Staging: android: add logging driver
authorRobert Love <rlove@google.com>
Sat, 20 Dec 2008 02:02:58 +0000 (18:02 -0800)
committerGreg Kroah-Hartman <gregkh@suse.de>
Tue, 6 Jan 2009 21:52:42 +0000 (13:52 -0800)
Signed-off-by: Robert Love <rlove@google.com>
Signed-off-by: Brian Swetland <swetland@google.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@suse.de>
drivers/staging/android/Kconfig
drivers/staging/android/Makefile
drivers/staging/android/logger.c [new file with mode: 0644]
drivers/staging/android/logger.h [new file with mode: 0644]

index 439a9fd..c889958 100644 (file)
@@ -10,4 +10,8 @@ config ANDROID_BINDER_IPC
        tristate "Android Binder IPC Driver"
        default y
 
+config ANDROID_LOGGER
+       tristate "Android log driver"
+       default n
+
 endmenu
index 57efdce..b354b5d 100644 (file)
@@ -1,2 +1,3 @@
 obj-$(CONFIG_ANDROID)                  += android.o
 obj-$(CONFIG_ANDROID_BINDER_IPC)       += binder.o
+obj-$(CONFIG_ANDROID_LOGGER)           += logger.o
diff --git a/drivers/staging/android/logger.c b/drivers/staging/android/logger.c
new file mode 100644 (file)
index 0000000..ab32003
--- /dev/null
@@ -0,0 +1,607 @@
+/*
+ * drivers/misc/logger.c
+ *
+ * A Logging Subsystem
+ *
+ * Copyright (C) 2007-2008 Google, Inc.
+ *
+ * Robert Love <rlove@google.com>
+ *
+ * This software is licensed under the terms of the GNU General Public
+ * License version 2, as published by the Free Software Foundation, and
+ * may be copied, distributed, and modified under those terms.
+ *
+ * This program is distributed in the hope that it will be useful,
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
+ * GNU General Public License for more details.
+ */
+
+#include <linux/module.h>
+#include <linux/fs.h>
+#include <linux/miscdevice.h>
+#include <linux/uaccess.h>
+#include <linux/poll.h>
+#include <linux/time.h>
+#include "logger.h"
+
+#include <asm/ioctls.h>
+
+/*
+ * struct logger_log - represents a specific log, such as 'main' or 'radio'
+ *
+ * This structure lives from module insertion until module removal, so it does
+ * not need additional reference counting. The structure is protected by the
+ * mutex 'mutex'.
+ */
+struct logger_log {
+       unsigned char *         buffer; /* the ring buffer itself */
+       struct miscdevice       misc;   /* misc device representing the log */
+       wait_queue_head_t       wq;     /* wait queue for readers */
+       struct list_head        readers; /* this log's readers */
+       struct mutex            mutex;  /* mutex protecting buffer */
+       size_t                  w_off;  /* current write head offset */
+       size_t                  head;   /* new readers start here */
+       size_t                  size;   /* size of the log */
+};
+
+/*
+ * struct logger_reader - a logging device open for reading
+ *
+ * This object lives from open to release, so we don't need additional
+ * reference counting. The structure is protected by log->mutex.
+ */
+struct logger_reader {
+       struct logger_log *     log;    /* associated log */
+       struct list_head        list;   /* entry in logger_log's list */
+       size_t                  r_off;  /* current read head offset */
+};
+
+/* logger_offset - returns index 'n' into the log via (optimized) modulus */
+#define logger_offset(n)       ((n) & (log->size - 1))
+
+/*
+ * file_get_log - Given a file structure, return the associated log
+ *
+ * This isn't aesthetic. We have several goals:
+ *
+ *     1) Need to quickly obtain the associated log during an I/O operation
+ *     2) Readers need to maintain state (logger_reader)
+ *     3) Writers need to be very fast (open() should be a near no-op)
+ *
+ * In the reader case, we can trivially go file->logger_reader->logger_log.
+ * For a writer, we don't want to maintain a logger_reader, so we just go
+ * file->logger_log. Thus what file->private_data points at depends on whether
+ * or not the file was opened for reading. This function hides that dirtiness.
+ */
+static inline struct logger_log * file_get_log(struct file *file)
+{
+       if (file->f_mode & FMODE_READ) {
+               struct logger_reader *reader = file->private_data;
+               return reader->log;
+       } else
+               return file->private_data;
+}
+
+/*
+ * get_entry_len - Grabs the length of the payload of the next entry starting
+ * from 'off'.
+ *
+ * Caller needs to hold log->mutex.
+ */
+static __u32 get_entry_len(struct logger_log *log, size_t off)
+{
+       __u16 val;
+
+       switch (log->size - off) {
+       case 1:
+               memcpy(&val, log->buffer + off, 1);
+               memcpy(((char *) &val) + 1, log->buffer, 1);
+               break;
+       default:
+               memcpy(&val, log->buffer + off, 2);
+       }
+
+       return sizeof(struct logger_entry) + val;
+}
+
+/*
+ * do_read_log_to_user - reads exactly 'count' bytes from 'log' into the
+ * user-space buffer 'buf'. Returns 'count' on success.
+ *
+ * Caller must hold log->mutex.
+ */
+static ssize_t do_read_log_to_user(struct logger_log *log,
+                                  struct logger_reader *reader,
+                                  char __user *buf,
+                                  size_t count)
+{
+       size_t len;
+
+       /*
+        * We read from the log in two disjoint operations. First, we read from
+        * the current read head offset up to 'count' bytes or to the end of
+        * the log, whichever comes first.
+        */
+       len = min(count, log->size - reader->r_off);
+       if (copy_to_user(buf, log->buffer + reader->r_off, len))
+               return -EFAULT;
+
+       /*
+        * Second, we read any remaining bytes, starting back at the head of
+        * the log.
+        */
+       if (count != len)
+               if (copy_to_user(buf + len, log->buffer, count - len))
+                       return -EFAULT;
+
+       reader->r_off = logger_offset(reader->r_off + count);
+
+       return count;
+}
+
+/*
+ * logger_read - our log's read() method
+ *
+ * Behavior:
+ *
+ *     - O_NONBLOCK works
+ *     - If there are no log entries to read, blocks until log is written to
+ *     - Atomically reads exactly one log entry
+ *
+ * Optimal read size is LOGGER_ENTRY_MAX_LEN. Will set errno to EINVAL if read
+ * buffer is insufficient to hold next entry.
+ */
+static ssize_t logger_read(struct file *file, char __user *buf,
+                          size_t count, loff_t *pos)
+{
+       struct logger_reader *reader = file->private_data;
+       struct logger_log *log = reader->log;
+       ssize_t ret;
+       DEFINE_WAIT(wait);
+
+start:
+       while (1) {
+               prepare_to_wait(&log->wq, &wait, TASK_INTERRUPTIBLE);
+
+               mutex_lock(&log->mutex);
+               ret = (log->w_off == reader->r_off);
+               mutex_unlock(&log->mutex);
+               if (!ret)
+                       break;
+
+               if (file->f_flags & O_NONBLOCK) {
+                       ret = -EAGAIN;
+                       break;
+               }
+
+               if (signal_pending(current)) {
+                       ret = -EINTR;
+                       break;
+               }
+
+               schedule();
+       }
+
+       finish_wait(&log->wq, &wait);
+       if (ret)
+               return ret;
+
+       mutex_lock(&log->mutex);
+
+       /* is there still something to read or did we race? */
+       if (unlikely(log->w_off == reader->r_off)) {
+               mutex_unlock(&log->mutex);
+               goto start;
+       }
+
+       /* get the size of the next entry */
+       ret = get_entry_len(log, reader->r_off);
+       if (count < ret) {
+               ret = -EINVAL;
+               goto out;
+       }
+
+       /* get exactly one entry from the log */
+       ret = do_read_log_to_user(log, reader, buf, ret);
+
+out:
+       mutex_unlock(&log->mutex);
+
+       return ret;
+}
+
+/*
+ * get_next_entry - return the offset of the first valid entry at least 'len'
+ * bytes after 'off'.
+ *
+ * Caller must hold log->mutex.
+ */
+static size_t get_next_entry(struct logger_log *log, size_t off, size_t len)
+{
+       size_t count = 0;
+
+       do {
+               size_t nr = get_entry_len(log, off);
+               off = logger_offset(off + nr);
+               count += nr;
+       } while (count < len);
+
+       return off;
+}
+
+/*
+ * clock_interval - is a < c < b in mod-space? Put another way, does the line
+ * from a to b cross c?
+ */
+static inline int clock_interval(size_t a, size_t b, size_t c)
+{
+       if (b < a) {
+               if (a < c || b >= c)
+                       return 1;
+       } else {
+               if (a < c && b >= c)
+                       return 1;
+       }
+
+       return 0;
+}
+
+/*
+ * fix_up_readers - walk the list of all readers and "fix up" any who were
+ * lapped by the writer; also do the same for the default "start head".
+ * We do this by "pulling forward" the readers and start head to the first
+ * entry after the new write head.
+ *
+ * The caller needs to hold log->mutex.
+ */
+static void fix_up_readers(struct logger_log *log, size_t len)
+{
+       size_t old = log->w_off;
+       size_t new = logger_offset(old + len);
+       struct logger_reader *reader;
+
+       if (clock_interval(old, new, log->head))
+               log->head = get_next_entry(log, log->head, len);
+
+       list_for_each_entry(reader, &log->readers, list)
+               if (clock_interval(old, new, reader->r_off))
+                       reader->r_off = get_next_entry(log, reader->r_off, len);
+}
+
+/*
+ * do_write_log - writes 'len' bytes from 'buf' to 'log'
+ *
+ * The caller needs to hold log->mutex.
+ */
+static void do_write_log(struct logger_log *log, const void *buf, size_t count)
+{
+       size_t len;
+
+       len = min(count, log->size - log->w_off);
+       memcpy(log->buffer + log->w_off, buf, len);
+
+       if (count != len)
+               memcpy(log->buffer, buf + len, count - len);
+
+       log->w_off = logger_offset(log->w_off + count);
+
+}
+
+/*
+ * do_write_log_user - writes 'len' bytes from the user-space buffer 'buf' to
+ * the log 'log'
+ *
+ * The caller needs to hold log->mutex.
+ *
+ * Returns 'count' on success, negative error code on failure.
+ */
+static ssize_t do_write_log_from_user(struct logger_log *log,
+                                     const void __user *buf, size_t count)
+{
+       size_t len;
+
+       len = min(count, log->size - log->w_off);
+       if (len && copy_from_user(log->buffer + log->w_off, buf, len))
+               return -EFAULT;
+
+       if (count != len)
+               if (copy_from_user(log->buffer, buf + len, count - len))
+                       return -EFAULT;
+
+       log->w_off = logger_offset(log->w_off + count);
+
+       return count;
+}
+
+/*
+ * logger_aio_write - our write method, implementing support for write(),
+ * writev(), and aio_write(). Writes are our fast path, and we try to optimize
+ * them above all else.
+ */
+ssize_t logger_aio_write(struct kiocb *iocb, const struct iovec *iov,
+                        unsigned long nr_segs, loff_t ppos)
+{
+       struct logger_log *log = file_get_log(iocb->ki_filp);
+       size_t orig = log->w_off;
+       struct logger_entry header;
+       struct timespec now;
+       ssize_t ret = 0;
+
+       now = current_kernel_time();
+
+       header.pid = current->tgid;
+       header.tid = current->pid;
+       header.sec = now.tv_sec;
+       header.nsec = now.tv_nsec;
+       header.len = min_t(size_t, iocb->ki_left, LOGGER_ENTRY_MAX_PAYLOAD);
+
+       /* null writes succeed, return zero */
+       if (unlikely(!header.len))
+               return 0;
+
+       mutex_lock(&log->mutex);
+
+       /*
+        * Fix up any readers, pulling them forward to the first readable
+        * entry after (what will be) the new write offset. We do this now
+        * because if we partially fail, we can end up with clobbered log
+        * entries that encroach on readable buffer.
+        */
+       fix_up_readers(log, sizeof(struct logger_entry) + header.len);
+
+       do_write_log(log, &header, sizeof(struct logger_entry));
+
+       while (nr_segs-- > 0) {
+               size_t len;
+               ssize_t nr;
+
+               /* figure out how much of this vector we can keep */
+               len = min_t(size_t, iov->iov_len, header.len - ret);
+
+               /* write out this segment's payload */
+               nr = do_write_log_from_user(log, iov->iov_base, len);
+               if (unlikely(nr < 0)) {
+                       log->w_off = orig;
+                       mutex_unlock(&log->mutex);
+                       return nr;
+               }
+
+               iov++;
+               ret += nr;
+       }
+
+       mutex_unlock(&log->mutex);
+
+       /* wake up any blocked readers */
+       wake_up_interruptible(&log->wq);
+
+       return ret;
+}
+
+static struct logger_log * get_log_from_minor(int);
+
+/*
+ * logger_open - the log's open() file operation
+ *
+ * Note how near a no-op this is in the write-only case. Keep it that way!
+ */
+static int logger_open(struct inode *inode, struct file *file)
+{
+       struct logger_log *log;
+       int ret;
+
+       ret = nonseekable_open(inode, file);
+       if (ret)
+               return ret;
+
+       log = get_log_from_minor(MINOR(inode->i_rdev));
+       if (!log)
+               return -ENODEV;
+
+       if (file->f_mode & FMODE_READ) {
+               struct logger_reader *reader;
+
+               reader = kmalloc(sizeof(struct logger_reader), GFP_KERNEL);
+               if (!reader)
+                       return -ENOMEM;
+
+               reader->log = log;
+               INIT_LIST_HEAD(&reader->list);
+
+               mutex_lock(&log->mutex);
+               reader->r_off = log->head;
+               list_add_tail(&reader->list, &log->readers);
+               mutex_unlock(&log->mutex);
+
+               file->private_data = reader;
+       } else
+               file->private_data = log;
+
+       return 0;
+}
+
+/*
+ * logger_release - the log's release file operation
+ *
+ * Note this is a total no-op in the write-only case. Keep it that way!
+ */
+static int logger_release(struct inode *ignored, struct file *file)
+{
+       if (file->f_mode & FMODE_READ) {
+               struct logger_reader *reader = file->private_data;
+               list_del(&reader->list);
+               kfree(reader);
+       }
+
+       return 0;
+}
+
+/*
+ * logger_poll - the log's poll file operation, for poll/select/epoll
+ *
+ * Note we always return POLLOUT, because you can always write() to the log.
+ * Note also that, strictly speaking, a return value of POLLIN does not
+ * guarantee that the log is readable without blocking, as there is a small
+ * chance that the writer can lap the reader in the interim between poll()
+ * returning and the read() request.
+ */
+static unsigned int logger_poll(struct file *file, poll_table *wait)
+{
+       struct logger_reader *reader;
+       struct logger_log *log;
+       unsigned int ret = POLLOUT | POLLWRNORM;
+
+       if (!(file->f_mode & FMODE_READ))
+               return ret;
+
+       reader = file->private_data;
+       log = reader->log;
+
+       poll_wait(file, &log->wq, wait);
+
+       mutex_lock(&log->mutex);
+       if (log->w_off != reader->r_off)
+               ret |= POLLIN | POLLRDNORM;
+       mutex_unlock(&log->mutex);
+
+       return ret;
+}
+
+static long logger_ioctl(struct file *file, unsigned int cmd, unsigned long arg)
+{
+       struct logger_log *log = file_get_log(file);
+       struct logger_reader *reader;
+       long ret = -ENOTTY;
+
+       mutex_lock(&log->mutex);
+
+       switch (cmd) {
+       case LOGGER_GET_LOG_BUF_SIZE:
+               ret = log->size;
+               break;
+       case LOGGER_GET_LOG_LEN:
+               if (!(file->f_mode & FMODE_READ)) {
+                       ret = -EBADF;
+                       break;
+               }
+               reader = file->private_data;
+               if (log->w_off >= reader->r_off)
+                       ret = log->w_off - reader->r_off;
+               else
+                       ret = (log->size - reader->r_off) + log->w_off;
+               break;
+       case LOGGER_GET_NEXT_ENTRY_LEN:
+               if (!(file->f_mode & FMODE_READ)) {
+                       ret = -EBADF;
+                       break;
+               }
+               reader = file->private_data;
+               if (log->w_off != reader->r_off)
+                       ret = get_entry_len(log, reader->r_off);
+               else
+                       ret = 0;
+               break;
+       case LOGGER_FLUSH_LOG:
+               if (!(file->f_mode & FMODE_WRITE)) {
+                       ret = -EBADF;
+                       break;
+               }
+               list_for_each_entry(reader, &log->readers, list)
+                       reader->r_off = log->w_off;
+               log->head = log->w_off;
+               ret = 0;
+               break;
+       }
+
+       mutex_unlock(&log->mutex);
+
+       return ret;
+}
+
+static struct file_operations logger_fops = {
+       .owner = THIS_MODULE,
+       .read = logger_read,
+       .aio_write = logger_aio_write,
+       .poll = logger_poll,
+       .unlocked_ioctl = logger_ioctl,
+       .compat_ioctl = logger_ioctl,
+       .open = logger_open,
+       .release = logger_release,
+};
+
+/*
+ * Defines a log structure with name 'NAME' and a size of 'SIZE' bytes, which
+ * must be a power of two, greater than LOGGER_ENTRY_MAX_LEN, and less than
+ * LONG_MAX minus LOGGER_ENTRY_MAX_LEN.
+ */
+#define DEFINE_LOGGER_DEVICE(VAR, NAME, SIZE) \
+static unsigned char _buf_ ## VAR[SIZE]; \
+static struct logger_log VAR = { \
+       .buffer = _buf_ ## VAR, \
+       .misc = { \
+               .minor = MISC_DYNAMIC_MINOR, \
+               .name = NAME, \
+               .fops = &logger_fops, \
+               .parent = NULL, \
+       }, \
+       .wq = __WAIT_QUEUE_HEAD_INITIALIZER(VAR .wq), \
+       .readers = LIST_HEAD_INIT(VAR .readers), \
+       .mutex = __MUTEX_INITIALIZER(VAR .mutex), \
+       .w_off = 0, \
+       .head = 0, \
+       .size = SIZE, \
+};
+
+DEFINE_LOGGER_DEVICE(log_main, LOGGER_LOG_MAIN, 64*1024)
+DEFINE_LOGGER_DEVICE(log_events, LOGGER_LOG_EVENTS, 256*1024)
+DEFINE_LOGGER_DEVICE(log_radio, LOGGER_LOG_RADIO, 64*1024)
+
+static struct logger_log * get_log_from_minor(int minor)
+{
+       if (log_main.misc.minor == minor)
+               return &log_main;
+       if (log_events.misc.minor == minor)
+               return &log_events;
+       if (log_radio.misc.minor == minor)
+               return &log_radio;
+       return NULL;
+}
+
+static int __init init_log(struct logger_log *log)
+{
+       int ret;
+
+       ret = misc_register(&log->misc);
+       if (unlikely(ret)) {
+               printk(KERN_ERR "logger: failed to register misc "
+                      "device for log '%s'!\n", log->misc.name);
+               return ret;
+       }
+
+       printk(KERN_INFO "logger: created %luK log '%s'\n",
+              (unsigned long) log->size >> 10, log->misc.name);
+
+       return 0;
+}
+
+static int __init logger_init(void)
+{
+       int ret;
+
+       ret = init_log(&log_main);
+       if (unlikely(ret))
+               goto out;
+
+       ret = init_log(&log_events);
+       if (unlikely(ret))
+               goto out;
+
+       ret = init_log(&log_radio);
+       if (unlikely(ret))
+               goto out;
+
+out:
+       return ret;
+}
+device_initcall(logger_init);
diff --git a/drivers/staging/android/logger.h b/drivers/staging/android/logger.h
new file mode 100644 (file)
index 0000000..a562434
--- /dev/null
@@ -0,0 +1,48 @@
+/* include/linux/logger.h
+ *
+ * Copyright (C) 2007-2008 Google, Inc.
+ * Author: Robert Love <rlove@android.com>
+ *
+ * This software is licensed under the terms of the GNU General Public
+ * License version 2, as published by the Free Software Foundation, and
+ * may be copied, distributed, and modified under those terms.
+ *
+ * This program is distributed in the hope that it will be useful,
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
+ * GNU General Public License for more details.
+ *
+ */
+
+#ifndef _LINUX_LOGGER_H
+#define _LINUX_LOGGER_H
+
+#include <linux/types.h>
+#include <linux/ioctl.h>
+
+struct logger_entry {
+       __u16           len;    /* length of the payload */
+       __u16           __pad;  /* no matter what, we get 2 bytes of padding */
+       __s32           pid;    /* generating process's pid */
+       __s32           tid;    /* generating process's tid */
+       __s32           sec;    /* seconds since Epoch */
+       __s32           nsec;   /* nanoseconds */
+       char            msg[0]; /* the entry's payload */
+};
+
+#define LOGGER_LOG_RADIO       "log_radio"     /* radio-related messages */
+#define LOGGER_LOG_EVENTS      "log_events"    /* system/hardware events */
+#define LOGGER_LOG_MAIN                "log_main"      /* everything else */
+
+#define LOGGER_ENTRY_MAX_LEN           (4*1024)
+#define LOGGER_ENTRY_MAX_PAYLOAD       \
+       (LOGGER_ENTRY_MAX_LEN - sizeof(struct logger_entry))
+
+#define __LOGGERIO     0xAE
+
+#define LOGGER_GET_LOG_BUF_SIZE                _IO(__LOGGERIO, 1) /* size of log */
+#define LOGGER_GET_LOG_LEN             _IO(__LOGGERIO, 2) /* used log len */
+#define LOGGER_GET_NEXT_ENTRY_LEN      _IO(__LOGGERIO, 3) /* next entry len */
+#define LOGGER_FLUSH_LOG               _IO(__LOGGERIO, 4) /* flush log */
+
+#endif /* _LINUX_LOGGER_H */