| 1 | /* |
| 2 | * arch/ubicom32/oprofile/profile.c |
| 3 | * Oprofile support for arch Ubicom32 |
| 4 | * |
| 5 | * (C) Copyright 2009, Ubicom, Inc. |
| 6 | * |
| 7 | * This file is part of the Ubicom32 Linux Kernel Port. |
| 8 | * |
| 9 | * The Ubicom32 Linux Kernel Port is free software: you can redistribute |
| 10 | * it and/or modify it under the terms of the GNU General Public License |
| 11 | * as published by the Free Software Foundation, either version 2 of the |
| 12 | * License, or (at your option) |
| 13 | * any later version. |
| 14 | * |
| 15 | * The Ubicom32 Linux Kernel Port is distributed in the hope that it will |
| 16 | * be useful, but WITHOUT ANY WARRANTY; without even the implied warranty |
| 17 | * of MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU |
| 18 | * General Public License for more details. |
| 19 | * |
| 20 | * You should have received a copy of the GNU General Public License |
| 21 | * along with the Ubicom32 Linux Kernel Port. If not, see |
| 22 | * <http://www.gnu.org/licenses/>. |
| 23 | * |
| 24 | * Ubicom32 implementation derived from (with many thanks): |
| 25 | * arch/m68knommu |
| 26 | * arch/blackfin |
| 27 | * arch/parisc |
| 28 | */ |
| 29 | |
| 30 | /** |
| 31 | * @file profile.c |
| 32 | * |
| 33 | * @remark Copyright 2002 OProfile authors |
| 34 | * @remark Read the file COPYING |
| 35 | * |
| 36 | * @author Hunyue Yau <hy@hy-research.com> |
| 37 | */ |
| 38 | |
| 39 | #include <linux/oprofile.h> |
| 40 | #include <linux/init.h> |
| 41 | #include <linux/errno.h> |
| 42 | #include <linux/interrupt.h> |
| 43 | #include <linux/module.h> |
| 44 | #include <linux/kernel.h> |
| 45 | |
| 46 | #include <asm/devtree.h> |
| 47 | #include <asm/thread.h> |
| 48 | |
| 49 | /* For identifying userland vs kernel address */ |
| 50 | #include <asm/stacktrace.h> |
| 51 | #include "ipProf.h" |
| 52 | |
| 53 | /* For communications with the backend */ |
| 54 | static struct profilenode *profile_node; |
| 55 | |
| 56 | /* Bitmask containing all Linux threads - as seen by the ROSR reg */ |
| 57 | static unsigned long th_all_mask; |
| 58 | |
| 59 | /* Lookup table to translate a hardware thread into a CPU identifier |
| 60 | * Table is indexed by the ROSR value which is assumed to be |
| 61 | * relatively small (0...15). |
| 62 | */ |
| 63 | unsigned int cpu_map[THREAD_ARCHITECTURAL_MAX]; |
| 64 | |
| 65 | static struct pt_regs regs; |
| 66 | |
| 67 | /* |
| 68 | * For each sample returned, checked to see if they are relevant to |
| 69 | * us. This is necessary as the ubicom32 architecture has other software |
| 70 | * running outside of Linux. Only then, put the sample into the relevant |
| 71 | * cpu bins. |
| 72 | * |
| 73 | * To minimize overhead, a global mask with all possible threads of in |
| 74 | * interest to us is used as a first check. Then a second mask identifying |
| 75 | * the thread is used to obtain an identifier for that "CPU". |
| 76 | */ |
| 77 | |
| 78 | /* |
| 79 | * ubicom32_build_cpu_th_mask() |
| 80 | * |
| 81 | * Build a lookup table for translation between hardware thread |
| 82 | * "ROSR" values and Linux CPU ids |
| 83 | * |
| 84 | * *** This gets executed on all CPUs at once! *** |
| 85 | */ |
| 86 | static void ubicom32_build_cpu_th_mask(void *mask) |
| 87 | { |
| 88 | thread_t self = thread_get_self(); |
| 89 | unsigned long *th_m = mask; |
| 90 | |
| 91 | BUG_ON(self <= 0 || self >= THREAD_ARCHITECTURAL_MAX); |
| 92 | cpu_map[self] = smp_processor_id(); |
| 93 | |
| 94 | set_bit(self, th_m); |
| 95 | } |
| 96 | |
| 97 | /* |
| 98 | * profile_interrupt() |
| 99 | * |
| 100 | * Process samples returned from the profiler backend. The backend |
| 101 | * may return samples that are irrelevant to us or may even return |
| 102 | * multiple samples for the same CPU. Note that the sames may be |
| 103 | * for ANY cpu. At this time, this is unique and to support this requires |
| 104 | * Oprofile to expose an interface to accept the CPU that the same came |
| 105 | * frome. |
| 106 | */ |
| 107 | static irqreturn_t profile_interrupt(int irq, void *arg) |
| 108 | { |
| 109 | int i, buf_entry; |
| 110 | int is_kernel; |
| 111 | unsigned int bit_th; |
| 112 | unsigned int th; |
| 113 | |
| 114 | if (!(profile_node->enabled) || profile_node->count < 0) { |
| 115 | printk(KERN_WARNING |
| 116 | "Unexpected interrupt, no samples or not enabled!\n"); |
| 117 | return IRQ_HANDLED; |
| 118 | } |
| 119 | |
| 120 | profile_node->busy = 1; /* Keep backend out */ |
| 121 | |
| 122 | for (i = 0; i < profile_node->count; i++) { |
| 123 | buf_entry = profile_node->tail; |
| 124 | profile_node->tail++; |
| 125 | profile_node->tail %= IPPROFILETIO_MAX_SAMPLES; |
| 126 | |
| 127 | /* Note - the "thread" ID is only the lower 4 bits */ |
| 128 | th = (0x0f & profile_node->samples[buf_entry].thread); |
| 129 | bit_th = (1 << th); |
| 130 | |
| 131 | if ((bit_th & th_all_mask) == 0) |
| 132 | continue; |
| 133 | |
| 134 | regs.pc = profile_node->samples[buf_entry].pc; |
| 135 | |
| 136 | is_kernel = ubicom32_is_kernel(regs.pc); |
| 137 | |
| 138 | oprofile_add_ext_sample_cpu(regs.pc, ®s, 0, is_kernel, |
| 139 | cpu_map[th]); |
| 140 | } |
| 141 | profile_node->count = 0; |
| 142 | profile_node->busy = 0; |
| 143 | |
| 144 | return IRQ_HANDLED; |
| 145 | } |
| 146 | |
| 147 | /* |
| 148 | * profile_start() |
| 149 | * |
| 150 | * Notification from oprofile to start the profiler |
| 151 | */ |
| 152 | static int profile_start(void) |
| 153 | { |
| 154 | if (!profile_node) |
| 155 | return -1; |
| 156 | |
| 157 | profile_node->enabled = 1; |
| 158 | |
| 159 | return 0; |
| 160 | } |
| 161 | |
| 162 | /* |
| 163 | * profile_stop() |
| 164 | * |
| 165 | * Notification from oprofile to stop the profiler |
| 166 | */ |
| 167 | static void profile_stop(void) |
| 168 | { |
| 169 | if (profile_node) |
| 170 | profile_node->enabled = 0; |
| 171 | } |
| 172 | |
| 173 | /* |
| 174 | * oprofile_arch_init() |
| 175 | * |
| 176 | * Attach to Oprofile after qualify the availability of the backend |
| 177 | * profiler support. |
| 178 | */ |
| 179 | int __init oprofile_arch_init(struct oprofile_operations *ops) |
| 180 | { |
| 181 | int r = -ENODEV; |
| 182 | |
| 183 | profile_node = (struct profilenode *)devtree_find_node("profiler"); |
| 184 | |
| 185 | if (profile_node == NULL) { |
| 186 | printk(KERN_WARNING "Cannot find profiler node\n"); |
| 187 | return r; |
| 188 | } |
| 189 | |
| 190 | r = request_irq(profile_node->dn.recvirq, profile_interrupt, |
| 191 | IRQF_DISABLED, "profiler", NULL); |
| 192 | |
| 193 | if (r < 0) { |
| 194 | profile_node = NULL; |
| 195 | printk(KERN_WARNING "Cannot get profiler IRQ\n"); |
| 196 | return r; |
| 197 | } |
| 198 | |
| 199 | ops->start = profile_start; |
| 200 | ops->stop = profile_stop; |
| 201 | ops->cpu_type = "timer"; |
| 202 | |
| 203 | memset(cpu_map, 0, sizeof(cpu_map)); |
| 204 | |
| 205 | on_each_cpu(ubicom32_build_cpu_th_mask, &th_all_mask, 1); |
| 206 | |
| 207 | memset(®s, 0, sizeof(regs)); |
| 208 | |
| 209 | return r; |
| 210 | } |
| 211 | |
| 212 | /* |
| 213 | * oprofile_arch_exit() |
| 214 | * |
| 215 | * External call to take outselves out. |
| 216 | * Make sure backend is not running. |
| 217 | */ |
| 218 | void oprofile_arch_exit(void) |
| 219 | { |
| 220 | BUG_ON(profile_node->enabled); |
| 221 | } |
| 222 | |
