32BIT) += div64.o lib-$(CONFIG_64BIT) += uaccess_mvcos.o lib-$(CONFIG_SMP) += spinlock.o diff -urpN linux-2.6/arch/s390/lib/qrnnd.S linux-2.6-patched/arch/s390/lib/qrnnd.S --- linux-2.6/arch/s390/lib/qrnnd.S 1970-01-01 01:00:00.000000000 +0100 +++ linux-2.6-patched/arch/s390/lib/qrnnd.S 2007-01-26 17:29:37.000000000 +0100 @@ -0,0 +1,77 @@ +# S/390 __udiv_qrnnd + +# r2 : &__r +# r3 : upper half of 64 bit word n +# r4 : lower half of 64 bit word n +# r5 : divisor d +# the reminder r of the division is to be stored to &__r and +# the quotient q is to be returned + + .text + .globl __udiv_qrnnd +__udiv_qrnnd: + st %r2,24(%r15) # store pointer to reminder for later + lr %r0,%r3 # reload n + lr %r1,%r4 + ltr %r2,%r5 # reload and test divisor + jp 5f + # divisor >= 0x80000000 + srdl %r0,2 # n/4 + srl %r2,1 # d/2 + slr %r1,%r2 # special case if last bit of d is set + brc 3,0f # (n/4) div (n/2) can overflow by 1 + ahi %r0,-1 # trick: subtract n/2, then divide +0: dr %r0,%r2 # signed division + ahi %r1,1 # trick part 2: add 1 to the quotient + # now (n >> 2) = (d >> 1) * %r1 + %r0 + lhi %r3,1 + nr %r3,%r1 # test last bit of q + jz 1f + alr %r0,%r2 # add (d>>1) to r +1: srl %r1,1 # q >>= 1 + # now (n >> 2) = (d&-2) * %r1 + %r0 + lhi %r3,1 + nr %r3,%r5 # test last bit of d + jz 2f + slr %r0,%r1 # r -= q + brc 3,2f # borrow ? + alr %r0,%r5 # r += d + ahi %r1,-1 +2: # now (n >> 2) = d * %r1 + %r0 + alr %r1,%r1 # q <<= 1 + alr %r0,%r0 # r <<= 1 + brc 12,3f # overflow on r ? + slr %r0,%r5 # r -= d + ahi %r1,1 # q += 1 +3: lhi %r3,2 + nr %r3,%r4 # test next to last bit of n + jz 4f + ahi %r0,1 # r += 1 +4: clr %r0,%r5 # r >= d ? + jl 6f + slr %r0,%r5 # r -= d + ahi %r1,1 # q += 1 + # now (n >> 1) = d * %r1 + %r0 + j 6f +5: # divisor < 0x80000000 + srdl %r0,1 + dr %r0,%r2 # signed division + # now (n >> 1) = d * %r1 + %r0 +6: alr %r1,%r1 # q <<= 1 + alr %r0,%r0 # r <<= 1 + brc 12,7f # overflow on r ? + slr %r0,%r5 # r -= d + ahi %r1,1 # q += 1 +7: lhi %r3,1 + nr %r3,%r4 # isolate last bit of n + alr %r0,%r3 # r += (n & 1) + clr %r0,%r5 # r >= d ? + jl 8f + slr %r0,%r5 # r -= d + ahi %r1,1 # q += 1 +8: # now n = d * %r1 + %r0 + l %r2,24(%r15) + st %r0,0(%r2) + lr %r2,%r1 + br %r14 + .end __udiv_qrnnd diff -urpN linux-2.6/arch/s390/math-emu/Makefile linux-2.6-patched/arch/s390/math-emu/Makefile --- linux-2.6/arch/s390/math-emu/Makefile 2006-11-29 22:57:37.000000000 +0100 +++ linux-2.6-patched/arch/s390/math-emu/Makefile 2007-01-26 17:29:37.000000000 +0100 @@ -2,7 +2,7 @@ # Makefile for the FPU instruction emulation. # -obj-$(CONFIG_MATHEMU) := math.o qrnnd.o +obj-$(CONFIG_MATHEMU) := math.o EXTRA_CFLAGS := -I$(src) -Iinclude/math-emu -w EXTRA_AFLAGS := -traditional diff -urpN linux-2.6/arch/s390/math-emu/math.c linux-2.6-patched/arch/s390/math-emu/math.c --- linux-2.6/arch/s390/math-emu/math.c 2006-11-29 22:57:37.000000000 +0100 +++ linux-2.6-patched/arch/s390/math-emu/math.c 2007-01-26 17:29:37.000000000 +0100 @@ -15,7 +15,7 @@ #include #include -#include "sfp-util.h" +#include #include #include #include diff -urpN linux-2.6/arch/s390/math-emu/qrnnd.S linux-2.6-patched/arch/s390/math-emu/qrnnd.S --- linux-2.6/arch/s390/math-emu/qrnnd.S 2006-11-29 22:57:37.000000000 +0100 +++ linux-2.6-patched/arch/s390/math-emu/qrnnd.S 1970-01-01 01:00:00.000000000 +0100 @@ -1,77 +0,0 @@ -# S/390 __udiv_qrnnd - -# r2 : &__r -# r3 : upper half of 64 bit word n -# r4 : lower half of 64 bit word n -# r5 : divisor d -# the reminder r of the division is to be stored to &__r and -# the quotient q is to be returned - - .text - .globl __udiv_qrnnd -__udiv_qrnnd: - st %r2,24(%r15) # store pointer to reminder for later - lr %r0,%r3 # reload n - lr %r1,%r4 - ltr %r2,%r5 # reload and test divisor - jp 5f - # divisor >= 0x80000000 - srdl %r0,2 # n/4 - srl %r2,1 # d/2 - slr %r1,%r2 # special case if last bit of d is set - brc 3,0f # (n/4) div (n/2) can overflow by 1 - ahi %r0,-1 # trick: subtract n/2, then divide -0: dr %r0,%r2 # signed division - ahi %r1,1 # trick part 2: add 1 to the quotient - # now (n >> 2) = (d >> 1) * %r1 + %r0 - lhi %r3,1 - nr %r3,%r1 # test last bit of q - jz 1f - alr %r0,%r2 # add (d>>1) to r -1: srl %r1,1 # q >>= 1 - # now (n >> 2) = (d&-2) * %r1 + %r0 - lhi %r3,1 - nr %r3,%r5 # test last bit of d - jz 2f - slr %r0,%r1 # r -= q - brc 3,2f # borrow ? - alr %r0,%r5 # r += d - ahi %r1,-1 -2: # now (n >> 2) = d * %r1 + %r0 - alr %r1,%r1 # q <<= 1 - alr %r0,%r0 # r <<= 1 - brc 12,3f # overflow on r ? - slr %r0,%r5 # r -= d - ahi %r1,1 # q += 1 -3: lhi %r3,2 - nr %r3,%r4 # test next to last bit of n - jz 4f - ahi %r0,1 # r += 1 -4: clr %r0,%r5 # r >= d ? - jl 6f - slr %r0,%r5 # r -= d - ahi %r1,1 # q += 1 - # now (n >> 1) = d * %r1 + %r0 - j 6f -5: # divisor < 0x80000000 - srdl %r0,1 - dr %r0,%r2 # signed division - # now (n >> 1) = d * %r1 + %r0 -6: alr %r1,%r1 # q <<= 1 - alr %r0,%r0 # r <<= 1 - brc 12,7f # overflow on r ? - slr %r0,%r5 # r -= d - ahi %r1,1 # q += 1 -7: lhi %r3,1 - nr %r3,%r4 # isolate last bit of n - alr %r0,%r3 # r += (n & 1) - clr %r0,%r5 # r >= d ? - jl 8f - slr %r0,%r5 # r -= d - ahi %r1,1 # q += 1 -8: # now n = d * %r1 + %r0 - l %r2,24(%r15) - st %r0,0(%r2) - lr %r2,%r1 - br %r14 - .end __udiv_qrnnd diff -urpN linux-2.6/arch/s390/math-emu/sfp-util.h linux-2.6-patched/arch/s390/math-emu/sfp-util.h --- linux-2.6/arch/s390/math-emu/sfp-util.h 2006-11-29 22:57:37.000000000 +0100 +++ linux-2.6-patched/arch/s390/math-emu/sfp-util.h 1970-01-01 01:00:00.000000000 +0100 @@ -1,66 +0,0 @@ -#include -#include -#include -#include - -#define add_ssaaaa(sh, sl, ah, al, bh, bl) ({ \ - unsigned int __sh = (ah); \ - unsigned int __sl = (al); \ - asm volatile( \ - " alr %1,%3\n" \ - " brc 12,0f\n" \ - " ahi %0,1\n" \ - "0: alr %0,%2" \ - : "+&d" (__sh), "+d" (__sl) \ - : "d" (bh), "d" (bl) : "cc"); \ - (sh) = __sh; \ - (sl) = __sl; \ -}) - -#define sub_ddmmss(sh, sl, ah, al, bh, bl) ({ \ - unsigned int __sh = (ah); \ - unsigned int __sl = (al); \ - asm volatile( \ - " slr %1,%3\n" \ - " brc 3,0f\n" \ - " ahi %0,-1\n" \ - "0: slr %0,%2" \ - : "+&d" (__sh), "+d" (__sl) \ - : "d" (bh), "d" (bl) : "cc"); \ - (sh) = __sh; \ - (sl) = __sl; \ -}) - -/* a umul b = a mul b + (a>=2<<31) ? b<<32:0 + (b>=2<<31) ? a<<32:0 */ -#define umul_ppmm(wh, wl, u, v) ({ \ - unsigned int __wh = u; \ - unsigned int __wl = v; \ - asm volatile( \ - " ltr 1,%0\n" \ - " mr 0,%1\n" \ - " jnm 0f\n" \ - " alr 0,%1\n" \ - "0: ltr %1,%1\n" \ - " jnm 1f\n" \ - " alr 0,%0\n" \ - "1: lr %0,0\n" \ - " lr %1,1\n" \ - : "+d" (__wh), "+d" (__wl) \ - : : "0", "1", "cc"); \ - wh = __wh; \ - wl = __wl; \ -}) - -#define udiv_qrnnd(q, r, n1, n0, d) \ - do { unsigned long __r; \ - (q) = __udiv_qrnnd (&__r, (n1), (n0), (d)); \ - (r) = __r; \ - } while (0) -extern unsigned long __udiv_qrnnd (unsigned long *, unsigned long, - unsigned long , unsigned long); - -#define UDIV_NEEDS_NORMALIZATION 0 - -#define abort() return 0 - -#define __BYTE_ORDER __BIG_ENDIAN diff -urpN linux-2.6/drivers/s390/Makefile linux-2.6-patched/drivers/s390/Makefile --- linux-2.6/drivers/s390/Makefile 2006-11-29 22:57:37.000000000 +0100 +++ linux-2.6-patched/drivers/s390/Makefile 2007-01-26 17:29:37.000000000 +0100 @@ -2,6 +2,8 @@ # Makefile for the S/390 specific device drivers # +CFLAGS_sysinfo.o += -Iinclude/math-emu -Iarch/s390/math-emu -w + obj-y += s390mach.o sysinfo.o s390_rdev.o obj-y += cio/ block/ char/ crypto/ net/ scsi/ diff -urpN linux-2.6/drivers/s390/sysinfo.c linux-2.6-patched/drivers/s390/sysinfo.c --- linux-2.6/drivers/s390/sysinfo.c 2006-11-29 22:57:37.000000000 +0100 +++ linux-2.6-patched/drivers/s390/sysinfo.c 2007-01-26 17:29:37.000000000 +0100 @@ -9,8 +9,14 @@ #include #include #include +#include #include +/* Sigh, math-emu. Don't ask. */ +#include +#include +#include + struct sysinfo_1_1_1 { char reserved_0[32]; char manufacturer[16]; @@ -198,7 +204,7 @@ static int stsi_1_2_2(struct sysinfo_1_2 * if the higher order 8 bits are not zero. Printing * a floating point number in the kernel is a no-no, * always print the number as 32 bit unsigned integer. - * The user-space needs to know about the stange + * The user-space needs to know about the strange * encoding of the alternate cpu capability. */ len += sprintf(page + len, "Capability: %u %u\n", @@ -351,3 +357,46 @@ static __init int create_proc_sysinfo(vo __initcall(create_proc_sysinfo); +/* + * calibrate the delay loop + */ +void __init calibrate_delay(void) +{ + struct sysinfo_1_2_2 *info = (void *) get_zeroed_page (GFP_KERNEL); + const unsigned int fmil = 0x4b189680; /* 1e7 as 32-bit float. */ + FP_DECL_S(SA); FP_DECL_S(SB); FP_DECL_S(SR); + FP_DECL_EX; + unsigned int capability; + + if (stsi(info, 1, 2, 2) != -ENOSYS) { + /* + * Major sigh. The cpu capability encoding is "special". + * If the first 9 bits of info->capability are 0 then it + * is a 32 bit unsigned integer in the range 0 .. 2^23. + * If the first 9 bits are != 0 then it is a 32 bit float. + * In addition a lower value indicates a proportionally + * higher cpu capacity. Bogomips are the other way round. + * To get to a halfway suitable number we divide 1e7 + * by the cpu capability number. Yes, that means a floating + * point division .. math-emu here we come :-) + */ + FP_UNPACK_SP(SA, &fmil); + if ((info->capability >> 23) == 0) + FP_FROM_INT_S(SB, info->capability, 32, int); + else + FP_UNPACK_SP(SB, &info->capability); + FP_DIV_S(SR, SA, SB); + FP_TO_INT_S(capability, SR, 32, 0); + } else + /* + * Really old machine without stsi block for basic + * cpu information. Report 42.0 bogomips. + */ + capability = 42; + loops_per_jiffy = capability * (500000/HZ); + free_page((unsigned long) info); + /* Print the good old Bogomips line .. */ + printk(KERN_DEBUG "Calibrating delay loop (skipped)... " + "%lu.%02lu BogoMIPS preset\n", loops_per_jiffy/(500000/HZ), + (loops_per_jiffy/(5000/HZ)) % 100); +} diff -urpN linux-2.6/include/asm-s390/sfp-util.h linux-2.6-patched/include/asm-s390/sfp-util.h --- linux-2.6/include/asm-s390/sfp-util.h 1970-01-01 01:00:00.000000000 +0100 +++ linux-2.6-patched/include/asm-s390/sfp-util.h 2007-01-26 17:29:37.000000000 +0100 @@ -0,0 +1,66 @@ +#include +#include +#include +#include + +#define add_ssaaaa(sh, sl, ah, al, bh, bl) ({ \ + unsigned int __sh = (ah); \ + unsigned int __sl = (al); \ + asm volatile( \ + " alr %1,%3\n" \ + " brc 12,0f\n" \ + " ahi %0,1\n" \ + "0: alr %0,%2" \ + : "+&d" (__sh), "+d" (__sl) \ + : "d" (bh), "d" (bl) : "cc"); \ + (sh) = __sh; \ + (sl) = __sl; \ +}) + +#define sub_ddmmss(sh, sl, ah, al, bh, bl) ({ \ + unsigned int __sh = (ah); \ + unsigned int __sl = (al); \ + asm volatile( \ + " slr %1,%3\n" \ + " brc 3,0f\n" \ + " ahi %0,-1\n" \ + "0: slr %0,%2" \ + : "+&d" (__sh), "+d" (__sl) \ + : "d" (bh), "d" (bl) : "cc"); \ + (sh) = __sh; \ + (sl) = __sl; \ +}) + +/* a umul b = a mul b + (a>=2<<31) ? b<<32:0 + (b>=2<<31) ? a<<32:0 */ +#define umul_ppmm(wh, wl, u, v) ({ \ + unsigned int __wh = u; \ + unsigned int __wl = v; \ + asm volatile( \ + " ltr 1,%0\n" \ + " mr 0,%1\n" \ + " jnm 0f\n" \ + " alr 0,%1\n" \ + "0: ltr %1,%1\n" \ + " jnm 1f\n" \ + " alr 0,%0\n" \ + "1: lr %0,0\n" \ + " lr %1,1\n" \ + : "+d" (__wh), "+d" (__wl) \ + : : "0", "1", "cc"); \ + wh = __wh; \ + wl = __wl; \ +}) + +#define udiv_qrnnd(q, r, n1, n0, d) \ + do { unsigned int __r; \ + (q) = __udiv_qrnnd (&__r, (n1), (n0), (d)); \ + (r) = __r; \ + } while (0) +extern unsigned long __udiv_qrnnd (unsigned int *, unsigned int, + unsigned int , unsigned int); + +#define UDIV_NEEDS_NORMALIZATION 0 + +#define abort() return 0 + +#define __BYTE_ORDER __BIG_ENDIAN - To unsubscribe from this list: send the line "unsubscribe linux-kernel" in the body of a message to majordomo@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html Please read the FAQ at http://www.tux.org/lkml/