On legacy pre-1.0 firmware versions so far the following message is
printed which may cause some confusion:
SCP Protocol 0.0 Firmware 0.0.0 version

Therefore replace the message with the following if firmware doesn't
provide usable version information:
SCP Protocol legacy pre-1.0 firmware
Tested-by: NKevin Hilman <khilman@baylibre.com>
Signed-off-by: NHeiner Kallweit <hkallweit1@gmail.com>
Signed-off-by: NSudeep Holla <sudeep.holla@arm.com>

Macro definitions can be simplified by making use of the FIELD_GET/_PREP
bitfield macros.
Tested-by: NKevin Hilman <khilman@baylibre.com>
Signed-off-by: NHeiner Kallweit <hkallweit1@gmail.com>
Signed-off-by: NSudeep Holla <sudeep.holla@arm.com>

One more single-element struct was left, remove it.
Tested-by: NKevin Hilman <khilman@baylibre.com>
Signed-off-by: NHeiner Kallweit <hkallweit1@gmail.com>
Signed-off-by: NSudeep Holla <sudeep.holla@arm.com>

At several positions in the code sparse complains about incorrect access
to __iomem annotated memory. Fix this and make sparse happy.
Tested-by: NKevin Hilman <khilman@baylibre.com>
Signed-off-by: NHeiner Kallweit <hkallweit1@gmail.com>
[Sudeep Holla: changed the patch title to describe the change]
Signed-off-by: NSudeep Holla <sudeep.holla@arm.com>

Both clk_get_value and sensor_value structures contains a single element
and hence needs no packing making the whole structure defination
unnecessary.

This patch gets rid of both those unnecessary structures.
Tested-by: NKevin Hilman <khilman@baylibre.com>
Reviewed-by: NHeiner Kallweit <hkallweit1@gmail.com>
Signed-off-by: NSudeep Holla <sudeep.holla@arm.com>

This patch drops the only present type cast of the SCPI payload pointer
to scpi_shared_mem inorder to align with other occurrences, IOW for
consistency.
Tested-by: NKevin Hilman <khilman@baylibre.com>
Reviewed-by: NHeiner Kallweit <hkallweit1@gmail.com>
Signed-off-by: NSudeep Holla <sudeep.holla@arm.com>

lo_val and hi_val together in this order are a little endian 64 bit value.
Therefore we can simplify struct sensor_value and the code by defining
it as a __le64 value and by using le64_to_cpu.
Tested-by: NKevin Hilman <khilman@baylibre.com>
Signed-off-by: NHeiner Kallweit <hkallweit1@gmail.com>
Signed-off-by: NSudeep Holla <sudeep.holla@arm.com>

By using FIELD_GET and proper masks we can avoid quite some shifting
and masking macro magic and make the code better readable.
Tested-by: NKevin Hilman <khilman@baylibre.com>
Signed-off-by: NHeiner Kallweit <hkallweit1@gmail.com>
Signed-off-by: NSudeep Holla <sudeep.holla@arm.com>

Making the header subfields members of struct dvfs_info allows to make
the code better readable and avoids some macro magic.

In addition remove a useless statement using info->latency.
Tested-by: NKevin Hilman <khilman@baylibre.com>
Signed-off-by: NHeiner Kallweit <hkallweit1@gmail.com>
Signed-off-by: NSudeep Holla <sudeep.holla@arm.com>

Replace two remaining functions in probe with their devm versions.
Tested-by: NKevin Hilman <khilman@baylibre.com>
Signed-off-by: NHeiner Kallweit <hkallweit1@gmail.com>
Signed-off-by: NSudeep Holla <sudeep.holla@arm.com>

Make freeing the mbox channels device-managed, thus further simplifying
scpi_remove and and one further step to get rid of scpi_remove.
Tested-by: NKevin Hilman <khilman@baylibre.com>
Signed-off-by: NHeiner Kallweit <hkallweit1@gmail.com>
Signed-off-by: NSudeep Holla <sudeep.holla@arm.com>

Both memory areas are free'd anyway when the device is destroyed,
so we don't have to do it manually.
Tested-by: NKevin Hilman <khilman@baylibre.com>
Signed-off-by: NHeiner Kallweit <hkallweit1@gmail.com>
Signed-off-by: NSudeep Holla <sudeep.holla@arm.com>

Revert "Merge tag 'scpi-updates-4.15' of git://git.kernel.org/pub/scm/linux/kernel/git/sudeep.holla/linux into next/drivers"

Paraphrased from email from Kevin Hilman:

Revert ARM SCPI changes since v4.14.

Untested changes caused regressions in SCPI and CPUfreq/DVFS failures
on most Amlogic SoCs.  Changes reverted for v4.15 so they can be better
reviewed and tested.

These ARM SCPI changes caused SCPI regressions resulting in CPUfreq
failures on most Amlogic SoCs (found by kernelci.org.)

Unfortunately, this was not caught in linux-next due to other bugs/panics
on these platforms masking this problem so we've only found it since
we've fixed the other issues.

Since we're already in the -rc cycle, I'd prefer to revert to a known
working state (that of v4.14) rather than finding/reverting a subset,
which would just lead to another untested state.

These changes can then have some time to be better reviewed and tested
and resubmitted for v4.16.

Kevin Hilman has tested this revert on the affected Amlogic SoCs and
verified that we're back to the previous (working) condition.

This reverts commit 6710acf2, reversing
changes made to 4b367f2e.
Reported-by: NKevin Hilman <khilman@baylibre.com>
Signed-off-by: NOlof Johansson <olof@lixom.net>

At several positions in the code sparse complains about incorrect access
to __iomem annotated memory. Fix this and make sparse happy.
Signed-off-by: NHeiner Kallweit <hkallweit1@gmail.com>
Signed-off-by: NSudeep Holla <sudeep.holla@arm.com>

Both clk_get_value and sensor_value structures contains a single element
and hence needs no packing making the whole structure defination
unnecessary.

This patch gets rid of both those unnecessary structures.
Reviewed-by: NHeiner Kallweit <hkallweit1@gmail.com>
Signed-off-by: NSudeep Holla <sudeep.holla@arm.com>

This patch drops the only present type cast of the SCPI payload pointer
to scpi_shared_mem inorder to align with other occurrences, IOW for
consistency.
Reviewed-by: NHeiner Kallweit <hkallweit1@gmail.com>
Signed-off-by: NSudeep Holla <sudeep.holla@arm.com>

lo_val and hi_val together in this order are a little endian 64 bit value.
Therefore we can simplify struct sensor_value and the code by defining
it as a __le64 value and by using le64_to_cpu.
Signed-off-by: NHeiner Kallweit <hkallweit1@gmail.com>
Signed-off-by: NSudeep Holla <sudeep.holla@arm.com>

By using FIELD_GET and proper masks we can avoid quite some shifting
and masking macro magic and make the code better readable.
Signed-off-by: NHeiner Kallweit <hkallweit1@gmail.com>
Signed-off-by: NSudeep Holla <sudeep.holla@arm.com>

Making the header subfields members of struct dvfs_info allows to make
the code better readable and avoids some macro magic.

In addition remove a useless statement using info->latency.
Signed-off-by: NHeiner Kallweit <hkallweit1@gmail.com>
Signed-off-by: NSudeep Holla <sudeep.holla@arm.com>

sysfs_create_groups and of_platform_populate can be replaced with the
device-managed versions what allows us to remove scpi_remove.
Signed-off-by: NHeiner Kallweit <hkallweit1@gmail.com>
Signed-off-by: NSudeep Holla <sudeep.holla@arm.com>

Make freeing the mbox channels device-managed, thus further simplifying
scpi_remove and and one further step to get rid of scpi_remove.
Signed-off-by: NHeiner Kallweit <hkallweit1@gmail.com>
Signed-off-by: NSudeep Holla <sudeep.holla@arm.com>

Pre-populating the dvfs info data in scpi_probe allows to make all
memory allocations device-managed. This helps to simplify scpi_remove
and eventually to get rid of scpi_remove completely.
Signed-off-by: NHeiner Kallweit <hkallweit1@gmail.com>
[sudeep.holla: changed to continue probe even if scpi_dvfs_populate_info fails]
Signed-off-by: NSudeep Holla <sudeep.holla@arm.com>

Both memory areas are free'd anyway when the device is destroyed,
so we don't have to do it manually.
Signed-off-by: NHeiner Kallweit <hkallweit1@gmail.com>
Signed-off-by: NSudeep Holla <sudeep.holla@arm.com>

There's no benefit using drvdata as variable scpi_info is global.

Setting scpi_info to NULL in scpi_remove isn't needed too. If arm_scpi
is built-in, then this code is never used. And if arm_scpi is built as
a module and some other module calls get_scpi_ops() then due to this
dependency scpi_remove is called only after the other module has been
removed.

Last but not least, users usually store the result of get_scpi_ops(),
therefore setting scpi_info to NULL wouldn't really help.
Signed-off-by: NHeiner Kallweit <hkallweit1@gmail.com>
[sudeep.holla: reworded the commit log a bit]
Signed-off-by: NSudeep Holla <sudeep.holla@arm.com>

scpi_device_{g,s}et_power_state correctly handles the conversion of
endianness for dev_id using cpu_to_le16. However dev_id is declared
as u16 in struct dev_pstate_set which is incorrect.

This patch fixes the endianness of dev_id in dev_pstate_set structure.

Fixes: 37a441dc ("firmware: arm_scpi: add support for device power state management")
Signed-off-by: NSudeep Holla <sudeep.holla@arm.com>

Now that we have a custom printf format specifier, convert users of
full_name to use %pOF instead. This is preparation to remove storing
of the full path string for each node.
Signed-off-by: NRob Herring <robh@kernel.org>
Acked-by: NSudeep Holla <sudeep.holla@arm.com>
Cc: Mark Rutland <mark.rutland@arm.com>
Cc: Lorenzo Pieralisi <lorenzo.pieralisi@arm.com>
Cc: linux-arm-kernel@lists.infradead.org
Signed-off-by: NArnd Bergmann <arnd@arndb.de>

Currently only CPU devices use the transition latency and the OPPs
populated in the SCPI driver. scpi-cpufreq has logic to handle these.
However, even GPU and other users of SCPI DVFS will need the same logic.

In order to avoid duplication, this patch adds support to get DVFS
transition latency and add all the OPPs to the device using OPP library
helper functions. The helper functions added here can be used for any
device whose DVFS are managed by SCPI.

Also, we also have incorrect dependency on the cluster identifier for
the CPUs. It's fundamentally wrong as the domain id need not match the
cluster id. This patch gets rid of that dependency by making use of the
clock bindings which are already in place.
Signed-off-by: NSudeep Holla <sudeep.holla@arm.com>

Instead of performing full initialization of the completion structure
on each transfer in scpi_send_message(), we initialize it at boot time
(more specifically, in the relevant probe() function) and use
reinit_completion() to reset ->done counter on each message transfer.
Signed-off-by: NAlexey Klimov <alexey.klimov@arm.com>
Signed-off-by: NSudeep Holla <sudeep.holla@arm.com>

The pre-1.0 SCPI firmwares are using single __le32 as sensor value,
while the SCPI v1.0 protocol uses two __le32 as sensor values(64bit)
split into 32bit upper and 32bit lower value.

Using an "struct sensor_value" to read the sensor value on a pre-1.0
SCPI firmware gives garbage in the "hi_val" field.

This patch fixes the issue by reading only the lower 32-bit value for
all pre-1.0 SCPI versions.
Suggested-by: NSudeep Holla <Sudeep.Holla@arm.com>
Signed-off-by: NMartin Blumenstingl <martin.blumenstingl@googlemail.com>
[sudeep.holla@arm.com: updated the commit log to reflect the implementation]
Signed-off-by: NSudeep Holla <sudeep.holla@arm.com>

This patch adds new DT match table to setup the support for SCPI protocol
versions prior to v1.0 releases. It also adds "arm,scpi-pre-1.0" to the
SCPI match entry.
Signed-off-by: NSudeep Holla <sudeep.holla@arm.com>

On Amlogic SCPI legacy implementation, the GET_CAPABILITIES command is
not supported, failover by using 0.0.0 version.
Signed-off-by: NNeil Armstrong <narmstrong@baylibre.com>
[sudeep.holla@arm.com: changed the subject]
Signed-off-by: NSudeep Holla <sudeep.holla@arm.com>

This patch adds support for the Legacy SCPI protocol that is available
in very early JUNO versions and shipped Amlogic ARMv8 based SoCs. Some
Rockchip SoC are also known to use this version of protocol with
extended vendor commands.

In order to support the legacy SCPI protocol variant, we need to add the
structures and macros definitions that varies against the final SCPI v1.0
specification.

We add the indirection table for legacy commands set so that it can
co-exist with the standard v1.0 command set. It also adds bitmap field
for channel selection since the legacy protocol mandates to send only
selected subset of the commands on the high priority channel.
Signed-off-by: NNeil Armstrong <narmstrong@baylibre.com>
[sudeep.holla@arm.com: Updated the changelog]
Signed-off-by: NSudeep Holla <sudeep.holla@arm.com>

Since Amlogic SoCs supports more than 8 OPPs per domains, we need increase
the OPP structure size.

This patch increases the MAX_DVFS_OPPS to 16.
Signed-off-by: NNeil Armstrong <narmstrong@baylibre.com>
Signed-off-by: NSudeep Holla <sudeep.holla@arm.com>

Since the legacy SCPI and the SCPI v1.0 differ in the command values,
it's better to create some sort of command indirection in the driver
to avoid repeated version check at multiple places.

This patch adds the indirection command table to allow different values
of the command across SCPI versions.

[narmstrong@baylibre.com: added cmd check in scpi_send_message]
Signed-off-by: NNeil Armstrong <narmstrong@baylibre.com>
Signed-off-by: NSudeep Holla <sudeep.holla@arm.com>

of_node_put needs to be called when the device node which is got
from of_parse_phandle has finished using it.

Besides, of_address_to_resource always returns -EINVAL for error, delete
the assignment for ret.
Signed-off-by: NPeter Chen <peter.chen@nxp.com>
Signed-off-by: NSudeep Holla <sudeep.holla@arm.com>

SCPI protocol supports device power state management. This deals with
power states of various peripheral devices in the system other than the
core compute subsystem.

This patch adds support for the power state management of those
peripheral devices.
Tested-by: NMathieu Poirier <mathieu.poirier@linaro.org>
Tested-by: NJon Medhurst <tixy@linaro.org>
Reviewed-by: NJon Medhurst <tixy@linaro.org>
Reviewed-by: NKevin Hilman <khilman@baylibre.com>
Reviewed-by: NUlf Hansson <ulf.hansson@linaro.org>
Signed-off-by: NSudeep Holla <sudeep.holla@arm.com>

scpi_get_sensor_value like other scpi operations needs to be static.
This patch marks it as static to be consistent with others.
Signed-off-by: NSudeep Holla <sudeep.holla@arm.com>

dvfs_get packed structure is used to read the DVFS/OPP index from the
firmware. It just contains a single byte that needs no packing making
the whole structure defination unnecessary.

This patch replaces the unnecessary dvfs_get packed structure with an
unsigned byte.
Signed-off-by: NSudeep Holla <sudeep.holla@arm.com>

SCPI specification version 1.1 extended the sensor from 32-bit to 64-bit
values in order to accommodate new sensor class with 64-bit requirements

Since the SCPI driver sets the higher 32-bit for older protocol version
to zeros, there's no need to explicitly check the SCPI protocol version
and the backward compatibility is maintainted.
Acked-by: NGuenter Roeck <linux@roeck-us.net>
Reviewed-by: NPunit Agrawal <punit.agrawal@arm.com>
Signed-off-by: NSudeep Holla <sudeep.holla@arm.com>

Currently we have Tx timeout of 50ms while Rx timeout of 20 ms. Tx state
machine is maintained by the mailbox framework and Rx by SCPI driver.

It is possible that before msg_submit call tx_prepare(because of other
message in the queue and the channel being active), wait for completion
in scpi_send_message times out and the buffers are freed. In that case
when Tx state machine timer goes off later, poll_txdone calls
scpi_tx_prepare on that message, which adds it to the rx_pending list,
corrupting link pointers.

This patch reduces the Tx timeout to 20ms and increases Rx timeout to
30ms to avoid the above mentioned issue.
Reported-by: NJon Medhurst (Tixy) <tixy@linaro.org>
Signed-off-by: NSudeep Holla <sudeep.holla@arm.com>