(adding linux-acpi) On Wed, Sep 06, 2017 at 04:02:10PM -0500, Pierre-Louis Bossart wrote: > On 9/6/17 3:42 PM, Johannes Stezenbach wrote: > > I'm poking around on Asus E200HA, based on Atom x5-Z8300 (Cherrytrail), > > trying to get S0ix working, atm with gross hacks that are recorded in > > https://bugzilla.kernel.org/show_bug.cgi?id=193891 > > > > > > I found commit d31fd43c0f9a4 "clk: x86: Do not gate clocks enabled by the firmware" > > prevents S0i3 entry. Unfortunately the commit message doesn't explain > > the problem it fixes, nor does it mention the platform. > > > > How can we fix d31fd43c0f9a4 to not break S0i3? > > In the existing audio machine drivers, I only use the clock framework for > Baytrail. on Cherrytrail, the pmc_platform_clk3 used for the audio MCLK is > actually managed by the BIOS (look for CLK3 in the DSDT) so there is no > activity in the drivers, the audio don't even get/prepare/enable the > pmc_platform_clk3 so I would guess there is an issue with a non-audio driver > or the BIOS. Put differently, it may be so that by allowing the drivers to > gate an unused clock we allowed you to enter S0i3 even if the bios didn't > enable this mode. I would remove all pmc-clock related stuff and see what > happens. Thanks for this info and Carlo's reply. I've been busy reading source code, DSDT and ACPI spec, and now I'm confused. ;-/ As you say, these clocks are managed by BIOS and clk-pmc-atom.c shouldn't mess with them, thus d31fd43c0f9a4 is correct. Nothing else in the system besides sound/soc/intel/boards/ uses pmc_plt_clk_3 (or other pmc_plt_clk*). Now the question is why ACPI doesn't handle the clocks. E200HA DSDT https://bugzilla.kernel.org/attachment.cgi?id=253921 has PowerResource defintions for the clocks (where _ON and _OFF just set the same registers that clk-pmc-atom.c uses, e.g. CKC3 at 0x6c for CLK3). However, these PowerResources are only listed as child nodes of the devices that use them, but are not referenced by _PR[0-3], with the exception of CLK3 used by the audio codec. The power resources exist in /sys as LNXPOWER:xx subnodes of the device (e.g. i2c 809622C1:02/LNXPOWER:05) and are also listed during boot. acpi_bus_init_power_state() sets device->power.flags.power_resources = 1 only if _PR[0-3] exist. Then acpi_device_set_power() only calls acpi_power_transition() if power_resource == 1, and acpi_power_transition() would only call acpi_power_off_list() for the resources according to _PR[0-3]. I think this is all OK, maybe the BIOS is not optimal in making the clocks available for device runtime pm. What doesn't seem to happen, though, is that the PowerResources are turned _OFF when system transitions to S0ix. The PowerResources have systemlevel == 0, meaning they should be turned off when the system transitions out of S0, if I'm understanding the ACPI (6.1) spec correctly in section 7.2 Declaring a Power Resource Object. Well, actually I'm not sure, the wording of the ACPI spec is: "Systemlevel is the deepest system sleep level OSPM must maintain to keep this power resource on (0 equates to S0, 1 equates to S1, and so on)." FWIW, I used "echo 0x00800000 >/sys/module/acpi/parameters/debug_layer" to confirm devices transition between D0 and D3hot, but clock resources are never turned off. Johannes -- To unsubscribe from this list: send the line "unsubscribe linux-acpi" in the body of a message to majordomo@xxxxxxxxxxxxxxx More majordomo info at http://vger.kernel.org/majordomo-info.html
