| Message ID | 20230710140337.1434060-1-michal.wilczynski@intel.com |
|---|---|
| Headers | show |
| Series | Prefer using _OSC method over deprecated _PDC | expand |
On Mon, Jul 10, 2023 at 4:06 PM Michal Wilczynski <michal.wilczynski@intel.com> wrote: > > ACPI 3.0 introduced a new Operating System Capabilities _OSC control > method. This method is similar to _PDC, which was marked as deprecated > in ACPI 3.0. > > Prefer using _OSC method over deprecated _PDC in the acpi_bus_init(). In > case of the failure of the _OSC, try using _PDC as a fallback. > > Testing done: > Tested on physical server with BIOS implementing _OSC methods. In this > case acpi_processor_osc() was executed for each CPU core. acpi_run_osc() > returned success indicating that _OSC method succeeded. > > Tested on qemu VM to check whether the code would work on a SeaBios (the > default for qemu, doesn't support _OSC methods, or _PDC). This way I was > able to see how code behaves in case BIOS doesn't implement _OSC. In > that case the function > acpi_run_osc() returned failure, which propagated all the way up to > acpi_early_processor_osc(). The logic responsible for triggering _PDC > execution was triggered correctly. > > Tested this using debug messages with printk. > > v4: > - move setting processor capabilities bits into arch code > - move workaround for mwait to acpi/x86 directory > - rename ACPI_PDC* constants to more generic ACPI_PROC_CAP* > - introduce new function acpi_early_processor_control_setup() > > v3: > - split into more commits to make review easier > - changed "_UID" to METHOD_NAME_UID > - changed hard-coded constant to ACPI_PDC_COLLAB_PROC_PERF > - added Suggested-by tags > - fixed style issues > - fixed whitespaces > - refactored code > v2: > - fixed compilation issues on ia64 and arm > > Michal Wilczynski (9): > acpi: Move mwait quirk out of acpi_processor.c > acpi: Move processor_physically_present() to acpi_processor.c > acpi: Refactor arch_acpi_set_pdc_bits() > acpi: Rename ACPI_PDC constants > acpi: Clear C_C2C3_FFH and C_C1_FFH in arch_acpi_set_proc_cap_bits() > acpi: Move setting CAP_SMP_T_SWCOORD to arch_acpi_set_proc_cap_bits() > acpi: Introduce acpi_processor_osc() > acpi: Use _OSC method to convey processor OSPM capabilities > acpi: Remove acpi_hwp_native_thermal_lvt_osc() I have made a number of changes to the patches (edited subjects, rewrote changelogs, fixed up whitespace in a few places, changed the return value of acpi_early_processor_osc() to bool and made it static etc.) and tentatively queued them up for 6.6. Please see the bleeding-edge branch of linux-pm.git. I have made sure that the series will still compile at least on x86-64, but testing it again would be appreciated. Thanks!
On 7/14/2023 6:12 PM, Rafael J. Wysocki wrote: > On Mon, Jul 10, 2023 at 4:06 PM Michal Wilczynski > <michal.wilczynski@intel.com> wrote: >> ACPI 3.0 introduced a new Operating System Capabilities _OSC control >> method. This method is similar to _PDC, which was marked as deprecated >> in ACPI 3.0. >> >> Prefer using _OSC method over deprecated _PDC in the acpi_bus_init(). In >> case of the failure of the _OSC, try using _PDC as a fallback. >> >> Testing done: >> Tested on physical server with BIOS implementing _OSC methods. In this >> case acpi_processor_osc() was executed for each CPU core. acpi_run_osc() >> returned success indicating that _OSC method succeeded. >> >> Tested on qemu VM to check whether the code would work on a SeaBios (the >> default for qemu, doesn't support _OSC methods, or _PDC). This way I was >> able to see how code behaves in case BIOS doesn't implement _OSC. In >> that case the function >> acpi_run_osc() returned failure, which propagated all the way up to >> acpi_early_processor_osc(). The logic responsible for triggering _PDC >> execution was triggered correctly. >> >> Tested this using debug messages with printk. >> >> v4: >> - move setting processor capabilities bits into arch code >> - move workaround for mwait to acpi/x86 directory >> - rename ACPI_PDC* constants to more generic ACPI_PROC_CAP* >> - introduce new function acpi_early_processor_control_setup() >> >> v3: >> - split into more commits to make review easier >> - changed "_UID" to METHOD_NAME_UID >> - changed hard-coded constant to ACPI_PDC_COLLAB_PROC_PERF >> - added Suggested-by tags >> - fixed style issues >> - fixed whitespaces >> - refactored code >> v2: >> - fixed compilation issues on ia64 and arm >> >> Michal Wilczynski (9): >> acpi: Move mwait quirk out of acpi_processor.c >> acpi: Move processor_physically_present() to acpi_processor.c >> acpi: Refactor arch_acpi_set_pdc_bits() >> acpi: Rename ACPI_PDC constants >> acpi: Clear C_C2C3_FFH and C_C1_FFH in arch_acpi_set_proc_cap_bits() >> acpi: Move setting CAP_SMP_T_SWCOORD to arch_acpi_set_proc_cap_bits() >> acpi: Introduce acpi_processor_osc() >> acpi: Use _OSC method to convey processor OSPM capabilities >> acpi: Remove acpi_hwp_native_thermal_lvt_osc() > I have made a number of changes to the patches (edited subjects, > rewrote changelogs, fixed up whitespace in a few places, changed the > return value of acpi_early_processor_osc() to bool and made it static > etc.) and tentatively queued them up for 6.6. Please see the > bleeding-edge branch of linux-pm.git. > > I have made sure that the series will still compile at least on > x86-64, but testing it again would be appreciated. > > Thanks! Thanks, will re-test on linux-pm