From 50ca6001808e13cf1f12bfdafc9d8c1422e8f1a4 Mon Sep 17 00:00:00 2001
From: Maxim Baryshnikov <barysmax@fel.cvut.cz>
Date: Wed, 25 May 2016 01:07:30 +0200
Subject: [PATCH] benchmarks: use better bench. bootstrap: Fix slow memory
 access bug.

---
 l4/pkg/benchmark/server/src/main.c            | 211 ++++--------------
 .../server/src/ARCH-x86/crt0.S                |   4 +-
 2 files changed, 51 insertions(+), 164 deletions(-)

diff --git a/l4/pkg/benchmark/server/src/main.c b/l4/pkg/benchmark/server/src/main.c
index 324d222fa..b790cc57f 100644
--- a/l4/pkg/benchmark/server/src/main.c
+++ b/l4/pkg/benchmark/server/src/main.c
@@ -1,183 +1,70 @@
-//taken from [..see below..] and modifed a little.
-/*
- * lib_mem.c - library of routines used to analyze the memory hierarchy
- *
- * @(#)lib_mem.c 1.15 staelin@hpliclu2.hpli.hpl.hp.com
- *
- * Copyright (c) 2000 Carl Staelin.
- * Copyright (c) 1994 Larry McVoy.  
- * Distributed under the FSF GPL with
- * additional restriction that results may published only if
- * (1) the benchmark is unmodified, and
- * (2) the version in the sccsid below is included in the report.
- * Support for this development by Sun Microsystems is gratefully acknowledged.
- */
+
+#define L4
 
 #include <stdio.h>
-#include <string.h>
-#include <stdlib.h>
+#include <stdint.h>
+#include <stdbool.h>
 
-// #include <l4/re/c/mem_alloc.h>
-// #include <l4/re/c/rm.h>
-// #include <l4/re/c/util/cap_alloc.h>
-// #include <l4/sys/err.h>
+#ifdef L4
 
 #include <l4/util/rdtsc.h>
-#include <l4/sys/l4int.h>
 #include <l4/sys/kip.h>
 #include <l4/re/env.h>
+#include <l4/util/util.h>
 
-#define DEF_SIZE (4*1024)
-#define DEF_STRIDE (1)
-#define TEST_TYPE l4_uint8_t
-#define POLLITE_CACHE
-
-void use_dummy(int result);
-TEST_TYPE* words_initialize(l4_uint64_t max, int scale);
-l4_uint64_t parse_param( char param, int argc, char* argv[]);
-//-------------------------------------------------------------
-//---------L4 memory allocation--------------------------------
-/*------from examples/libs/l4re/c/ma+rm.c example-------------*/
-// static int allocate_mem(unsigned long size_in_bytes, unsigned long flags,
-//                         void **virt_addr)
-// {
-//   int r;
-//   l4re_ds_t ds;
-
-//   /* Allocate a free capability index for our data space */
-//   ds = l4re_util_cap_alloc();
-//   if (l4_is_invalid_cap(ds))
-//     return -L4_ENOMEM;
-
-//   size_in_bytes = l4_trunc_page(size_in_bytes);
-
-//   /* Allocate memory via a dataspace */
-//   if ((r = l4re_ma_alloc(size_in_bytes, ds, flags)))
-//     return r;
-
-//   /* Make the dataspace visible in our address space */
-//   *virt_addr = 0;
-//   if ((r = l4re_rm_attach(virt_addr, size_in_bytes,
-//                           L4RE_RM_SEARCH_ADDR, ds, 0,
-//                           flags & L4RE_MA_SUPER_PAGES
-//                              ? L4_SUPERPAGESHIFT : L4_PAGESHIFT)))
-//     return r;
-
-//   /* Done, virtual address is in virt_addr */
-//   return 0;
-// }
+#endif//L4
 
-// /**
-//  * \brief Free previously allocated memory.
-//  *
-//  * \param virt_addr    Virtual address return by allocate_mem
-//  *
-//  * \return 0 on success, error code otherwise
-//  */
-// static int free_mem(void *virt_addr)
-// {
-//   int r;
-//   l4re_ds_t ds;
+#define WORKSET_SIZE (12*1024*1024)
+#define ALL_WORKSETS_BENCH 
 
-//   /* Detach memory from our address space */
-//   if ((r = l4re_rm_detach_ds(virt_addr, &ds)))
-//     return r;
+struct s {
+	int dummy[56];
+	struct s *ptr;
+};
 
-//   /* Free memory at our memory allocator */
-//   if ((r = l4re_ma_free(ds)))
-//     return r;
+struct s array[0x1000000/sizeof(struct s)];
+#define REPEATS (0x20000000)
 
-//   l4re_util_cap_free(ds);
-
-//   /* All went ok */
-//   return 0;
-// }
-
-//-------------------------------------------------------------
-static volatile l4_uint64_t use_result_dummy;
-void use_dummy(int result) { use_result_dummy += result; }
-
-/*
- * words_initialize
- *
- * This is supposed to create the order in which the words in a 
- * "cache line" are used.  Since we rarely know the cache line
- * size with any real reliability, we need to jump around so
- * as to maximize the number of potential cache misses, and to
- * minimize the possibility of re-using a cache line.
- */
-
-TEST_TYPE* words_initialize(l4_uint64_t max, int scale)
+static __inline__ uint64_t rdtsc(void)
 {
-	l4_uint64_t	i, j, nbits;
-	TEST_TYPE*	words = (TEST_TYPE*) malloc(max);
-	printf("%llu bytes allocated.\n", max*sizeof(TEST_TYPE));
-	memset(words, 0, max);
-
-	if (!words) return NULL;
+	uint32_t a, d;
+	//asm("cpuid");
+	asm volatile("rdtsc" : "=a" (a), "=d" (d));
 
-	printf("Start init.");
-	//bzero(words, max * sizeof(size_t));
-#ifdef POLLITE_CACHE
-	for (i = max>>1, nbits = 0; i != 0; i >>= 1, nbits++);
-	 for (i = 0; i < max; ++i) {
-	 	/* now reverse the bits */
-	 	for (j = 0; j < nbits; j++) {
-	 		if (i & (1<<j)) {
-	 			words[i] |= (1<<(nbits-j-1));
-			}
-	 	}
-	 	words[i] *= scale;
-	 }
- #endif
-	return words;
+	return (((uint64_t)a) | (((uint64_t)d) << 32));
 }
 
-l4_uint64_t parse_param( char param, int argc, char* argv[])
+int main(int argc, char *argv[])
 {
-	int i;
-	for (i = 1; i < argc; i++){
-		if ('-' == argv[i][0] && param == argv[i][1] ){
-			return atoll(argv[i+1]);		
-		}
-	}
-	return 0;
-}
-
-
-int main(int argc, char* argv[])
-{
-	printf("Benchmark started.\n");
-	l4_uint64_t time_start, time_end, time_diff, j=0, i, stride = DEF_STRIDE;
-	l4_uint64_t size = DEF_SIZE;
-	// //size = parse_param('s', argc, argv);
-	// //if ( !size ) 
-	// //stride = parse_param('t', argc, argv); 
-	// //if ( !stride ) 
 	
-	TEST_TYPE * arry = words_initialize(size, 5);
-	printf("Done init.\n");
-	 if (!arry) {
-	 	printf("Init failed.");
-	 	return -1;
-	 }	
-	
-	printf("Start benchmark: array size is %llu, stride is %llu.\n", size, stride);
-	l4_calibrate_tsc(l4re_kip());
-
+	unsigned int size;
 	while(1){
-	 	
-	 	time_start = l4_tsc_to_us(l4_rdtsc());              
-		for (i = 0; i < size; i += stride) {
-	 		j += arry[i];
-	 	}
-	 	time_end = l4_tsc_to_us(l4_rdtsc());
-	 	use_dummy(j);
-
-	 	time_diff = (time_end-time_start);
-	    double baud = ((double) size*(sizeof(TEST_TYPE)))/ ((double) time_diff);
-
-	 	printf("time: %llu us, baud: %lf bytes/us\n", time_diff, baud);
-	 	sleep(1);
+#ifdef ALL_WORKSETS_BENCH
+		for (size = 1024; size <= sizeof(array); size *= 2)
+#else //!ALL_WORKSETS_BENCH
+		size = WORKSET_SIZE;
+		while (1)	
+#endif
+		{
+			unsigned int i;
+			for (i=0; i < size / sizeof(array[0]); i++)
+				array[i].ptr = &array[i+1];
+			array[ size / sizeof(array[0]) - 1].ptr = &array[0];
+
+			i = REPEATS;
+			volatile struct s *p = &array[0];
+			uint64_t tic, tac;
+			tic = rdtsc();
+			while (i--) {
+				p = p->ptr;
+				//printf("%p\n", p);
+			}
+			tac = rdtsc();
+			printf("%d %llu\n", size, (tac - tic) / REPEATS);
+			fflush(stdout);
+		}
+		printf("10 sec wait and do again.\n");
+		l4_sleep(10000);
 	}
+	return 0;	
 }
diff --git a/l4/pkg/bootstrap_custom/server/src/ARCH-x86/crt0.S b/l4/pkg/bootstrap_custom/server/src/ARCH-x86/crt0.S
index 3595b3497..4e3b04e15 100644
--- a/l4/pkg/bootstrap_custom/server/src/ARCH-x86/crt0.S
+++ b/l4/pkg/bootstrap_custom/server/src/ARCH-x86/crt0.S
@@ -74,12 +74,12 @@ _start:
 
 	mov $(X86_CR0_PG | X86_CR0_WP | X86_CR0_PE),%eax
 	mov %eax,%cr0
-	
+	*/
 	movl $MSR_MTRR_DEF_TYPE,%ecx
 	rdmsr
 	or $MTRR_ENABLE,%eax
 	wrmsr
-	*/
+	
 	cld
 	cli
 	mov	$(3 * 8), %eax
-- 
2.39.2