[PATCH v2 0/6] Add support for a fake, para-virtualised machine

Hello again,

This is version two of the patches originally posted here:

  http://lists.infradead.org/pipermail/linux-arm-kernel/2012-December/135870.html

Given the lively discussion sparked in that thread, there have been a
fair number of changes since the RFC:

	* A client-side implementation of PSCI and proposed DT binding
	* Use of PSCI for SMP boot
	* Removal of the SMP pen code
	* Dropped the RFC tag

Marc hacked up the KVM side separately and this has been tested
successfully with his code using a magic build of kvmtool.

As usual, all feedback welcome.

Cheers,

Will


Marc Zyngier (1):
  ARM: Dummy Virtual Machine platform support

Will Deacon (5):
  ARM: opcodes: add missing include of linux/linkage.h
  ARM: opcodes: add opcodes definitions for ARM security extensions
  ARM: psci: add devicetree binding for describing PSCI firmware
  ARM: psci: add support for PSCI invocations from the kernel
  ARM: mach-virt: add SMP support using PSCI

 Documentation/devicetree/bindings/arm/psci.txt |  58 +++++++
 arch/arm/Kconfig                               |  12 ++
 arch/arm/Makefile                              |   1 +
 arch/arm/include/asm/opcodes-sec.h             |  24 +++
 arch/arm/include/asm/opcodes.h                 |   1 +
 arch/arm/include/asm/psci.h                    |  36 +++++
 arch/arm/kernel/Makefile                       |   1 +
 arch/arm/kernel/psci.c                         | 214 +++++++++++++++++++++++++
 arch/arm/mach-virt/Kconfig                     |  10 ++
 arch/arm/mach-virt/Makefile                    |   6 +
 arch/arm/mach-virt/platsmp.c                   |  76 +++++++++
 arch/arm/mach-virt/virt.c                      |  71 ++++++++
 12 files changed, 510 insertions(+)
 create mode 100644 Documentation/devicetree/bindings/arm/psci.txt
 create mode 100644 arch/arm/include/asm/opcodes-sec.h
 create mode 100644 arch/arm/include/asm/psci.h
 create mode 100644 arch/arm/kernel/psci.c
 create mode 100644 arch/arm/mach-virt/Kconfig
 create mode 100644 arch/arm/mach-virt/Makefile
 create mode 100644 arch/arm/mach-virt/platsmp.c
 create mode 100644 arch/arm/mach-virt/virt.c

-- 
1.8.0

opcodes.h wants to declare an asmlinkage function, so we need to include
linux/linkage.h

Acked-by: Dave Martin <redacted>
Signed-off-by: Will Deacon <redacted>
---
 arch/arm/include/asm/opcodes.h | 1 +
 1 file changed, 1 insertion(+)

diff --git a/arch/arm/include/asm/opcodes.h b/arch/arm/include/asm/opcodes.h
index 74e211a..e796c59 100644
--- a/arch/arm/include/asm/opcodes.h
+++ b/arch/arm/include/asm/opcodes.h

@@ -10,6 +10,7 @@
 #define __ASM_ARM_OPCODES_H
 
 #ifndef __ASSEMBLY__
+#include <linux/linkage.h>
 extern asmlinkage unsigned int arm_check_condition(u32 opcode, u32 psr);
 #endif
 

-- 
1.8.0

The ARM security extensions introduced the smc instruction, which is not
supported by all versions of GAS.

This patch introduces opcodes-sec.h, so that smc is made available in a
similar manner to hvc.

Acked-by: Dave Martin <redacted>
Signed-off-by: Will Deacon <redacted>
---
 arch/arm/include/asm/opcodes-sec.h | 24 ++++++++++++++++++++++++
 1 file changed, 24 insertions(+)
 create mode 100644 arch/arm/include/asm/opcodes-sec.h

diff --git a/arch/arm/include/asm/opcodes-sec.h b/arch/arm/include/asm/opcodes-sec.h
new file mode 100644
index 0000000..bc3a917
--- /dev/null
+++ b/arch/arm/include/asm/opcodes-sec.h

@@ -0,0 +1,24 @@
+/*
+ * This program is free software; you can redistribute it and/or modify
+ * it under the terms of the GNU General Public License version 2 as
+ * published by the Free Software Foundation.
+ *
+ * This program is distributed in the hope that it will be useful,
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
+ * GNU General Public License for more details.
+ *
+ * Copyright (C) 2012 ARM Limited
+ */
+
+#ifndef __ASM_ARM_OPCODES_SEC_H
+#define __ASM_ARM_OPCODES_SEC_H
+
+#include <asm/opcodes.h>
+
+#define __SMC(imm4) __inst_arm_thumb32(					\
+	0xE1600070 | (((imm4) & 0xF) << 0),				\
+	0xF7F08000 | (((imm4) & 0xF) << 16)				\
+)
+
+#endif /* __ASM_ARM_OPCODES_SEC_H */

-- 
1.8.0

This patch adds a new devicetree binding for describing PSCI firmware
to Linux.

Signed-off-by: Will Deacon <redacted>
---
 Documentation/devicetree/bindings/arm/psci.txt | 58 ++++++++++++++++++++++++++
 1 file changed, 58 insertions(+)
 create mode 100644 Documentation/devicetree/bindings/arm/psci.txt

diff --git a/Documentation/devicetree/bindings/arm/psci.txt b/Documentation/devicetree/bindings/arm/psci.txt
new file mode 100644
index 0000000..904d0d3
--- /dev/null
+++ b/Documentation/devicetree/bindings/arm/psci.txt

@@ -0,0 +1,58 @@
+* Power State Coordination Interface (PSCI)
+
+Firmware implementing the PSCI functions described in ARM document number
+ARM DEN 0022A ("Power State Coordination Interface System Software on ARM
+processors") can be used by Linux to initiate various CPU-centric power
+operations.
+
+Issue A of the specification describes functions for CPU suspend, hotplug
+and migration of secure software.
+
+Functions are invoked by trapping to the privilege level of the PSCI
+firmware (specified as part of the binding below) and passing arguments
+in a manner similar to that specified by AAPCS:
+
+	 r0		=> 32-bit Function ID / return value
+	{r1 - r3}	=> Parameters
+
+Note that the immediate field of the trapping instruction must be set
+to #0.
+
+
+Main node required properties:
+
+ - compatible    : Must be "arm,psci"
+
+ - method        : The method of calling the PSCI firmware. Permitted
+                   values are:
+
+                   "smc" : SMC #0, with the register assignments specified
+		           in this binding.
+
+                   "hvc" : HVC #0, with the register assignments specified
+		           in this binding.
+
+ - function-base : The base ID from which the functions are offset.
+
+Main node optional properties:
+
+ - cpu_suspend   : Offset of CPU_SUSPEND ID from function-base
+
+ - cpu_off       : Offset of CPU_OFF ID from function-base
+
+ - cpu_on        : Offset of CPU_ON ID from function-base
+
+ - migrate       : Offset of MIGRATE ID from function-base
+
+
+Example:
+
+	psci {
+		compatible	= "arm,psci";
+		method		= "smc";
+		function-base	= <0x95c1ba5e>;
+		cpu_suspend	= <0>;
+		cpu_off		= <1>;
+		cpu_on		= <2>;
+		migrate		= <3>;
+	};

-- 
1.8.0

This patch adds support for the Power State Coordination Interface
defined by ARM, allowing Linux to request CPU-centric power-management
operations from firmware implementing the PSCI protocol.

Signed-off-by: Will Deacon <redacted>
---
 arch/arm/Kconfig            |  10 +++
 arch/arm/include/asm/psci.h |  36 ++++++++
 arch/arm/kernel/Makefile    |   1 +
 arch/arm/kernel/psci.c      | 214 ++++++++++++++++++++++++++++++++++++++++++++
 4 files changed, 261 insertions(+)
 create mode 100644 arch/arm/include/asm/psci.h
 create mode 100644 arch/arm/kernel/psci.c

diff --git a/arch/arm/Kconfig b/arch/arm/Kconfig
index e4017ea..2a04375 100644
--- a/arch/arm/Kconfig
+++ b/arch/arm/Kconfig

@@ -1611,6 +1611,16 @@ config HOTPLUG_CPU
 	  Say Y here to experiment with turning CPUs off and on.  CPUs
 	  can be controlled through /sys/devices/system/cpu.
 
+config ARM_PSCI
+	bool "Support for the ARM Power State Coordination Interface (PSCI)"
+	depends on CPU_V7
+	help
+	  Say Y here if you want Linux to communicate with system firmware
+	  implementing the PSCI specification for CPU-centric power
+	  management operations described in ARM document number ARM DEN
+	  0022A ("Power State Coordination Interface System Software on
+	  ARM processors").
+
 config LOCAL_TIMERS
 	bool "Use local timer interrupts"
 	depends on SMP

diff --git a/arch/arm/include/asm/psci.h b/arch/arm/include/asm/psci.h
new file mode 100644
index 0000000..ce0dbe7
--- /dev/null
+++ b/arch/arm/include/asm/psci.h

@@ -0,0 +1,36 @@
+/*
+ * This program is free software; you can redistribute it and/or modify
+ * it under the terms of the GNU General Public License version 2 as
+ * published by the Free Software Foundation.
+ *
+ * This program is distributed in the hope that it will be useful,
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
+ * GNU General Public License for more details.
+ *
+ * Copyright (C) 2012 ARM Limited
+ */
+
+#ifndef __ASM_ARM_PSCI_H
+#define __ASM_ARM_PSCI_H
+
+#define PSCI_POWER_STATE_TYPE_STANDBY		0
+#define PSCI_POWER_STATE_TYPE_POWER_DOWN	1
+
+struct psci_power_state {
+	u16	id;
+	u8	type;
+	u8	affinity_level;
+};
+
+struct psci_operations {
+	int (*cpu_suspend)(struct psci_power_state state,
+			   unsigned long entry_point);
+	int (*cpu_off)(struct psci_power_state state);
+	int (*cpu_on)(unsigned long cpuid, unsigned long entry_point);
+	int (*migrate)(unsigned long cpuid);
+};
+
+extern struct psci_operations psci_ops;
+
+#endif /* __ASM_ARM_PSCI_H */

diff --git a/arch/arm/kernel/Makefile b/arch/arm/kernel/Makefile
index 5bbec7b..5f3338e 100644
--- a/arch/arm/kernel/Makefile
+++ b/arch/arm/kernel/Makefile

@@ -82,5 +82,6 @@ obj-$(CONFIG_DEBUG_LL)	+= debug.o
 obj-$(CONFIG_EARLY_PRINTK)	+= early_printk.o
 
 obj-$(CONFIG_ARM_VIRT_EXT)	+= hyp-stub.o
+obj-$(CONFIG_ARM_PSCI)		+= psci.o
 
 extra-y := $(head-y) vmlinux.lds

diff --git a/arch/arm/kernel/psci.c b/arch/arm/kernel/psci.c
new file mode 100644
index 0000000..e7fe909
--- /dev/null
+++ b/arch/arm/kernel/psci.c

@@ -0,0 +1,214 @@
+/*
+ * This program is free software; you can redistribute it and/or modify
+ * it under the terms of the GNU General Public License version 2 as
+ * published by the Free Software Foundation.
+ *
+ * This program is distributed in the hope that it will be useful,
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
+ * GNU General Public License for more details.
+ *
+ * Copyright (C) 2012 ARM Limited
+ *
+ * Author: Will Deacon <will.deacon@arm.com>
+ */
+
+#define pr_fmt(fmt) "psci: " fmt
+
+#include <linux/init.h>
+#include <linux/of.h>
+
+#include <asm/compiler.h>
+#include <asm/errno.h>
+#include <asm/opcodes-sec.h>
+#include <asm/opcodes-virt.h>
+#include <asm/psci.h>
+
+struct psci_operations psci_ops;
+
+static int (*invoke_psci_fn)(u32, u32, u32, u32);
+
+enum psci_function {
+	PSCI_FN_CPU_SUSPEND,
+	PSCI_FN_CPU_ON,
+	PSCI_FN_CPU_OFF,
+	PSCI_FN_MIGRATE,
+	PSCI_FN_MAX,
+};
+
+static u32 psci_function_id[PSCI_FN_MAX];
+
+#define PSCI_RET_SUCCESS		0
+#define PSCI_RET_EOPNOTSUPP		-1
+#define PSCI_RET_EINVAL			-2
+#define PSCI_RET_EPERM			-3
+
+static int psci_to_linux_errno(int errno)
+{
+	switch (errno) {
+	case PSCI_RET_SUCCESS:
+		return 0;
+	case PSCI_RET_EOPNOTSUPP:
+		return -EOPNOTSUPP;
+	case PSCI_RET_EINVAL:
+		return -EINVAL;
+	case PSCI_RET_EPERM:
+		return -EPERM;
+	};
+
+	return -EINVAL;
+}
+
+#define PSCI_POWER_STATE_ID_MASK	0xffff
+#define PSCI_POWER_STATE_ID_SHIFT	0
+#define PSCI_POWER_STATE_TYPE_MASK	0x1
+#define PSCI_POWER_STATE_TYPE_SHIFT	16
+#define PSCI_POWER_STATE_AFFL_MASK	0x3
+#define PSCI_POWER_STATE_AFFL_SHIFT	24
+
+static u32 psci_power_state_pack(struct psci_power_state state)
+{
+	return	((state.id & PSCI_POWER_STATE_ID_MASK)
+			<< PSCI_POWER_STATE_ID_SHIFT)	|
+		((state.type & PSCI_POWER_STATE_TYPE_MASK)
+			<< PSCI_POWER_STATE_TYPE_SHIFT)	|
+		((state.affinity_level & PSCI_POWER_STATE_AFFL_MASK)
+			<< PSCI_POWER_STATE_AFFL_SHIFT);
+}
+
+/*
+ * The following two functions are invoked via the invoke_psci_fn pointer
+ * and will not be inlined, allowing us to piggyback on the AAPCS.
+ */
+static int __invoke_psci_fn_hvc(u32 function_id, u32 arg0, u32 arg1, u32 arg2)
+{
+	asm volatile(
+			__asmeq("%0", "r0")
+			__asmeq("%1", "r1")
+			__asmeq("%2", "r2")
+			__asmeq("%3", "r3")
+			__HVC(0)
+		: "+r" (function_id)
+		: "r" (arg0), "r" (arg1), "r" (arg2));
+
+	return function_id;
+}
+
+static int __invoke_psci_fn_smc(u32 function_id, u32 arg0, u32 arg1, u32 arg2)
+{
+	asm volatile(
+			__asmeq("%0", "r0")
+			__asmeq("%1", "r1")
+			__asmeq("%2", "r2")
+			__asmeq("%3", "r3")
+			__SMC(0)
+		: "+r" (function_id)
+		: "r" (arg0), "r" (arg1), "r" (arg2));
+
+	return function_id;
+}
+
+static int psci_cpu_suspend(struct psci_power_state state,
+			    unsigned long entry_point)
+{
+	int err;
+	u32 fn, power_state;
+
+	fn = psci_function_id[PSCI_FN_CPU_SUSPEND];
+	power_state = psci_power_state_pack(state);
+	err = invoke_psci_fn(fn, power_state, (u32)entry_point, 0);
+	return psci_to_linux_errno(err);
+}
+
+static int psci_cpu_off(struct psci_power_state state)
+{
+	int err;
+	u32 fn, power_state;
+
+	fn = psci_function_id[PSCI_FN_CPU_OFF];
+	power_state = psci_power_state_pack(state);
+	err = invoke_psci_fn(fn, power_state, 0, 0);
+	return psci_to_linux_errno(err);
+}
+
+static int psci_cpu_on(unsigned long cpuid, unsigned long entry_point)
+{
+	int err;
+	u32 fn;
+
+	fn = psci_function_id[PSCI_FN_CPU_ON];
+	err = invoke_psci_fn(fn, cpuid, entry_point, 0);
+	return psci_to_linux_errno(err);
+}
+
+static int psci_migrate(unsigned long cpuid)
+{
+	int err;
+	u32 fn;
+
+	fn = psci_function_id[PSCI_FN_MIGRATE];
+	err = invoke_psci_fn(fn, cpuid, 0, 0);
+	return psci_to_linux_errno(err);
+}
+
+static const struct of_device_id psci_of_match[] __initconst = {
+	{ .compatible = "arm,psci",	},
+	{},
+};
+
+static int __init psci_init(void)
+{
+	struct device_node *np;
+	const char *method;
+	u32 base, id;
+
+	np = of_find_matching_node(NULL, psci_of_match);
+	if (!np)
+		return 0;
+
+	pr_info("probing function IDs from device-tree\n");
+
+	if (of_property_read_u32(np, "function-base", &base)) {
+		pr_warning("missing \"function-base\" property\n");
+		goto out_put_node;
+	}
+
+	if (of_property_read_string(np, "method", &method)) {
+		pr_warning("missing \"method\" property\n");
+		goto out_put_node;
+	}
+
+	if (!strcmp("hvc", method)) {
+		invoke_psci_fn = __invoke_psci_fn_hvc;
+	} else if (!strcmp("smc", method)) {
+		invoke_psci_fn = __invoke_psci_fn_smc;
+	} else {
+		pr_warning("invalid \"method\" property: %s\n", method);
+		goto out_put_node;
+	}
+
+	if (!of_property_read_u32(np, "cpu_suspend", &id)) {
+		psci_function_id[PSCI_FN_CPU_SUSPEND] = base + id;
+		psci_ops.cpu_suspend = psci_cpu_suspend;
+	}
+
+	if (!of_property_read_u32(np, "cpu_off", &id)) {
+		psci_function_id[PSCI_FN_CPU_OFF] = base + id;
+		psci_ops.cpu_off = psci_cpu_off;
+	}
+
+	if (!of_property_read_u32(np, "cpu_on", &id)) {
+		psci_function_id[PSCI_FN_CPU_ON] = base + id;
+		psci_ops.cpu_on = psci_cpu_on;
+	}
+
+	if (!of_property_read_u32(np, "migrate", &id)) {
+		psci_function_id[PSCI_FN_MIGRATE] = base + id;
+		psci_ops.migrate = psci_migrate;
+	}
+
+out_put_node:
+	of_node_put(np);
+	return 0;
+}
+early_initcall(psci_init);

-- 
1.8.0

From: Marc Zyngier <redacted>

Add support for the smallest, dumbest possible platform, to be
used as a guest for KVM or other hypervisors.

It only mandates a GIC and architected timers. Fits nicely with
a multiplatform zImage. Uses very little silicon area.

Signed-off-by: Marc Zyngier <redacted>
Signed-off-by: Will Deacon <redacted>
---
 arch/arm/Kconfig            |  2 ++
 arch/arm/Makefile           |  1 +
 arch/arm/mach-virt/Kconfig  |  9 +++++++
 arch/arm/mach-virt/Makefile |  5 ++++
 arch/arm/mach-virt/virt.c   | 65 +++++++++++++++++++++++++++++++++++++++++++++
 5 files changed, 82 insertions(+)
 create mode 100644 arch/arm/mach-virt/Kconfig
 create mode 100644 arch/arm/mach-virt/Makefile
 create mode 100644 arch/arm/mach-virt/virt.c

diff --git a/arch/arm/Kconfig b/arch/arm/Kconfig
index 2a04375..552d72e0 100644
--- a/arch/arm/Kconfig
+++ b/arch/arm/Kconfig

@@ -1127,6 +1127,8 @@ source "arch/arm/mach-versatile/Kconfig"
 source "arch/arm/mach-vexpress/Kconfig"
 source "arch/arm/plat-versatile/Kconfig"
 
+source "arch/arm/mach-virt/Kconfig"
+
 source "arch/arm/mach-w90x900/Kconfig"
 
 # Definitions to make life easier

diff --git a/arch/arm/Makefile b/arch/arm/Makefile
index 5f914fc..e8232ad 100644
--- a/arch/arm/Makefile
+++ b/arch/arm/Makefile

@@ -192,6 +192,7 @@ machine-$(CONFIG_ARCH_SOCFPGA)		+= socfpga
 machine-$(CONFIG_ARCH_SPEAR13XX)	+= spear13xx
 machine-$(CONFIG_ARCH_SPEAR3XX)		+= spear3xx
 machine-$(CONFIG_MACH_SPEAR600)		+= spear6xx
+machine-$(CONFIG_ARCH_VIRT)		+= virt
 machine-$(CONFIG_ARCH_ZYNQ)		+= zynq
 
 # Platform directory name.  This list is sorted alphanumerically

diff --git a/arch/arm/mach-virt/Kconfig b/arch/arm/mach-virt/Kconfig
new file mode 100644
index 0000000..a568a2a
--- /dev/null
+++ b/arch/arm/mach-virt/Kconfig

@@ -0,0 +1,9 @@
+config ARCH_VIRT
+	bool "Dummy Virtual Machine" if ARCH_MULTI_V7
+	select ARCH_WANT_OPTIONAL_GPIOLIB
+	select ARM_GIC
+	select ARM_ARCH_TIMER
+	select HAVE_SMP
+	select CPU_V7
+	select SPARSE_IRQ
+	select USE_OF

diff --git a/arch/arm/mach-virt/Makefile b/arch/arm/mach-virt/Makefile
new file mode 100644
index 0000000..7ddbfa6
--- /dev/null
+++ b/arch/arm/mach-virt/Makefile

@@ -0,0 +1,5 @@
+#
+# Makefile for the linux kernel.
+#
+
+obj-y					:= virt.o

diff --git a/arch/arm/mach-virt/virt.c b/arch/arm/mach-virt/virt.c
new file mode 100644
index 0000000..174b9da
--- /dev/null
+++ b/arch/arm/mach-virt/virt.c

@@ -0,0 +1,65 @@
+/*
+ * Dummy Virtual Machine - does what it says on the tin.
+ *
+ * Copyright (C) 2012 ARM Ltd
+ * Authors: Will Deacon <will.deacon@arm.com>,
+ *          Marc Zyngier <marc.zyngier@arm.com>
+ *
+ * This program is free software; you can redistribute it and/or modify
+ * it under the terms of the GNU General Public License version 2 as
+ * published by the Free Software Foundation.
+ *
+ * This program is distributed in the hope that it will be useful,
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
+ * GNU General Public License for more details.
+ *
+ * You should have received a copy of the GNU General Public License
+ * along with this program.  If not, see <http://www.gnu.org/licenses/>.
+ */
+
+#include <linux/of_irq.h>
+#include <linux/of_platform.h>
+
+#include <asm/arch_timer.h>
+#include <asm/hardware/gic.h>
+#include <asm/mach/arch.h>
+#include <asm/mach/time.h>
+
+const static struct of_device_id irq_match[] = {
+	{ .compatible = "arm,cortex-a15-gic", .data = gic_of_init, },
+	{}
+};
+
+static void __init gic_init_irq(void)
+{
+	of_irq_init(irq_match);
+}
+
+static void __init virt_init(void)
+{
+	of_platform_populate(NULL, of_default_bus_match_table, NULL, NULL);
+}
+
+static void __init virt_timer_init(void)
+{
+	WARN_ON(arch_timer_of_register() != 0);
+	WARN_ON(arch_timer_sched_clock_init() != 0);
+}
+
+static const char *virt_dt_match[] = {
+	"linux,dummy-virt",
+	NULL
+};
+
+static struct sys_timer virt_timer = {
+	.init = virt_timer_init,
+};
+
+DT_MACHINE_START(VIRT, "Dummy Virtual Machine")
+	.init_irq	= gic_init_irq,
+	.handle_irq     = gic_handle_irq,
+	.timer		= &virt_timer,
+	.init_machine	= virt_init,
+	.dt_compat	= virt_dt_match,
+MACHINE_END

-- 
1.8.0

This patch adds support for SMP to mach-virt using the PSCI
infrastructure.

Signed-off-by: Will Deacon <redacted>
---
 arch/arm/mach-virt/Kconfig   |  1 +
 arch/arm/mach-virt/Makefile  |  1 +
 arch/arm/mach-virt/platsmp.c | 76 ++++++++++++++++++++++++++++++++++++++++++++
 arch/arm/mach-virt/virt.c    |  6 ++++
 4 files changed, 84 insertions(+)
 create mode 100644 arch/arm/mach-virt/platsmp.c

diff --git a/arch/arm/mach-virt/Kconfig b/arch/arm/mach-virt/Kconfig
index a568a2a..8958f0d 100644
--- a/arch/arm/mach-virt/Kconfig
+++ b/arch/arm/mach-virt/Kconfig

@@ -3,6 +3,7 @@ config ARCH_VIRT
 	select ARCH_WANT_OPTIONAL_GPIOLIB
 	select ARM_GIC
 	select ARM_ARCH_TIMER
+	select ARM_PSCI
 	select HAVE_SMP
 	select CPU_V7
 	select SPARSE_IRQ

diff --git a/arch/arm/mach-virt/Makefile b/arch/arm/mach-virt/Makefile
index 7ddbfa6..042afc1 100644
--- a/arch/arm/mach-virt/Makefile
+++ b/arch/arm/mach-virt/Makefile

@@ -3,3 +3,4 @@
 #
 
 obj-y					:= virt.o
+obj-$(CONFIG_SMP)			+= platsmp.o

diff --git a/arch/arm/mach-virt/platsmp.c b/arch/arm/mach-virt/platsmp.c
new file mode 100644
index 0000000..930362b
--- /dev/null
+++ b/arch/arm/mach-virt/platsmp.c

@@ -0,0 +1,76 @@
+/*
+ * Dummy Virtual Machine - does what it says on the tin.
+ *
+ * Copyright (C) 2012 ARM Ltd
+ * Author: Will Deacon <will.deacon@arm.com>
+ *
+ * This program is free software; you can redistribute it and/or modify
+ * it under the terms of the GNU General Public License version 2 as
+ * published by the Free Software Foundation.
+ *
+ * This program is distributed in the hope that it will be useful,
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
+ * GNU General Public License for more details.
+ *
+ * You should have received a copy of the GNU General Public License
+ * along with this program.  If not, see <http://www.gnu.org/licenses/>.
+ */
+
+#include <linux/init.h>
+#include <linux/smp.h>
+#include <linux/of.h>
+
+#include <asm/psci.h>
+#include <asm/smp_plat.h>
+#include <asm/hardware/gic.h>
+
+extern void secondary_startup(void);
+
+/*
+ * Enumerate the possible CPU set from the device tree.
+ */
+static void __init virt_smp_init_cpus(void)
+{
+	struct device_node *dn = NULL;
+	int cpu = 0;
+
+	while ((dn = of_find_node_by_type(dn, "cpu"))) {
+		if (cpu < NR_CPUS)
+			set_cpu_possible(cpu, true);
+		cpu++;
+	}
+
+	/* sanity check */
+	if (cpu > NR_CPUS)
+		pr_warning("no. of cores (%d) greater than configured maximum "
+			   "of %d - clipping\n",
+			   cpu, NR_CPUS);
+
+	set_smp_cross_call(gic_raise_softirq);
+}
+
+static void __init virt_smp_prepare_cpus(unsigned int max_cpus)
+{
+}
+
+static int __cpuinit virt_boot_secondary(unsigned int cpu,
+					 struct task_struct *idle)
+{
+	if (psci_ops.cpu_on)
+		return psci_ops.cpu_on(cpu_logical_map(cpu),
+				       __pa(secondary_startup));
+	return -ENODEV;
+}
+
+static void __cpuinit virt_secondary_init(unsigned int cpu)
+{
+	gic_secondary_init(0);
+}
+
+struct smp_operations __initdata virt_smp_ops = {
+	.smp_init_cpus		= virt_smp_init_cpus,
+	.smp_prepare_cpus	= virt_smp_prepare_cpus,
+	.smp_secondary_init	= virt_secondary_init,
+	.smp_boot_secondary	= virt_boot_secondary,
+};

diff --git a/arch/arm/mach-virt/virt.c b/arch/arm/mach-virt/virt.c
index 174b9da..d764835 100644
--- a/arch/arm/mach-virt/virt.c
+++ b/arch/arm/mach-virt/virt.c

@@ -20,6 +20,7 @@
 
 #include <linux/of_irq.h>
 #include <linux/of_platform.h>
+#include <linux/smp.h>
 
 #include <asm/arch_timer.h>
 #include <asm/hardware/gic.h>

@@ -56,10 +57,15 @@ static struct sys_timer virt_timer = {
 	.init = virt_timer_init,
 };
 
+#ifdef CONFIG_SMP
+extern struct smp_operations virt_smp_ops;
+#endif
+
 DT_MACHINE_START(VIRT, "Dummy Virtual Machine")
 	.init_irq	= gic_init_irq,
 	.handle_irq     = gic_handle_irq,
 	.timer		= &virt_timer,
 	.init_machine	= virt_init,
+	.smp		= smp_ops(virt_smp_ops),
 	.dt_compat	= virt_dt_match,
 MACHINE_END

-- 
1.8.0

On Monday 17 December 2012, Will Deacon wrote:

+
+ - function-base : The base ID from which the functions are offset.
+
+Main node optional properties:
+
+ - cpu_suspend   : Offset of CPU_SUSPEND ID from function-base
+
+ - cpu_off       : Offset of CPU_OFF ID from function-base
+
+ - cpu_on        : Offset of CPU_ON ID from function-base
+
+ - migrate       : Offset of MIGRATE ID from function-base

What is the benefit of the "function-base" property over just having
32 bit IDs for each function. For all I can tell, the interface does
not rely on the numbers to be consecutive, so removing the function-base
attribute would make the binding simpler as well as more flexible.

	Arnd

On Mon, 17 Dec 2012, Will Deacon wrote:

This patch adds support for the Power State Coordination Interface
defined by ARM, allowing Linux to request CPU-centric power-management
operations from firmware implementing the PSCI protocol.

Signed-off-by: Will Deacon <redacted>

[...]

+/*
+ * The following two functions are invoked via the invoke_psci_fn pointer
+ * and will not be inlined, allowing us to piggyback on the AAPCS.
+ */

To make sure the code is always in sync with the intent, you could mark 
those with  noinline as well.

+static int __invoke_psci_fn_hvc(u32 function_id, u32 arg0, u32 arg1, u32 arg2)
+{
+	asm volatile(
+			__asmeq("%0", "r0")
+			__asmeq("%1", "r1")
+			__asmeq("%2", "r2")
+			__asmeq("%3", "r3")
+			__HVC(0)
+		: "+r" (function_id)
+		: "r" (arg0), "r" (arg1), "r" (arg2));
+
+	return function_id;
+}
+
+static int __invoke_psci_fn_smc(u32 function_id, u32 arg0, u32 arg1, u32 arg2)
+{
+	asm volatile(
+			__asmeq("%0", "r0")
+			__asmeq("%1", "r1")
+			__asmeq("%2", "r2")
+			__asmeq("%3", "r3")
+			__SMC(0)
+		: "+r" (function_id)
+		: "r" (arg0), "r" (arg1), "r" (arg2));
+
+	return function_id;
+}
+
+static int psci_cpu_suspend(struct psci_power_state state,
+			    unsigned long entry_point)
+{
+	int err;
+	u32 fn, power_state;
+
+	fn = psci_function_id[PSCI_FN_CPU_SUSPEND];
+	power_state = psci_power_state_pack(state);
+	err = invoke_psci_fn(fn, power_state, (u32)entry_point, 0);

Why do you need the u32 cast here?

+	return psci_to_linux_errno(err);
+}
+
+static int psci_cpu_off(struct psci_power_state state)
+{
+	int err;
+	u32 fn, power_state;
+
+	fn = psci_function_id[PSCI_FN_CPU_OFF];
+	power_state = psci_power_state_pack(state);
+	err = invoke_psci_fn(fn, power_state, 0, 0);
+	return psci_to_linux_errno(err);
+}
+
+static int psci_cpu_on(unsigned long cpuid, unsigned long entry_point)
+{
+	int err;
+	u32 fn;
+
+	fn = psci_function_id[PSCI_FN_CPU_ON];
+	err = invoke_psci_fn(fn, cpuid, entry_point, 0);
+	return psci_to_linux_errno(err);
+}
+
+static int psci_migrate(unsigned long cpuid)
+{
+	int err;
+	u32 fn;
+
+	fn = psci_function_id[PSCI_FN_MIGRATE];
+	err = invoke_psci_fn(fn, cpuid, 0, 0);
+	return psci_to_linux_errno(err);
+}
+
+static const struct of_device_id psci_of_match[] __initconst = {
+	{ .compatible = "arm,psci",	},
+	{},
+};
+
+static int __init psci_init(void)
+{
+	struct device_node *np;
+	const char *method;
+	u32 base, id;
+
+	np = of_find_matching_node(NULL, psci_of_match);
+	if (!np)
+		return 0;
+
+	pr_info("probing function IDs from device-tree\n");

Having "probing function IDs from device-tree" in the middle of a kernel 
log isn't very informative.  Better make this more useful or remove it.

+
+	if (of_property_read_u32(np, "function-base", &base)) {
+		pr_warning("missing \"function-base\" property\n");

Same thing here: this lacks context in a kernel log.
And so on for the other occurrences.


Nicolas

On Mon, 17 Dec 2012, Will Deacon wrote:

Since commit 5587164eea you shouldn't need any of the above.

You don't need to surround prototype declaration here, unless your goal 
was to define a dummy virt_smp_ops when CONFIG_SMP is not selected?
Otherwise the reference below would break compilation.

+
 DT_MACHINE_START(VIRT, "Dummy Virtual Machine")
 	.init_irq	= gic_init_irq,
 	.handle_irq     = gic_handle_irq,
 	.timer		= &virt_timer,
 	.init_machine	= virt_init,
+	.smp		= smp_ops(virt_smp_ops),
 	.dt_compat	= virt_dt_match,
 MACHINE_END
-- 
1.8.0

Hi Arnd,

On Mon, Dec 17, 2012 at 08:00:11PM +0000, Arnd Bergmann wrote:

On Monday 17 December 2012, Will Deacon wrote:

quoted

+
+ - function-base : The base ID from which the functions are offset.
+
+Main node optional properties:
+
+ - cpu_suspend   : Offset of CPU_SUSPEND ID from function-base
+
+ - cpu_off       : Offset of CPU_OFF ID from function-base
+
+ - cpu_on        : Offset of CPU_ON ID from function-base
+
+ - migrate       : Offset of MIGRATE ID from function-base

What is the benefit of the "function-base" property over just having
32 bit IDs for each function. For all I can tell, the interface does
not rely on the numbers to be consecutive, so removing the function-base
attribute would make the binding simpler as well as more flexible.

Sure, happy to change that for v3.

Will

On Mon, Dec 17, 2012 at 08:51:27PM +0000, Nicolas Pitre wrote:

On Mon, 17 Dec 2012, Will Deacon wrote:

quoted

This patch adds support for the Power State Coordination Interface
defined by ARM, allowing Linux to request CPU-centric power-management
operations from firmware implementing the PSCI protocol.

Signed-off-by: Will Deacon <redacted>

[...]

quoted

+/*
+ * The following two functions are invoked via the invoke_psci_fn pointer
+ * and will not be inlined, allowing us to piggyback on the AAPCS.
+ */

To make sure the code is always in sync with the intent, you could mark 
those with  noinline as well.

Can do.

quoted

+static int __invoke_psci_fn_hvc(u32 function_id, u32 arg0, u32 arg1, u32 arg2)
+{
+	asm volatile(
+			__asmeq("%0", "r0")
+			__asmeq("%1", "r1")
+			__asmeq("%2", "r2")
+			__asmeq("%3", "r3")
+			__HVC(0)
+		: "+r" (function_id)
+		: "r" (arg0), "r" (arg1), "r" (arg2));
+
+	return function_id;
+}
+
+static int __invoke_psci_fn_smc(u32 function_id, u32 arg0, u32 arg1, u32 arg2)
+{
+	asm volatile(
+			__asmeq("%0", "r0")
+			__asmeq("%1", "r1")
+			__asmeq("%2", "r2")
+			__asmeq("%3", "r3")
+			__SMC(0)
+		: "+r" (function_id)
+		: "r" (arg0), "r" (arg1), "r" (arg2));
+
+	return function_id;
+}
+
+static int psci_cpu_suspend(struct psci_power_state state,
+			    unsigned long entry_point)
+{
+	int err;
+	u32 fn, power_state;
+
+	fn = psci_function_id[PSCI_FN_CPU_SUSPEND];
+	power_state = psci_power_state_pack(state);
+	err = invoke_psci_fn(fn, power_state, (u32)entry_point, 0);

Why do you need the u32 cast here?

That's a hangover from when entry_point was a void *. I'll fix that, thanks.

quoted

+static int __init psci_init(void)
+{
+	struct device_node *np;
+	const char *method;
+	u32 base, id;
+
+	np = of_find_matching_node(NULL, psci_of_match);
+	if (!np)
+		return 0;
+
+	pr_info("probing function IDs from device-tree\n");

Having "probing function IDs from device-tree" in the middle of a kernel 
log isn't very informative.  Better make this more useful or remove it.

quoted

+
+	if (of_property_read_u32(np, "function-base", &base)) {
+		pr_warning("missing \"function-base\" property\n");

Same thing here: this lacks context in a kernel log.
And so on for the other occurrences.

Actually, these are all prefixed with "psci: " thanks to the pr_fmt
definition at the top of the file. I can remove them if you like, but then
it's not obvious which parts of the PSCI code are available from looking at
a kernel boot log.

Cheers for the review,

Will

On Mon, Dec 17, 2012 at 09:45:52PM +0000, Nicolas Pitre wrote:

On Mon, 17 Dec 2012, Will Deacon wrote:

quoted

This patch adds support for SMP to mach-virt using the PSCI
infrastructure.

Signed-off-by: Will Deacon <redacted>

[...]

quoted

+/*
+ * Enumerate the possible CPU set from the device tree.
+ */
+static void __init virt_smp_init_cpus(void)
+{
+	struct device_node *dn = NULL;
+	int cpu = 0;
+
+	while ((dn = of_find_node_by_type(dn, "cpu"))) {
+		if (cpu < NR_CPUS)
+			set_cpu_possible(cpu, true);
+		cpu++;
+	}
+
+	/* sanity check */
+	if (cpu > NR_CPUS)
+		pr_warning("no. of cores (%d) greater than configured maximum "
+			   "of %d - clipping\n",
+			   cpu, NR_CPUS);

Since commit 5587164eea you shouldn't need any of the above.

There's going to be nothing left at this rate! Thanks.

quoted

+#ifdef CONFIG_SMP
+extern struct smp_operations virt_smp_ops;
+#endif

You don't need to surround prototype declaration here, unless your goal 
was to define a dummy virt_smp_ops when CONFIG_SMP is not selected?
Otherwise the reference below would break compilation.

Right you are, the smp_ops macro does the magic for us. I'll put together a
v3.

Cheers,

Will

quoted

+
 DT_MACHINE_START(VIRT, "Dummy Virtual Machine")
 	.init_irq	= gic_init_irq,
 	.handle_irq     = gic_handle_irq,
 	.timer		= &virt_timer,
 	.init_machine	= virt_init,
+	.smp		= smp_ops(virt_smp_ops),
 	.dt_compat	= virt_dt_match,
 MACHINE_END
-- 
1.8.0

On Mon, 17 Dec 2012, Will Deacon wrote:

From: Marc Zyngier <redacted>

Add support for the smallest, dumbest possible platform, to be
used as a guest for KVM or other hypervisors.

It only mandates a GIC and architected timers. Fits nicely with
a multiplatform zImage. Uses very little silicon area.

Signed-off-by: Marc Zyngier <redacted>
Signed-off-by: Will Deacon <redacted>
---
 arch/arm/Kconfig            |  2 ++
 arch/arm/Makefile           |  1 +
 arch/arm/mach-virt/Kconfig  |  9 +++++++
 arch/arm/mach-virt/Makefile |  5 ++++
 arch/arm/mach-virt/virt.c   | 65 +++++++++++++++++++++++++++++++++++++++++++++
 5 files changed, 82 insertions(+)
 create mode 100644 arch/arm/mach-virt/Kconfig
 create mode 100644 arch/arm/mach-virt/Makefile
 create mode 100644 arch/arm/mach-virt/virt.c

Should it come along with a DTS?

On Mon, 17 Dec 2012, Will Deacon wrote:

Considering that PSCI is actually needed only to boot secondary cpus,
maybe we want to select it if CONFIG_SMP is enabled?

Isn't there a better way to check whether PSCI is actually "enabled", as
in present in the device tree and initialized correctly?

Maybe we need a pcsi_enabled() static inline of some sort?

On 18/12/12 12:19, Stefano Stabellini wrote:

On Mon, 17 Dec 2012, Will Deacon wrote:

quoted

This patch adds support for SMP to mach-virt using the PSCI
infrastructure.

Signed-off-by: Will Deacon <redacted>

 arch/arm/mach-virt/Kconfig   |  1 +
 arch/arm/mach-virt/Makefile  |  1 +
 arch/arm/mach-virt/platsmp.c | 76 ++++++++++++++++++++++++++++++++++++++++++++
 arch/arm/mach-virt/virt.c    |  6 ++++
 4 files changed, 84 insertions(+)
 create mode 100644 arch/arm/mach-virt/platsmp.c

diff --git a/arch/arm/mach-virt/Kconfig b/arch/arm/mach-virt/Kconfig
index a568a2a..8958f0d 100644
--- a/arch/arm/mach-virt/Kconfig
+++ b/arch/arm/mach-virt/Kconfig

@@ -3,6 +3,7 @@ config ARCH_VIRT
 	select ARCH_WANT_OPTIONAL_GPIOLIB
 	select ARM_GIC
 	select ARM_ARCH_TIMER
+	select ARM_PSCI
 	select HAVE_SMP
 	select CPU_V7
 	select SPARSE_IRQ

Considering that PSCI is actually needed only to boot secondary cpus,
maybe we want to select it if CONFIG_SMP is enabled?

Well, I was considering using it to "power-off" the VM when the last CPU
powers itself off, and this would apply to UP as well.

quoted

diff --git a/arch/arm/mach-virt/Makefile b/arch/arm/mach-virt/Makefile
index 7ddbfa6..042afc1 100644
--- a/arch/arm/mach-virt/Makefile
+++ b/arch/arm/mach-virt/Makefile

@@ -3,3 +3,4 @@
 #
 
 obj-y					:= virt.o
+obj-$(CONFIG_SMP)			+= platsmp.o

diff --git a/arch/arm/mach-virt/platsmp.c b/arch/arm/mach-virt/platsmp.c
new file mode 100644
index 0000000..930362b
--- /dev/null
+++ b/arch/arm/mach-virt/platsmp.c

@@ -0,0 +1,76 @@
+/*
+ * Dummy Virtual Machine - does what it says on the tin.
+ *
+ * Copyright (C) 2012 ARM Ltd
+ * Author: Will Deacon <will.deacon@arm.com>
+ *
+ * This program is free software; you can redistribute it and/or modify
+ * it under the terms of the GNU General Public License version 2 as
+ * published by the Free Software Foundation.
+ *
+ * This program is distributed in the hope that it will be useful,
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
+ * GNU General Public License for more details.
+ *
+ * You should have received a copy of the GNU General Public License
+ * along with this program.  If not, see <http://www.gnu.org/licenses/>.
+ */
+
+#include <linux/init.h>
+#include <linux/smp.h>
+#include <linux/of.h>
+
+#include <asm/psci.h>
+#include <asm/smp_plat.h>
+#include <asm/hardware/gic.h>
+
+extern void secondary_startup(void);
+
+/*
+ * Enumerate the possible CPU set from the device tree.
+ */
+static void __init virt_smp_init_cpus(void)
+{
+	struct device_node *dn = NULL;
+	int cpu = 0;
+
+	while ((dn = of_find_node_by_type(dn, "cpu"))) {
+		if (cpu < NR_CPUS)
+			set_cpu_possible(cpu, true);
+		cpu++;
+	}
+
+	/* sanity check */
+	if (cpu > NR_CPUS)
+		pr_warning("no. of cores (%d) greater than configured maximum "
+			   "of %d - clipping\n",
+			   cpu, NR_CPUS);
+
+	set_smp_cross_call(gic_raise_softirq);
+}
+
+static void __init virt_smp_prepare_cpus(unsigned int max_cpus)
+{
+}
+
+static int __cpuinit virt_boot_secondary(unsigned int cpu,
+					 struct task_struct *idle)
+{
+	if (psci_ops.cpu_on)
+		return psci_ops.cpu_on(cpu_logical_map(cpu),
+				       __pa(secondary_startup));
+	return -ENODEV;
+}

Isn't there a better way to check whether PSCI is actually "enabled", as
in present in the device tree and initialized correctly?

All methods are optional, so I'm afraid you have to check for their
validity each time you want to access one.

Maybe we need a pcsi_enabled() static inline of some sort?

quoted

+static void __cpuinit virt_secondary_init(unsigned int cpu)
+{
+	gic_secondary_init(0);
+}
+
+struct smp_operations __initdata virt_smp_ops = {
+	.smp_init_cpus		= virt_smp_init_cpus,
+	.smp_prepare_cpus	= virt_smp_prepare_cpus,
+	.smp_secondary_init	= virt_secondary_init,
+	.smp_boot_secondary	= virt_boot_secondary,
+};

diff --git a/arch/arm/mach-virt/virt.c b/arch/arm/mach-virt/virt.c
index 174b9da..d764835 100644
--- a/arch/arm/mach-virt/virt.c
+++ b/arch/arm/mach-virt/virt.c

@@ -20,6 +20,7 @@
 
 #include <linux/of_irq.h>
 #include <linux/of_platform.h>
+#include <linux/smp.h>
 
 #include <asm/arch_timer.h>
 #include <asm/hardware/gic.h>

@@ -56,10 +57,15 @@ static struct sys_timer virt_timer = {
 	.init = virt_timer_init,
 };
 
+#ifdef CONFIG_SMP
+extern struct smp_operations virt_smp_ops;
+#endif
+
 DT_MACHINE_START(VIRT, "Dummy Virtual Machine")
 	.init_irq	= gic_init_irq,
 	.handle_irq     = gic_handle_irq,
 	.timer		= &virt_timer,
 	.init_machine	= virt_init,
+	.smp		= smp_ops(virt_smp_ops),
 	.dt_compat	= virt_dt_match,
 MACHINE_END

-- 
1.8.0

-- 
Jazz is not dead. It just smells funny...

Hi Stefano,

On Tue, Dec 18, 2012 at 12:04:38PM +0000, Stefano Stabellini wrote:

On Mon, 17 Dec 2012, Will Deacon wrote:

quoted

From: Marc Zyngier <redacted>

Add support for the smallest, dumbest possible platform, to be
used as a guest for KVM or other hypervisors.

It only mandates a GIC and architected timers. Fits nicely with
a multiplatform zImage. Uses very little silicon area.

Signed-off-by: Marc Zyngier <redacted>
Signed-off-by: Will Deacon <redacted>
---
 arch/arm/Kconfig            |  2 ++
 arch/arm/Makefile           |  1 +
 arch/arm/mach-virt/Kconfig  |  9 +++++++
 arch/arm/mach-virt/Makefile |  5 ++++
 arch/arm/mach-virt/virt.c   | 65 +++++++++++++++++++++++++++++++++++++++++++++
 5 files changed, 82 insertions(+)
 create mode 100644 arch/arm/mach-virt/Kconfig
 create mode 100644 arch/arm/mach-virt/Makefile
 create mode 100644 arch/arm/mach-virt/virt.c

Should it come along with a DTS?

The only things the platform needs are GIC, timers, memory and a CPU.
Furthermore, the location, size, frequency etc properties of these aren't
fixed, so a dts would be fairly useless because it will probably not match
the particular mach-virt instance you're targetting.

For kvmtool, I've been generating the device-tree at runtime based on how
kvmtool is invoked and it's been working pretty well so far.

Will

On Tue, 18 Dec 2012, Will Deacon wrote:

Hi Stefano,

On Tue, Dec 18, 2012 at 12:04:38PM +0000, Stefano Stabellini wrote:

quoted

On Mon, 17 Dec 2012, Will Deacon wrote:

quoted

From: Marc Zyngier <redacted>

Add support for the smallest, dumbest possible platform, to be
used as a guest for KVM or other hypervisors.

It only mandates a GIC and architected timers. Fits nicely with
a multiplatform zImage. Uses very little silicon area.

Signed-off-by: Marc Zyngier <redacted>
Signed-off-by: Will Deacon <redacted>
---
 arch/arm/Kconfig            |  2 ++
 arch/arm/Makefile           |  1 +
 arch/arm/mach-virt/Kconfig  |  9 +++++++
 arch/arm/mach-virt/Makefile |  5 ++++
 arch/arm/mach-virt/virt.c   | 65 +++++++++++++++++++++++++++++++++++++++++++++
 5 files changed, 82 insertions(+)
 create mode 100644 arch/arm/mach-virt/Kconfig
 create mode 100644 arch/arm/mach-virt/Makefile
 create mode 100644 arch/arm/mach-virt/virt.c

Should it come along with a DTS?

The only things the platform needs are GIC, timers, memory and a CPU.
Furthermore, the location, size, frequency etc properties of these aren't
fixed, so a dts would be fairly useless because it will probably not match
the particular mach-virt instance you're targetting.

For kvmtool, I've been generating the device-tree at runtime based on how
kvmtool is invoked and it's been working pretty well so far.