When syncing blocks of data using raw writes combine the writes into a
single block write, saving us bus overhead for setup, addressing and
teardown.
Currently the block write is done unconditionally as it is expected that
hardware which has a register format which can support raw writes will
support auto incrementing writes, this decision may need to be revised in
future.
Signed-off-by: Mark Brown <broonie@opensource.wolfsonmicro.com>
Reviewed-by: Dimitris Papastamos <dp@opensource.wolfsonmicro.com>
For code clarity after implementing block writes split out the raw and
non-raw I/O sync implementations.
Signed-off-by: Mark Brown <broonie@opensource.wolfsonmicro.com>
Reviewed-by: Dimitris Papastamos <dp@opensource.wolfsonmicro.com>
The idea of holding blocks of registers in device format is shared between
at least rbtree and lzo cache formats so split out the loop that does the
sync from the rbtree code so optimisations on it can be reused.
Signed-off-by: Mark Brown <broonie@opensource.wolfsonmicro.com>
Reviewed-by: Dimitris Papastamos <dp@opensource.wolfsonmicro.com>
The idea of maintaining a bitmap of present registers is something that
can usefully be used by other cache types that maintain blocks of cached
registers so move the code out of the rbtree cache and into the generic
regcache code.
Refactor the interface slightly as we go to wrap the set bit and enlarge
bitmap operations (since we never do one without the other) and make it
more robust for reads of uncached registers by bounds checking before we
look at the bitmap.
Signed-off-by: Mark Brown <broonie@opensource.wolfsonmicro.com>
Reviewed-by: Dimitris Papastamos <dp@opensource.wolfsonmicro.com>
Provide a helper to do the size based index into a block of registers and
use it when reading a value.
Signed-off-by: Mark Brown <broonie@opensource.wolfsonmicro.com>
It's more idiomatic to pass the map structure around and this means we
can use other bits of information from the map.
Signed-off-by: Mark Brown <broonie@opensource.wolfsonmicro.com>
While for I2C and SPI devices the overhead of using rbtree for devices with
only one block of registers is negligible the same isn't always going to
be true for MMIO devices where the I/O costs are very much lower. Cater
for these devices by adding a simple flat array type for them where the
lookups are simple array accesses, taking us right back to the original
ASoC cache implementation.
Thanks to Magnus Damm for the discussion which prompted this.
Signed-off-by: Mark Brown <broonie@opensource.wolfsonmicro.com>
regmap_config.reg_stride is introduced. All extant register addresses
are a multiple of this value. Users of serial-oriented regmap busses will
typically set this to 1. Users of the MMIO regmap bus will typically set
this based on the value size of their registers, in bytes, so 4 for a
32-bit register.
Throughout the regmap code, actual register addresses are used. Wherever
the register address is used to index some array of values, the address
is divided by the stride to determine the index, or vice-versa. Error-
checking is added to all entry-points for register address data to ensure
that register addresses actually satisfy the specified stride. The MMIO
bus ensures that the specified stride is large enough for the register
size.
Signed-off-by: Stephen Warren <swarren@nvidia.com>
Signed-off-by: Mark Brown <broonie@opensource.wolfsonmicro.com>
Some bus types have very fast IO. For these, acquiring a mutex for every
IO operation is a significant overhead. Allow busses to indicate their IO
is fast, and enhance regmap to use a spinlock for those busses.
[Currently limited to native endian registers -- broonie]
Signed-off-by: Stephen Warren <swarren@nvidia.com>
Signed-off-by: Mark Brown <broonie@opensource.wolfsonmicro.com>
regcache_sync_region() isn't going to be useful to most drivers if we
don't export it since otherwise they can't use it when built modular.
Signed-off-by: Mark Brown <broonie@opensource.wolfsonmicro.com>
"[RFC PATCH 0/2] audit of linux/device.h users in include/*"
https://lkml.org/lkml/2012/3/4/159
--
Nearly every subsystem has some kind of header with a proto like:
void foo(struct device *dev);
and yet there is no reason for most of these guys to care about the
sub fields within the device struct. This allows us to significantly
reduce the scope of headers including headers. For this instance, a
reduction of about 40% is achieved by replacing the include with the
simple fact that the device is some kind of a struct.
Unlike the much larger module.h cleanup, this one is simply two
commits. One to fix the implicit <linux/device.h> users, and then
one to delete the device.h includes from the linux/include/ dir
wherever possible.
-----BEGIN PGP SIGNATURE-----
Version: GnuPG v1.4.11 (GNU/Linux)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=3j4+
-----END PGP SIGNATURE-----
Merge tag 'device-for-3.4' of git://git.kernel.org/pub/scm/linux/kernel/git/paulg/linux
Pull <linux/device.h> avoidance patches from Paul Gortmaker:
"Nearly every subsystem has some kind of header with a proto like:
void foo(struct device *dev);
and yet there is no reason for most of these guys to care about the
sub fields within the device struct. This allows us to significantly
reduce the scope of headers including headers. For this instance, a
reduction of about 40% is achieved by replacing the include with the
simple fact that the device is some kind of a struct.
Unlike the much larger module.h cleanup, this one is simply two
commits. One to fix the implicit <linux/device.h> users, and then one
to delete the device.h includes from the linux/include/ dir wherever
possible."
* tag 'device-for-3.4' of git://git.kernel.org/pub/scm/linux/kernel/git/paulg/linux:
device.h: audit and cleanup users in main include dir
device.h: cleanup users outside of linux/include (C files)
For files that are actively using linux/device.h, make sure
that they call it out. This will allow us to clean up some
of the implicit uses of linux/device.h within include/*
without introducing build regressions.
Yes, this was created by "cheating" -- i.e. the headers were
cleaned up, and then the fallout was found and fixed, and then
the two commits were reordered. This ensures we don't introduce
build regressions into the git history.
Signed-off-by: Paul Gortmaker <paul.gortmaker@windriver.com>
Provide a regcache_sync_region() operation which allows drivers to
write only part of the cache back to the hardware. This is intended
for use in cases like power domains or DSP memories where part of the
device register map may be reset without fully resetting the device.
Fully supporting these devices is likely to require additional work to
make specific regions of the register map cache only while they are in
reset, but this is enough for most devices.
Signed-off-by: Mark Brown <broonie@opensource.wolfsonmicro.com>
In order to allow us to support partial sync operations add minimum and
maximum register arguments to the sync operation and update the rbtree
and lzo caches to use this new information. The LZO implementation is
obviously not good, we could exit the iteration earlier, but there may
be room for more wide reaching optimisation there.
Signed-off-by: Mark Brown <broonie@opensource.wolfsonmicro.com>
Previously the cache would never be marked clean, meaning syncs would
never be suppressed which isn't the end of the world but could be
inefficient.
Signed-off-by: Mark Brown <broonie@opensource.wolfsonmicro.com>
It's not used as all cache types have sync operations so it's just dead
code which never gets tested.
Signed-off-by: Mark Brown <broonie@opensource.wolfsonmicro.com>
During regcache_init, if client has not passed the
default data of cached register then it is directly
read from the hw to initialize cache. This hw register
read happens before cache ops are initialized and hence
avoiding register read to check for the data available
on cache or not by enabling flag of cache_bypass.
Signed-off-by: Laxman Dewangan <ldewangan@nvidia.com>
Signed-off-by: Mark Brown <broonie@opensource.wolfsonmicro.com>
Currently registers with a value of 0 are ignored when initializing the register
defaults from raw defaults. This worked in the past, because registers without a
explicit default were assumed to have a default value of 0. This was changed in
commit b03622a8 ("regmap: Ensure rbtree syncs registers set to zero properly").
As a result registers, which have a raw default value of 0 are now assumed to
have no default. This again can result in unnecessary writes when syncing the
cache. It will also result in unnecessary reads for e.g. the first update
operation. In the case where readback is not possible this will even let the
update operation fail, if the register has not been written to before.
So this patch removes the check. Instead it adds a check to ignore raw defaults
for registers which are volatile, since those registers are not cached.
Signed-off-by: Lars-Peter Clausen <lars@metafoo.de>
Signed-off-by: Mark Brown <broonie@opensource.wolfsonmicro.com>
Cc: stable@vger.kernel.org
regcache_set_val() returns false if cache[idx] != val.
Thus it actually is not unreachable.
Signed-off-by: Axel Lin <axel.lin@gmail.com>
Signed-off-by: Mark Brown <broonie@opensource.wolfsonmicro.com>
On the basis that if we don't actually need to resync the cache then the
patches are probably also already applied.
Signed-off-by: Mark Brown <broonie@opensource.wolfsonmicro.com>
They have no current users which is fortunate as they don't take the lock
and therefore aren't safe to use externally. We'll need to add new
operations if direct cache access is needed.
Signed-off-by: Mark Brown <broonie@opensource.wolfsonmicro.com>
Device manufacturers frequently provide register sequences, usually not
fully documented, to be run at startup in order to provide better defaults
for devices (for example, improving performance in the light of silicon
evaluation). Support such updates by allowing drivers to register update
sets with the core. These updates will be written to the device immediately
and will also be rewritten when the cache is synced.
The assumption is that the reason for resyncing the cache will always be
that the device has been powered off. If this turns out to not be the case
then a separate operation can be provided.
Currently the implementation only allows a single set of updates to be
specified for a device, this could be extended in future.
Signed-off-by: Mark Brown <broonie@opensource.wolfsonmicro.com>
Currently we only trace physical reads, there's no instrumentation if
the read is satisfied from cache.
Signed-off-by: Mark Brown <broonie@opensource.wolfsonmicro.com>
There should be no situation where it offers any advantage over rbtree
and there are no current users so remove the code for simplicity.
Signed-off-by: Mark Brown <broonie@opensource.wolfsonmicro.com>
One of the reasons for using a cache is to have a software shadow of a register
which is writable but not readable. This allows us to do a read-modify-write
operation on such a register.
Currently regcache checks whether a register is readable when performing a
cached read and returns an error if it is not. Drop this check, since it will
prevent us from using the cache for registers where read-back is not possible.
Signed-off-by: Lars-Peter Clausen <lars@metafoo.de>
Signed-off-by: Mark Brown <broonie@opensource.wolfsonmicro.com>
regcache currently only properly works with val bit sizes of 8 or 16, since
it will, when calculating the cache word size, round down. This causes the
cache storage to be too small to hold the full register value. Fix this by
rounding up instead.
Signed-off-by: Lars-Peter Clausen <lars@metafoo.de>
Signed-off-by: Mark Brown <broonie@opensource.wolfsonmicro.com>
The reg_defaults field usually points to a static per driver array, which should
not be modified. Make requirement this explicit by making reg_defaults const.
To allow this the regcache_init code needs some minor changes. Previoulsy the
reg_config was not available in regcache_init and regmap->reg_defaults was used
to pass the default register set to regcache_init. Now that the reg_config is
available we can work on it directly.
Signed-off-by: Lars-Peter Clausen <lars@metafoo.de>
Signed-off-by: Mark Brown <broonie@opensource.wolfsonmicro.com>
Move the initialization regcache related fields of the regmap struct to
regcache_init. This allows us to keep regmap and regcache code better
separated.
Signed-off-by: Lars-Peter Clausen <lars@metafoo.de>
Signed-off-by: Mark Brown <broonie@opensource.wolfsonmicro.com>
Make sure all allocated memory gets freed again in case initializing the cache
failed.
Signed-off-by: Lars-Peter Clausen <lars@metafoo.de>
Acked-by: Dimitris Papastamos <dp@opensource.wolfsonmicro.com>
Signed-off-by: Mark Brown <broonie@opensource.wolfsonmicro.com>
Make sure reg_defaults_raw gets freed in case of an error.
Signed-off-by: Lars-Peter Clausen <lars@metafoo.de>
Acked-by: Dimitris Papastamos <dp@opensource.wolfsonmicro.com>
Signed-off-by: Mark Brown <broonie@opensource.wolfsonmicro.com>
Allow drivers to optimise out the register cache sync if they didn't need
to do one. If the hardware is desynced from the register cache (by power
loss for example) then the driver should call regcache_mark_dirty() to
let the core know about this.
Signed-off-by: Mark Brown <broonie@opensource.wolfsonmicro.com>
Most of these files were implicitly getting EXPORT_SYMBOL via
device.h which was including module.h, but that path will be broken
soon.
[ with input from Stephen Rothwell <sfr@canb.auug.org.au> ]
Signed-off-by: Paul Gortmaker <paul.gortmaker@windriver.com>
We only really need the defaults in order to cut down the number of
registers we sync and to satisfy reads while the device is powered off
but not all devices are going to need to do that (always on devices like
PMICs being the prime example) so don't require those devices to supply
a default. Instead only try to fall back to hardware defaults if the
driver told us to.
Devices using LZO won't be able to instantiate with this, that will require
some updates in the LZO code to handle this case.
Signed-off-by: Mark Brown <broonie@opensource.wolfsonmicro.com>
Acked-by: Dimitris Papastamos <dp@opensource.wolfsonmicro.com>
If a register isn't cached then let callers know that so they can fall
back or error handle appropriately.
Signed-off-by: Mark Brown <broonie@opensource.wolfsonmicro.com>
Acked-by: Dimitris Papastamos <dp@opensource.wolfsonmicro.com>
Rather than open coding a binary search use the standard bsearch() using
the comparison function we're already using for sort() on insert. This
fixes a lockup I was observing due to iterating on min <= max rather
than min < max when we fail to look up.
Signed-off-by: Mark Brown <broonie@opensource.wolfsonmicro.com>
Acked-by: Dimitris Papastamos <dp@opensource.wolfsonmicro.com>
Since there are more lookups than insertions in a typical
scenario, optimize the linear search into a binary search. For
this to work, we need to keep reg_defaults sorted upon
insertions, for now be lazy and use sort().
Signed-off-by: Dimitris Papastamos <dp@opensource.wolfsonmicro.com>
Signed-off-by: Mark Brown <broonie@opensource.wolfsonmicro.com>
Ensure we've got a function so users can enable/disable the
cache bypass option.
Signed-off-by: Dimitris Papastamos <dp@opensource.wolfsonmicro.com>
Signed-off-by: Mark Brown <broonie@opensource.wolfsonmicro.com>
In preperation for the upcoming patches, modify map->cache_bypass
directly. The helper functions will grab an exclusive lock. Because
we'll have acquired the same lock we need to avoid a deadlock.
Signed-off-by: Dimitris Papastamos <dp@opensource.wolfsonmicro.com>
Signed-off-by: Mark Brown <broonie@opensource.wolfsonmicro.com>
We want to use regmap_write() to actually write anything
to the HW.
Signed-off-by: Dimitris Papastamos <dp@opensource.wolfsonmicro.com>
Signed-off-by: Mark Brown <broonie@opensource.wolfsonmicro.com>
In the absence of a sync callback, do it manually. This of course
can't take advantange of the specific optimizations of each
cache type but it will do well enough in most cases.
Signed-off-by: Dimitris Papastamos <dp@opensource.wolfsonmicro.com>
Signed-off-by: Mark Brown <broonie@opensource.wolfsonmicro.com>
This patch adds support for LZO compression when storing the register
cache.
For a typical device whose register map would normally occupy 25kB or 50kB
by using the LZO compression technique, one can get down to ~5-7kB. There
might be a performance penalty associated with each individual read/write
due to decompressing/compressing the underlying cache, however that should not
be noticeable. These memory benefits depend on whether the target architecture
can get rid of the memory occupied by the original register defaults cache
which is marked as __devinitconst. Nevertheless there will be some memory
gain even if the target architecture can't get rid of the original register
map, this should be around ~30-32kB instead of 50kB.
Signed-off-by: Dimitris Papastamos <dp@opensource.wolfsonmicro.com>
Signed-off-by: Mark Brown <broonie@opensource.wolfsonmicro.com>
This patch adds support for the rbtree cache compression type.
Each rbnode manages a variable length block of registers. There can be no
two nodes with overlapping blocks. Each block has a base register and a
currently top register, all the other registers, if any, lie in between these
two and in ascending order.
The reasoning behind the construction of this rbtree is simple. In the
snd_soc_rbtree_cache_init() function, we iterate over the register defaults
provided by the regcache core. For each register value that is non-zero we
insert it in the rbtree. In order to determine in which rbnode we need
to add the register, we first look if there is another register already
added that is adjacent to the one we are about to add. If that is the case
we append it in that rbnode block, otherwise we create a new rbnode
with a single register in its block and add it to the tree.
There are various optimizations across the implementation to speed up lookups
by caching the most recently used rbnode.
Signed-off-by: Dimitris Papastamos <dp@opensource.wolfsonmicro.com>
Tested-by: Lars-Peter Clausen <lars@metafoo.de>
Signed-off-by: Mark Brown <broonie@opensource.wolfsonmicro.com>
This is the simplest form of a cache available in regcache. Any
registers whose default value is 0 are ignored. If any of those
registers are modified in the future, they will be placed in the
cache on demand. The cache layout is essentially using the provided
register defaults by the regcache core directly and does not re-map
it to another representation.
Signed-off-by: Dimitris Papastamos <dp@opensource.wolfsonmicro.com>
Signed-off-by: Mark Brown <broonie@opensource.wolfsonmicro.com>
This patch introduces caching support for regmap. The regcache API
has evolved essentially out of ASoC soc-cache so most of the actual
caching types (except LZO) have been tested in the past.
The purpose of regcache is to optimize in time and space the handling
of register caches. Time optimization is achieved by not having to go
over a slow bus like I2C to read the value of a register, instead it is
cached locally in memory and can be retrieved faster. Regarding space
optimization, some of the cache types are better at packing the caches,
for e.g. the rbtree and the LZO caches. By doing this the sacrifice in
time still wins over doing I2C transactions.
Signed-off-by: Dimitris Papastamos <dp@opensource.wolfsonmicro.com>
Tested-by: Lars-Peter Clausen <lars@metafoo.de>
Signed-off-by: Mark Brown <broonie@opensource.wolfsonmicro.com>
Mark Brown