Re: [PATCH v4 04/14] x86/rtc: replace paravirt rtc check with platform legacy quirk

["Luis R. Rodriguez" <mcgrof@kernel.org>]

On Thu, Apr 07, 2016 at 08:55:54AM -0400, Boris Ostrovsky wrote:
> On 04/06/2016 08:06 PM, Luis R. Rodriguez wrote:
> >We have 4 types of x86 platforms that disable RTC:
> >
> >   * Intel MID
> >   * Lguest - uses paravirt
> >   * Xen dom-U - uses paravirt
> >   * x86 on legacy systems annotated with an ACPI legacy flag
> >
> >We can consolidate all of these into a platform specific legacy
> >quirk set early in boot through i386_start_kernel() and through
> >x86_64_start_reservations(). This deals with the RTC quirks which
> >we can rely on through the hardware subarch, the ACPI check can
> >be dealt with separately.
> >
> >v2: split the subarch check from the ACPI check, clarify
> >     on the ACPI change commit log why ordering works
> >
> >Suggested-by: Ingo Molnar <mingo@kernel.org>
> >Signed-off-by: Luis R. Rodriguez <mcgrof@kernel.org>

<-- snip -->

> >diff --git a/arch/x86/kernel/platform-quirks.c b/arch/x86/kernel/platform-quirks.c
> >new file mode 100644
> >index 000000000000..1b114ac5996f
> >--- /dev/null
> >+++ b/arch/x86/kernel/platform-quirks.c
> >@@ -0,0 +1,18 @@
> >+#include <linux/kernel.h>
> >+#include <linux/init.h>
> >+
> >+#include <asm/setup.h>
> >+#include <asm/bios_ebda.h>
> >+
> >+void __init x86_early_init_platform_quirks(void)
> >+{
> >+	x86_platform.legacy.rtc = 1;
> >+
> >+	switch (boot_params.hdr.hardware_subarch) {
> >+	case X86_SUBARCH_XEN:
> >+	case X86_SUBARCH_LGUEST:
> >+	case X86_SUBARCH_INTEL_MID:
> >+		x86_platform.legacy.rtc = 0;
> >+		break;
> >+	}
> >+}
> 
> What about Xen dom0 (aka initial domain)?

Indeed, thanks for catching this, the hunk below removes the re-enablement of
the the RTC for dom0:

> >--- a/arch/x86/xen/enlighten.c
> >+++ b/arch/x86/xen/enlighten.c
> >@@ -1192,7 +1192,6 @@ static const struct pv_info xen_info __initconst = {
> >  #ifdef CONFIG_X86_64
> >  	.extra_user_64bit_cs = FLAT_USER_CS64,
> >  #endif
> >-	.features = 0,
> >  	.name = "Xen",
> >  };
> >@@ -1525,8 +1524,6 @@ asmlinkage __visible void __init xen_start_kernel(void)
> >  	/* Install Xen paravirt ops */
> >  	pv_info = xen_info;
> >-	if (xen_initial_domain())
> >-		pv_info.features |= PV_SUPPORTED_RTC;
> >  	pv_init_ops = xen_init_ops;
> >  	if (!xen_pvh_domain()) {
> >  		pv_cpu_ops = xen_cpu_ops;

This should then break dom0 unless of course you have the respective next
patch applied and that disabled the RTC due to an ACPI setting on your
platform. Juergen, can you check to see if that was the case for your
testing platform on dom0 ?

This highlights a semantic gap issue. From a quick cursory review, I think
we can address this temporarily by just using a check:

void __init x86_early_init_platform_quirks(void)
{
	x86_platform.legacy.rtc = 1;

	switch (boot_params.hdr.hardware_subarch) {
	case X86_SUBARCH_XEN:
	case X86_SUBARCH_LGUEST:
	case X86_SUBARCH_INTEL_MID:
-		x86_platform.legacy.rtc = 0;
+		if (x86_init.mpparse.get_smp_config != x86_init_uint_noop)
+			x86_platform.legacy.rtc = 0;
		break;
	}
}

This would work given x86_early_init_platform_quirks() is called prior
to the any code that sets up x86_init.mpparse*, and the only code that
would have set this is the PV guest path, and the dom0 override. Is
would be replacing one hack with another though so I'm not exactly happy
with it as a compromise.

Another x86 standard thing dom0 populates on the PV path is
&boot_params.screen_info, done through xen_init_vga() -- but it
was not clear exactly what domU does with this ? Does domU never
have VGA set up? Given that the PV path is the only thing that would
ever set up the x86 zero page I take it domU leaves that empty,
and dom0 *always* seems to set up the screen_info->orig_video_isVGA,
so another mechanism might be something like:

void __init x86_early_init_platform_quirks(void)
{
+	struct screen_info *screen_info = &boot_params.screen_info;
+
	x86_platform.legacy.rtc = 1;

	switch (boot_params.hdr.hardware_subarch) {
	case X86_SUBARCH_XEN:
	case X86_SUBARCH_LGUEST:
	case X86_SUBARCH_INTEL_MID:
-		x86_platform.legacy.rtc = 0;
+		if (!screen_info->orig_video_isVGA)
+			x86_platform.legacy.rtc = 0;
		break;
	}
}

If the semantics of requiring VGA through the x86 boot params
suffice to annotate PV path dom0 then we have a win. Specially
if this might be useful to other virtualization environments
to do some of their own virtualization quirks in this path.

Additionally -- if domU never sets the screen info stuff, should
it or does it always set ACPI_FADT_NO_VGA as well ?

Is Xen the only guest type we have that has a notion of dom0 and need dom0 type
of quirks ? Also how would this work for HVMLite for domU and dom0 ? I think
the ARM folks are doing some things with EFI configuration tables to pass the
screen_info stuff, so perhaps that should be looked at.

  Luis