drivers: apanic: Android kernel panic handler.
authorSan Mehat <san@google.com>
Tue, 1 Sep 2009 20:43:01 +0000 (13:43 -0700)
committerColin Cross <ccross@android.com>
Tue, 14 Jun 2011 16:09:18 +0000 (09:09 -0700)
This driver triggers when the kernel panics and attempts to
write critical debug data to the flash.

Signed-off-by: San Mehat <san@google.com>
drivers: apanic: checkpatch fixes

Signed-off-by: San Mehat <san@google.com>
apanic: Fix a few cases of calling non-atomic things from atomic

We need to pay special care to not enrage cond_resched(), and the
base nand bb stuff calls schedule() so thats out.

Signed-off-by: San Mehat <san@google.com>
drivers/misc/Kconfig
drivers/misc/Makefile
drivers/misc/apanic.c [new file with mode: 0644]

index 0da53642a40e3c1f9cee59bf98204615bd5a1671..1c79dbf72b099f5a20f854461ff1d219ba6a37b8 100644 (file)
@@ -512,7 +512,21 @@ config WL127X_RFKILL
        ---help---
         Creates an rfkill entry in sysfs for power control of Bluetooth
         TI wl127x chips.
-        
+
+config APANIC
+       bool "Android kernel panic diagnostics driver"
+       default n
+       ---help---
+        Driver which handles kernel panics and attempts to write
+        critical debugging data to flash.
+
+config APANIC_PLABEL
+       string "Android panic dump flash partition label"
+       default "kpanic"
+       ---help---
+        If your platform uses a different flash partition label for storing
+        crashdumps, enter it here.
+
 source "drivers/misc/c2port/Kconfig"
 source "drivers/misc/eeprom/Kconfig"
 source "drivers/misc/cb710/Kconfig"
index 2b880207dfcfcff4b83045e7693f0874bd04f775..42d698081a5304a0bd80e7a3a6d929ffe4177f46 100644 (file)
@@ -50,3 +50,4 @@ obj-$(CONFIG_AB8500_PWM)      += ab8500-pwm.o
 obj-y                          += lis3lv02d/
 obj-y                          += carma/
 obj-$(CONFIG_WL127X_RFKILL)    += wl127x-rfkill.o
+obj-$(CONFIG_APANIC)           += apanic.o
diff --git a/drivers/misc/apanic.c b/drivers/misc/apanic.c
new file mode 100644 (file)
index 0000000..fde0871
--- /dev/null
@@ -0,0 +1,541 @@
+/* drivers/misc/apanic.c
+ *
+ * Copyright (C) 2009 Google, Inc.
+ * Author: San Mehat <san@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.
+ *
+ */
+
+#include <linux/module.h>
+#include <linux/kernel.h>
+#include <linux/string.h>
+#include <linux/errno.h>
+#include <linux/init.h>
+#include <linux/interrupt.h>
+#include <linux/device.h>
+#include <linux/types.h>
+#include <linux/delay.h>
+#include <linux/sched.h>
+#include <linux/wait.h>
+#include <linux/wakelock.h>
+#include <linux/platform_device.h>
+#include <linux/uaccess.h>
+#include <linux/mtd/mtd.h>
+#include <linux/notifier.h>
+#include <linux/mtd/mtd.h>
+#include <linux/debugfs.h>
+#include <linux/fs.h>
+#include <linux/proc_fs.h>
+#include <linux/mutex.h>
+#include <linux/workqueue.h>
+#include <linux/preempt.h>
+
+struct panic_header {
+       u32 magic;
+#define PANIC_MAGIC 0xdeadf00d
+
+       u32 version;
+#define PHDR_VERSION   0x01
+
+       u32 console_offset;
+       u32 console_length;
+
+       u32 threads_offset;
+       u32 threads_length;
+};
+
+#define CHECK_BB       0
+
+struct apanic_data {
+       struct mtd_info         *mtd;
+       struct panic_header     curr;
+       void                    *bounce;
+       struct proc_dir_entry   *apanic_console;
+       struct proc_dir_entry   *apanic_threads;
+};
+
+static struct apanic_data drv_ctx;
+static struct work_struct proc_removal_work;
+static DEFINE_MUTEX(drv_mutex);
+
+static void apanic_erase_callback(struct erase_info *done)
+{
+       wait_queue_head_t *wait_q = (wait_queue_head_t *) done->priv;
+       wake_up(wait_q);
+}
+
+static int apanic_proc_read(char *buffer, char **start, off_t offset,
+                              int count, int *peof, void *dat)
+{
+       struct apanic_data *ctx = &drv_ctx;
+       size_t file_length;
+       off_t file_offset;
+       unsigned int page_no;
+       off_t page_offset;
+       int rc;
+       size_t len;
+
+       if (!count)
+               return 0;
+
+       mutex_lock(&drv_mutex);
+
+       switch ((int) dat) {
+       case 1: /* apanic_console */
+               file_length = ctx->curr.console_length;
+               file_offset = ctx->curr.console_offset;
+               break;
+       case 2: /* apanic_threads */
+               file_length = ctx->curr.threads_length;
+               file_offset = ctx->curr.threads_offset;
+               break;
+       default:
+               pr_err("Bad dat (%d)\n", (int) dat);
+               mutex_unlock(&drv_mutex);
+               return -EINVAL;
+       }
+
+       if ((offset + count) > file_length) {
+               mutex_unlock(&drv_mutex);
+               return 0;
+       }
+
+       /* We only support reading a maximum of a flash page */
+       if (count > ctx->mtd->writesize)
+               count = ctx->mtd->writesize;
+
+       page_no = (file_offset + offset) / ctx->mtd->writesize;
+       page_offset = (file_offset + offset) % ctx->mtd->writesize;
+
+       rc = ctx->mtd->read(ctx->mtd,
+                           (page_no * ctx->mtd->writesize),
+                           ctx->mtd->writesize,
+                           &len, ctx->bounce);
+
+       if (page_offset)
+               count -= page_offset;
+       memcpy(buffer, ctx->bounce + page_offset, count);
+
+       *start = count;
+
+       if ((offset + count) == file_length)
+               *peof = 1;
+
+       mutex_unlock(&drv_mutex);
+       return count;
+}
+
+static void mtd_panic_erase(void)
+{
+       struct apanic_data *ctx = &drv_ctx;
+       struct erase_info erase;
+       DECLARE_WAITQUEUE(wait, current);
+       wait_queue_head_t wait_q;
+       int rc, i;
+
+       init_waitqueue_head(&wait_q);
+       erase.mtd = ctx->mtd;
+       erase.callback = apanic_erase_callback;
+       erase.len = ctx->mtd->erasesize;
+       erase.priv = (u_long)&wait_q;
+       for (i = 0; i < ctx->mtd->size; i += ctx->mtd->erasesize) {
+               erase.addr = i;
+               set_current_state(TASK_INTERRUPTIBLE);
+               add_wait_queue(&wait_q, &wait);
+
+               rc = ctx->mtd->block_isbad(ctx->mtd, erase.addr);
+               if (rc < 0) {
+                       printk(KERN_ERR
+                              "apanic: Bad block check "
+                              "failed (%d)\n", rc);
+                       goto out;
+               }
+               if (rc) {
+                       printk(KERN_WARNING
+                              "apanic: Skipping erase of bad "
+                              "block @%llx\n", erase.addr);
+                       set_current_state(TASK_RUNNING);
+                       remove_wait_queue(&wait_q, &wait);
+                       continue;
+               }
+
+               rc = ctx->mtd->erase(ctx->mtd, &erase);
+               if (rc) {
+                       set_current_state(TASK_RUNNING);
+                       remove_wait_queue(&wait_q, &wait);
+                       printk(KERN_ERR
+                              "apanic: Erase of 0x%llx, 0x%llx failed\n",
+                              (unsigned long long) erase.addr,
+                              (unsigned long long) erase.len);
+                       if (rc == -EIO) {
+                               if (ctx->mtd->block_markbad(ctx->mtd,
+                                                           erase.addr)) {
+                                       printk(KERN_ERR
+                                              "apanic: Err marking blk bad\n");
+                                       goto out;
+                               }
+                               printk(KERN_INFO
+                                      "apanic: Marked a bad block"
+                                      " @%llx\n", erase.addr);
+                               continue;
+                       }
+                       goto out;
+               }
+               schedule();
+               remove_wait_queue(&wait_q, &wait);
+       }
+       printk(KERN_DEBUG "apanic: %s partition erased\n",
+              CONFIG_APANIC_PLABEL);
+out:
+       return;
+}
+
+static void apanic_remove_proc_work(struct work_struct *work)
+{
+       struct apanic_data *ctx = &drv_ctx;
+
+       mutex_lock(&drv_mutex);
+       mtd_panic_erase();
+       memset(&ctx->curr, 0, sizeof(struct panic_header));
+       if (ctx->apanic_console) {
+               remove_proc_entry("apanic_console", NULL);
+               ctx->apanic_console = NULL;
+       }
+       if (ctx->apanic_threads) {
+               remove_proc_entry("apanic_threads", NULL);
+               ctx->apanic_threads = NULL;
+       }
+       mutex_unlock(&drv_mutex);
+}
+
+static int apanic_proc_write(struct file *file, const char __user *buffer,
+                               unsigned long count, void *data)
+{
+       schedule_work(&proc_removal_work);
+       return count;
+}
+
+static void mtd_panic_notify_add(struct mtd_info *mtd)
+{
+       struct apanic_data *ctx = &drv_ctx;
+       struct panic_header *hdr = ctx->bounce;
+       size_t len;
+       int rc;
+
+       if (strcmp(mtd->name, CONFIG_APANIC_PLABEL))
+               return;
+
+       ctx->mtd = mtd;
+
+       if (mtd->block_isbad(mtd, 0)) {
+               printk(KERN_ERR "apanic: Offset 0 bad block. Boourns!\n");
+               goto out_err;
+       }
+
+       rc = mtd->read(mtd, 0, mtd->writesize, &len, ctx->bounce);
+       if (rc && rc == -EBADMSG) {
+               printk(KERN_WARNING
+                      "apanic: Bad ECC on block 0 (ignored)\n");
+       } else if (rc && rc != -EUCLEAN) {
+               printk(KERN_ERR "apanic: Error reading block 0 (%d)\n", rc);
+               goto out_err;
+       }
+
+       if (len != mtd->writesize) {
+               printk(KERN_ERR "apanic: Bad read size (%d)\n", rc);
+               goto out_err;
+       }
+
+       printk(KERN_INFO "apanic: Bound to mtd partition '%s'\n", mtd->name);
+
+       if (hdr->magic != PANIC_MAGIC) {
+               printk(KERN_INFO "apanic: No panic data available\n");
+               mtd_panic_erase();
+               return;
+       }
+
+       if (hdr->version != PHDR_VERSION) {
+               printk(KERN_INFO "apanic: Version mismatch (%d != %d)\n",
+                      hdr->version, PHDR_VERSION);
+               mtd_panic_erase();
+               return;
+       }
+
+       memcpy(&ctx->curr, hdr, sizeof(struct panic_header));
+
+       printk(KERN_INFO "apanic: c(%u, %u) t(%u, %u)\n",
+              hdr->console_offset, hdr->console_length,
+              hdr->threads_offset, hdr->threads_length);
+
+       if (hdr->console_length) {
+               ctx->apanic_console = create_proc_entry("apanic_console",
+                                                     S_IFREG | S_IRUGO, NULL);
+               if (!ctx->apanic_console)
+                       printk(KERN_ERR "%s: failed creating procfile\n",
+                              __func__);
+               else {
+                       ctx->apanic_console->read_proc = apanic_proc_read;
+                       ctx->apanic_console->write_proc = apanic_proc_write;
+                       ctx->apanic_console->size = hdr->console_length;
+                       ctx->apanic_console->data = (void *) 1;
+               }
+       }
+
+       if (hdr->threads_length) {
+               ctx->apanic_threads = create_proc_entry("apanic_threads",
+                                                      S_IFREG | S_IRUGO, NULL);
+               if (!ctx->apanic_threads)
+                       printk(KERN_ERR "%s: failed creating procfile\n",
+                              __func__);
+               else {
+                       ctx->apanic_threads->read_proc = apanic_proc_read;
+                       ctx->apanic_threads->write_proc = apanic_proc_write;
+                       ctx->apanic_threads->size = hdr->threads_length;
+                       ctx->apanic_threads->data = (void *) 2;
+               }
+       }
+
+       return;
+out_err:
+       ctx->mtd = NULL;
+}
+
+static void mtd_panic_notify_remove(struct mtd_info *mtd)
+{
+       struct apanic_data *ctx = &drv_ctx;
+       if (mtd == ctx->mtd) {
+               ctx->mtd = NULL;
+               printk(KERN_INFO "apanic: Unbound from %s\n", mtd->name);
+       }
+}
+
+static struct mtd_notifier mtd_panic_notifier = {
+       .add    = mtd_panic_notify_add,
+       .remove = mtd_panic_notify_remove,
+};
+
+static int in_panic = 0;
+
+static int apanic_writeflashpage(struct mtd_info *mtd, loff_t to,
+                                const u_char *buf)
+{
+       int rc;
+       size_t wlen;
+       int panic = in_interrupt() | in_atomic();
+
+       if (panic && !mtd->panic_write) {
+               printk(KERN_EMERG "%s: No panic_write available\n", __func__);
+               return 0;
+       } else if (!panic && !mtd->write) {
+               printk(KERN_EMERG "%s: No write available\n", __func__);
+               return 0;
+       }
+
+       if (panic)
+               rc = mtd->panic_write(mtd, to, mtd->writesize, &wlen, buf);
+       else
+               rc = mtd->write(mtd, to, mtd->writesize, &wlen, buf);
+
+       if (rc) {
+               printk(KERN_EMERG
+                      "%s: Error writing data to flash (%d)\n",
+                      __func__, rc);
+               return rc;
+       }
+
+       return wlen;
+}
+
+extern int log_buf_copy(char *dest, int idx, int len);
+extern void log_buf_clear(void);
+
+/*
+ * Writes the contents of the console to the specified offset in flash.
+ * Returns number of bytes written
+ */
+static int apanic_write_console(struct mtd_info *mtd, unsigned int off)
+{
+       struct apanic_data *ctx = &drv_ctx;
+       int saved_oip;
+       int idx = 0;
+       int rc, rc2;
+       unsigned int last_chunk = 0;
+
+       while (!last_chunk) {
+               saved_oip = oops_in_progress;
+               oops_in_progress = 1;
+               rc = log_buf_copy(ctx->bounce, idx, mtd->writesize);
+               if (rc < 0)
+                       break;
+
+               if (rc != mtd->writesize)
+                       last_chunk = rc;
+
+               oops_in_progress = saved_oip;
+               if (rc <= 0)
+                       break;
+               if (rc != mtd->writesize)
+                       memset(ctx->bounce + rc, 0, mtd->writesize - rc);
+#if CHECK_BB
+check_badblock:
+               rc = mtd->block_isbad(mtd, off);
+               if (rc < 0) {
+                       printk(KERN_ERR
+                              "apanic: Bad block check "
+                              "failed (%d)\n", rc);
+               }
+               if (rc) {
+                       printk(KERN_WARNING
+                              "apanic: Skipping over bad "
+                              "block @%x\n", off);
+                       off += mtd->erasesize;
+                       printk("chk %u %llu\n", off, mtd->size);
+                       if (off >= mtd->size) {
+                               printk(KERN_EMERG
+                                      "apanic: Too many bad blocks!\n");
+                                      return -EIO;
+                       }
+                       goto check_badblock;
+               }
+#endif
+
+               rc2 = apanic_writeflashpage(mtd, off, ctx->bounce);
+               if (rc2 <= 0) {
+                       printk(KERN_EMERG
+                              "apanic: Flash write failed (%d)\n", rc2);
+                       return rc2;
+               }
+               if (!last_chunk)
+                       idx += rc2;
+               else
+                       idx += last_chunk;
+               off += rc2;
+       }
+       return idx;
+}
+
+static int apanic(struct notifier_block *this, unsigned long event,
+                       void *ptr)
+{
+       struct apanic_data *ctx = &drv_ctx;
+       struct panic_header *hdr = (struct panic_header *) ctx->bounce;
+       int console_offset = 0;
+       int console_len = 0;
+       int threads_offset = 0;
+       int threads_len = 0;
+       int rc;
+
+       if (in_panic)
+               return NOTIFY_DONE;
+       in_panic = 1;
+#ifdef CONFIG_PREEMPT
+       /* Ensure that cond_resched() won't try to preempt anybody */
+       add_preempt_count(PREEMPT_ACTIVE);
+#endif
+
+       if (!ctx->mtd)
+               goto out;
+
+       if (ctx->curr.magic) {
+               printk(KERN_EMERG "Crash partition in use!\n");
+               goto out;
+       }
+       console_offset = ctx->mtd->writesize;
+
+       /*
+        * Write out the console
+        */
+       console_len = apanic_write_console(ctx->mtd, console_offset);
+       if (console_len < 0) {
+               printk(KERN_EMERG "Error writing console to panic log! (%d)\n",
+                      console_len);
+               console_len = 0;
+       }
+
+       /*
+        * Write out all threads
+        */
+       threads_offset = ALIGN(console_offset + console_len,
+                              ctx->mtd->writesize);
+       if (!threads_offset)
+               threads_offset = ctx->mtd->writesize;
+
+       log_buf_clear();
+       show_state_filter(0);
+       threads_len = apanic_write_console(ctx->mtd, threads_offset);
+       if (threads_len < 0) {
+               printk(KERN_EMERG "Error writing threads to panic log! (%d)\n",
+                      threads_len);
+               threads_len = 0;
+       }
+
+       /*
+        * Finally write the panic header
+        */
+       memset(ctx->bounce, 0, PAGE_SIZE);
+       hdr->magic = PANIC_MAGIC;
+       hdr->version = PHDR_VERSION;
+
+       hdr->console_offset = console_offset;
+       hdr->console_length = console_len;
+
+       hdr->threads_offset = threads_offset;
+       hdr->threads_length = threads_len;
+
+       rc = apanic_writeflashpage(ctx->mtd, 0, ctx->bounce);
+       if (rc <= 0) {
+               printk(KERN_EMERG "apanic: Header write failed (%d)\n",
+                      rc);
+               goto out;
+       }
+
+       printk(KERN_EMERG "apanic: Panic dump sucessfully written to flash\n");
+
+ out:
+#ifdef CONFIG_PREEMPT
+       sub_preempt_count(PREEMPT_ACTIVE);
+#endif
+       in_panic = 0;
+       return NOTIFY_DONE;
+}
+
+static struct notifier_block panic_blk = {
+       .notifier_call  = apanic,
+};
+
+static int panic_dbg_get(void *data, u64 *val)
+{
+       apanic(NULL, 0, NULL);
+       return 0;
+}
+
+static int panic_dbg_set(void *data, u64 val)
+{
+       BUG();
+       return -1;
+}
+
+DEFINE_SIMPLE_ATTRIBUTE(panic_dbg_fops, panic_dbg_get, panic_dbg_set, "%llu\n");
+
+int __init apanic_init(void)
+{
+       register_mtd_user(&mtd_panic_notifier);
+       atomic_notifier_chain_register(&panic_notifier_list, &panic_blk);
+       debugfs_create_file("apanic", 0644, NULL, NULL, &panic_dbg_fops);
+       memset(&drv_ctx, 0, sizeof(drv_ctx));
+       drv_ctx.bounce = (void *) __get_free_page(GFP_KERNEL);
+       INIT_WORK(&proc_removal_work, apanic_remove_proc_work);
+       printk(KERN_INFO "Android kernel panic handler initialized (bind=%s)\n",
+              CONFIG_APANIC_PLABEL);
+       return 0;
+}
+
+module_init(apanic_init);