[RFCv2 0/1] x86: allow to notify host about guest entering s2idle

[RFCv2 0/1] x86: allow to notify host about guest entering s2idle

[Grzegorz Jaszczyk]

According to the mailing list discussion [1] about the preferred approach
for notifying hypervisor/VMM about guest entering s2idle state this RFC was
implemented.

Instead of original hypercall based approach, which involves KVM change [2]
and makes it hypervisor specific, implement different mechanism, which
takes advantage of MMIO/PIO trapping and makes it hypervisor independent.

For the RFCv1 [3]:
  Patch #1 extends S2Idle ops by new notify handler which will be invoked as a
  very last command before system actually enters S2Idle states. It also allows
  to register and use driver specific notification hook which is used in
  patch #2.

  Patch #2 introduces new driver for virtual PMC, which registers
  acpi_s2idle_dev_ops's notify handler. Its implementation is based on an
  ACPI _DSM evaluation, which in turn can perform MMIO access and allow to
  trap and therefore notify the VMM about guest entering S2Idle state.

For the RFCv2: the patch #1 was dropped as in the meantime Mario Limonciello
introduced a very similar patch [4] which uses s/notify/check and invokes the
callback a bit earlier just before s2idle_entry. Mentioned patch has already
been merged.

This patchset is marked as RFC since patch #2 implements driver for non
existing device "HYPE0001", which ACPI ID was not registered yet.
Furthermore the required registration process [5] will not be started
before getting positive feedback about this patchset.

[1] https://patchwork.kernel.org/project/linux-pm/patch/20220609110337.1238762-2-jaz@semihalf.com/
[2] https://patchwork.kernel.org/project/linux-pm/patch/20220609110337.1238762-3-jaz@semihalf.com/
[3] https://patchwork.kernel.org/project/linux-pm/cover/20220707125329.378277-1-jaz@semihalf.com/
[4] https://patchwork.kernel.org/project/linux-pm/patch/20220829162953.5947-2-mario.limonciello@amd.com
[5] https://uefi.org/PNP_ACPI_Registry

Grzegorz Jaszczyk (1):
  platform/x86: Add virtual PMC driver used for S2Idle

 drivers/platform/x86/Kconfig    |  7 ++++
 drivers/platform/x86/Makefile   |  3 ++
 drivers/platform/x86/virt_pmc.c | 73 +++++++++++++++++++++++++++++++++
 3 files changed, 83 insertions(+)
 create mode 100644 drivers/platform/x86/virt_pmc.c

-- 
2.39.1.519.gcb327c4b5f-goog

[RFCv2 1/1] platform/x86: Add virtual PMC driver used for S2Idle

[Grzegorz Jaszczyk]

Virtual PMC driver is meant for the guest VMs for the S2Idle
notification. Its purpose is to register S2Idle dev ops check handler,
which will evaluate ACPI _DSM just before the guest enters S2Idle power
state.

This allows to trap on MMIO access done as a consequence of _DSM
evaluation and therefore notify the VMM about the guest entering S2Idle
state.

Signed-off-by: Grzegorz Jaszczyk <jaz@semihalf.com>
---
Changelog v1..v2:
- Take advantage of acpi_s2idle_dev_ops's check() instead of notify()
---
 drivers/platform/x86/Kconfig    |  7 ++++
 drivers/platform/x86/Makefile   |  3 ++
 drivers/platform/x86/virt_pmc.c | 73 +++++++++++++++++++++++++++++++++
 3 files changed, 83 insertions(+)
 create mode 100644 drivers/platform/x86/virt_pmc.c

diff --git a/drivers/platform/x86/Kconfig b/drivers/platform/x86/Kconfig
index 5692385e2d26..b7c3f98031d7 100644
--- a/drivers/platform/x86/Kconfig
+++ b/drivers/platform/x86/Kconfig
@@ -1099,6 +1099,13 @@ config WINMATE_FM07_KEYS
 	  buttons below the display. This module adds an input device
 	  that delivers key events when these buttons are pressed.
 
+config VIRT_PMC
+	tristate "Virt PMC"
+	depends on ACPI && SUSPEND
+	help
+	  The Virtual PMC driver is meant for the guest VMs and its main
+	  purpose is to notify about guest entering s2idle state.
+
 endif # X86_PLATFORM_DEVICES
 
 config P2SB
diff --git a/drivers/platform/x86/Makefile b/drivers/platform/x86/Makefile
index 1d3d1b02541b..c4d3056cf4ea 100644
--- a/drivers/platform/x86/Makefile
+++ b/drivers/platform/x86/Makefile
@@ -129,6 +129,9 @@ obj-$(CONFIG_INTEL_SCU_WDT)		+= intel_scu_wdt.o
 obj-$(CONFIG_INTEL_SCU_IPC_UTIL)	+= intel_scu_ipcutil.o
 obj-$(CONFIG_X86_INTEL_LPSS)		+= pmc_atom.o
 
+# Virtual PMC
+obj-$(CONFIG_VIRT_PMC)			+= virt_pmc.o
+
 # Siemens Simatic Industrial PCs
 obj-$(CONFIG_SIEMENS_SIMATIC_IPC)	+= simatic-ipc.o
 
diff --git a/drivers/platform/x86/virt_pmc.c b/drivers/platform/x86/virt_pmc.c
new file mode 100644
index 000000000000..daf9c9ed86e5
--- /dev/null
+++ b/drivers/platform/x86/virt_pmc.c
@@ -0,0 +1,73 @@
+// SPDX-License-Identifier: GPL-2.0-only
+/*
+ * Virtual Power Management Controller Driver
+ *
+ * Author: Grzegorz Jaszczyk <jaz@semihalf.com>
+ */
+
+#include <linux/acpi.h>
+#include <linux/platform_device.h>
+
+#define ACPI_VIRT_PMC_DSM_UUID	"9ea49ba3-434a-49a6-be30-37cc55c4d397"
+#define ACPI_VIRT_PMC_NOTIFY 1
+
+static acpi_handle virt_pmc_handle;
+
+static void virt_pmc_s2idle_notify(void)
+{
+	union acpi_object *out_obj;
+	static guid_t dsm_guid;
+
+	guid_parse(ACPI_VIRT_PMC_DSM_UUID, &dsm_guid);
+
+	out_obj = acpi_evaluate_dsm(virt_pmc_handle, &dsm_guid,
+					0, ACPI_VIRT_PMC_NOTIFY, NULL);
+
+	acpi_handle_debug(virt_pmc_handle, "_DSM function %u evaluation %s\n",
+			  ACPI_VIRT_PMC_NOTIFY, out_obj ? "successful" : "failed");
+
+	ACPI_FREE(out_obj);
+}
+
+static struct acpi_s2idle_dev_ops amd_pmc_s2idle_dev_ops = {
+	.check = virt_pmc_s2idle_notify,
+};
+
+static int virt_pmc_probe(struct platform_device *pdev)
+{
+	int err = 0;
+
+	virt_pmc_handle = ACPI_HANDLE(&pdev->dev);
+
+	err = acpi_register_lps0_dev(&amd_pmc_s2idle_dev_ops);
+	if (err)
+		dev_warn(&pdev->dev, "failed to register LPS0 sleep handler\n");
+
+	return err;
+}
+
+static int virt_pmc_remove(struct platform_device *pdev)
+{
+	acpi_unregister_lps0_dev(&amd_pmc_s2idle_dev_ops);
+
+	return 0;
+}
+
+static const struct acpi_device_id virt_pmc_acpi_ids[] = {
+	{"HYPE0001", 0}, /* _HID for XXX Power Engine, _CID PNP0D80*/
+	{ }
+};
+MODULE_DEVICE_TABLE(acpi, virt_pmc_acpi_ids);
+
+static struct platform_driver virt_pmc_driver = {
+	.driver = {
+		.name = "virtual_pmc",
+		.acpi_match_table = ACPI_PTR(virt_pmc_acpi_ids),
+	},
+	.probe = virt_pmc_probe,
+	.remove = virt_pmc_remove,
+};
+
+module_platform_driver(virt_pmc_driver);
+
+MODULE_DESCRIPTION("Virtual PMC Driver");
-- 
2.39.1.519.gcb327c4b5f-goog

Re: [RFCv2 0/1] x86: allow to notify host about guest entering s2idle

[Hans de Goede]

Hi Grzegorz,

Can you please resend this with the linux-pm@vger.kernel.org,
x86@kernel.org and platform-driver-x86@vger.kernel.org email-lists
in the Cc to make sure this gets seen by some more people ?

(not everyone reads linux-kernel) 

And since you mention Mario Limonciello, maybe also add him the the Cc?

Regards,

Hans




On 2/9/23 14:56, Grzegorz Jaszczyk wrote:
> According to the mailing list discussion [1] about the preferred approach
> for notifying hypervisor/VMM about guest entering s2idle state this RFC was
> implemented.
> 
> Instead of original hypercall based approach, which involves KVM change [2]
> and makes it hypervisor specific, implement different mechanism, which
> takes advantage of MMIO/PIO trapping and makes it hypervisor independent.
> 
> For the RFCv1 [3]:
>   Patch #1 extends S2Idle ops by new notify handler which will be invoked as a
>   very last command before system actually enters S2Idle states. It also allows
>   to register and use driver specific notification hook which is used in
>   patch #2.
> 
>   Patch #2 introduces new driver for virtual PMC, which registers
>   acpi_s2idle_dev_ops's notify handler. Its implementation is based on an
>   ACPI _DSM evaluation, which in turn can perform MMIO access and allow to
>   trap and therefore notify the VMM about guest entering S2Idle state.
> 
> For the RFCv2: the patch #1 was dropped as in the meantime Mario Limonciello
> introduced a very similar patch [4] which uses s/notify/check and invokes the
> callback a bit earlier just before s2idle_entry. Mentioned patch has already
> been merged.
> 
> This patchset is marked as RFC since patch #2 implements driver for non
> existing device "HYPE0001", which ACPI ID was not registered yet.
> Furthermore the required registration process [5] will not be started
> before getting positive feedback about this patchset.
> 
> [1] https://patchwork.kernel.org/project/linux-pm/patch/20220609110337.1238762-2-jaz@semihalf.com/
> [2] https://patchwork.kernel.org/project/linux-pm/patch/20220609110337.1238762-3-jaz@semihalf.com/
> [3] https://patchwork.kernel.org/project/linux-pm/cover/20220707125329.378277-1-jaz@semihalf.com/
> [4] https://patchwork.kernel.org/project/linux-pm/patch/20220829162953.5947-2-mario.limonciello@amd.com
> [5] https://uefi.org/PNP_ACPI_Registry
> 
> Grzegorz Jaszczyk (1):
>   platform/x86: Add virtual PMC driver used for S2Idle
> 
>  drivers/platform/x86/Kconfig    |  7 ++++
>  drivers/platform/x86/Makefile   |  3 ++
>  drivers/platform/x86/virt_pmc.c | 73 +++++++++++++++++++++++++++++++++
>  3 files changed, 83 insertions(+)
>  create mode 100644 drivers/platform/x86/virt_pmc.c
>