]> pilppa.org Git - linux-2.6-omap-h63xx.git/blobdiff - kernel/trace/trace.c
ftrace: modulize the number of CPU buffers
[linux-2.6-omap-h63xx.git] / kernel / trace / trace.c
index ea11f4ebfae1c8e6c9e3b34f4b9d4feecda27fef..5da391c5fb0dc25a3e0ccf75460a22046d0b0c62 100644 (file)
 unsigned long __read_mostly    tracing_max_latency = (cycle_t)ULONG_MAX;
 unsigned long __read_mostly    tracing_thresh;
 
+static unsigned long __read_mostly     tracing_nr_buffers;
+static cpumask_t __read_mostly         tracing_buffer_mask;
+
+#define for_each_tracing_cpu(cpu)      \
+       for_each_cpu_mask(cpu, tracing_buffer_mask)
+
+/* dummy trace to disable tracing */
+static struct tracer no_tracer __read_mostly = {
+       .name           = "none",
+};
+
+static int trace_alloc_page(void);
+static int trace_free_page(void);
+
 static int tracing_disabled = 1;
 
-static long
+long
 ns2usecs(cycle_t nsec)
 {
        nsec += 500;
@@ -84,9 +98,16 @@ void trace_wake_up(void)
 
 static int __init set_nr_entries(char *str)
 {
+       unsigned long nr_entries;
+       int ret;
+
        if (!str)
                return 0;
-       trace_nr_entries = simple_strtoul(str, &str, 0);
+       ret = strict_strtoul(str, 0, &nr_entries);
+       /* nr_entries can not be zero */
+       if (ret < 0 || nr_entries == 0)
+               return 0;
+       trace_nr_entries = nr_entries;
        return 1;
 }
 __setup("trace_entries=", set_nr_entries);
@@ -96,18 +117,6 @@ unsigned long nsecs_to_usecs(unsigned long nsecs)
        return nsecs / 1000;
 }
 
-enum trace_type {
-       __TRACE_FIRST_TYPE = 0,
-
-       TRACE_FN,
-       TRACE_CTX,
-       TRACE_WAKE,
-       TRACE_STACK,
-       TRACE_SPECIAL,
-
-       __TRACE_LAST_TYPE
-};
-
 enum trace_flag_type {
        TRACE_FLAG_IRQS_OFF             = 0x01,
        TRACE_FLAG_NEED_RESCHED         = 0x02,
@@ -190,7 +199,7 @@ void *head_page(struct trace_array_cpu *data)
        return page_address(page);
 }
 
-static int
+int
 trace_seq_printf(struct trace_seq *s, const char *fmt, ...)
 {
        int len = (PAGE_SIZE - 1) - s->len;
@@ -205,7 +214,7 @@ trace_seq_printf(struct trace_seq *s, const char *fmt, ...)
        va_end(ap);
 
        /* If we can't write it all, don't bother writing anything */
-       if (ret > len)
+       if (ret >= len)
                return 0;
 
        s->len += ret;
@@ -251,19 +260,18 @@ trace_seq_putmem(struct trace_seq *s, void *mem, size_t len)
 }
 
 #define HEX_CHARS 17
+static const char hex2asc[] = "0123456789abcdef";
 
 static int
 trace_seq_putmem_hex(struct trace_seq *s, void *mem, size_t len)
 {
        unsigned char hex[HEX_CHARS];
-       unsigned char *data;
+       unsigned char *data = mem;
        unsigned char byte;
        int i, j;
 
        BUG_ON(len >= HEX_CHARS);
 
-       data = mem;
-
 #ifdef __BIG_ENDIAN
        for (i = 0, j = 0; i < len; i++) {
 #else
@@ -271,22 +279,10 @@ trace_seq_putmem_hex(struct trace_seq *s, void *mem, size_t len)
 #endif
                byte = data[i];
 
-               hex[j]   = byte & 0x0f;
-               if (hex[j] >= 10)
-                       hex[j] += 'a' - 10;
-               else
-                       hex[j] += '0';
-               j++;
-
-               hex[j] = byte >> 4;
-               if (hex[j] >= 10)
-                       hex[j] += 'a' - 10;
-               else
-                       hex[j] += '0';
-               j++;
+               hex[j++] = hex2asc[byte & 0x0f];
+               hex[j++] = hex2asc[byte >> 4];
        }
-       hex[j] = ' ';
-       j++;
+       hex[j++] = ' ';
 
        return trace_seq_putmem(s, hex, j);
 }
@@ -338,7 +334,7 @@ update_max_tr(struct trace_array *tr, struct task_struct *tsk, int cpu)
        WARN_ON_ONCE(!irqs_disabled());
        __raw_spin_lock(&ftrace_max_lock);
        /* clear out all the previous traces */
-       for_each_possible_cpu(i) {
+       for_each_tracing_cpu(i) {
                data = tr->data[i];
                flip_trace(max_tr.data[i], data);
                tracing_reset(data);
@@ -362,7 +358,7 @@ update_max_tr_single(struct trace_array *tr, struct task_struct *tsk, int cpu)
 
        WARN_ON_ONCE(!irqs_disabled());
        __raw_spin_lock(&ftrace_max_lock);
-       for_each_possible_cpu(i)
+       for_each_tracing_cpu(i)
                tracing_reset(max_tr.data[i]);
 
        flip_trace(max_tr.data[cpu], data);
@@ -408,7 +404,7 @@ int register_tracer(struct tracer *type)
                 * internal tracing to verify that everything is in order.
                 * If we fail, we do not register this tracer.
                 */
-               for_each_possible_cpu(i) {
+               for_each_tracing_cpu(i) {
                        data = tr->data[i];
                        if (!head_page(data))
                                continue;
@@ -427,7 +423,7 @@ int register_tracer(struct tracer *type)
                        goto out;
                }
                /* Only reset on passing, to avoid touching corrupted buffers */
-               for_each_possible_cpu(i) {
+               for_each_tracing_cpu(i) {
                        data = tr->data[i];
                        if (!head_page(data))
                                continue;
@@ -638,7 +634,7 @@ tracing_generic_entry_update(struct trace_entry *entry, unsigned long flags)
        pc = preempt_count();
 
        entry->preempt_count    = pc & 0xff;
-       entry->pid              = tsk->pid;
+       entry->pid              = (tsk) ? tsk->pid : 0;
        entry->t                = ftrace_now(raw_smp_processor_id());
        entry->flags = (irqs_disabled_flags(flags) ? TRACE_FLAG_IRQS_OFF : 0) |
                ((pc & HARDIRQ_MASK) ? TRACE_FLAG_HARDIRQ : 0) |
@@ -857,7 +853,7 @@ find_next_entry(struct trace_iterator *iter, int *ent_cpu)
        int next_cpu = -1;
        int cpu;
 
-       for_each_possible_cpu(cpu) {
+       for_each_tracing_cpu(cpu) {
                if (!head_page(tr->data[cpu]))
                        continue;
                ent = trace_entry_idx(tr, tr->data[cpu], iter, cpu);
@@ -982,7 +978,7 @@ static void *s_start(struct seq_file *m, loff_t *pos)
                iter->prev_ent = NULL;
                iter->prev_cpu = -1;
 
-               for_each_possible_cpu(i) {
+               for_each_tracing_cpu(i) {
                        iter->next_idx[i] = 0;
                        iter->next_page[i] = NULL;
                }
@@ -1099,7 +1095,7 @@ print_trace_header(struct seq_file *m, struct trace_iterator *iter)
        if (type)
                name = type->name;
 
-       for_each_possible_cpu(cpu) {
+       for_each_tracing_cpu(cpu) {
                if (head_page(tr->data[cpu])) {
                        total += tr->data[cpu]->trace_idx;
                        if (tr->data[cpu]->trace_idx > tr->entries)
@@ -1171,12 +1167,12 @@ lat_print_generic(struct trace_seq *s, struct trace_entry *entry, int cpu)
 
        hardirq = entry->flags & TRACE_FLAG_HARDIRQ;
        softirq = entry->flags & TRACE_FLAG_SOFTIRQ;
-       if (hardirq && softirq)
+       if (hardirq && softirq) {
                trace_seq_putc(s, 'H');
-       else {
-               if (hardirq)
+       else {
+               if (hardirq) {
                        trace_seq_putc(s, 'h');
-               else {
+               else {
                        if (softirq)
                                trace_seq_putc(s, 's');
                        else
@@ -1220,6 +1216,7 @@ print_lat_fmt(struct trace_iterator *iter, unsigned int trace_idx, int cpu)
        char *comm;
        int S, T;
        int i;
+       unsigned state;
 
        if (!next_entry)
                next_entry = entry;
@@ -1250,11 +1247,11 @@ print_lat_fmt(struct trace_iterator *iter, unsigned int trace_idx, int cpu)
                break;
        case TRACE_CTX:
        case TRACE_WAKE:
-               S = entry->ctx.prev_state < sizeof(state_to_char) ?
-                       state_to_char[entry->ctx.prev_state] : 'X';
                T = entry->ctx.next_state < sizeof(state_to_char) ?
                        state_to_char[entry->ctx.next_state] : 'X';
 
+               state = entry->ctx.prev_state ? __ffs(entry->ctx.prev_state) + 1 : 0;
+               S = state < sizeof(state_to_char) - 1 ? state_to_char[state] : 'X';
                comm = trace_find_cmdline(entry->ctx.next_pid);
                trace_seq_printf(s, " %5d:%3d:%c %s %5d:%3d:%c %s\n",
                                 entry->ctx.prev_pid,
@@ -1528,7 +1525,7 @@ static int trace_empty(struct trace_iterator *iter)
        struct trace_array_cpu *data;
        int cpu;
 
-       for_each_possible_cpu(cpu) {
+       for_each_tracing_cpu(cpu) {
                data = iter->tr->data[cpu];
 
                if (head_page(data) && data->trace_idx &&
@@ -1541,6 +1538,9 @@ static int trace_empty(struct trace_iterator *iter)
 
 static int print_trace_line(struct trace_iterator *iter)
 {
+       if (iter->trace && iter->trace->print_line)
+               return iter->trace->print_line(iter);
+
        if (trace_flags & TRACE_ITER_BIN)
                return print_bin_fmt(iter);
 
@@ -1837,7 +1837,7 @@ tracing_cpumask_write(struct file *filp, const char __user *ubuf,
 
        raw_local_irq_disable();
        __raw_spin_lock(&ftrace_max_lock);
-       for_each_possible_cpu(cpu) {
+       for_each_tracing_cpu(cpu) {
                /*
                 * Increase/decrease the disabled counter if we are
                 * about to flip a bit in the cpumask:
@@ -1918,8 +1918,8 @@ tracing_iter_ctrl_write(struct file *filp, const char __user *ubuf,
        int neg = 0;
        int i;
 
-       if (cnt > 63)
-               cnt = 63;
+       if (cnt >= sizeof(buf))
+               return -EINVAL;
 
        if (copy_from_user(&buf, ubuf, cnt))
                return -EFAULT;
@@ -2008,18 +2008,21 @@ tracing_ctrl_write(struct file *filp, const char __user *ubuf,
                   size_t cnt, loff_t *ppos)
 {
        struct trace_array *tr = filp->private_data;
-       long val;
        char buf[64];
+       long val;
+       int ret;
 
-       if (cnt > 63)
-               cnt = 63;
+       if (cnt >= sizeof(buf))
+               return -EINVAL;
 
        if (copy_from_user(&buf, ubuf, cnt))
                return -EFAULT;
 
        buf[cnt] = 0;
 
-       val = simple_strtoul(buf, NULL, 10);
+       ret = strict_strtoul(buf, 10, &val);
+       if (ret < 0)
+               return ret;
 
        val = !!val;
 
@@ -2111,10 +2114,10 @@ tracing_max_lat_read(struct file *filp, char __user *ubuf,
        char buf[64];
        int r;
 
-       r = snprintf(buf, 64, "%ld\n",
+       r = snprintf(buf, sizeof(buf), "%ld\n",
                     *ptr == (unsigned long)-1 ? -1 : nsecs_to_usecs(*ptr));
-       if (r > 64)
-               r = 64;
+       if (r > sizeof(buf))
+               r = sizeof(buf);
        return simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
 }
 
@@ -2123,18 +2126,21 @@ tracing_max_lat_write(struct file *filp, const char __user *ubuf,
                      size_t cnt, loff_t *ppos)
 {
        long *ptr = filp->private_data;
-       long val;
        char buf[64];
+       long val;
+       int ret;
 
-       if (cnt > 63)
-               cnt = 63;
+       if (cnt >= sizeof(buf))
+               return -EINVAL;
 
        if (copy_from_user(&buf, ubuf, cnt))
                return -EFAULT;
 
        buf[cnt] = 0;
 
-       val = simple_strtoul(buf, NULL, 10);
+       ret = strict_strtoul(buf, 10, &val);
+       if (ret < 0)
+               return ret;
 
        *ptr = val * 1000;
 
@@ -2162,6 +2168,7 @@ static int tracing_open_pipe(struct inode *inode, struct file *filp)
                return -ENOMEM;
 
        iter->tr = &global_trace;
+       iter->trace = current_trace;
 
        filp->private_data = iter;
 
@@ -2188,8 +2195,7 @@ tracing_poll_pipe(struct file *filp, poll_table *poll_table)
                 * Always select as readable when in blocking mode
                 */
                return POLLIN | POLLRDNORM;
-       }
-       else {
+       } else {
                if (!trace_empty(iter))
                        return POLLIN | POLLRDNORM;
                poll_wait(filp, &trace_wait, poll_table);
@@ -2209,6 +2215,8 @@ tracing_read_pipe(struct file *filp, char __user *ubuf,
 {
        struct trace_iterator *iter = filp->private_data;
        struct trace_array_cpu *data;
+       struct trace_array *tr = iter->tr;
+       struct tracer *tracer = iter->trace;
        static cpumask_t mask;
        static int start;
        unsigned long flags;
@@ -2238,8 +2246,10 @@ tracing_read_pipe(struct file *filp, char __user *ubuf,
        start = 0;
 
        while (trace_empty(iter)) {
-               if (!(trace_flags & TRACE_ITER_BLOCK))
-                       return -EWOULDBLOCK;
+
+               if ((filp->f_flags & O_NONBLOCK))
+                       return -EAGAIN;
+
                /*
                 * This is a make-shift waitqueue. The reason we don't use
                 * an actual wait queue is because:
@@ -2260,6 +2270,9 @@ tracing_read_pipe(struct file *filp, char __user *ubuf,
                if (signal_pending(current))
                        return -EINTR;
 
+               if (iter->trace != current_trace)
+                       return 0;
+
                /*
                 * We block until we read something and tracing is disabled.
                 * We still block if tracing is disabled, but we have never
@@ -2283,7 +2296,8 @@ tracing_read_pipe(struct file *filp, char __user *ubuf,
                cnt = PAGE_SIZE - 1;
 
        memset(iter, 0, sizeof(*iter));
-       iter->tr = &global_trace;
+       iter->tr = tr;
+       iter->trace = tracer;
        iter->pos = -1;
 
        /*
@@ -2300,7 +2314,7 @@ tracing_read_pipe(struct file *filp, char __user *ubuf,
        ftrace_enabled = 0;
 #endif
        smp_wmb();
-       for_each_possible_cpu(cpu) {
+       for_each_tracing_cpu(cpu) {
                data = iter->tr->data[cpu];
 
                if (!head_page(data) || !data->trace_idx)
@@ -2363,6 +2377,73 @@ tracing_read_pipe(struct file *filp, char __user *ubuf,
        return read;
 }
 
+static ssize_t
+tracing_entries_read(struct file *filp, char __user *ubuf,
+                    size_t cnt, loff_t *ppos)
+{
+       struct trace_array *tr = filp->private_data;
+       char buf[64];
+       int r;
+
+       r = sprintf(buf, "%lu\n", tr->entries);
+       return simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
+}
+
+static ssize_t
+tracing_entries_write(struct file *filp, const char __user *ubuf,
+                     size_t cnt, loff_t *ppos)
+{
+       unsigned long val;
+       char buf[64];
+       int ret;
+
+       if (cnt >= sizeof(buf))
+               return -EINVAL;
+
+       if (copy_from_user(&buf, ubuf, cnt))
+               return -EFAULT;
+
+       buf[cnt] = 0;
+
+       ret = strict_strtoul(buf, 10, &val);
+       if (ret < 0)
+               return ret;
+
+       /* must have at least 1 entry */
+       if (!val)
+               return -EINVAL;
+
+       mutex_lock(&trace_types_lock);
+
+       if (current_trace != &no_tracer) {
+               cnt = -EBUSY;
+               pr_info("ftrace: set current_tracer to none"
+                       " before modifying buffer size\n");
+               goto out;
+       }
+
+       if (val > global_trace.entries) {
+               while (global_trace.entries < val) {
+                       if (trace_alloc_page()) {
+                               cnt = -ENOMEM;
+                               goto out;
+                       }
+               }
+       } else {
+               /* include the number of entries in val (inc of page entries) */
+               while (global_trace.entries > val + (ENTRIES_PER_PAGE - 1))
+                       trace_free_page();
+       }
+
+       filp->f_pos += cnt;
+
+ out:
+       max_tr.entries = global_trace.entries;
+       mutex_unlock(&trace_types_lock);
+
+       return cnt;
+}
+
 static struct file_operations tracing_max_lat_fops = {
        .open           = tracing_open_generic,
        .read           = tracing_max_lat_read,
@@ -2388,6 +2469,12 @@ static struct file_operations tracing_pipe_fops = {
        .release        = tracing_release_pipe,
 };
 
+static struct file_operations tracing_entries_fops = {
+       .open           = tracing_open_generic,
+       .read           = tracing_entries_read,
+       .write          = tracing_entries_write,
+};
+
 #ifdef CONFIG_DYNAMIC_FTRACE
 
 static ssize_t
@@ -2499,6 +2586,12 @@ static __init void tracer_init_debugfs(void)
                pr_warning("Could not create debugfs "
                           "'tracing_threash' entry\n");
 
+       entry = debugfs_create_file("trace_entries", 0644, d_tracer,
+                                   &global_trace, &tracing_entries_fops);
+       if (!entry)
+               pr_warning("Could not create debugfs "
+                          "'tracing_threash' entry\n");
+
 #ifdef CONFIG_DYNAMIC_FTRACE
        entry = debugfs_create_file("dyn_ftrace_total_info", 0444, d_tracer,
                                    &ftrace_update_tot_cnt,
@@ -2509,12 +2602,6 @@ static __init void tracer_init_debugfs(void)
 #endif
 }
 
-/* dummy trace to disable tracing */
-static struct tracer no_tracer __read_mostly =
-{
-       .name           = "none",
-};
-
 static int trace_alloc_page(void)
 {
        struct trace_array_cpu *data;
@@ -2524,7 +2611,7 @@ static int trace_alloc_page(void)
        int i;
 
        /* first allocate a page for each CPU */
-       for_each_possible_cpu(i) {
+       for_each_tracing_cpu(i) {
                array = (void *)__get_free_page(GFP_KERNEL);
                if (array == NULL) {
                        printk(KERN_ERR "tracer: failed to allocate page"
@@ -2549,9 +2636,8 @@ static int trace_alloc_page(void)
        }
 
        /* Now that we successfully allocate a page per CPU, add them */
-       for_each_possible_cpu(i) {
+       for_each_tracing_cpu(i) {
                data = global_trace.data[i];
-               data->lock = (raw_spinlock_t)__RAW_SPIN_LOCK_UNLOCKED;
                page = list_entry(pages.next, struct page, lru);
                list_del_init(&page->lru);
                list_add_tail(&page->lru, &data->trace_pages);
@@ -2559,7 +2645,6 @@ static int trace_alloc_page(void)
 
 #ifdef CONFIG_TRACER_MAX_TRACE
                data = max_tr.data[i];
-               data->lock = (raw_spinlock_t)__RAW_SPIN_LOCK_UNLOCKED;
                page = list_entry(pages.next, struct page, lru);
                list_del_init(&page->lru);
                list_add_tail(&page->lru, &data->trace_pages);
@@ -2578,6 +2663,55 @@ static int trace_alloc_page(void)
        return -ENOMEM;
 }
 
+static int trace_free_page(void)
+{
+       struct trace_array_cpu *data;
+       struct page *page;
+       struct list_head *p;
+       int i;
+       int ret = 0;
+
+       /* free one page from each buffer */
+       for_each_tracing_cpu(i) {
+               data = global_trace.data[i];
+               p = data->trace_pages.next;
+               if (p == &data->trace_pages) {
+                       /* should never happen */
+                       WARN_ON(1);
+                       tracing_disabled = 1;
+                       ret = -1;
+                       break;
+               }
+               page = list_entry(p, struct page, lru);
+               ClearPageLRU(page);
+               list_del(&page->lru);
+               __free_page(page);
+
+               tracing_reset(data);
+
+#ifdef CONFIG_TRACER_MAX_TRACE
+               data = max_tr.data[i];
+               p = data->trace_pages.next;
+               if (p == &data->trace_pages) {
+                       /* should never happen */
+                       WARN_ON(1);
+                       tracing_disabled = 1;
+                       ret = -1;
+                       break;
+               }
+               page = list_entry(p, struct page, lru);
+               ClearPageLRU(page);
+               list_del(&page->lru);
+               __free_page(page);
+
+               tracing_reset(data);
+#endif
+       }
+       global_trace.entries -= ENTRIES_PER_PAGE;
+
+       return ret;
+}
+
 __init static int tracer_alloc_buffers(void)
 {
        struct trace_array_cpu *data;
@@ -2589,8 +2723,12 @@ __init static int tracer_alloc_buffers(void)
 
        global_trace.ctrl = tracer_enabled;
 
+       /* TODO: make the number of buffers hot pluggable with CPUS */
+       tracing_nr_buffers = num_possible_cpus();
+       tracing_buffer_mask = cpu_possible_map;
+
        /* Allocate the first page for all buffers */
-       for_each_possible_cpu(i) {
+       for_each_tracing_cpu(i) {
                data = global_trace.data[i] = &per_cpu(global_trace_cpu, i);
                max_tr.data[i] = &per_cpu(max_data, i);
 
@@ -2608,6 +2746,9 @@ __init static int tracer_alloc_buffers(void)
                /* use the LRU flag to differentiate the two buffers */
                ClearPageLRU(page);
 
+               data->lock = (raw_spinlock_t)__RAW_SPIN_LOCK_UNLOCKED;
+               max_tr.data[i]->lock = (raw_spinlock_t)__RAW_SPIN_LOCK_UNLOCKED;
+
 /* Only allocate if we are actually using the max trace */
 #ifdef CONFIG_TRACER_MAX_TRACE
                array = (void *)__get_free_page(GFP_KERNEL);