Fix fastpath issues
Since mmci_request() can be called from a non-interrupt
context, and does, during kernel init, causing a host
of debug messages during boot if you enable spinlock debugging,
we need to use the spinlock calls that save IRQ flags and
restore them.
Signed-off-by: Linus Walleij <triad@df.lth.se>
Signed-off-by: Russell King <rmk+kernel@arm.linux.org.uk>
To keep the size of changesets sane we split the switch by drivers;
to keep the damn thing bisectable we do the following:
1) rename the affected methods, add ones with correct
prototypes, make (few) callers handle both. That's this changeset.
2) for each driver convert to new methods. *ALL* drivers
are converted in this series.
3) kill the old (renamed) methods.
Note that it _is_ a flagday; all in-tree drivers are converted and by the
end of this series no trace of old methods remain. The only reason why
we do that this way is to keep the damn thing bisectable and allow per-driver
debugging if anything goes wrong.
New methods:
open(bdev, mode)
release(disk, mode)
ioctl(bdev, mode, cmd, arg) /* Called without BKL */
compat_ioctl(bdev, mode, cmd, arg)
locked_ioctl(bdev, mode, cmd, arg) /* Called with BKL, legacy */
Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
Since the original authour (Thomas Kleffel) has been too busy to
merge the s3cmci driver and keep it up to date, I (mostly as part
of my role with Simtec Electronics) got the driver to a mergable
state and have been maintaining it since I think that I should
be added to the header. Also add a copyright statement for the
new work.
Signed-off-by: Ben Dooks <ben-linux@fluff.org>
Signed-off-by: Pierre Ossman <drzeus@drzeus.cx>
The s3cmci driver uses the host->pio_ptr field to
point to the current position into the buffer for data
transfer. During the transfers it does the following:
while (fifo_words--)
*(host->pio_ptr++) = readl(from_ptr);
This is inefficent, as host->pio_ptr is not used in any
other part of the transfer but the compiler emits code
which does the following:
while (fifo_words--) {
u32 *ptr = host->pio_ptr;
*ptr = readl(from_ptr);
ptr++;
host->pio_ptr = ptr;
}
This is obviously a waste of a load and store each time
around the loop, which could be up to 16 times depending
on how much needs to be transfered.
Move the ptr accesses to outside the while loop so that
we do not end up reloading/re-writing the pointer.
Note, this seems to make the code 16 bytes larger.
Signed-off-by: Ben Dooks <ben-linux@fluff.org>
Signed-off-by: Pierre Ossman <drzeus@drzeus.cx>
To be able to do SDIO the s3cmci driver has to support non-word-sized
transfers. Change pio_words into pio_bytes and fix up all the places
where it is used.
This variant of the patch will not overrun the buffer when reading an
odd number of bytes. When writing, this variant will still read past
the end of the buffer, but since the driver can't support non-word-
aligned transfers anyway, this should not be a problem, since a
word-aligned transfer will never cross a page boundary.
This has been tested with a CSR SDIO Bluetooth Type A device on a
Samsung S3C24A0 processor.
Signed-off-by: Christer Weinigel <christer@weinigel.se>
Signed-off-by: Ben Dooks <ben-linux@fluff.org>
Signed-off-by: Pierre Ossman <drzeus@drzeus.cx>
General errors, such as timeouts during probe do not need to
be sent to the console, so move them down to be included if the
debug is enabled.
Such errors include:
s3c2440-sdi s3c2440-sdi: s3cmci_request: no medium present
Signed-off-by: Ben Dooks <ben-linux@fluff.org>
Signed-off-by: Pierre Ossman <drzeus@drzeus.cx>
Hard-coded timeout values of 250ms for writes and 100ms for reads are
currently used for MMC transactions over SPI. The spec states that the
timeout values from the card should be used.
Signed-off-by: Matthew Fleming <matthew.fleming@imgtec.com>
Signed-off-by: Pierre Ossman <drzeus@drzeus.cx>
The MMC spec states that the timeout for accessing the CSD and CID
registers is 64 clock cycles.
Signed-off-by: Matthew Fleming <matthew.fleming@imgtec.com>
Signed-off-by: Pierre Ossman <drzeus@drzeus.cx>
The variable 'scratch' is always initialized before it's used. The
conditional which is responsible for initialization of 'scratch' will
always evaluate 'true' when the first loop iteration occurs, and thus,
it's properly initialized. GCC doesn't see this, of course, so using
the uninitialized_var() macro seems to work for silencing this case.
Signed-off-by: Steven Noonan <steven@uplinklabs.net>
Signed-off-by: Pierre Ossman <drzeus@drzeus.cx>
We use 512 byte blocks on all cards, and newer cards support nothing
else, so hard code it and make the code less complex.
Signed-off-by: Pierre Ossman <drzeus@drzeus.cx>
Some high speed capable controllers forget to set the high speed
capability bit. Make sure we enable the functionality anyway.
Signed-off-by: Pierre Ossman <drzeus@drzeus.cx>
The MMC block driver services requests one at a time and in strict
order. Indicate this to the block layer so that it can handle barriers
in an efficient manner.
Signed-off-by: Pierre Ossman <drzeus@drzeus.cx>
Make sure we consider the maximum block count when we tell the block
layer about the maximum sector count. That way we don't have to chop
up the request ourselves.
Signed-off-by: Pierre Ossman <drzeus@drzeus.cx>
Make sure we can be woken from the forced sleep that is done on errors.
Removing a card often results in -ENOMEDIUM or -EILSEQ so we previously
locked up the removal process for a second.
We could completely exit on -ENOMEDIUM, but it might be a transient
glitch so treat it like any other error.
Signed-off-by: Pierre Ossman <drzeus@drzeus.cx>
The card detection delay was added early when the behaviour of the
card interrupt was still very much unknown (i.e. before there was a
public specification). As it is now known that it is a debounced signal,
reduce the delay to something more sensible.
Signed-off-by: Pierre Ossman <drzeus@drzeus.cx>
The sdhci controllers can interrupt us when the busy state from the
card has ended, saving CPU cycles and power.
Signed-off-by: Pierre Ossman <drzeus@drzeus.cx>
* Implement disk_devt() and part_devt() and use them to directly
access devt instead of computing it from ->major and ->first_minor.
Note that all references to ->major and ->first_minor outside of
block layer is used to determine devt of the disk (the part0) and as
->major and ->first_minor will continue to represent devt for the
disk, converting these users aren't strictly necessary. However,
convert them for consistency.
* Implement disk_max_parts() to avoid directly deferencing
genhd->minors.
* Update bdget_disk() such that it doesn't assume consecutive minor
space.
* Move devt computation from register_disk() to add_disk() and make it
the only one (all other usages use the initially determined value).
These changes clean up the code and will help disk->part dereference
fix and extended block device numbers.
Signed-off-by: Tejun Heo <tj@kernel.org>
Signed-off-by: Jens Axboe <jens.axboe@oracle.com>
The atmel-mci driver sometimes fails data transfers like this:
mmcblk0: error -5 transferring data
end_request: I/O error, dev mmcblk0, sector 2749769
end_request: I/O error, dev mmcblk0, sector 2749777
It turns out that this might be caused by the BLKR register (which
contains the block size and the number of blocks being transfered) being
initialized too late. This patch moves the initialization of BLKR so
that it contains the correct value before the block transfer command is
sent.
This error is difficult to reproduce, but if you insert a long delay
(mdelay(10) or thereabouts) between the calls to atmci_start_command()
and atmci_submit_data(), all transfers seem to fail without this patch,
while I haven't seen any failures with this patch.
Reported-by: Hein_Tibosch <hein_tibosch@yahoo.es>
Signed-off-by: Haavard Skinnemoen <haavard.skinnemoen@atmel.com>
Signed-off-by: Pierre Ossman <drzeus@drzeus.cx>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
After a data error, we wait for the NOTBUSY bit to be set so that we can
be sure the data transfer is completely finished. However, when NOTBUSY
is set, the interrupt handler copies the contents of SR into
data_status, overwriting any error bits we may have detected earlier.
To avoid this, initialize data_status to 0 before starting a request, and
don't overwrite it unless it still contains 0.
Signed-off-by: Haavard Skinnemoen <haavard.skinnemoen@atmel.com>
This adds support for DMA transfers through the generic DMA engine
framework with the DMA slave extensions.
The driver has been tested using mmc-block and ext3fs on several SD,
SDHC and MMC+ cards. Reads and writes work fine, with read transfer
rates up to 7.5 MiB/s on fast cards with debugging disabled.
Unfortunately, the driver has been known to lock up from time to time
with DMA enabled, so DMA support is currently optional and marked
EXPERIMENTAL. However, I didn't see any problems while testing 13
different cards (MMC, SD and SDHC of different brands and sizes), so I
suspect the "Initialize BLKR before sending data transfer command" fix
that was posted earlier fixed this as well.
Signed-off-by: Haavard Skinnemoen <haavard.skinnemoen@atmel.com>
The Atmel MCI controller can drive multiple cards through separate sets
of pins, but only one at a time. This patch adds support for
multiplexing access to the controller so that multiple card slots can be
used as if they were hooked up to separate mmc controllers.
The atmel-mci driver registers each slot as a separate mmc_host. Both
access the same common controller state, but they also have some state
on their own for card detection/write protect handling, and separate
shadows of the MR and SDCR registers.
When one of the slots receives a request from the mmc core, the common
controller state is checked. If it's idle, the request is submitted
immediately. If not, the request is added to a queue. When a request is
done, the queue is checked and if there is a queued request, it is
submitted before the completion callback is called.
This patch also includes a few cleanups and fixes, including a locking
overhaul. I had to change the locking extensively in any case, so I
might as well try to get it right. The driver no longer takes any
irq-safe locks, which may or may not improve the overall system
performance.
This patch also adds a bit of documentation of the internal data
structures.
Signed-off-by: Haavard Skinnemoen <haavard.skinnemoen@atmel.com>
Add the necessary platform infrastructure to support multiple mmc/sdcard
slots all at once through a single controller. Currently, the driver
will use the first valid slot it finds and stick with that, but later
patches will add support for switching between several slots on the fly.
Extend the platform data structure with per-slot information: MMC/SDcard
bus width and card detect/write protect pins. This will affect the pin
muxing as well as the capabilities announced to the mmc core.
Note that board code is now required to supply a mci_platform_data
struct to at32_add_device_mci().
Signed-off-by: Haavard Skinnemoen <haavard.skinnemoen@atmel.com>
Some cards might get upset if we turn off the clock for extended periods
of time. So keep the clock running until the mmc core tells us to turn
it off.
Also, don't reset the controller between each transfer. That was an
attempt to work around earlier bugs, and it never really worked very
well.
Signed-off-by: Haavard Skinnemoen <haavard.skinnemoen@atmel.com>
With the current system of completed/pending events, things may get
handled in different order depending on which event triggers first. For
example, if the data transfer is complete before the command, the stop
command must be sent after the command is complete, not the data. This
creates a bit of complexity around the stop command.
By having the tasklet go through a sequence of clearly defined states,
things always happen in a certain order even if the events come at
different times, so the stop command can simply be sent when we exit the
"sending data" state because we will never enter that state before the
command has been sent successfully.
Signed-off-by: Haavard Skinnemoen <haavard.skinnemoen@atmel.com>
The atmel-mci driver sometimes fails data transfers like this:
mmcblk0: error -5 transferring data
end_request: I/O error, dev mmcblk0, sector 2749769
end_request: I/O error, dev mmcblk0, sector 2749777
It turns out that this might be caused by the BLKR register (which
contains the block size and the number of blocks being transfered) being
initialized too late. This patch moves the initialization of BLKR so
that it contains the correct value before the block transfer command is
sent.
This error is difficult to reproduce, but if you insert a long delay
(mdelay(10) or thereabouts) between the calls to atmci_start_command()
and atmci_submit_data(), all transfers seem to fail without this patch,
while I haven't seen any failures with this patch.
Reported-by: Hein_Tibosch <hein_tibosch@yahoo.es>
Signed-off-by: Haavard Skinnemoen <haavard.skinnemoen@atmel.com>
1. add a CPUID table in the comment
2. make cpu_is_pxa25x() true for PXA210/250/255/26x
3. PXA210 is treated as PXA25x, all related code modified to
reflect this
Signed-off-by: Eric Miao <eric.miao@marvell.com>
Signed-off-by: Russell King <rmk+kernel@arm.linux.org.uk>
The mutex mmc_test_lock is initialized at every time mmc_test device
is probed. Probing another mmc_test device may break the mutex, if
the probe function is called while the mutex is locked.
This patch fixes it by statically initializing mmc_test_lock.
Signed-off-by: Akinobu Mita <akinobu.mita@gmail.com>
Signed-off-by: Pierre Ossman <drzeus@drzeus.cx>
Check error from mmc_register_driver() and properly unwind
block device registration.
Signed-off-by: Akinobu Mita <akinobu.mita@gmail.com>
Signed-off-by: Pierre Ossman <drzeus@drzeus.cx>
This allows the mmc core to detect card insertion/removal for slots that
don't have any CD pin wired up.
Signed-off-by: Haavard Skinnemoen <haavard.skinnemoen@atmel.com>
Signed-off-by: Pierre Ossman <drzeus@drzeus.cx>
We used to store a binary register snapshot in the "regs" file, so we
set the file size to be the size of this snapshot. This is no longer
valid since we switched to using seq_file.
Signed-off-by: Haavard Skinnemoen <haavard.skinnemoen@atmel.com>
Signed-off-by: Pierre Ossman <drzeus@drzeus.cx>
The debugfs hook atmci_regs_show allocates a temporary buffer for
storing a register snapshot, but it doesn't free it before returning.
Plug this leak.
Signed-off-by: Haavard Skinnemoen <haavard.skinnemoen@atmel.com>
Signed-off-by: Pierre Ossman <drzeus@drzeus.cx>
Make sure that the peripheral clock is enabled before reading the MMIO
registers for the debugfs "regs" dump.
Signed-off-by: Haavard Skinnemoen <haavard.skinnemoen@atmel.com>
Signed-off-by: Pierre Ossman <drzeus@drzeus.cx>
mmc_block_open() increments md->usage although it returns with -EROFS when
default mounting a MMC/SD card with write protect switch on. This
reference counting bug prevents /dev/mmcblkX from being released on card
removal, and situation worsen with reinsertion until the minor number
range runs out.
Reported-by: <sasin@solomon-systech.com>
Acked-by: Pierre Ossman <drzeus-list@drzeus.cx>
Cc: <stable@kernel.org> [2.6.25.x, 2.6.26.x]
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
At91_mci is abusing dma_free_coherent(), which may not be called with IRQs
disabled. I saw "mkfs.ext3" on an MMC card objecting voluminously as each
write completed:
WARNING: at arch/arm/mm/consistent.c:368 dma_free_coherent+0x2c/0x224()
[<c002726c>] (dump_stack+0x0/0x14) from [<c00387d4>] (warn_on_slowpath+0x4c/0x68)
[<c0038788>] (warn_on_slowpath+0x0/0x68) from [<c0028768>] (dma_free_coherent+0x2c/0x224)
r6:00008008 r5:ffc06000 r4:00000000
[<c002873c>] (dma_free_coherent+0x0/0x224) from [<c01918ac>] (at91_mci_irq+0x374/0x420)
[<c0191538>] (at91_mci_irq+0x0/0x420) from [<c0065d9c>] (handle_IRQ_event+0x2c/0x6c)
...
This bug has been around for a LONG time. The MM warning is from late
2005, but the driver merged a year later ... so I'm puzzled why nobody
noticed this before now.
The fix involves noting that this buffer shouldn't be DMA-coherent; it's
just used for normal DMA writes. So replace it with standard kmalloc()
buffering and DMA mapping calls.
This is the quickie fix. A better one would not rely on allocating large
bounce buffers. (Note that dma_alloc_coherent could have failed too, but
that case was ignored... kmalloc is a bit more likely to fail though.)
Signed-off-by: David Brownell <dbrownell@users.sourceforge.net>
Acked-by: Pierre Ossman <drzeus-mmc@drzeus.cx>
Cc: Andrew Victor <linux@maxim.org.za>
Acked-by: Nicolas Ferre <nicolas.ferre@atmel.com>
Cc: Russell King <rmk@arm.linux.org.uk>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
The drivers below do not use LINUX_VERSION_CODE nor KERNEL_VERSION.
drivers/mmc/host/sdricoh_cs.c
This patch removes the said #include <version.h>.
Signed-off-by: Huang Weiyi <weiyi.huang@gmail.com>
Signed-off-by: Pierre Ossman <drzeus@drzeus.cx>
Attach the routine to get_cd to allow the MMC core to find out whether
there is a card present or not without the tedious process of trying to
send commands to the card or not.
Signed-off-by: Ben Dooks <ben-linux@fluff.org>
Signed-off-by: Pierre Ossman <drzeus@drzeus.cx>
Linus Walleij