On 2/26/2020 4:09 AM, Mark Rutland wrote: > On Mon, Feb 24, 2020 at 12:57:36PM -0800, Elliot Berman wrote: >> Some implementors of PSCI may relax the requirements of the PSCI >> architectural warm reset. In order to comply with PSCI specification, a >> different reset_type value must be used. > > This reads as-if you're saying the firmware isn't spec compliant, and > this is a workaround in order to get the expected behaviour. > > Can you please elaborate on what you mean by "relax the requirements" > here? What's your firmware doing or not doing that you want to avoid? > >> The alternate PSCI SYSTEM_RESET2 may be used in all warm/soft reboot >> scenarios, replacing the architectural warm reset. > > I assume you mean SYSTEM_REET2's SYSTEM_WARM_RESET reset? Please call > that out explicitly by name -- it makes this easier to look up, and > if/when more architectural resets are added the commit message won't > become ambiguous. I can reword to: Some implementors of PSCI may wish to generally use a different reset type than SYSTEM_WARM_RESET. For instance, Qualcomm SoCs support an alternate reset_type which may be used in more warm reboot scenarios than SYSTEM_WARM_RESET permits (e.g. to reboot into recovery mode). > >> >> Signed-off-by: Elliot Berman <eberman@xxxxxxxxxxxxxx> >> --- >> Documentation/devicetree/bindings/arm/psci.yaml | 5 +++++ >> 1 file changed, 5 insertions(+) >> >> diff --git a/Documentation/devicetree/bindings/arm/psci.yaml b/Documentation/devicetree/bindings/arm/psci.yaml >> index 8ef8542..469256a2 100644 >> --- a/Documentation/devicetree/bindings/arm/psci.yaml >> +++ b/Documentation/devicetree/bindings/arm/psci.yaml >> @@ -102,6 +102,11 @@ properties: >> [1] Kernel documentation - ARM idle states bindings >> Documentation/devicetree/bindings/arm/idle-states.txt >> >> + arm,psci-sys-reset2-param: >> + $ref: /schemas/types.yaml#/definitions/uint32 >> + description: | >> + reset_param value to use during a warm or soft reboot. > > A "soft" reboot isn't a PSCI concept, so I'm worried this is just > hooking up magic values for Linux internals.> > I'd like to better understand what you're trying to achieve here. In Qualcomm use cases, we do not always want to preserve memory to caller's (i.e. Linux) exception level. For instance, crash recovery mode runs in higher exception level and would not continue booting into Linux except through a hard reset. Also, this early firmware doesn't have the ability to understand device tree or ACPI tables to know what memory to preserve. Per discussion with Sudeep and Charles, this use case violates PSCI specification for SYSTEM_WARM_RESET reset type, but would be appropriate for a vendor-specific reset type. Thus, Qualcomm firmware supports a vendor-specific reset type which does not have the requirement to preserve memory to caller's EL or to describe what memory is to be preserved in DT or ACPI. If this vendor-specific reset type is used, then firmware checks various registers (e.g. download mode [1]) to alter the restart flow (e.g. to enter recovery mode). If no alternate flow is requested, then firmware would boot back into Linux, preserving memory. [1]: https://git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git/tree/drivers/firmware/qcom_scm.c?h=v5.6-rc3#n1120 Thanks, Elliot -- The Qualcomm Innovation Center, Inc. is a member of the Code Aurora Forum, a Linux Foundation Collaborative Project
