SCMI Clock Test 17, Attributes Parameter is incorrect

See merge request !53

Balaji Gontumukkala authored Feb 19, 2025 and balaji-gontumukkala committed Feb 19, 2025

 - Added mask to attributes parameter to get only clock state from
   clock attributes saved value.

Handle unsupported SCMI message

See merge request !52

 - In the SCMI handler table, unsupported messages are indicated by leaving the
   corresponding function pointer unassigned. However, there is no validation
   check on entry which leads to potential null pointer dereference while handling
   message.
 - This change explicitly initializes the handler table and validates
   entry in runtime to determine whether a given message is supported

Signed-off-by: Balaji Gontumukkala <balaji.gontumukkala@arm.com>

tc3: update cpu names to public ones

See merge request !51

update the expected SCP version.
update the expected performance domain names.
update the expected performance domain number for GPU on tc4.

Signed-off-by: Gourish Kondeti <Gourish.kondeti@arm.com>

Fixed static checks

See merge request !50

Signed-off-by: Balaji Gontumukkala <balaji.gontumukkala@arm.com>

tc: Update expected num_power_domains to be 1

See merge request !49

Miscellaneous fixes targeting SCMI v3.2

See merge request !47

Since OSPM will no longer have access to CPUs,
change expected power domains to be 1; the GPU.

Signed-off-by: Gourish Kondeti <Gourish.kondeti@arm.com>

FUNCTION_SELECT does NOT exist anymore in v3.2 while SETTINGS_GET and
SETTINGS_CONFIGURE have a new syntax to accomodate also FUNCTION_SELECT.

Signed-off-by: Cristian Marussi <cristian.marussi@arm.com>

Use version to configure the expected number or zeroed reserved bits in
power test_p007, test_p008, test_p0012, test_p0013.

Signed-off-by: Cristian Marussi <cristian.marussi@arm.com>

Squash 2 different test cases, previously only slightly differing
between versions, into one single testcase that takes version as
a parameter.

Signed-off-by: Cristian Marussi <cristian.marussi@arm.com>

Squash 3 different test cases, previously only slightly differing
between versions, into one single testcase that takes version as
a parameter.

Signed-off-by: Cristian Marussi <cristian.marussi@arm.com>

When Powercap protocol implementation is greater than or equal to
0x20000, CAP_SET can accept a zero value to signify disable powercapping
on that domain.

Signed-off-by: Cristian Marussi <cristian.marussi@arm.com>

Skip asynchronous CAP_SET test for powercap domains NOT supporting
asynchronous cap setting and make sure that no delayed response is
set when an asynchronus CAP_SET with the Ignore flag is requested.
Al fix the test to check tagainst powercap protocol instead of clock.

Signed-off-by: Cristian Marussi <cristian.marussi@arm.com>

A powercap domain is invalid if BOTH cap_config and monitoring supports
are false as specified in 4.10.3.5; pai_config, though, is allowed to be
unsupported.

Signed-off-by: Cristian Marussi <cristian.marussi@arm.com>

Receiving a DENY on a CLOCK_RATE_SET around a specific domain is an
acceptable PASS: in such a case, keep looping on the remaining clock
domains, carrying on more tests, instead of bailing out with PASS.

Also check header consistency at first, like in other test cases.

Signed-off-by: Cristian Marussi <cristian.marussi@arm.com>

Choose the invalid skip_index to use depending on how the test suite was
invoked: when running with relaxed version checking use the total number
of protocols advertised by the platform.
This will allow this test to be run successfully also against platform
implementing new, unknown protocols.

Signed-off-by: Cristian Marussi <cristian.marussi@arm.com>

Helper to check if the test suite has been invoked with relaxed version
checking enabled.

Signed-off-by: Cristian Marussi <cristian.marussi@arm.com>

Some testcases vary depending on the version advertised by the platform;
when running with relaxed version checking (-r), checking the current
version with val_protocol_version_relaxed() returns true when the version
advertised by the platform is equal or greater than the expected one.

In this way you can run a test even against a newer SCMI server, which is
possibly still not fully supported by the ACS test-suite: full backward
compatibility cannot be assured, but you can run the testcases.

In order for all of this to work, relaxed protocol versions should be
checked starting from the latest/newest version supported.

When running without using relaxed checks, val_protocol_version_relaxed()
fallbacks to strict version-checking, so the ordering  of th checks is
unimportant.

Signed-off-by: Cristian Marussi <cristian.marussi@arm.com>

Issuing a POWER_DOMAIN_NAME_GET against an existent domain which does
not support extended names should return SCMI_NOT_SUPPORTED.

Fix accordingly.

Signed-off-by: Cristian Marussi <cristian.marussi@arm.com>

Test was mistakenly using POWERCAP protocol.

Signed-off-by: Cristian Marussi <cristian.marussi@arm.com>

VOLTAGE_DOMAIN_ATTRIBUTES returns, amongst other things, if a domain
supports an extended name: in such a case the reply from will contain ONLY
the first 15 chars of the extended name.

When supported, instead, the extended domain name can be queried by issuing
an additional dedicated VOLTAGE_DOMAIN_NAME_GET command.

The current check against ext_name_supp makes no sense from the
specification point of view and it also causes a segfault when running
against TARGET generic_scmi.

Fix by removing the needless check.

Signed-off-by: Cristian Marussi <cristian.marussi@arm.com>

POWERCAP_DOMAIN_ATTRIBUTES returns, amongst other things, if a domain
supports an extended name: in such a case the reply from will contain ONLY
the first 15 chars of the extended name.

When supported, instead, the extended domain name can be queried by issuing
an additional dedicated POWERCAP_DOMAIN_NAME_GET command.

The current check against ext_name_supp makes no sense from the
specification point of view and it also causes a segfault when running
against TARGET generic_scmi.

Fix by removing the needless check.

Signed-off-by: Cristian Marussi <cristian.marussi@arm.com>

A missing initializer on a local automatic variable causes the linux app
to segfault when running these clock testcases.

Signed-off-by: Cristian Marussi <cristian.marussi@arm.com>

Update JUNO expectation to latest v2.15.0 SCP fw.

Signed-off-by: Cristian Marussi <cristian.marussi@arm.com>

Add minimal dummy Powercap expectations on JUNO to avoid break the build
when compiling Powercap against JUNO.

Signed-off-by: Cristian Marussi <cristian.marussi@arm.com>

On a POSIX system read/write syscalls can return successfully even when
not all the requested bytes were not read or written.

Introduce 2 wrapper functions to make sure that the read completes only
on EOF (0 bytes) and the write only when the full buffer has been
written out, while still bailing on errors other than -EINTR.

Make use of these new helpers in place of the bare read/write invocation.

Signed-off-by: Cristian Marussi <cristian.marussi@arm.com>

Changes for SCMI v3.2 BET-1 release

See merge request !46

 - Added NEGOTIATE_PROTOCOL_VERSION command to all protocols and incremented version of all protocols
 - Updated Clock tests to support clock protocol version 3.0
 - Changes in Pin control protocol to support SCMI v3.2 updates

Signed-off-by: Balaji Gontumukkala <balaji.gontumukkala@arm.com>

Added warning message for num of domains

See merge request !44

 - warning message to highlight number of domains
   in system exceeds the number supported by ACS
 - Added a  print when a platform return platform specific
   protocol ID

Signed-off-by: Balaji Gontumukkala <balaji.gontumukkala@arm.com>

Changes for SCMI v3.2 BETA-0 release

See merge request !43

Update .gitlab-ci.yml file

See merge request !42