Skip to contentSkip to conversation

ARM64: consolidate ACPI/DT SMP initialization

PATCH New Lorenzo Pieralisi 2 participants 6 messages May 13, 2015 – Jul 13, 2015 arm64

lore
Share & tools
lore
#0 lore raw

v2 cover ARM64: consolidate ACPI/DT SMP initialization

This patch is a v2 of a previous posting:

http://lists.infradead.org/pipermail/linux-arm-kernel/2015-May/341579.html

v1 -> v2:

- Reinstated/clarified enable-method comment in asm/cpu_ops.h header
- Changed ACPI enable-method error log message
- Changed u32 to __be32 pointer in of_get_cpu_mpidr()

Cc Will Deacon <[email protected]>
Cc Hanjun Guo <[email protected]>
Cc Sudeep Holla <[email protected]>
Cc Catalin Marinas <[email protected]>
Cc Mark Rutland <[email protected]>

Lorenzo Pieralisi (2):
ARM64: kernel: make cpu_ops hooks DT agnostic
ARM64: kernel: unify ACPI and DT cpus initialization

 arch/arm64/include/asm/acpi.h      |   4 +
 arch/arm64/include/asm/cpu_ops.h   |  27 +++--
 arch/arm64/include/asm/smp.h       |   2 +-
 arch/arm64/kernel/acpi.c           | 124 --------------------
 arch/arm64/kernel/cpu_ops.c        |  72 +++++++-----
 arch/arm64/kernel/cpuidle.c        |   7 +-
 arch/arm64/kernel/psci.c           |  11 +-
 arch/arm64/kernel/setup.c          |   7 +-
 arch/arm64/kernel/smp.c            | 230 ++++++++++++++++++++++++++++---------
 arch/arm64/kernel/smp_spin_table.c |   8 +-
 10 files changed, 257 insertions(+), 235 deletions(-)
Signature · 1 line
2.2.1
#1 lore raw

PATCH v2 1/2 ARM64: kernel: make cpu_ops hooks DT agnostic

ARM64 CPU operations such as cpu_init and cpu_init_idle take a struct device_node pointer as a parameter, which corresponds to the device tree node of the logical cpu on which the operation has to be applied.

With the advent of ACPI on arm64, where MADT static table entries are used to initialize cpus, the device tree node parameter in cpu_ops hooks become useless when booting with ACPI, since in that case cpu device tree nodes are not present and can not be used for cpu initialization.

The current cpu_init hook requires a struct device_node pointer parameter because it is called while parsing the device tree to initialize CPUs, when the cpu_logical_map (that is used to match a cpu node reg property to a device tree node) for a given logical cpu id is not set up yet. This means that the cpu_init hook cannot rely on the of_get_cpu_node function to retrieve the device tree node corresponding to the logical cpu id passed in as parameter, so the cpu device tree node must be passed in as a parameter to fix this catch-22 dependency cycle.

This patch reshuffles the cpu_logical_map initialization code so that the cpu_init cpu_ops hook can safely use the of_get_cpu_node function to retrieve the cpu device tree node, removing the need for the device tree node pointer parameter.

In the process, the patch removes device tree node parameters from all cpu_ops hooks, in preparation for SMP DT/ACPI cpus initialization consolidation.

---
 arch/arm64/include/asm/cpu_ops.h   | 26 ++++++++++++++-----------
 arch/arm64/kernel/acpi.c           |  2 +-
 arch/arm64/kernel/cpu_ops.c        | 23 +++++++++++-----------
 arch/arm64/kernel/cpuidle.c        |  7 +------
 arch/arm64/kernel/psci.c           | 11 +++++++----
 arch/arm64/kernel/smp.c            | 39 +++++++++++++++++++++++++++-----------
 arch/arm64/kernel/smp_spin_table.c |  8 +++++++-
 7 files changed, 70 insertions(+), 46 deletions(-)
arch/arm64/include/asm/cpu_ops.h +15 -11
index 5a31d67..7d8b999 100644
@@ -19,15 +19,15 @@
 #include <linux/init.h>
 #include <linux/threads.h>
 
-struct device_node;
-
 /**
  * struct cpu_operations - Callback operations for hotplugging CPUs.
  *
  * @name:	Name of the property as appears in a devicetree cpu node's
- *		enable-method property.
- * @cpu_init:	Reads any data necessary for a specific enable-method from the
- *		devicetree, for a given cpu node and proposed logical id.
+ *		enable-method property. On systems booting with ACPI, @name
+ *		identifies the struct cpu_operations entry corresponding to
+ *		the boot protocol specified in the ACPI MADT table.
+ * @cpu_init:	Reads any data necessary for a specific enable-method for a
+ *		proposed logical id.
  * @cpu_prepare: Early one-time preparation step for a cpu. If there is a
  *		mechanism for doing so, tests whether it is possible to boot
  *		the given CPU.
@@ -40,15 +40,15 @@ struct device_node;
  * @cpu_die:	Makes a cpu leave the kernel. Must not fail. Called from the
  *		cpu being killed.
  * @cpu_kill:  Ensures a cpu has left the kernel. Called from another cpu.
- * @cpu_init_idle: Reads any data necessary to initialize CPU idle states from
- *		devicetree, for a given cpu node and proposed logical id.
+ * @cpu_init_idle: Reads any data necessary to initialize CPU idle states for
+ *		   a proposed logical id.
  * @cpu_suspend: Suspends a cpu and saves the required context. May fail owing
  *               to wrong parameters or error conditions. Called from the
  *               CPU being suspended. Must be called with IRQs disabled.
  */
 struct cpu_operations {
 	const char	*name;
-	int		(*cpu_init)(struct device_node *, unsigned int);
+	int		(*cpu_init)(unsigned int);
 	int		(*cpu_prepare)(unsigned int);
 	int		(*cpu_boot)(unsigned int);
 	void		(*cpu_postboot)(void);
@@ -58,14 +58,18 @@ struct cpu_operations {
 	int		(*cpu_kill)(unsigned int cpu);
 #endif
 #ifdef CONFIG_CPU_IDLE
-	int		(*cpu_init_idle)(struct device_node *, unsigned int);
+	int		(*cpu_init_idle)(unsigned int);
 	int		(*cpu_suspend)(unsigned long);
 #endif
 };
 
 extern const struct cpu_operations *cpu_ops[NR_CPUS];
-int __init cpu_read_ops(struct device_node *dn, int cpu);
-void __init cpu_read_bootcpu_ops(void);
+int __init cpu_read_ops(int cpu);
 const struct cpu_operations *cpu_get_ops(const char *name);
 
+static inline void __init cpu_read_bootcpu_ops(void)
+{
+	cpu_read_ops(0);
+}
+
 #endif /* ifndef __ASM_CPU_OPS_H */
arch/arm64/kernel/acpi.c +1 -1
index 8b83955..945cc1f 100644
@@ -154,7 +154,7 @@ acpi_map_gic_cpu_interface(struct acpi_madt_generic_interrupt *processor)
 		if (!cpu_ops[enabled_cpus])
 			return;
 
-		if (cpu_ops[enabled_cpus]->cpu_init(NULL, enabled_cpus))
+		if (cpu_ops[enabled_cpus]->cpu_init(enabled_cpus))
 			return;
 
 		/* map the logical cpu id to cpu MPIDR */
arch/arm64/kernel/cpu_ops.c +11 -12
index fb8ff9b..bbda2d2 100644
@@ -52,9 +52,18 @@ const struct cpu_operations * __init cpu_get_ops(const char *name)
 /*
  * Read a cpu's enable method from the device tree and record it in cpu_ops.
  */
-int __init cpu_read_ops(struct device_node *dn, int cpu)
+int __init cpu_read_ops(int cpu)
 {
-	const char *enable_method = of_get_property(dn, "enable-method", NULL);
+	const char *enable_method;
+	struct device_node *dn = of_get_cpu_node(cpu, NULL);
+
+	if (!dn) {
+		if (!cpu)
+			pr_err("Failed to find device node for boot cpu\n");
+		return -ENODEV;
+	}
+
+	enable_method = of_get_property(dn, "enable-method", NULL);
 	if (!enable_method) {
 		/*
 		 * The boot CPU may not have an enable method (e.g. when
@@ -75,13 +84,3 @@ int __init cpu_read_ops(struct device_node *dn, int cpu)
 
 	return 0;
 }
-
-void __init cpu_read_bootcpu_ops(void)
-{
-	struct device_node *dn = of_get_cpu_node(0, NULL);
-	if (!dn) {
-		pr_err("Failed to find device node for boot cpu\n");
-		return;
-	}
-	cpu_read_ops(dn, 0);
-}
arch/arm64/kernel/cpuidle.c +1 -6
index a78143a..f8aa169 100644
@@ -18,15 +18,10 @@
 int arm_cpuidle_init(unsigned int cpu)
 {
 	int ret = -EOPNOTSUPP;
-	struct device_node *cpu_node = of_cpu_device_node_get(cpu);
-
-	if (!cpu_node)
-		return -ENODEV;
 
 	if (cpu_ops[cpu] && cpu_ops[cpu]->cpu_init_idle)
-		ret = cpu_ops[cpu]->cpu_init_idle(cpu_node, cpu);
+		ret = cpu_ops[cpu]->cpu_init_idle(cpu);
 
-	of_node_put(cpu_node);
 	return ret;
 }
 
arch/arm64/kernel/psci.c +7 -4
index ea18cb5..efe3480 100644
@@ -186,12 +186,15 @@ static int psci_migrate_info_type(void)
 	return err;
 }
 
-static int __maybe_unused cpu_psci_cpu_init_idle(struct device_node *cpu_node,
-						 unsigned int cpu)
+static int __maybe_unused cpu_psci_cpu_init_idle(unsigned int cpu)
 {
 	int i, ret, count = 0;
 	struct psci_power_state *psci_states;
-	struct device_node *state_node;
+	struct device_node *state_node, *cpu_node;
+
+	cpu_node = of_get_cpu_node(cpu, NULL);
+	if (!cpu_node)
+		return -ENODEV;
 
 	/*
 	 * If the PSCI cpu_suspend function hook has not been initialized
@@ -444,7 +447,7 @@ int __init psci_acpi_init(void)
 
 #ifdef CONFIG_SMP
 
-static int __init cpu_psci_cpu_init(struct device_node *dn, unsigned int cpu)
+static int __init cpu_psci_cpu_init(unsigned int cpu)
 {
 	return 0;
 }
arch/arm64/kernel/smp.c +28 -11
index 2cb0081..98eb68b 100644
@@ -319,6 +319,23 @@ void __init smp_prepare_boot_cpu(void)
 }
 
 /*
+ * Initialize cpu operations for a logical cpu and
+ * set it in the possible mask on success
+ */
+static int __init smp_cpu_setup(int cpu)
+{
+	if (cpu_read_ops(cpu))
+		return -ENODEV;
+
+	if (cpu_ops[cpu]->cpu_init(cpu))
+		return -ENODEV;
+
+	set_cpu_possible(cpu, true);
+
+	return 0;
+}
+
+/*
  * Enumerate the possible CPU set from the device tree and build the
  * cpu logical map array containing MPIDR values related to logical
  * cpus. Assumes that cpu_logical_map(0) has already been initialized.
@@ -395,12 +412,6 @@ void __init of_smp_init_cpus(void)
 		if (cpu >= NR_CPUS)
 			goto next;
 
-		if (cpu_read_ops(dn, cpu) != 0)
-			goto next;
-
-		if (cpu_ops[cpu]->cpu_init(dn, cpu))
-			goto next;
-
 		pr_debug("cpu logical map 0x%llx\n", hwid);
 		cpu_logical_map(cpu) = hwid;
 next:
@@ -418,12 +429,18 @@ next:
 	}
 
 	/*
-	 * All the cpus that made it to the cpu_logical_map have been
-	 * validated so set them as possible cpus.
+	 * We need to set the cpu_logical_map entries before enabling
+	 * the cpus so that cpu processor description entries (DT cpu nodes
+	 * and ACPI MADT entries) can be retrieved by matching the cpu hwid
+	 * with entries in cpu_logical_map while initializing the cpus.
+	 * If the cpu set-up fails, invalidate the cpu_logical_map entry.
 	 */
-	for (i = 0; i < NR_CPUS; i++)
-		if (cpu_logical_map(i) != INVALID_HWID)
-			set_cpu_possible(i, true);
+	for (i = 1; i < NR_CPUS; i++) {
+		if (cpu_logical_map(i) != INVALID_HWID) {
+			if (smp_cpu_setup(i))
+				cpu_logical_map(i) = INVALID_HWID;
+		}
+	}
 }
 
 void __init smp_prepare_cpus(unsigned int max_cpus)
arch/arm64/kernel/smp_spin_table.c +7 -1
index 14944e5..aef3605 100644
@@ -49,8 +49,14 @@ static void write_pen_release(u64 val)
 }
 
 
-static int smp_spin_table_cpu_init(struct device_node *dn, unsigned int cpu)
+static int smp_spin_table_cpu_init(unsigned int cpu)
 {
+	struct device_node *dn;
+
+	dn = of_get_cpu_node(cpu, NULL);
+	if (!dn)
+		return -ENODEV;
+
 	/*
 	 * Determine the address from which the CPU is polling.
 	 */
Signature · 1 line
2.2.1
#2 lore raw

PATCH v2 2/2 ARM64: kernel: unify ACPI and DT cpus initialization

In reply to Lorenzo Pieralisi

The code that initializes cpus on arm64 is currently split in two different code paths that carry out DT and ACPI cpus initialization.

Most of the code executing SMP initialization is common and should be merged to reduce discrepancies between ACPI and DT initialization and to have code initializing cpus in a single common place in the kernel.

This patch refactors arm64 SMP cpus initialization code to merge ACPI and DT boot paths in a common file and to create sanity checks that can be reused by both boot methods.

Current code assumes PSCI is the only available boot method when arm64 boots with ACPI; this can be easily extended if/when the ACPI parking protocol is merged into the kernel.

---
 arch/arm64/include/asm/acpi.h    |   4 +
 arch/arm64/include/asm/cpu_ops.h |   1 -
 arch/arm64/include/asm/smp.h     |   2 +-
 arch/arm64/kernel/acpi.c         | 124 -------------------------
 arch/arm64/kernel/cpu_ops.c      |  65 ++++++++-----
 arch/arm64/kernel/setup.c        |   7 +-
 arch/arm64/kernel/smp.c          | 191 ++++++++++++++++++++++++++++++---------
 7 files changed, 196 insertions(+), 198 deletions(-)
arch/arm64/include/asm/acpi.h +4 -0
index 59c05d8..1240b86 100644
@@ -93,4 +93,8 @@ static inline bool acpi_psci_use_hvc(void) { return false; }
 static inline void acpi_init_cpus(void) { }
 #endif /* CONFIG_ACPI */
 
+static inline const char *acpi_get_enable_method(int cpu)
+{
+	return acpi_psci_present() ? "psci" : NULL;
+}
 #endif /*_ASM_ACPI_H*/
arch/arm64/include/asm/cpu_ops.h +0 -1
index 7d8b999..8f03446 100644
@@ -65,7 +65,6 @@ struct cpu_operations {
 
 extern const struct cpu_operations *cpu_ops[NR_CPUS];
 int __init cpu_read_ops(int cpu);
-const struct cpu_operations *cpu_get_ops(const char *name);
 
 static inline void __init cpu_read_bootcpu_ops(void)
 {
arch/arm64/include/asm/smp.h +1 -1
index bf22650..db02be8 100644
@@ -42,7 +42,7 @@ extern void handle_IPI(int ipinr, struct pt_regs *regs);
  * Discover the set of possible CPUs and determine their
  * SMP operations.
  */
-extern void of_smp_init_cpus(void);
+extern void smp_init_cpus(void);
 
 /*
  * Provide a function to raise an IPI cross call on CPUs in callmap.
arch/arm64/kernel/acpi.c +0 -124

136 lines not on this page · raw message

arch/arm64/kernel/cpu_ops.c +41 -24

95 lines not on this page · raw message

arch/arm64/kernel/setup.c +2 -5
index 7475313..508cca1 100644
@@ -408,16 +408,13 @@ void __init setup_arch(char **cmdline_p)
 	if (acpi_disabled) {
 		unflatten_device_tree();
 		psci_dt_init();
-		cpu_read_bootcpu_ops();
-#ifdef CONFIG_SMP
-		of_smp_init_cpus();
-#endif
 	} else {
 		psci_acpi_init();
-		acpi_init_cpus();
 	}
 
+	cpu_read_bootcpu_ops();
 #ifdef CONFIG_SMP
+	smp_init_cpus();
 	smp_build_mpidr_hash();
 #endif
 
arch/arm64/kernel/smp.c +148 -43

239 lines not on this page · raw message

Signature · 1 line
2.2.1
#3 lore raw

PATCH v2 2/2 ARM64: kernel: unify ACPI and DT cpus initialization

Quoted 16 lines · Lorenzo Pieralisi

The code that initializes cpus on arm64 is currently split in two different code paths that carry out DT and ACPI cpus initialization.

Most of the code executing SMP initialization is common and should be merged to reduce discrepancies between ACPI and DT initialization and to have code initializing cpus in a single common place in the kernel.

This patch refactors arm64 SMP cpus initialization code to merge ACPI and DT boot paths in a common file and to create sanity checks that can be reused by both boot methods.

Current code assumes PSCI is the only available boot method when arm64 boots with ACPI; this can be easily extended if/when the ACPI parking protocol is merged into the kernel.

Quoted 8 lines

[snip...]

These patches look pretty good so far, but I've only just started reviewing them. This one raised the question: do we know if there is someone already working on parking protocol? I've seen a form of it working on AMD Seattles but I don't recall seeing any patches being submitted -- I'll search archives again, just in case I missed it the first time, but if there is someone working on it, that would be good to know.

Thanks.

Signature · 8 lines
ciao,
al
-----------------------------------
Al Stone
Software Engineer
Red Hat, Inc.
ahs3 at redhat.com
-----------------------------------
#4 lore raw

PATCH v2 2/2 ARM64: kernel: unify ACPI and DT cpus initialization

[restored CC list]

Quoted 27 lines · Al Stone
Quoted 24 lines · Lorenzo Pieralisi

The code that initializes cpus on arm64 is currently split in two different code paths that carry out DT and ACPI cpus initialization.

Most of the code executing SMP initialization is common and should be merged to reduce discrepancies between ACPI and DT initialization and to have code initializing cpus in a single common place in the kernel.

This patch refactors arm64 SMP cpus initialization code to merge ACPI and DT boot paths in a common file and to create sanity checks that can be reused by both boot methods.

Current code assumes PSCI is the only available boot method when arm64 boots with ACPI; this can be easily extended if/when the ACPI parking protocol is merged into the kernel.

[snip...]

Quoted 7 lines

These patches look pretty good so far, but I've only just started reviewing them. This one raised the question: do we know if there is someone already working on parking protocol? I've seen a form of it working on AMD Seattles but I don't recall seeing any patches being submitted -- I'll search archives again, just in case I missed it the first time, but if there is someone working on it, that would be good to know.

I have seen patches for the ACPI parking protocol but nothing posted publicly. This patch consolidates what exists today, honestly I do not think that putting together a patch to enable the ACPI parking protocol is a big deal on top of this set, keeping in mind that it should be used as a fallback solution if PSCI can't be implemented (ie AMD Seattle must boot with PSCI).

Lorenzo

#5 lore raw

PATCH v2 2/2 ARM64: kernel: unify ACPI and DT cpus initialization

In reply to Al Stone
Quoted 27 lines · Al Stone
Quoted 24 lines · Lorenzo Pieralisi

The code that initializes cpus on arm64 is currently split in two different code paths that carry out DT and ACPI cpus initialization.

Most of the code executing SMP initialization is common and should be merged to reduce discrepancies between ACPI and DT initialization and to have code initializing cpus in a single common place in the kernel.

This patch refactors arm64 SMP cpus initialization code to merge ACPI and DT boot paths in a common file and to create sanity checks that can be reused by both boot methods.

Current code assumes PSCI is the only available boot method when arm64 boots with ACPI; this can be easily extended if/when the ACPI parking protocol is merged into the kernel.

[snip...]

Quoted 7 lines

These patches look pretty good so far, but I've only just started reviewing them. This one raised the question: do we know if there is someone already working on parking protocol? I've seen a form of it working on AMD Seattles but I don't recall seeing any patches being submitted -- I'll search archives again, just in case I missed it the first time, but if there is someone working on it, that would be good to know.

FYI, I took M.Salter's initial patch for parking protocol and refactored it for upstream on top of 4.2-rc2, I will drop it on the list asap.

Thanks,
Lorenzo