mtd: Standardising prints in onenand_base.c
authorAmul Kumar Saha <amul.saha@samsung.com>
Fri, 2 Oct 2009 11:29:11 +0000 (16:59 +0530)
committerDavid Woodhouse <David.Woodhouse@intel.com>
Mon, 5 Oct 2009 06:53:39 +0000 (07:53 +0100)
This patch resolves all the prints present in onenand_base.c
Primarily, it replaces the hard-coded function names in the prints,
and makes use of __func__.

Signed-off-by: Amul Kumar Saha <amul.saha@samsung.com>
Signed-off-by: David Woodhouse <David.Woodhouse@intel.com>
drivers/mtd/onenand/onenand_base.c

index ff66e43..8935e63 100644 (file)
@@ -500,25 +500,28 @@ static int onenand_wait(struct mtd_info *mtd, int state)
                int ecc = onenand_read_ecc(this);
                if (ecc) {
                        if (ecc & ONENAND_ECC_2BIT_ALL) {
-                               printk(KERN_ERR "onenand_wait: ECC error = 0x%04x\n", ecc);
+                               printk(KERN_ERR "%s: ECC error = 0x%04x\n",
+                                       __func__, ecc);
                                mtd->ecc_stats.failed++;
                                return -EBADMSG;
                        } else if (ecc & ONENAND_ECC_1BIT_ALL) {
-                               printk(KERN_DEBUG "onenand_wait: correctable ECC error = 0x%04x\n", ecc);
+                               printk(KERN_DEBUG "%s: correctable ECC error = 0x%04x\n",
+                                       __func__, ecc);
                                mtd->ecc_stats.corrected++;
                        }
                }
        } else if (state == FL_READING) {
-               printk(KERN_ERR "onenand_wait: read timeout! ctrl=0x%04x intr=0x%04x\n", ctrl, interrupt);
+               printk(KERN_ERR "%s: read timeout! ctrl=0x%04x intr=0x%04x\n",
+                       __func__, ctrl, interrupt);
                return -EIO;
        }
 
        /* If there's controller error, it's a real error */
        if (ctrl & ONENAND_CTRL_ERROR) {
-               printk(KERN_ERR "onenand_wait: controller error = 0x%04x\n",
-                       ctrl);
+               printk(KERN_ERR "%s: controller error = 0x%04x\n",
+                       __func__, ctrl);
                if (ctrl & ONENAND_CTRL_LOCK)
-                       printk(KERN_ERR "onenand_wait: it's locked error.\n");
+                       printk(KERN_ERR "%s: it's locked error.\n", __func__);
                return -EIO;
        }
 
@@ -1015,7 +1018,8 @@ static int onenand_recover_lsb(struct mtd_info *mtd, loff_t addr, int status)
        /* We are attempting to reread, so decrement stats.failed
         * which was incremented by onenand_wait due to read failure
         */
-       printk(KERN_INFO "onenand_recover_lsb: Attempting to recover from uncorrectable read\n");
+       printk(KERN_INFO "%s: Attempting to recover from uncorrectable read\n",
+               __func__);
        mtd->ecc_stats.failed--;
 
        /* Issue the LSB page recovery command */
@@ -1046,7 +1050,8 @@ static int onenand_mlc_read_ops_nolock(struct mtd_info *mtd, loff_t from,
        int ret = 0;
        int writesize = this->writesize;
 
-       DEBUG(MTD_DEBUG_LEVEL3, "onenand_mlc_read_ops_nolock: from = 0x%08x, len = %i\n", (unsigned int) from, (int) len);
+       DEBUG(MTD_DEBUG_LEVEL3, "%s: from = 0x%08x, len = %i\n",
+               (unsigned int) from, (int) len);
 
        if (ops->mode == MTD_OOB_AUTO)
                oobsize = this->ecclayout->oobavail;
@@ -1057,7 +1062,8 @@ static int onenand_mlc_read_ops_nolock(struct mtd_info *mtd, loff_t from,
 
        /* Do not allow reads past end of device */
        if (from + len > mtd->size) {
-               printk(KERN_ERR "onenand_mlc_read_ops_nolock: Attempt read beyond end of device\n");
+               printk(KERN_ERR "%s: Attempt read beyond end of device\n",
+                       __func__);
                ops->retlen = 0;
                ops->oobretlen = 0;
                return -EINVAL;
@@ -1146,7 +1152,8 @@ static int onenand_read_ops_nolock(struct mtd_info *mtd, loff_t from,
        int ret = 0, boundary = 0;
        int writesize = this->writesize;
 
-       DEBUG(MTD_DEBUG_LEVEL3, "onenand_read_ops_nolock: from = 0x%08x, len = %i\n", (unsigned int) from, (int) len);
+       DEBUG(MTD_DEBUG_LEVEL3, "%s: from = 0x%08x, len = %i\n",
+                       __func__, (unsigned int) from, (int) len);
 
        if (ops->mode == MTD_OOB_AUTO)
                oobsize = this->ecclayout->oobavail;
@@ -1157,7 +1164,8 @@ static int onenand_read_ops_nolock(struct mtd_info *mtd, loff_t from,
 
        /* Do not allow reads past end of device */
        if ((from + len) > mtd->size) {
-               printk(KERN_ERR "onenand_read_ops_nolock: Attempt read beyond end of device\n");
+               printk(KERN_ERR "%s: Attempt read beyond end of device\n",
+                       __func__);
                ops->retlen = 0;
                ops->oobretlen = 0;
                return -EINVAL;
@@ -1275,7 +1283,8 @@ static int onenand_read_oob_nolock(struct mtd_info *mtd, loff_t from,
 
        from += ops->ooboffs;
 
-       DEBUG(MTD_DEBUG_LEVEL3, "onenand_read_oob_nolock: from = 0x%08x, len = %i\n", (unsigned int) from, (int) len);
+       DEBUG(MTD_DEBUG_LEVEL3, "%s: from = 0x%08x, len = %i\n",
+               __func__, (unsigned int) from, (int) len);
 
        /* Initialize return length value */
        ops->oobretlen = 0;
@@ -1288,7 +1297,8 @@ static int onenand_read_oob_nolock(struct mtd_info *mtd, loff_t from,
        column = from & (mtd->oobsize - 1);
 
        if (unlikely(column >= oobsize)) {
-               printk(KERN_ERR "onenand_read_oob_nolock: Attempted to start read outside oob\n");
+               printk(KERN_ERR "%s: Attempted to start read outside oob\n",
+                       __func__);
                return -EINVAL;
        }
 
@@ -1296,7 +1306,8 @@ static int onenand_read_oob_nolock(struct mtd_info *mtd, loff_t from,
        if (unlikely(from >= mtd->size ||
                     column + len > ((mtd->size >> this->page_shift) -
                                     (from >> this->page_shift)) * oobsize)) {
-               printk(KERN_ERR "onenand_read_oob_nolock: Attempted to read beyond end of device\n");
+               printk(KERN_ERR "%s: Attempted to read beyond end of device\n",
+                       __func__);
                return -EINVAL;
        }
 
@@ -1319,7 +1330,8 @@ static int onenand_read_oob_nolock(struct mtd_info *mtd, loff_t from,
                        ret = onenand_recover_lsb(mtd, from, ret);
 
                if (ret && ret != -EBADMSG) {
-                       printk(KERN_ERR "onenand_read_oob_nolock: read failed = 0x%x\n", ret);
+                       printk(KERN_ERR "%s: read failed = 0x%x\n",
+                               __func__, ret);
                        break;
                }
 
@@ -1450,20 +1462,21 @@ static int onenand_bbt_wait(struct mtd_info *mtd, int state)
        if (interrupt & ONENAND_INT_READ) {
                int ecc = onenand_read_ecc(this);
                if (ecc & ONENAND_ECC_2BIT_ALL) {
-                       printk(KERN_INFO "onenand_bbt_wait: ecc error = 0x%04x"
-                               ", controller error 0x%04x\n", ecc, ctrl);
+                       printk(KERN_WARNING "%s: ecc error = 0x%04x, "
+                               "controller error 0x%04x\n",
+                               __func__, ecc, ctrl);
                        return ONENAND_BBT_READ_ECC_ERROR;
                }
        } else {
-               printk(KERN_ERR "onenand_bbt_wait: read timeout!"
-                       "ctrl=0x%04x intr=0x%04x\n", ctrl, interrupt);
+               printk(KERN_ERR "%s: read timeout! ctrl=0x%04x intr=0x%04x\n",
+                       __func__, ctrl, interrupt);
                return ONENAND_BBT_READ_FATAL_ERROR;
        }
 
        /* Initial bad block case: 0x2400 or 0x0400 */
        if (ctrl & ONENAND_CTRL_ERROR) {
-               printk(KERN_DEBUG "onenand_bbt_wait: "
-                       "controller error = 0x%04x\n", ctrl);
+               printk(KERN_DEBUG "%s: controller error = 0x%04x\n",
+                       __func__, ctrl);
                return ONENAND_BBT_READ_ERROR;
        }
 
@@ -1487,14 +1500,16 @@ int onenand_bbt_read_oob(struct mtd_info *mtd, loff_t from,
        size_t len = ops->ooblen;
        u_char *buf = ops->oobbuf;
 
-       DEBUG(MTD_DEBUG_LEVEL3, "onenand_bbt_read_oob: from = 0x%08x, len = %zi\n", (unsigned int) from, len);
+       DEBUG(MTD_DEBUG_LEVEL3, "%s: from = 0x%08x, len = %zi\n",
+               __func__, (unsigned int) from, len);
 
        /* Initialize return value */
        ops->oobretlen = 0;
 
        /* Do not allow reads past end of device */
        if (unlikely((from + len) > mtd->size)) {
-               printk(KERN_ERR "onenand_bbt_read_oob: Attempt read beyond end of device\n");
+               printk(KERN_ERR "%s: Attempt read beyond end of device\n",
+                       __func__);
                return ONENAND_BBT_READ_FATAL_ERROR;
        }
 
@@ -1661,21 +1676,23 @@ static int onenand_panic_write(struct mtd_info *mtd, loff_t to, size_t len,
        /* Wait for any existing operation to clear */
        onenand_panic_wait(mtd);
 
-       DEBUG(MTD_DEBUG_LEVEL3, "onenand_panic_write: to = 0x%08x, len = %i\n",
-             (unsigned int) to, (int) len);
+       DEBUG(MTD_DEBUG_LEVEL3, "%s: to = 0x%08x, len = %i\n",
+               __func__, (unsigned int) to, (int) len);
 
        /* Initialize retlen, in case of early exit */
        *retlen = 0;
 
        /* Do not allow writes past end of device */
        if (unlikely((to + len) > mtd->size)) {
-               printk(KERN_ERR "onenand_panic_write: Attempt write to past end of device\n");
+               printk(KERN_ERR "%s: Attempt write to past end of device\n",
+                       __func__);
                return -EINVAL;
        }
 
        /* Reject writes, which are not page aligned */
         if (unlikely(NOTALIGNED(to) || NOTALIGNED(len))) {
-                printk(KERN_ERR "onenand_panic_write: Attempt to write not page aligned data\n");
+               printk(KERN_ERR "%s: Attempt to write not page aligned data\n",
+                       __func__);
                 return -EINVAL;
         }
 
@@ -1711,7 +1728,7 @@ static int onenand_panic_write(struct mtd_info *mtd, loff_t to, size_t len,
                }
 
                if (ret) {
-                       printk(KERN_ERR "onenand_panic_write: write failed %d\n", ret);
+                       printk(KERN_ERR "%s: write failed %d\n", __func__, ret);
                        break;
                }
 
@@ -1792,7 +1809,8 @@ static int onenand_write_ops_nolock(struct mtd_info *mtd, loff_t to,
        u_char *oobbuf;
        int ret = 0;
 
-       DEBUG(MTD_DEBUG_LEVEL3, "onenand_write_ops_nolock: to = 0x%08x, len = %i\n", (unsigned int) to, (int) len);
+       DEBUG(MTD_DEBUG_LEVEL3, "%s: to = 0x%08x, len = %i\n",
+               __func__, (unsigned int) to, (int) len);
 
        /* Initialize retlen, in case of early exit */
        ops->retlen = 0;
@@ -1800,13 +1818,15 @@ static int onenand_write_ops_nolock(struct mtd_info *mtd, loff_t to,
 
        /* Do not allow writes past end of device */
        if (unlikely((to + len) > mtd->size)) {
-               printk(KERN_ERR "onenand_write_ops_nolock: Attempt write to past end of device\n");
+               printk(KERN_ERR "%s: Attempt write to past end of device\n",
+                       __func__);
                return -EINVAL;
        }
 
        /* Reject writes, which are not page aligned */
         if (unlikely(NOTALIGNED(to) || NOTALIGNED(len))) {
-                printk(KERN_ERR "onenand_write_ops_nolock: Attempt to write not page aligned data\n");
+               printk(KERN_ERR "%s: Attempt to write not page aligned data\n",
+                       __func__);
                 return -EINVAL;
         }
 
@@ -1879,7 +1899,8 @@ static int onenand_write_ops_nolock(struct mtd_info *mtd, loff_t to,
                        onenand_update_bufferram(mtd, prev, !ret && !prev_subpage);
                        if (ret) {
                                written -= prevlen;
-                               printk(KERN_ERR "onenand_write_ops_nolock: write failed %d\n", ret);
+                               printk(KERN_ERR "%s: write failed %d\n",
+                                       __func__, ret);
                                break;
                        }
 
@@ -1887,7 +1908,8 @@ static int onenand_write_ops_nolock(struct mtd_info *mtd, loff_t to,
                                /* Only check verify write turn on */
                                ret = onenand_verify(mtd, buf - len, to - len, len);
                                if (ret)
-                                       printk(KERN_ERR "onenand_write_ops_nolock: verify failed %d\n", ret);
+                                       printk(KERN_ERR "%s: verify failed %d\n",
+                                               __func__, ret);
                                break;
                        }
 
@@ -1905,14 +1927,16 @@ static int onenand_write_ops_nolock(struct mtd_info *mtd, loff_t to,
                        /* In partial page write we don't update bufferram */
                        onenand_update_bufferram(mtd, to, !ret && !subpage);
                        if (ret) {
-                               printk(KERN_ERR "onenand_write_ops_nolock: write failed %d\n", ret);
+                               printk(KERN_ERR "%s: write failed %d\n",
+                                       __func__, ret);
                                break;
                        }
 
                        /* Only check verify write turn on */
                        ret = onenand_verify(mtd, buf, to, thislen);
                        if (ret) {
-                               printk(KERN_ERR "onenand_write_ops_nolock: verify failed %d\n", ret);
+                               printk(KERN_ERR "%s: verify failed %d\n",
+                                       __func__, ret);
                                break;
                        }
 
@@ -1968,7 +1992,8 @@ static int onenand_write_oob_nolock(struct mtd_info *mtd, loff_t to,
 
        to += ops->ooboffs;
 
-       DEBUG(MTD_DEBUG_LEVEL3, "onenand_write_oob_nolock: to = 0x%08x, len = %i\n", (unsigned int) to, (int) len);
+       DEBUG(MTD_DEBUG_LEVEL3, "%s: to = 0x%08x, len = %i\n",
+               __func__, (unsigned int) to, (int) len);
 
        /* Initialize retlen, in case of early exit */
        ops->oobretlen = 0;
@@ -1981,14 +2006,15 @@ static int onenand_write_oob_nolock(struct mtd_info *mtd, loff_t to,
        column = to & (mtd->oobsize - 1);
 
        if (unlikely(column >= oobsize)) {
-               printk(KERN_ERR "onenand_write_oob_nolock: Attempted to start write outside oob\n");
+               printk(KERN_ERR "%s: Attempted to start write outside oob\n",
+                       __func__);
                return -EINVAL;
        }
 
        /* For compatibility with NAND: Do not allow write past end of page */
        if (unlikely(column + len > oobsize)) {
-               printk(KERN_ERR "onenand_write_oob_nolock: "
-                     "Attempt to write past end of page\n");
+               printk(KERN_ERR "%s: Attempt to write past end of page\n",
+                       __func__);
                return -EINVAL;
        }
 
@@ -1996,7 +2022,8 @@ static int onenand_write_oob_nolock(struct mtd_info *mtd, loff_t to,
        if (unlikely(to >= mtd->size ||
                     column + len > ((mtd->size >> this->page_shift) -
                                     (to >> this->page_shift)) * oobsize)) {
-               printk(KERN_ERR "onenand_write_oob_nolock: Attempted to write past end of device\n");
+               printk(KERN_ERR "%s: Attempted to write past end of device\n"
+                       __func__);
                return -EINVAL;
        }
 
@@ -2038,13 +2065,14 @@ static int onenand_write_oob_nolock(struct mtd_info *mtd, loff_t to,
 
                ret = this->wait(mtd, FL_WRITING);
                if (ret) {
-                       printk(KERN_ERR "onenand_write_oob_nolock: write failed %d\n", ret);
+                       printk(KERN_ERR "%s: write failed %d\n", __func__, ret);
                        break;
                }
 
                ret = onenand_verify_oob(mtd, oobbuf, to);
                if (ret) {
-                       printk(KERN_ERR "onenand_write_oob_nolock: verify failed %d\n", ret);
+                       printk(KERN_ERR "%s: verify failed %d\n",
+                               __func__, ret);
                        break;
                }
 
@@ -2161,7 +2189,7 @@ static int onenand_erase(struct mtd_info *mtd, struct erase_info *instr)
 
        /* Do not allow erase past end of device */
        if (unlikely((len + addr) > mtd->size)) {
-               printk(KERN_ERR "onenand_erase: Erase past end of device\n");
+               printk(KERN_ERR "%s: Erase past end of device\n", __func__);
                return -EINVAL;
        }
 
@@ -2177,7 +2205,7 @@ static int onenand_erase(struct mtd_info *mtd, struct erase_info *instr)
                 * Erase region's start offset is always block start address.
                 */
                if (unlikely((addr - region->offset) & (block_size - 1))) {
-                       printk(KERN_ERR "onenand_erase: Unaligned address\n");
+                       printk(KERN_ERR "%s: Unaligned address\n", __func__);
                        return -EINVAL;
                }
        } else {
@@ -2185,14 +2213,14 @@ static int onenand_erase(struct mtd_info *mtd, struct erase_info *instr)
 
                /* Start address must align on block boundary */
                if (unlikely(addr & (block_size - 1))) {
-                       printk(KERN_ERR "onenand_erase: Unaligned address\n");
+                       printk(KERN_ERR "%s: Unaligned address\n", __func__);
                        return -EINVAL;
                }
        }
 
        /* Length must align on block boundary */
        if (unlikely(len & (block_size - 1))) {
-               printk(KERN_ERR "onenand_erase: Length not block aligned\n");
+               printk(KERN_ERR "%s: Length not block aligned\n", __func__);
                return -EINVAL;
        }
 
@@ -2209,7 +2237,9 @@ static int onenand_erase(struct mtd_info *mtd, struct erase_info *instr)
 
                /* Check if we have a bad block, we do not erase bad blocks */
                if (onenand_block_isbad_nolock(mtd, addr, 0)) {
-                       printk (KERN_WARNING "onenand_erase: attempt to erase a bad block at addr 0x%012llx\n", (unsigned long long) addr);
+                       printk(KERN_WARNING "%s: attempt to erase a bad block "
+                                       "at addr 0x%012llx\n",
+                                       __func__, (unsigned long long) addr);
                        instr->state = MTD_ERASE_FAILED;
                        goto erase_exit;
                }
@@ -2221,8 +2251,8 @@ static int onenand_erase(struct mtd_info *mtd, struct erase_info *instr)
                ret = this->wait(mtd, FL_ERASING);
                /* Check, if it is write protected */
                if (ret) {
-                       printk(KERN_ERR "onenand_erase: Failed erase, block %d\n",
-                                                onenand_block(this, addr));
+                       printk(KERN_ERR "%s: Failed erase, block %d\n",
+                               __func__, onenand_block(this, addr));
                        instr->state = MTD_ERASE_FAILED;
                        instr->fail_addr = addr;
                        goto erase_exit;
@@ -2241,7 +2271,8 @@ static int onenand_erase(struct mtd_info *mtd, struct erase_info *instr)
 
                        if (len & (block_size - 1)) {
                                /* FIXME: This should be handled at MTD partitioning level. */
-                               printk(KERN_ERR "onenand_erase: Unaligned address\n");
+                               printk(KERN_ERR "%s: Unaligned address\n",
+                                       __func__);
                                goto erase_exit;
                        }
                }
@@ -2272,7 +2303,7 @@ erase_exit:
  */
 static void onenand_sync(struct mtd_info *mtd)
 {
-       DEBUG(MTD_DEBUG_LEVEL3, "onenand_sync: called\n");
+       DEBUG(MTD_DEBUG_LEVEL3, "%s: called\n", __func__);
 
        /* Grab the lock and see if the device is available */
        onenand_get_device(mtd, FL_SYNCING);
@@ -2406,7 +2437,8 @@ static int onenand_do_lock_cmd(struct mtd_info *mtd, loff_t ofs, size_t len, int
                /* Check lock status */
                status = this->read_word(this->base + ONENAND_REG_WP_STATUS);
                if (!(status & wp_status_mask))
-                       printk(KERN_ERR "wp status = 0x%x\n", status);
+                       printk(KERN_ERR "%s: wp status = 0x%x\n",
+                               __func__, status);
 
                return 0;
        }
@@ -2435,7 +2467,8 @@ static int onenand_do_lock_cmd(struct mtd_info *mtd, loff_t ofs, size_t len, int
                /* Check lock status */
                status = this->read_word(this->base + ONENAND_REG_WP_STATUS);
                if (!(status & wp_status_mask))
-                       printk(KERN_ERR "block = %d, wp status = 0x%x\n", block, status);
+                       printk(KERN_ERR "%s: block = %d, wp status = 0x%x\n",
+                               __func__, block, status);
        }
 
        return 0;
@@ -2502,7 +2535,8 @@ static int onenand_check_lock_status(struct onenand_chip *this)
                /* Check lock status */
                status = this->read_word(this->base + ONENAND_REG_WP_STATUS);
                if (!(status & ONENAND_WP_US)) {
-                       printk(KERN_ERR "block = %d, wp status = 0x%x\n", block, status);
+                       printk(KERN_ERR "%s: block = %d, wp status = 0x%x\n",
+                               __func__, block, status);
                        return 0;
                }
        }
@@ -3172,7 +3206,8 @@ static int flexonenand_check_blocks_erased(struct mtd_info *mtd, int start, int
                                break;
 
                if (i != mtd->oobsize) {
-                       printk(KERN_WARNING "Block %d not erased.\n", block);
+                       printk(KERN_WARNING "%s: Block %d not erased.\n",
+                               __func__, block);
                        return 1;
                }
        }
@@ -3204,8 +3239,8 @@ int flexonenand_set_boundary(struct mtd_info *mtd, int die,
        blksperdie >>= ONENAND_IS_DDP(this) ? 1 : 0;
 
        if (boundary >= blksperdie) {
-               printk(KERN_ERR "flexonenand_set_boundary: Invalid boundary value. "
-                               "Boundary not changed.\n");
+               printk(KERN_ERR "%s: Invalid boundary value. "
+                               "Boundary not changed.\n", __func__);
                return -EINVAL;
        }
 
@@ -3214,7 +3249,8 @@ int flexonenand_set_boundary(struct mtd_info *mtd, int die,
        new = boundary + (die * this->density_mask);
        ret = flexonenand_check_blocks_erased(mtd, min(old, new) + 1, max(old, new));
        if (ret) {
-               printk(KERN_ERR "flexonenand_set_boundary: Please erase blocks before boundary change\n");
+               printk(KERN_ERR "%s: Please erase blocks "
+                               "before boundary change\n", __func__);
                return ret;
        }
 
@@ -3227,12 +3263,12 @@ int flexonenand_set_boundary(struct mtd_info *mtd, int die,
 
        thisboundary = this->read_word(this->base + ONENAND_DATARAM);
        if ((thisboundary >> FLEXONENAND_PI_UNLOCK_SHIFT) != 3) {
-               printk(KERN_ERR "flexonenand_set_boundary: boundary locked\n");
+               printk(KERN_ERR "%s: boundary locked\n", __func__);
                ret = 1;
                goto out;
        }
 
-       printk(KERN_INFO "flexonenand_set_boundary: Changing die %d boundary: %d%s\n",
+       printk(KERN_INFO "Changing die %d boundary: %d%s\n",
                        die, boundary, lock ? "(Locked)" : "(Unlocked)");
 
        addr = die ? this->diesize[0] : 0;
@@ -3243,7 +3279,8 @@ int flexonenand_set_boundary(struct mtd_info *mtd, int die,
        this->command(mtd, ONENAND_CMD_ERASE, addr, 0);
        ret = this->wait(mtd, FL_ERASING);
        if (ret) {
-               printk(KERN_ERR "flexonenand_set_boundary: Failed PI erase for Die %d\n", die);
+               printk(KERN_ERR "%s: flexonenand_set_boundary: "
+                               "Failed PI erase for Die %d\n", __func__, die);
                goto out;
        }
 
@@ -3251,7 +3288,8 @@ int flexonenand_set_boundary(struct mtd_info *mtd, int die,
        this->command(mtd, ONENAND_CMD_PROG, addr, 0);
        ret = this->wait(mtd, FL_WRITING);
        if (ret) {
-               printk(KERN_ERR "flexonenand_set_boundary: Failed PI write for Die %d\n", die);
+               printk(KERN_ERR "%s: Failed PI write for Die %d\n",
+                       __func__, die);
                goto out;
        }
 
@@ -3408,8 +3446,8 @@ static void onenand_resume(struct mtd_info *mtd)
        if (this->state == FL_PM_SUSPENDED)
                onenand_release_device(mtd);
        else
-               printk(KERN_ERR "resume() called for the chip which is not"
-                               "in suspended state\n");
+               printk(KERN_ERR "%s: resume() called for the chip which is not "
+                               "in suspended state\n", __func__);
 }
 
 /**
@@ -3464,7 +3502,8 @@ int onenand_scan(struct mtd_info *mtd, int maxchips)
        if (!this->page_buf) {
                this->page_buf = kzalloc(mtd->writesize, GFP_KERNEL);
                if (!this->page_buf) {
-                       printk(KERN_ERR "onenand_scan(): Can't allocate page_buf\n");
+                       printk(KERN_ERR "%s: Can't allocate page_buf\n",
+                               __func__);
                        return -ENOMEM;
                }
                this->options |= ONENAND_PAGEBUF_ALLOC;
@@ -3472,7 +3511,8 @@ int onenand_scan(struct mtd_info *mtd, int maxchips)
        if (!this->oob_buf) {
                this->oob_buf = kzalloc(mtd->oobsize, GFP_KERNEL);
                if (!this->oob_buf) {
-                       printk(KERN_ERR "onenand_scan(): Can't allocate oob_buf\n");
+                       printk(KERN_ERR "%s: Can't allocate oob_buf\n",
+                               __func__);
                        if (this->options & ONENAND_PAGEBUF_ALLOC) {
                                this->options &= ~ONENAND_PAGEBUF_ALLOC;
                                kfree(this->page_buf);
@@ -3505,8 +3545,8 @@ int onenand_scan(struct mtd_info *mtd, int maxchips)
                break;
 
        default:
-               printk(KERN_WARNING "No OOB scheme defined for oobsize %d\n",
-                       mtd->oobsize);
+               printk(KERN_WARNING "%s: No OOB scheme defined for oobsize %d\n",
+                       __func__, mtd->oobsize);
                mtd->subpage_sft = 0;
                /* To prevent kernel oops */
                this->ecclayout = &onenand_oob_32;