In preparation to enable building scmi as a single module, let us move
the scmi bus {de-,}initialisation call into the driver.

The main reason for this is to keep it simple instead of maintaining
it as separate modules and dealing with all possible initcall races
and deferred probe handling. We can move it as separate modules if
needed in future.

Link: https://lore.kernel.org/r/20200907195046.56615-3-sudeep.holla@arm.comTested-by: NCristian Marussi <cristian.marussi@arm.com>
Signed-off-by: NSudeep Holla <sudeep.holla@arm.com>

Add SCMI device for system power protocol by just adding the name "syspower"
to the list of supported core protocol devices.

Link: https://lore.kernel.org/r/20200907174657.32466-3-cristian.marussi@arm.comSigned-off-by: NCristian Marussi <cristian.marussi@arm.com>
Signed-off-by: NSudeep Holla <sudeep.holla@arm.com>

Event reports are used to convey information describing events to the
registered user-callbacks: they are necessarily derived from the underlying
raw SCMI events' messages but they are not meant to expose or directly
mirror any of those messages data layout, which belong to the protocol
layer.

Using fixed size types for report fields, mirroring messages structure,
is at odd with this: get rid of them using more generic, equivalent,
typing.

Substitute scmi_event_header fixed size fields with generic types too and
shuffle around fields definitions to minimize implicit padding while
adapting involved functions.

Link: https://lore.kernel.org/r/20200710133919.39792-3-cristian.marussi@arm.comSigned-off-by: NCristian Marussi <cristian.marussi@arm.com>
Signed-off-by: NSudeep Holla <sudeep.holla@arm.com>

Initialize and enable SCMI notifications core support during bus/driver
probe phase, so that protocols can start registering their supported
events during their initialization.

Link: https://lore.kernel.org/r/20200701155348.52864-5-cristian.marussi@arm.comReviewed-by: NJonathan Cameron <Jonathan.Cameron@huawei.com>
Signed-off-by: NCristian Marussi <cristian.marussi@arm.com>
Signed-off-by: NSudeep Holla <sudeep.holla@arm.com>

Commit e5bfb21d ("firmware: smccc: Add HAVE_ARM_SMCCC_DISCOVERY to
identify SMCCC v1.1 and above") introduced new config option to identify
the availability of SMCCC discoverability of version and features
transparently hiding the indirect dependency on ARM_PSCI_FW.

Commit 5a897e3a ("firmware: arm_scmi: fix psci dependency") just
worked around the build dependency making SCMI SMC/HVC transport depend
on ARM_PSCI_FW at the time. Since it really just relies on SMCCC directly
and not on ARM_PSCI_FW, let us move to use CONFIG_HAVE_ARM_SMCCC_DISCOVERY
instead of CONFIG_ARM_PSCI_FW.

Link: https://lore.kernel.org/r/20200625101937.51939-1-sudeep.holla@arm.com
Cc: Peng Fan <peng.fan@nxp.com>
Cc: Arnd Bergmann <arnd@arndb.de>
Signed-off-by: NSudeep Holla <sudeep.holla@arm.com>

Currently the trace event 'scmi_xfer_end' reports the status of the
transfer using the unsigned status field read from the firmware which
may not be easy to interpret. It may also miss to emit any timeouts
that happen in the driver resulting in emitting garbage in the status
field in those scenarios.

Let us use signed integer so that error values are emitted out after
they are mapped from firmware error formats to standard linux error
codes. While at this, also include any timeouts in the driver itself.

Link: https://lore.kernel.org/r/20200609134503.55860-1-sudeep.holla@arm.com
Cc: Jim Quinlan <james.quinlan@broadcom.com>
Cc: Lukasz Luba <lukasz.luba@arm.com>
Reviewed-by: NLukasz Luba <lukasz.luba@arm.com>
Signed-off-by: NSudeep Holla <sudeep.holla@arm.com>

When CONFIG_ARM_PSCI_FW is disabled but CONFIG_HAVE_ARM_SMCCC is enabled,
arm-scmi runs into a link failure:

arm-linux-gnueabi-ld: drivers/firmware/arm_scmi/smc.o: in function `smc_send_message':
smc.c:(.text+0x200): undefined reference to `arm_smccc_1_1_get_conduit'

Change from HAVE_ARM_SMCCC to ARM_PSCI_FW config dependency for now.
We rely on PSCI bindings anyways for the conduit and this should be
fine.

Link: https://lore.kernel.org/r/20200507144905.11397-1-sudeep.holla@arm.com
Fixes: 1dc65580 ("firmware: arm_scmi: Add smc/hvc transport")
Signed-off-by: NSudeep Holla <sudeep.holla@arm.com>

Upon reception of an unexpected bogus delayed response, clear the channel
and bail-out safely.

Link: https://lore.kernel.org/r/20200420152315.21008-5-cristian.marussi@arm.com
Fixes: 4d09852b ("firmware: arm_scmi: Add support for notifications message processing")
Signed-off-by: NCristian Marussi <cristian.marussi@arm.com>
Signed-off-by: NSudeep Holla <sudeep.holla@arm.com>

Clear channel properly when done processing a delayed response.
This will let the platform firmware know that the channel is now free to
use it for any new delayed response or notification.

Link: https://lore.kernel.org/r/20200420152315.21008-4-cristian.marussi@arm.comSigned-off-by: NCristian Marussi <cristian.marussi@arm.com>
(sudeep.holla: Updated commit log to reflect that channel is now free for
 platform to use)
Signed-off-by: NSudeep Holla <sudeep.holla@arm.com>

When an unexpected response message is received we currently warn the user
and bail-out, ensure to also free the channel by invoking the transport
independent operation .clear_channel()

Link: https://lore.kernel.org/r/20200420152315.21008-3-cristian.marussi@arm.comSigned-off-by: NCristian Marussi <cristian.marussi@arm.com>
Signed-off-by: NSudeep Holla <sudeep.holla@arm.com>

SCMI transport operation .clear_notification() is indeed a generic method
to clear the channel in a transport dependent way, as such it could be a
useful helper also in other contexts.

Rename such method as .clear_channel(), renaming accordingly also its
already existent call-sites.

No functional change.

Link: https://lore.kernel.org/r/20200420152315.21008-2-cristian.marussi@arm.comSigned-off-by: NCristian Marussi <cristian.marussi@arm.com>
Signed-off-by: NSudeep Holla <sudeep.holla@arm.com>

Add the mechanisms to distinguish notifications from delayed responses
and command responses. Also add support to properly fetch notification
messages upon reception. Notifications processing does not continue any
further after the fetch phase.

Link: https://lore.kernel.org/r/20200327143438.5382-5-cristian.marussi@arm.comReviewed-by: NJonathan Cameron <Jonathan.Cameron@huawei.com>
[Reworked/renamed scmi_handle_xfer_delayed_resp()]
Signed-off-by: NCristian Marussi <cristian.marussi@arm.com>
Signed-off-by: NSudeep Holla <sudeep.holla@arm.com>

With all the plumbing in place, let's just add the separate dedicated
receive buffers to handle notifications that can arrive asynchronously
from the platform firmware to OS.

Also add one check to see if the platform supports any receive channels
before allocating the receive buffers: since those buffers are optionally
supported though, the whole xfer initialization is also postponed to be
able to check for their existence in advance.

Link: https://lore.kernel.org/r/20200327143438.5382-2-cristian.marussi@arm.comReviewed-by: NJonathan Cameron <Jonathan.Cameron@huawei.com>
[Changed parameters in __scmi_xfer_info_init()]
Signed-off-by: NCristian Marussi <cristian.marussi@arm.com>
Signed-off-by: NSudeep Holla <sudeep.holla@arm.com>

Use the value of "arm,smc-id" property from the device tree as the first
argument for SMCCC call leaving all the other arguments as zero for now.

There is no Rx, only Tx because of smc/hvc not support Rx.

Link: https://lore.kernel.org/r/1583673879-20714-3-git-send-email-peng.fan@nxp.comSigned-off-by: NPeng Fan <peng.fan@nxp.com>
[sudeep.holla: reworded commit log/subject and fixed !HAVE_ARM_SMCCC build]
Signed-off-by: NSudeep Holla <sudeep.holla@arm.com>

The SCMI specification is fairly independent of the transport protocol,
which can be a simple mailbox (already implemented) or anything else.
The current Linux implementation however is very much dependent on the
mailbox transport layer.

This patch makes the SCMI core code (driver.c) independent of the
mailbox transport layer and moves all mailbox related code to a new
file: mailbox.c and all struct shared_mem related code to a new file:
shmem.c.

We can now implement more transport protocols to transport SCMI
messages.

The transport protocols just need to provide struct scmi_transport_ops,
with its version of the callbacks to enable exchange of SCMI messages.
Signed-off-by: NViresh Kumar <viresh.kumar@linaro.org>
Link: https://lore.kernel.org/r/8698a3cec199b8feab35c2339f02dc232bfd773b.1580448239.git.viresh.kumar@linaro.orgSigned-off-by: NSudeep Holla <sudeep.holla@arm.com>

Move message header specific macros and helper routines to common.h as
they will be used outside of driver.c in a later commit.
Signed-off-by: NViresh Kumar <viresh.kumar@linaro.org>
Link: https://lore.kernel.org/r/6615db480370719b0a0241447a5f3feb8eea421f.1580448239.git.viresh.kumar@linaro.orgSigned-off-by: NSudeep Holla <sudeep.holla@arm.com>

Fix minor formatting issues with the doc style comments.
Signed-off-by: NViresh Kumar <viresh.kumar@linaro.org>
Link: https://lore.kernel.org/r/1bff7c0d1ad2c8b6eeff9660421f414f8c612eb2.1580448239.git.viresh.kumar@linaro.orgSigned-off-by: NSudeep Holla <sudeep.holla@arm.com>

The SCMI transport layer communicates via mailboxes and shared memory with
firmware running on a microcontroller. It is platform specific how long it
takes to pass a SCMI message. The most sensitive requests are coming from
CPUFreq subsystem, which might be used by the scheduler.
Thus, there is a need to measure these delays and capture anomalies.
This change introduces trace events wrapped around transfer code.

According to Jim's suggestion a unique transfer_id is to distinguish
similar entries which might have the same message id, protocol id and
sequence. This is a case then there are some timeouts in transfers.
Suggested-by: NJim Quinlan <james.quinlan@broadcom.com>
Signed-off-by: NLukasz Luba <lukasz.luba@arm.com>
Signed-off-by: NSudeep Holla <sudeep.holla@arm.com>

Platform drivers now have the option to have the platform core create
and remove any needed sysfs attribute files. Using the same, let's add
the scmi firmware and protocol version attributes as well as vendor and
sub-vendor identifiers to sysfs.

It helps to identify the firmware details from the sysfs entries similar
to ARM SCPI implementation.
Reviewed-by: NCristian Marussi <cristian.marussi@arm.com>
Signed-off-by: NSudeep Holla <sudeep.holla@arm.com>

Now that scmi bus provides option to create named scmi device, let us
create the default devices with names. This will help to add names for
matching to respective drivers and eventually to add multiple devices
and drivers per protocol.
Reviewed-by: NCristian Marussi <cristian.marussi@arm.com>
Signed-off-by: NSudeep Holla <sudeep.holla@arm.com>

Now that the scmi bus supports adding multiple devices per protocol,
and since scmi_create_protocol_device calls scmi_mbox_chan_setup,
we must avoid allocating and initialising the mbox channel if it is
already initialised.
Reviewed-by: NCristian Marussi <cristian.marussi@arm.com>
Signed-off-by: NSudeep Holla <sudeep.holla@arm.com>

Currently only one scmi device is created for each protocol enumerated.
However, there is requirement to make use of some procotols by multiple
kernel subsystems/frameworks. One such example is SCMI PERFORMANCE
protocol which can be used by both cpufreq and devfreq drivers.
Similarly, SENSOR protocol may be used by hwmon and iio subsystems,
and POWER protocol may be used by genpd and regulator drivers.

In order to achieve that, let us extend the scmi bus to match based
not only protocol id but also the scmi device name if one is available.
Reviewed-by: NCristian Marussi <cristian.marussi@arm.com>
Signed-off-by: NSudeep Holla <sudeep.holla@arm.com>

Messages that are sent to platform, also known as commands and can be:

1. Synchronous commands that block the channel until the requested work
has been completed. The platform responds to these commands over the
same channel and hence can't be used to send another command until the
previous command has completed.

2. Asynchronous commands on the other hand, the platform schedules the
requested work to complete later in time and returns almost immediately
freeing the channel for new commands. The response indicates the success
or failure in the ability to schedule the requested work. When the work
has completed, the platform sends an additional delayed response message.

Using the same transmit buffer used for sending the asynchronous command
even for the delayed response corresponding to it simplifies handling of
the delayed response. It's the caller of asynchronous command that is
responsible for allocating the completion flag that scmi driver can
complete to indicate the arrival of delayed response.
Signed-off-by: NSudeep Holla <sudeep.holla@arm.com>

In order to identify the message type when a response arrives, we need
a mechanism to unpack the message header similar to packing. Let's
add one.
Signed-off-by: NSudeep Holla <sudeep.holla@arm.com>

Currently we pre-allocate transmit buffers only and use the first free
slot in that pre-allocated buffer for transmitting any new message that
are generally originated from OS to the platform firmware.

Notifications or the delayed responses on the other hand are originated
from the platform firmware and consumes by the OS. It's better to have
separate and dedicated pre-allocated buffers to handle the notifications.
We can still use the transmit buffers for the delayed responses.

In addition, let's prepare existing scmi_xfer_{get,put} for acquiring
and releasing a slot to identify the right(tx/rx) buffers.
Signed-off-by: NSudeep Holla <sudeep.holla@arm.com>

With scmi_mbox_chan_setup enabled to identify and setup both Tx and Rx,
let's consolidate setting up of both the channels under the function
scmi_mbox_txrx_setup.

Since some platforms may opt not to support notifications or delayed
response, they may not need support for Rx. Hence Rx is optional and
failure of setting one up is not considered fatal.
Signed-off-by: NSudeep Holla <sudeep.holla@arm.com>

The transmit(Tx) channels are specified as the first entry and the
receive(Rx) channels are the second entry as per the device tree
bindings. Since we currently just support Tx, index 0 is hardcoded at
all required callsites.

In order to prepare for adding Rx support, let's remove those hardcoded
index and add boolean parameter to identify Tx/Rx channels when setting
them up.
Signed-off-by: NSudeep Holla <sudeep.holla@arm.com>

Re-shuffling few functions to keep definitions and their usages close.
This is also needed to avoid too many unnecessary forward declarations
while adding new features(delayed response and notifications).

Keeping this separate to avoid mixing up of these trivial change that
doesn't affect functionality into the ones that does.
Signed-off-by: NSudeep Holla <sudeep.holla@arm.com>

Sometimes platfom may take too long to respond to the command and OS
might timeout before platform transfer the ownership of the shared
memory region to the OS with the response.

Since the mailbox channel associated with the channel is freed and new
commands are dispatch on the same channel, OS needs to wait until it
gets back the ownership. If not, either OS may end up overwriting the
platform response for the last command(which is fine as OS timed out
that command) or platform might overwrite the payload for the next
command with the response for the old.

The latter is problematic as platform may end up interpretting the
response as the payload. In order to avoid such race, let's wait until
the OS gets back the ownership before we prepare the shared memory with
the payload for the next command.
Reported-by: NJim Quinlan <james.quinlan@broadcom.com>
Signed-off-by: NSudeep Holla <sudeep.holla@arm.com>

In preparation to adding support for other two types of messages that
SCMI specification mentions, let's replace the term 'command' with the
correct term 'message'.

As per the specification the messages are of 3 types:
commands(synchronous or asynchronous), delayed responses and notifications.
Signed-off-by: NSudeep Holla <sudeep.holla@arm.com>

While adding new comments found couple of typos that are better fixed.

s/informfation/information/
s/statues/status/
Signed-off-by: NSudeep Holla <sudeep.holla@arm.com>

scmi_xfer_get_init ensures both transmit and receive buffer lengths are
within the maximum limits. If receive buffer length is not supplied by
the caller, it's set to the maximum limit value. Receive buffer length
is never modified after that. So there's no need for the extra check
when receive transmit completion for a command essage.

Further, if the response header length is greater than the prescribed
receive buffer length, the response buffer is truncated to the latter.
Reported-by: NJim Quinlan <james.quinlan@broadcom.com>
Signed-off-by: NSudeep Holla <sudeep.holla@arm.com>

of_match_device can return NULL if no matching device is found though
it's highly unlikely to happen in scmi_probe as it's called only if
a valid match is found.

However we can use of_device_get_match_data() instead of
of_match_device()->data to handle NULL pointer checks and return -EINVAL
in such a scenario.
Reviewed-by: NSteven Price <steven.price@arm.com>
Signed-off-by: NAditya Pakki <pakki001@umn.edu>
Signed-off-by: NSudeep Holla <sudeep.holla@arm.com>

of_parse_phandle_with_args() requires the caller to call of_node_put() on
the returned args->np pointer. Otherwise the reference count will remain
incremented.

However, in this case, since we don't actually use the returned pointer,
we can simply pass in NULL.

Fixes: aa4f886f ("firmware: arm_scmi: add basic driver infrastructure for SCMI")
Signed-off-by: NSteven Price <steven.price@arm.com>
Signed-off-by: NSudeep Holla <sudeep.holla@arm.com>

Yet another nasty indentation left out during code restructuring. It's
must simpler to return on error instead of having unnecessary indentation.
Suggested-by: NJonathan Cameron <Jonathan.Cameron@huawei.com>
Signed-off-by: NSudeep Holla <sudeep.holla@arm.com>

kcalloc zeros the memory and it's totally unnecessary to zero the bitmap
again using bitmap_zero. This patch just drops the unnecessary use of
the bitmap_zero in the context.
Signed-off-by: NSudeep Holla <sudeep.holla@arm.com>

Just after the initial patches were queued, Jonathan Cameron mentioned
that scmi_one_xfer_{get_put} were not very clear and suggested to use
scmi_xfer_{alloc,free}. While I agree to some extent, the reason not to
have alloc/free as these are preallocated buffers and these functions
just returns a reference to free slot in that preallocated array.
However it was agreed to drop "_one" as it's implicit that we are always
dealing with one slot anyways.

This patch updates the name accordingly dropping "_one" in both {get,put}
functions. Also scmi_one_xfer_init is renamed as scmi_xfer_get_init to
reflect the fact that it gets the free slots and then initialise it.
Reported-by: NJonathan Cameron <jonathan.cameron@huawei.com>
Signed-off-by: NSudeep Holla <sudeep.holla@arm.com>

There are few missing descriptions for function parameters and structure
members along with certain instances where excessive function parameters
or structure members are described.

This patch fixes all of those warnings.
Reported-by: NJonathan Cameron <Jonathan.Cameron@huawei.com>
Signed-off-by: NSudeep Holla <sudeep.holla@arm.com>

By using FIELD_{FIT,GET,PREP} and GENMASK macro accessors we can avoid
some clumpsy custom shifting and masking macros and also improve the
code better readability.
Signed-off-by: NSudeep Holla <sudeep.holla@arm.com>

We need to return here instead of setting up the freed sdev device as a
transport.

Fixes: 907b6d14 ("firmware: arm_scmi: add per-protocol channels support using idr objects")
Signed-off-by: NDan Carpenter <dan.carpenter@oracle.com>
Signed-off-by: NSudeep Holla <sudeep.holla@arm.com>