]> www.pilppa.org Git - linux-2.6-omap-h63xx.git/commitdiff
cfi_cmdset_0001: factorize code to wait for flash status
authorNicolas Pitre <nico@cam.org>
Thu, 30 Mar 2006 14:52:41 +0000 (15:52 +0100)
committerDavid Woodhouse <dwmw2@infradead.org>
Wed, 17 May 2006 00:25:37 +0000 (01:25 +0100)
This allows for much better abstraction and separation of the XIP and
non-XIP cases with their own specific implementations.  This fixes the
case where a timeout was tripped on in the XIP case by the code that
was meant for the non-XIP case only.

This also makes for a nice code reduction.

Signed-off-by: Nicolas Pitre <nico@cam.org>
CC: "Alexey, Korolev" <alexey.korolev@intel.com>
Signed-off-by: David Woodhouse <dwmw2@infradead.org>
drivers/mtd/chips/cfi_cmdset_0001.c

index ab3888f5e4645bf4d264174d4179a637c285b6d7..d0d5e521b564a7387f95f9590d90780d4313aab8 100644 (file)
@@ -399,7 +399,7 @@ struct mtd_info *cfi_cmdset_0001(struct map_info *map, int primary)
        for (i=0; i< cfi->numchips; i++) {
                cfi->chips[i].word_write_time = 1<<cfi->cfiq->WordWriteTimeoutTyp;
                cfi->chips[i].buffer_write_time = 1<<cfi->cfiq->BufWriteTimeoutTyp;
-               cfi->chips[i].erase_time = 1<<cfi->cfiq->BlockEraseTimeoutTyp;
+               cfi->chips[i].erase_time = 1000<<cfi->cfiq->BlockEraseTimeoutTyp;
                cfi->chips[i].ref_point_counter = 0;
                init_waitqueue_head(&(cfi->chips[i].wq));
        }
@@ -894,26 +894,33 @@ static void __xipram xip_enable(struct map_info *map, struct flchip *chip,
 
 /*
  * When a delay is required for the flash operation to complete, the
- * xip_udelay() function is polling for both the given timeout and pending
- * (but still masked) hardware interrupts.  Whenever there is an interrupt
- * pending then the flash erase or write operation is suspended, array mode
- * restored and interrupts unmasked.  Task scheduling might also happen at that
- * point.  The CPU eventually returns from the interrupt or the call to
- * schedule() and the suspended flash operation is resumed for the remaining
- * of the delay period.
+ * xip_wait_for_operation() function is polling for both the given timeout
+ * and pending (but still masked) hardware interrupts.  Whenever there is an
+ * interrupt pending then the flash erase or write operation is suspended,
+ * array mode restored and interrupts unmasked.  Task scheduling might also
+ * happen at that point.  The CPU eventually returns from the interrupt or
+ * the call to schedule() and the suspended flash operation is resumed for
+ * the remaining of the delay period.
  *
  * Warning: this function _will_ fool interrupt latency tracing tools.
  */
 
-static void __xipram xip_udelay(struct map_info *map, struct flchip *chip,
-                               unsigned long adr, int usec)
+static int __xipram xip_wait_for_operation(
+               struct map_info *map, struct flchip *chip,
+               unsigned long adr, int *chip_op_time )
 {
        struct cfi_private *cfi = map->fldrv_priv;
        struct cfi_pri_intelext *cfip = cfi->cmdset_priv;
        map_word status, OK = CMD(0x80);
-       unsigned long suspended, start = xip_currtime();
+       unsigned long usec, suspended, start, done;
        flstate_t oldstate, newstate;
 
+               start = xip_currtime();
+       usec = *chip_op_time * 8;
+       if (usec == 0)
+               usec = 500000;
+       done = 0;
+
        do {
                cpu_relax();
                if (xip_irqpending() && cfip &&
@@ -930,9 +937,9 @@ static void __xipram xip_udelay(struct map_info *map, struct flchip *chip,
                         * we resume the whole thing at once).  Yes, it
                         * can happen!
                         */
+                       usec -= done;
                        map_write(map, CMD(0xb0), adr);
                        map_write(map, CMD(0x70), adr);
-                       usec -= xip_elapsed_since(start);
                        suspended = xip_currtime();
                        do {
                                if (xip_elapsed_since(suspended) > 100000) {
@@ -942,7 +949,7 @@ static void __xipram xip_udelay(struct map_info *map, struct flchip *chip,
                                         * This is a critical error but there
                                         * is not much we can do here.
                                         */
-                                       return;
+                                       return -EIO;
                                }
                                status = map_read(map, adr);
                        } while (!map_word_andequal(map, status, OK, OK));
@@ -1002,65 +1009,107 @@ static void __xipram xip_udelay(struct map_info *map, struct flchip *chip,
                        xip_cpu_idle();
                }
                status = map_read(map, adr);
+               done = xip_elapsed_since(start);
        } while (!map_word_andequal(map, status, OK, OK)
-                && xip_elapsed_since(start) < usec);
-}
+                && done < usec);
 
-#define UDELAY(map, chip, adr, usec)  xip_udelay(map, chip, adr, usec)
+       return (done >= usec) ? -ETIME : 0;
+}
 
 /*
  * The INVALIDATE_CACHED_RANGE() macro is normally used in parallel while
  * the flash is actively programming or erasing since we have to poll for
  * the operation to complete anyway.  We can't do that in a generic way with
  * a XIP setup so do it before the actual flash operation in this case
- * and stub it out from INVALIDATE_CACHE_UDELAY.
+ * and stub it out from INVAL_CACHE_AND_WAIT.
  */
 #define XIP_INVAL_CACHED_RANGE(map, from, size)  \
        INVALIDATE_CACHED_RANGE(map, from, size)
 
-#define INVALIDATE_CACHE_UDELAY(map, chip, cmd_adr, adr, len, usec)  \
-       UDELAY(map, chip, cmd_adr, usec)
-
-/*
- * Extra notes:
- *
- * Activating this XIP support changes the way the code works a bit.  For
- * example the code to suspend the current process when concurrent access
- * happens is never executed because xip_udelay() will always return with the
- * same chip state as it was entered with.  This is why there is no care for
- * the presence of add_wait_queue() or schedule() calls from within a couple
- * xip_disable()'d  areas of code, like in do_erase_oneblock for example.
- * The queueing and scheduling are always happening within xip_udelay().
- *
- * Similarly, get_chip() and put_chip() just happen to always be executed
- * with chip->state set to FL_READY (or FL_XIP_WHILE_*) where flash state
- * is in array mode, therefore never executing many cases therein and not
- * causing any problem with XIP.
- */
+#define INVAL_CACHE_AND_WAIT(map, chip, cmd_adr, inval_adr, inval_len, p_usec) \
+       xip_wait_for_operation(map, chip, cmd_adr, p_usec)
 
 #else
 
 #define xip_disable(map, chip, adr)
 #define xip_enable(map, chip, adr)
 #define XIP_INVAL_CACHED_RANGE(x...)
+#define INVAL_CACHE_AND_WAIT inval_cache_and_wait_for_operation
+
+static int inval_cache_and_wait_for_operation(
+               struct map_info *map, struct flchip *chip,
+               unsigned long cmd_adr, unsigned long inval_adr, int inval_len,
+               int *chip_op_time )
+{
+       struct cfi_private *cfi = map->fldrv_priv;
+       map_word status, status_OK = CMD(0x80);
+       int z, chip_state = chip->state;
+       unsigned long timeo;
+
+       spin_unlock(chip->mutex);
+       if (inval_len)
+               INVALIDATE_CACHED_RANGE(map, inval_adr, inval_len);
+       if (*chip_op_time)
+               cfi_udelay(*chip_op_time);
+       spin_lock(chip->mutex);
+
+       timeo = *chip_op_time * 8 * HZ / 1000000;
+       if (timeo < HZ/2)
+               timeo = HZ/2;
+       timeo += jiffies;
+
+       z = 0;
+       for (;;) {
+               if (chip->state != chip_state) {
+                       /* Someone's suspended the operation: sleep */
+                       DECLARE_WAITQUEUE(wait, current);
+
+                       set_current_state(TASK_UNINTERRUPTIBLE);
+                       add_wait_queue(&chip->wq, &wait);
+                       spin_unlock(chip->mutex);
+                       schedule();
+                       remove_wait_queue(&chip->wq, &wait);
+                       timeo = jiffies + (HZ / 2); /* FIXME */
+                       spin_lock(chip->mutex);
+                       continue;
+               }
+
+               status = map_read(map, cmd_adr);
+               if (map_word_andequal(map, status, status_OK, status_OK))
+                       break;
 
-#define UDELAY(map, chip, adr, usec)  \
-do {  \
-       spin_unlock(chip->mutex);  \
-       cfi_udelay(usec);  \
-       spin_lock(chip->mutex);  \
-} while (0)
-
-#define INVALIDATE_CACHE_UDELAY(map, chip, cmd_adr, adr, len, usec)  \
-do {  \
-       spin_unlock(chip->mutex);  \
-       INVALIDATE_CACHED_RANGE(map, adr, len);  \
-       cfi_udelay(usec);  \
-       spin_lock(chip->mutex);  \
-} while (0)
+               /* OK Still waiting */
+               if (time_after(jiffies, timeo)) {
+                       map_write(map, CMD(0x70), cmd_adr);
+                       chip->state = FL_STATUS;
+                       return -ETIME;
+               }
+
+               /* Latency issues. Drop the lock, wait a while and retry */
+               z++;
+               spin_unlock(chip->mutex);
+               cfi_udelay(1);
+               spin_lock(chip->mutex);
+       }
+
+       if (!z) {
+               if (!--(*chip_op_time))
+                       *chip_op_time = 1;
+       } else if (z > 1)
+               ++(*chip_op_time);
+
+       /* Done and happy. */
+       chip->state = FL_STATUS;
+       return 0;
+}
 
 #endif
 
+#define WAIT_TIMEOUT(map, chip, adr, udelay) \
+       ({ int __udelay = (udelay); \
+          INVAL_CACHE_AND_WAIT(map, chip, adr, 0, 0, &__udelay); })
+
+
 static int do_point_onechip (struct map_info *map, struct flchip *chip, loff_t adr, size_t len)
 {
        unsigned long cmd_addr;
@@ -1250,14 +1299,11 @@ static int __xipram do_write_oneword(struct map_info *map, struct flchip *chip,
                                     unsigned long adr, map_word datum, int mode)
 {
        struct cfi_private *cfi = map->fldrv_priv;
-       map_word status, status_OK, write_cmd;
-       unsigned long timeo;
-       int z, ret=0;
+       map_word status, write_cmd;
+       int ret=0;
 
        adr += chip->start;
 
-       /* Let's determine those according to the interleave only once */
-       status_OK = CMD(0x80);
        switch (mode) {
        case FL_WRITING:
                write_cmd = (cfi->cfiq->P_ID != 0x0200) ? CMD(0x40) : CMD(0x41);
@@ -1283,57 +1329,17 @@ static int __xipram do_write_oneword(struct map_info *map, struct flchip *chip,
        map_write(map, datum, adr);
        chip->state = mode;
 
-       INVALIDATE_CACHE_UDELAY(map, chip, adr,
-                               adr, map_bankwidth(map),
-                               chip->word_write_time);
-
-       timeo = jiffies + (HZ/2);
-       z = 0;
-       for (;;) {
-               if (chip->state != mode) {
-                       /* Someone's suspended the write. Sleep */
-                       DECLARE_WAITQUEUE(wait, current);
-
-                       set_current_state(TASK_UNINTERRUPTIBLE);
-                       add_wait_queue(&chip->wq, &wait);
-                       spin_unlock(chip->mutex);
-                       schedule();
-                       remove_wait_queue(&chip->wq, &wait);
-                       timeo = jiffies + (HZ / 2); /* FIXME */
-                       spin_lock(chip->mutex);
-                       continue;
-               }
-
-               status = map_read(map, adr);
-               if (map_word_andequal(map, status, status_OK, status_OK))
-                       break;
-
-               /* OK Still waiting */
-               if (time_after(jiffies, timeo)) {
-                       map_write(map, CMD(0x70), adr);
-                       chip->state = FL_STATUS;
-                       xip_enable(map, chip, adr);
-                       printk(KERN_ERR "%s: word write error (status timeout)\n", map->name);
-                       ret = -EIO;
-                       goto out;
-               }
-
-               /* Latency issues. Drop the lock, wait a while and retry */
-               z++;
-               UDELAY(map, chip, adr, 1);
-       }
-       if (!z) {
-               chip->word_write_time--;
-               if (!chip->word_write_time)
-                       chip->word_write_time = 1;
+       ret = INVAL_CACHE_AND_WAIT(map, chip, adr,
+                                  adr, map_bankwidth(map),
+                                  &chip->word_write_time);
+       if (ret) {
+               xip_enable(map, chip, adr);
+               printk(KERN_ERR "%s: word write error (status timeout)\n", map->name);
+               goto out;
        }
-       if (z > 1)
-               chip->word_write_time++;
-
-       /* Done and happy. */
-       chip->state = FL_STATUS;
 
        /* check for errors */
+       status = map_read(map, adr);
        if (map_word_bitsset(map, status, CMD(0x1a))) {
                unsigned long chipstatus = MERGESTATUS(status);
 
@@ -1450,9 +1456,9 @@ static int __xipram do_write_buffer(struct map_info *map, struct flchip *chip,
                                    unsigned long *pvec_seek, int len)
 {
        struct cfi_private *cfi = map->fldrv_priv;
-       map_word status, status_OK, write_cmd, datum;
-       unsigned long cmd_adr, timeo;
-       int wbufsize, z, ret=0, word_gap, words;
+       map_word status, write_cmd, datum;
+       unsigned long cmd_adr;
+       int ret, wbufsize, word_gap, words;
        const struct kvec *vec;
        unsigned long vec_seek;
 
@@ -1461,7 +1467,6 @@ static int __xipram do_write_buffer(struct map_info *map, struct flchip *chip,
        cmd_adr = adr & ~(wbufsize-1);
 
        /* Let's determine this according to the interleave only once */
-       status_OK = CMD(0x80);
        write_cmd = (cfi->cfiq->P_ID != 0x0200) ? CMD(0xe8) : CMD(0xe9);
 
        spin_lock(chip->mutex);
@@ -1493,32 +1498,20 @@ static int __xipram do_write_buffer(struct map_info *map, struct flchip *chip,
        }
 
        chip->state = FL_WRITING_TO_BUFFER;
-
-       z = 0;
-       for (;;) {
-               map_write(map, write_cmd, cmd_adr);
-
+       map_write(map, write_cmd, cmd_adr);
+       ret = WAIT_TIMEOUT(map, chip, cmd_adr, 0);
+       if (ret) {
+               /* Argh. Not ready for write to buffer */
+               map_word Xstatus = map_read(map, cmd_adr);
+               map_write(map, CMD(0x70), cmd_adr);
+               chip->state = FL_STATUS;
                status = map_read(map, cmd_adr);
-               if (map_word_andequal(map, status, status_OK, status_OK))
-                       break;
-
-               UDELAY(map, chip, cmd_adr, 1);
-
-               if (++z > 20) {
-                       /* Argh. Not ready for write to buffer */
-                       map_word Xstatus;
-                       map_write(map, CMD(0x70), cmd_adr);
-                       chip->state = FL_STATUS;
-                       Xstatus = map_read(map, cmd_adr);
-                       /* Odd. Clear status bits */
-                       map_write(map, CMD(0x50), cmd_adr);
-                       map_write(map, CMD(0x70), cmd_adr);
-                       xip_enable(map, chip, cmd_adr);
-                       printk(KERN_ERR "%s: Chip not ready for buffer write. status = %lx, Xstatus = %lx\n",
-                              map->name, status.x[0], Xstatus.x[0]);
-                       ret = -EIO;
-                       goto out;
-               }
+               map_write(map, CMD(0x50), cmd_adr);
+               map_write(map, CMD(0x70), cmd_adr);
+               xip_enable(map, chip, cmd_adr);
+               printk(KERN_ERR "%s: Chip not ready for buffer write. Xstatus = %lx, status = %lx\n",
+                               map->name, Xstatus.x[0], status.x[0]);
+               goto out;
        }
 
        /* Figure out the number of words to write */
@@ -1573,56 +1566,19 @@ static int __xipram do_write_buffer(struct map_info *map, struct flchip *chip,
        map_write(map, CMD(0xd0), cmd_adr);
        chip->state = FL_WRITING;
 
-       INVALIDATE_CACHE_UDELAY(map, chip, cmd_adr,
-                               adr, len,
-                               chip->buffer_write_time);
-
-       timeo = jiffies + (HZ/2);
-       z = 0;
-       for (;;) {
-               if (chip->state != FL_WRITING) {
-                       /* Someone's suspended the write. Sleep */
-                       DECLARE_WAITQUEUE(wait, current);
-                       set_current_state(TASK_UNINTERRUPTIBLE);
-                       add_wait_queue(&chip->wq, &wait);
-                       spin_unlock(chip->mutex);
-                       schedule();
-                       remove_wait_queue(&chip->wq, &wait);
-                       timeo = jiffies + (HZ / 2); /* FIXME */
-                       spin_lock(chip->mutex);
-                       continue;
-               }
-
-               status = map_read(map, cmd_adr);
-               if (map_word_andequal(map, status, status_OK, status_OK))
-                       break;
-
-               /* OK Still waiting */
-               if (time_after(jiffies, timeo)) {
-                       map_write(map, CMD(0x70), cmd_adr);
-                       chip->state = FL_STATUS;
-                       xip_enable(map, chip, cmd_adr);
-                       printk(KERN_ERR "%s: buffer write error (status timeout)\n", map->name);
-                       ret = -EIO;
-                       goto out;
-               }
-
-               /* Latency issues. Drop the lock, wait a while and retry */
-               z++;
-               UDELAY(map, chip, cmd_adr, 1);
-       }
-       if (!z) {
-               chip->buffer_write_time--;
-               if (!chip->buffer_write_time)
-                       chip->buffer_write_time = 1;
+       ret = INVAL_CACHE_AND_WAIT(map, chip, cmd_adr,
+                                  adr, len,
+                                  &chip->buffer_write_time);
+       if (ret) {
+               map_write(map, CMD(0x70), cmd_adr);
+               chip->state = FL_STATUS;
+               xip_enable(map, chip, cmd_adr);
+               printk(KERN_ERR "%s: buffer write error (status timeout)\n", map->name);
+               goto out;
        }
-       if (z > 1)
-               chip->buffer_write_time++;
-
-       /* Done and happy. */
-       chip->state = FL_STATUS;
 
        /* check for errors */
+       status = map_read(map, cmd_adr);
        if (map_word_bitsset(map, status, CMD(0x1a))) {
                unsigned long chipstatus = MERGESTATUS(status);
 
@@ -1718,17 +1674,12 @@ static int __xipram do_erase_oneblock(struct map_info *map, struct flchip *chip,
                                      unsigned long adr, int len, void *thunk)
 {
        struct cfi_private *cfi = map->fldrv_priv;
-       map_word status, status_OK;
-       unsigned long timeo;
+       map_word status;
        int retries = 3;
-       DECLARE_WAITQUEUE(wait, current);
-       int ret = 0;
+       int ret;
 
        adr += chip->start;
 
-       /* Let's determine this according to the interleave only once */
-       status_OK = CMD(0x80);
-
  retry:
        spin_lock(chip->mutex);
        ret = get_chip(map, chip, adr, FL_ERASING);
@@ -1750,48 +1701,15 @@ static int __xipram do_erase_oneblock(struct map_info *map, struct flchip *chip,
        chip->state = FL_ERASING;
        chip->erase_suspended = 0;
 
-       INVALIDATE_CACHE_UDELAY(map, chip, adr,
-                               adr, len,
-                               chip->erase_time*1000/2);
-
-       /* FIXME. Use a timer to check this, and return immediately. */
-       /* Once the state machine's known to be working I'll do that */
-
-       timeo = jiffies + (HZ*20);
-       for (;;) {
-               if (chip->state != FL_ERASING) {
-                       /* Someone's suspended the erase. Sleep */
-                       set_current_state(TASK_UNINTERRUPTIBLE);
-                       add_wait_queue(&chip->wq, &wait);
-                       spin_unlock(chip->mutex);
-                       schedule();
-                       remove_wait_queue(&chip->wq, &wait);
-                       spin_lock(chip->mutex);
-                       continue;
-               }
-               if (chip->erase_suspended) {
-                       /* This erase was suspended and resumed.
-                          Adjust the timeout */
-                       timeo = jiffies + (HZ*20); /* FIXME */
-                       chip->erase_suspended = 0;
-               }
-
-               status = map_read(map, adr);
-               if (map_word_andequal(map, status, status_OK, status_OK))
-                       break;
-
-               /* OK Still waiting */
-               if (time_after(jiffies, timeo)) {
-                       map_write(map, CMD(0x70), adr);
-                       chip->state = FL_STATUS;
-                       xip_enable(map, chip, adr);
-                       printk(KERN_ERR "%s: block erase error: (status timeout)\n", map->name);
-                       ret = -EIO;
-                       goto out;
-               }
-
-               /* Latency issues. Drop the lock, wait a while and retry */
-               UDELAY(map, chip, adr, 1000000/HZ);
+       ret = INVAL_CACHE_AND_WAIT(map, chip, adr,
+                                  adr, len,
+                                  &chip->erase_time);
+       if (ret) {
+               map_write(map, CMD(0x70), adr);
+               chip->state = FL_STATUS;
+               xip_enable(map, chip, adr);
+               printk(KERN_ERR "%s: block erase error: (status timeout)\n", map->name);
+               goto out;
        }
 
        /* We've broken this before. It doesn't hurt to be safe */
@@ -1820,7 +1738,6 @@ static int __xipram do_erase_oneblock(struct map_info *map, struct flchip *chip,
                        ret = -EIO;
                } else if (chipstatus & 0x20 && retries--) {
                        printk(KERN_DEBUG "block erase failed at 0x%08lx: status 0x%lx. Retrying...\n", adr, chipstatus);
-                       timeo = jiffies + HZ;
                        put_chip(map, chip, adr);
                        spin_unlock(chip->mutex);
                        goto retry;
@@ -1926,15 +1843,11 @@ static int __xipram do_xxlock_oneblock(struct map_info *map, struct flchip *chip
 {
        struct cfi_private *cfi = map->fldrv_priv;
        struct cfi_pri_intelext *extp = cfi->cmdset_priv;
-       map_word status, status_OK;
-       unsigned long timeo = jiffies + HZ;
+       int udelay;
        int ret;
 
        adr += chip->start;
 
-       /* Let's determine this according to the interleave only once */
-       status_OK = CMD(0x80);
-
        spin_lock(chip->mutex);
        ret = get_chip(map, chip, adr, FL_LOCKING);
        if (ret) {
@@ -1959,41 +1872,21 @@ static int __xipram do_xxlock_oneblock(struct map_info *map, struct flchip *chip
         * If Instant Individual Block Locking supported then no need
         * to delay.
         */
+       udelay = (!extp || !(extp->FeatureSupport & (1 << 5))) ? 1000000/HZ : 0;
 
-       if (!extp || !(extp->FeatureSupport & (1 << 5)))
-               UDELAY(map, chip, adr, 1000000/HZ);
-
-       /* FIXME. Use a timer to check this, and return immediately. */
-       /* Once the state machine's known to be working I'll do that */
-
-       timeo = jiffies + (HZ*20);
-       for (;;) {
-
-               status = map_read(map, adr);
-               if (map_word_andequal(map, status, status_OK, status_OK))
-                       break;
-
-               /* OK Still waiting */
-               if (time_after(jiffies, timeo)) {
-                       map_write(map, CMD(0x70), adr);
-                       chip->state = FL_STATUS;
-                       xip_enable(map, chip, adr);
-                       printk(KERN_ERR "%s: block unlock error: (status timeout)\n", map->name);
-                       put_chip(map, chip, adr);
-                       spin_unlock(chip->mutex);
-                       return -EIO;
-               }
-
-               /* Latency issues. Drop the lock, wait a while and retry */
-               UDELAY(map, chip, adr, 1);
+       ret = WAIT_TIMEOUT(map, chip, adr, udelay);
+       if (ret) {
+               map_write(map, CMD(0x70), adr);
+               chip->state = FL_STATUS;
+               xip_enable(map, chip, adr);
+               printk(KERN_ERR "%s: block unlock error: (status timeout)\n", map->name);
+               goto out;
        }
 
-       /* Done and happy. */
-       chip->state = FL_STATUS;
        xip_enable(map, chip, adr);
-       put_chip(map, chip, adr);
+out:   put_chip(map, chip, adr);
        spin_unlock(chip->mutex);
-       return 0;
+       return ret;
 }
 
 static int cfi_intelext_lock(struct mtd_info *mtd, loff_t ofs, size_t len)