added comments and @author

apps/membench.c

@@ -16,8 +16,10 @@
  *
  * This file is part of MetalSVM.
  */
-/* 
+/**
- * Based on:
+ * @author Steffen Vogel <steffen.vogel@rwth-aachen.de>
+ *
+ *  Based on:
  *   www.cs.berkeley.edu/~richie/bebop/notes/matmul/files/membench/
  */
 
@@ -28,67 +30,146 @@
 #include <asm/processor.h>
 #include <asm/pmc.h>
 
-#define SAMPLES		40
+#define SAMPLES		1000
-#define CLKS		1e9
+#define CLKS		1000000000L
-#define RANGE_MIN	1024		/* 1 KiB */
+
-#define RANGE_MAX	(16*1024*1024)	/* 16 MiB */
+#define STRIDE_MIN	1
+
+#define RANGE_MIN	(32)		/* 32 Byte */
+#define RANGE_MAX	(32*1024*1024)	/* 32 MiB */
+
+#define FLUSH		1
+
+static inline void warmup(char* memory)
+{
+#ifdef FLUSH
+	tlb_flush();
+	flush_cache();
+#else
+	int index;
+	for (index = 0; index < RANGE_MAX; index++) {
+		memory[index] = 0;
+	}
+#endif
+}
 
 int membench(void* arg)
 {
 	kprintf("======= Starting membench\n");
 
+	/* Init PMCs */
+	struct pmc_caps* cap = pmc_init();
+
+	kprintf("PMC architecural version: %u\n", cap->version);
+	kprintf("There are %u general purpose PMCs (%u bit wide) available\n", cap->gp_count, cap->gp_width);
+	kprintf("There are %u fixed function PMCs (%u bit wide) available\n", cap->ff_count, cap->ff_width);
+
+	int i;
+
+	uint16_t tests[][2] = {
+#if 0
+		{PMC_EVT_MEM_LOAD_RETIRED_DTLB_MISS, 0}
+		{PMC_EVT_MEM_LOAD_RETIRED_L1D_MISS, 0}
+		{PMC_EVT_MEM_LOAD_RETIRED_L2_MISS, 0}
+#elif 0
+		{PMC_EVT_PAGE_WALK_CLKS, PMC_EVT_PAGE_WALK_COUNT}
+#else
+		{PMC_EVT_DTLB_MISS_ANY, PMC_EVT_DTLB_MISS_LD},
+		{PMC_EVT_DTLB_MISS_ANY, PMC_EVT_DTLB_MISS_L0_LD},
+		{PMC_EVT_DTLB_MISS_ANY, PMC_EVT_DTLB_MISS_ST},
+#endif
+	};
+
+
 	char *memory = palloc(RANGE_MAX, 0);
 	kprintf("Allocated test memory: %#lx bytes at %p\n", RANGE_MAX, memory);
 
-	kprintf("%12s%12s%12s%12s\n", "reads", "range", "stride", "clks");
+	kprintf("#%12s%12s%12s%12s%12s%12s\n", "range", "stride", "steps", "runs", "reads", "results");
+
+	irq_disable();
+
+	/* Setup PMCs */
+	pmc_stop_all();
+	pmc_ff_config(1, PMC_FIXED_OS); // CPU_CLK_UNHALTED.CORE
+
+	/* Variables for PMC values */
+	uint64_t gp[2], ff[3];
+	uint64_t gp_ovh[2], ff_ovh[3];
+	int64_t gp_real[2], ff_real[3];
+	uint64_t run, steps, tsteps, index, runs, reads;
+	uint64_t range, stride;
+	volatile char temp;
 
 	/* Run the timing experiments */
-	register uint64_t range, stride;
 	for (range = RANGE_MIN; range <= RANGE_MAX; range *= 2) {
-		for (stride = 1; stride < range/2; stride *= 2) {
+		for (stride = STRIDE_MIN; stride < range; stride *= 2) {
-			register uint64_t t0, t1, t2;
+			runs = SAMPLES * stride / range + 1;
-			register uint64_t run, steps, tsteps, temp, index;
 
-			/* Time the loop with strided access + loop overhead */
+			for (i = 0; i < sizeof(tests) / 4; i++) {
-			steps = 0;
+				pmc_gp_config(0, tests[i][0], PMC_EVTSEL_OS | PMC_EVTSEL_EN, 0, 0);
-			tsteps = 0;
+				pmc_gp_config(1, tests[i][1], PMC_EVTSEL_OS | PMC_EVTSEL_EN, 0, 0);
 
-			// TODO: setup and start PMCs
+				steps = 0;
+				tsteps = 0;
 
-			t0 = rdtsc();
+				/* Time the loop with strided access + loop overhead */
-			do {
+				warmup(memory);
-				for (run = 0; run < SAMPLES; run++) {
+				pmc_reset_all();
-					for (index = 0; index < range; index += stride) {
+				pmc_start_all();
-						memory[index]++;
+				do {
+					for (run = 0; run < runs; run++) {
+						for (index = 0; index < range; index += stride) {
+							temp = memory[index];
+						}
 					}
-				}
 
-				steps++;
+					steps++;
-				t1 = rdtsc() - t0;
+				} while (pmc_ff_read(1) < CLKS);
-			} while (t1 < CLKS);
 
-			// TODO: read and stop PMCs
+				pmc_stop_all();
 
-			/* Try to time just the overheads */
+				gp[0] = pmc_gp_read(0);
-			t0 = rdtsc();
+				gp[1] = pmc_gp_read(1);
-			do {
+				ff[1] = pmc_ff_read(1);
-				for (run = 0; run < SAMPLES; run++) {
+
-					for (index = 0; index < range; index += stride) {
+				/* Try to time just the overheads */
-						temp++;
+				warmup(memory);
+				pmc_reset_all();
+				pmc_start_all();
+				do {
+					for (run = 0; run < runs; run++) {
+						for (index = 0; index < range; index += stride) {
+							temp++;
+						}
 					}
-				}
 
-				tsteps++;
+					tsteps++;
-				t2 = rdtsc() - t0;
+				} while (tsteps < steps);
-			} while (tsteps < steps);
+				pmc_stop_all();
 
-			/* Report on the average time per read/write */
+				gp_ovh[0] = pmc_gp_read(0);
-			int64_t clks = t1 - t2;
+				gp_ovh[1] = pmc_gp_read(1);
-			uint64_t reads = SAMPLES * steps * range / stride;
+				ff_ovh[1] = pmc_ff_read(1);
 
-			kprintf("%12d%12d%12d%12llu\n", reads, range, stride, 1000 * clks / reads);
+				gp_real[0] = gp[0] - gp_ovh[0];
+				gp_real[1] = gp[1] - gp_ovh[1];
+				ff_real[1] = ff[1] - ff_ovh[1];
+
+				reads = runs * steps * range / stride;
+
+				if (i == 0)
+					kprintf("%12llu%12llu%12llu%12llu%12llu%12llu", range, stride, steps, runs, reads, ff_real[1]);
+
+				kprintf("%12llu", gp[0]);
+				kprintf("%12llu", gp[1]);
+				kprintf("\t");
+			}
+			kprintf("\n");
 		}
+		kprintf("\n");
 	}
 
+	irq_enable();
+
 	return 0;
 }

apps/memory.c

@@ -17,6 +17,10 @@
  * This file is part of MetalSVM.
  */
 
+/**
+ * @author Steffen Vogel <steffen.vogel@rwth-aachen.de>
+ */
+
 #include <metalsvm/stdlib.h>
 #include <metalsvm/stdio.h>
 #include <metalsvm/stdarg.h>

arch/x86/kernel/uart.c

@@ -20,6 +20,10 @@
 #include <asm/uart.h>
 #include <asm/io.h>
 
+/**
+ * @author Steffen Vogel <steffen.vogel@rwth-aachen.de>
+ */
+
 static inline int is_transmit_empty()
 {
 	return inportb(UART_PORT + 5) & 0x20;

arch/x86/mm/page64.c

@@ -33,6 +33,10 @@
 #include <asm/multiboot.h>
 #include <asm/apic.h>
 
+/**
+ * @author Steffen Vogel <steffen.vogel@rwth-aachen.de>
+ */
+
 /*
  * Virtual Memory Layout of the standard configuration
  * (1 GB kernel space)

include/metalsvm/malloc.h

@@ -17,6 +17,10 @@
  * This file is part of MetalSVM. 
  */
 
+/**
+ * @author Steffen Vogel <steffen.vogel@rwth-aachen.de>
+ */
+
 #ifndef __MALLOC_H__
 #define __MALLOC_H__
 

include/metalsvm/vma.h

@@ -19,6 +19,7 @@
 
 /**
  * @author Stefan Lankes
+ * @author Steffen Vogel <steffen.vogel@rwth-aachen.de>
  * @file include/metalsvm/vma.h
  * @brief VMA related sructure and functions
  */