mmc: make number of mmcblk minors configurable

The old limit of number of minor numbers per mmcblk device was hardcoded at 8. This isn't enough for some of the more elaborate partitioning schemes, for example those used by Chrome OS.

Since there might be a bunch of systems out there with static /dev contents that relies on the old numbering scheme, let's make it a build-time option with the default set to the previous 8.

---
 drivers/mmc/card/Kconfig |   10 ++++++++++
 drivers/mmc/card/block.c |   20 ++++++++------------
 2 files changed, 18 insertions(+), 12 deletions(-)

drivers/mmc/card/Kconfig +0 -0

index 3f2a912..e939dcf 100644
@@ -14,6 +14,16 @@ config MMC_BLOCK
 	  mount the filesystem. Almost everyone wishing MMC support
 	  should say Y or M here.

drivers/mmc/card/Kconfig +0 -0

 config MMC_BLOCK_BOUNCE
 	bool "Use bounce buffer for simple hosts"
 	depends on MMC_BLOCK

drivers/mmc/card/block.c +0 -0

index d545f79..524e232 100644

Quoted 4 lines · On Tue, 17 Aug 2010 23:13:33 -0500 Olof Johansson <[email protected]> wrote:

The old limit of number of minor numbers per mmcblk device was hardcoded at 8. This isn't enough for some of the more elaborate partitioning schemes, for example those used by Chrome OS.

Quoted 4 lines

Since there might be a bunch of systems out there with static /dev contents that relies on the old numbering scheme, let's make it a build-time option with the default set to the previous 8.

How does the numbering scheme change? What's the user-visible effect of setting CONFIG_MMC_BLOCK_MINORS to, say, 16?

Quoted 13 lines

---
 drivers/mmc/card/Kconfig |   10 ++++++++++
 drivers/mmc/card/block.c |   20 ++++++++------------
 2 files changed, 18 insertions(+), 12 deletions(-)

drivers/mmc/card/Kconfig +0 -0

index 3f2a912..e939dcf 100644
@@ -14,6 +14,16 @@ config MMC_BLOCK
 	  mount the filesystem. Almost everyone wishing MMC support
 	  should say Y or M here.
 

drivers/mmc/card/Kconfig config MMC_BLOCK · line 25

Quoted 9 lines

drivers/mmc/card/Kconfig +9 -0

+config MMC_BLOCK_MINORS
+	int "Number of minors per block device"
+	range 4 32
+	default 8
+	help
+	  Number of minors per block device. One is needed for every
+	  partition (plus one for the whole device).
+	  Default is 8 to be backwards compatible with previous
+	  hardcoded device numbering.

If possible it would be better to do this with a kernel boot (or modprobe) parameter. That permits vendors to ship packaged kernels with full functionality, etc. Possible? The default value of the runtime-settable parameter could be Kconfigurable is there's a real need for that.

Quoted 11 lines

drivers/mmc/card/Kconfig +0 -0

 config MMC_BLOCK_BOUNCE
 	bool "Use bounce buffer for simple hosts"
 	depends on MMC_BLOCK

drivers/mmc/card/block.c +0 -0

index d545f79..524e232 100644
@@ -45,13 +45,9 @@
 
 MODULE_ALIAS("mmc:block");
 

drivers/mmc/card/block.c line 48

Quoted 6 lines

drivers/mmc/card/block.c +1 -5

-/*
- * max 8 partitions per card
- */
-#define MMC_SHIFT	3
-#define MMC_NUM_MINORS	(256 >> MMC_SHIFT)
+#define MMC_MAX_DEVICES	((255 + CONFIG_MMC_BLOCK_MINORS) / CONFIG_MMC_BLOCK_MINORS)

So if I increase CONFIG_MMC_BLOCK_MINORS, I can no longer support as many devices? That'd be worth mentioning somewhere?

fwiw, the above could use DIV_ROUND_UP().

hi all
I find sys_init_module system call in kernel/module.c file. but it did not find any.
the source code version is linux-2.6.35

does the kernel have changed?

the book I am reading is LDD 3rd edition

Quoted 8 lines · On Sat, 21 Aug 2010 06:18:04 +0800 runcoderen wrote:

hi all
I find sys_init_module system call in kernel/module.c file. but it did not find any.
the source code version is linux-2.6.35

does the kernel have changed?

the book I am reading is LDD 3rd edition

You won't find a function in kernel/module.c that looks like it has the name "sys_init_module". Instead, the function is defined by using a macro, and it looks like this:

/* This is where the real work happens */ SYSCALL_DEFINE3(init_module, void __user *, umod,
unsigned long, len, const char __user *, uargs)
{

The use of SYSCALL_DEFINEx(func_name, args) is a fairly recent change (well, January, 2009) that may be the reason why you think that there has been a change. See
http://git.kernel.org/?p=linux/kernel/git/torvalds/linux-2.6.git;a=commit;h=1a94bc34768e463a93cb3751819709ab0ea80a01

---
~Randy
*** Remember to use Documentation/SubmitChecklist when testing your code ***

The old limit of number of minor numbers per mmcblk device was hardcoded at 8. This isn't enough for some of the more elaborate partitioning schemes, for example those used by Chrome OS.

Since there might be a bunch of systems out there with static /dev contents that relies on the old numbering scheme, let's make it a build-time option with the default set to the previous 8.

Also provide a boot/modprobe-time parameter to override the config
default: mmcblk.perdev_minors.

Signed-off-by Olof Johansson <[email protected]>

Cc Mandeep Baines <[email protected]>

---

Andrew, sorry for the churn but Mandeep had a couple of good points that needed addressing.

Changes since v1:
       * Runtime override of config default
       * Better help text
       * DIV_ROUND_UP for max_devices calculation
       * Clarify mmcblk device count limitations

Changes since v2 based on feedback from Mandeep Baines:
       * DIV_ROUND_UP is doing the wrong thing, we'll end up using the last fractional range of minors which we shouldn't.
       * Documentation/devices.txt update
       * No need to compute max_devices twice, just do it at runtime.
       * Permission fix for the module_param -- it's readonly.

 Documentation/devices.txt |    6 ++++++ drivers/mmc/card/Kconfig  |   17 +++++++++++++++++ drivers/mmc/card/block.c  |   41 ++++++++++++++++++++++++++++++-----------

 3 files changed, 53 insertions(+), 11 deletions(-)

Documentation/devices.txt +0 -0

index 1d83d12..fdf3821 100644
@@ -2518,6 +2518,12 @@ Your cooperation is appreciated.

                 8 = /dev/mmcblk1      Second SD/MMC card
                   ...

+               The start of next SD/MMC card can be configured with
+               CONFIG_MMC_BLOCK_MINORS, or overridden at boot/modprobe
+               time using the mmcblk.perdev_minors option. That would
+               bump the offset between each card to be the configured
+               value instead of the default 8.
+
 179 char       CCube DVXChip-based PCI products
                 0 = /dev/dvxirq0      First DVX device
                 1 = /dev/dvxirq1      Second DVX device

drivers/mmc/card/Kconfig +0 -0

index 3f2a912..57e4416 100644
@@ -14,6 +14,23 @@ config MMC_BLOCK

         mount the filesystem. Almost everyone wishing MMC support should say Y or M here.

drivers/mmc/card/Kconfig +17 -0

+config MMC_BLOCK_MINORS
+       int "Number of minors per block device"
+       range 4 256
+       default 8
+       help
+         Number of minors per block device. One is needed for every
+         partition on the disk (plus one for the whole disk).
+
+         Number of total MMC minors available is 256, so your number
+         of supported block devices will be limited to 256 divided
+         by this number.
+
+         Default is 8 to be backwards compatible with previous
+         hardwired device numbering.
+
+         If unsure, say 8 here.
+

 config MMC_BLOCK_BOUNCE
       bool "Use bounce buffer for simple hosts"
       depends on MMC_BLOCK

drivers/mmc/card/block.c +30 -12

index cb9fbc8..ec94f56 100644
@@ -43,14 +43,26 @@
  #include "queue.h"
 
  MODULE_ALIAS("mmc:block");
+#ifdef MODULE_PARAM_PREFIX
+#undef MODULE_PARAM_PREFIX
+#endif
+#define MODULE_PARAM_PREFIX "mmcblk."
+
+
+/*
+ * The defaults come from config options but can be overriden by module
+ * or bootarg options.
+ */
+static int perdev_minors = CONFIG_MMC_BLOCK_MINORS;
 
  /*
- * max 8 partitions per card
+ * We've only got one major, so number of mmcblk devices is
+ * limited to 256 / number of minors per device.
  */
-#define MMC_SHIFT      3
-#define MMC_NUM_MINORS (256 >> MMC_SHIFT)
+static int max_devices;
 
-static DECLARE_BITMAP(dev_use, MMC_NUM_MINORS);
+/* 256 minors, so at most 256 separate devices */
+static DECLARE_BITMAP(dev_use, 256);
 
  /*
  * There is one mmc_blk_data per slot.

@@ -66,6 +78,9 @@ struct mmc_blk_data {
 
  static DEFINE_MUTEX(open_lock);
 
+module_param(perdev_minors, int, 0444);
+MODULE_PARM_DESC(perdev_minors, "Minors numbers to allocate per device");
+
  static struct mmc_blk_data *mmc_blk_get(struct gendisk *disk)
  {
        struct mmc_blk_data *md;

@@ -87,10 +102,10 @@ static void mmc_blk_put(struct mmc_blk_data *md)
        md->usage--;
        if (md->usage == 0) {
                int devmaj = MAJOR(disk_devt(md->disk));
-               int devidx = MINOR(disk_devt(md->disk)) >> MMC_SHIFT;
+               int devidx = MINOR(disk_devt(md->disk)) / perdev_minors;
 
                if (!devmaj)
-                       devidx = md->disk->first_minor >> MMC_SHIFT;
+                       devidx = md->disk->first_minor / perdev_minors;
 
                blk_cleanup_queue(md->queue.queue);
 

@@ -482,8 +497,8 @@ static struct mmc_blk_data *mmc_blk_alloc(struct mmc_card *card)
        struct mmc_blk_data *md;
        int devidx, ret;
 
-       devidx = find_first_zero_bit(dev_use, MMC_NUM_MINORS);
-       if (devidx >= MMC_NUM_MINORS)
+       devidx = find_first_zero_bit(dev_use, max_devices);
+       if (devidx >= max_devices)
                return ERR_PTR(-ENOSPC);
        __set_bit(devidx, dev_use);
 

@@ -500,7 +515,7 @@ static struct mmc_blk_data *mmc_blk_alloc(struct mmc_card *card)
         */
        md->read_only = mmc_blk_readonly(card);
 
-       md->disk = alloc_disk(1 << MMC_SHIFT);
+       md->disk = alloc_disk(perdev_minors);
        if (md->disk == NULL) {
                ret = -ENOMEM;
                goto err_kfree;

@@ -517,7 +532,7 @@ static struct mmc_blk_data *mmc_blk_alloc(struct mmc_card *card)
        md->queue.data = md;
 
        md->disk->major = MMC_BLOCK_MAJOR;
-       md->disk->first_minor = devidx << MMC_SHIFT;
+       md->disk->first_minor = devidx * perdev_minors;
        md->disk->fops = &mmc_bdops;
        md->disk->private_data = md;
        md->disk->queue = md->queue.queue;

@@ -593,7 +608,6 @@ static int mmc_blk_probe(struct mmc_card *card)
  {
        struct mmc_blk_data *md;
        int err;
-
        char cap_str[10];
 
        /*

@@ -683,6 +697,11 @@ static int __init mmc_blk_init(void)
  {
        int res;
 
+       if (perdev_minors != CONFIG_MMC_BLOCK_MINORS)
+               pr_info("mmcblk: using %d minors per device\n", perdev_minors);
+
+       max_devices = 256 / perdev_minors;
+
        res = register_blkdev(MMC_BLOCK_MAJOR, "mmc");
        if (res)
                goto out;
--

1.5.6.5

The patch's purpose is good. As modern sd&mmc is used to host the file system, the 8 partition limitation is becoming a kind of bottleneck... But why not just add GENHD_FL_EXT_DEVT flag to allow mmc use extended partition numbers?

Quoted 4 lines · Lei Wen

The patch's purpose is good. As modern sd&mmc is used to host the file system, the 8 partition limitation is becoming a kind of bottleneck... But why not just add GENHD_FL_EXT_DEVT flag to allow mmc use extended partition numbers?

No need to quote the whole patch next time. :)

Given that MMC is sometimes used in deeply embedded environments where udev might not be running to take care of dynamic device numbering, I chose to stay with a static layout.

I could be convinced otherwise though. It would cause some additional hassles for me since we start udev lateish during boot and have a prepopulated /dev before that, but that can be dealt with.

Quoted 13 lines · Olof Johansson

Quoted 4 lines · Lei Wen

The patch's purpose is good. As modern sd&mmc is used to host the file system, the 8 partition limitation is becoming a kind of bottleneck... But why not just add GENHD_FL_EXT_DEVT flag to allow mmc use extended partition numbers?

No need to quote the whole patch next time. :)

Given that MMC is sometimes used in deeply embedded environments where udev might not be running to take care of dynamic device numbering, I chose to stay with a static layout.

Quoted 3 lines

I could be convinced otherwise though. It would cause some additional hassles for me since we start udev lateish during boot and have a prepopulated /dev before that, but that can be dealt with.

I think the extended number are only used for stuff larger than the static 8?

Apart from that, you can not reliably or securely use a static /dev these days, you never know which device you talk to, because the kernel has far too many dynamically assigned numbers. For that reason, most embedded setups use the busybox hack to populate /dev. Recent kernels have a devfs again, and there is no reason today to continue any static /dev experiments -- unless someone is going over the entire tree and fixes all the dynamic assignments, which is unlikely to happen ever.

Quoted 14 lines · Kay Sievers

Quoted 13 lines · Olof Johansson

Quoted 4 lines · Lei Wen

The patch's purpose is good. As modern sd&mmc is used to host the file system, the 8 partition limitation is becoming a kind of bottleneck... But why not just add GENHD_FL_EXT_DEVT flag to allow mmc use extended partition numbers?

No need to quote the whole patch next time. :)

Given that MMC is sometimes used in deeply embedded environments where udev might not be running to take care of dynamic device numbering, I chose to stay with a static layout.

Quoted 5 lines

Quoted 3 lines

I could be convinced otherwise though. It would cause some additional hassles for me since we start udev lateish during boot and have a prepopulated /dev before that, but that can be dealt with.

I think the extended number are only used for stuff larger than the static 8?

Ah, yes, of course.

Quoted 8 lines

Apart from that, you can not reliably or securely use a static /dev these days, you never know which device you talk to, because the kernel has far too many dynamically assigned numbers. For that reason, most embedded setups use the busybox hack to populate /dev. Recent kernels have a devfs again, and there is no reason today to continue any static /dev experiments -- unless someone is going over the entire tree and fixes all the dynamic assignments, which is unlikely to happen ever.

Yeah, it's a weak argument but I wanted to go for the least surprising one for anyone who still relies on static numbering of mmc, since it's used as root device, etc.

But with the first 8 minors being the same as before, there's little reason to care about that -- the EXT_DEVT approach will actually be the smaller change, since it won't renumber anything, just extend in a different part of the address space.

Lei, want to submit a proper patch for it to replace mine?

Set the GENHD_FL_EXT_DEVT flag, which will allocate minor numbers in major 259 for partitions past disk->minors.

Also remove the use of disk_devt to determine devidx from md->disk. md->disk->first_minor is always initialized from devidx and can always be used to recover it.

Signed-off-by Colin Cross <[email protected]>

---
 drivers/mmc/card/block.c |    7 ++-----

 1 files changed, 2 insertions(+), 5 deletions(-)

drivers/mmc/card/block.c +2 -6

index d545f79..07d8eb0 100644
@@ -87,11 +87,7 @@ static void mmc_blk_put(struct mmc_blk_data *md)
        mutex_lock(&open_lock);
        md->usage--;
        if (md->usage == 0) {
-               int devmaj = MAJOR(disk_devt(md->disk));
-               int devidx = MINOR(disk_devt(md->disk)) >> MMC_SHIFT;
-
-               if (!devmaj)
-                       devidx = md->disk->first_minor >> MMC_SHIFT;
+               int devidx = md->disk->first_minor >> MMC_SHIFT;
 
                blk_cleanup_queue(md->queue.queue);
 

@@ -607,6 +603,7 @@ static struct mmc_blk_data *mmc_blk_alloc(struct mmc_card *card)
        md->disk->private_data = md;
        md->disk->queue = md->queue.queue;
        md->disk->driverfs_dev = &card->dev;
+       md->disk->flags = GENHD_FL_EXT_DEVT;
 
        /*
         * As discussed on lkml, GENHD_FL_REMOVABLE should:
--

1.7.1

Set the GENHD_FL_EXT_DEVT flag, which will allocate minor numbers in major 259 for partitions past disk->minors.

Also remove the use of disk_devt to determine devidx from md->disk. md->disk->first_minor is always initialized from devidx and can always be used to recover it.

Signed-off-by Colin Cross <[email protected]>

---
 drivers/mmc/card/block.c |    7 ++-----
 1 files changed, 2 insertions(+), 5 deletions(-)

drivers/mmc/card/block.c +2 -5

index d545f79..07d8eb0 100644
@@ -87,11 +87,7 @@ static void mmc_blk_put(struct mmc_blk_data *md)
 	mutex_lock(&open_lock);
 	md->usage--;
 	if (md->usage == 0) {
-		int devmaj = MAJOR(disk_devt(md->disk));
-		int devidx = MINOR(disk_devt(md->disk)) >> MMC_SHIFT;
-
-		if (!devmaj)
-			devidx = md->disk->first_minor >> MMC_SHIFT;
+		int devidx = md->disk->first_minor >> MMC_SHIFT;
 
 		blk_cleanup_queue(md->queue.queue);
 

@@ -607,6 +603,7 @@ static struct mmc_blk_data *mmc_blk_alloc(struct mmc_card *card)
 	md->disk->private_data = md;
 	md->disk->queue = md->queue.queue;
 	md->disk->driverfs_dev = &card->dev;
+	md->disk->flags = GENHD_FL_EXT_DEVT;
 
 	/*
 	 * As discussed on lkml, GENHD_FL_REMOVABLE should: