Merge branch 'refs/heads/bits/220-tso' into asahi-wip

Author: marcan

arch/arm64/Kconfig

@@ -381,6 +381,9 @@ config KASAN_SHADOW_OFFSET
 config UNWIND_TABLES
 	bool
 
+config ARM64_ACTLR_STATE
+	bool
+
 source "arch/arm64/Kconfig.platforms"
 
 menu "Kernel Features"
@@ -2128,6 +2131,17 @@ config ARM64_DEBUG_PRIORITY_MASKING
 	  If unsure, say N
 endif # ARM64_PSEUDO_NMI
 
+config ARM64_MEMORY_MODEL_CONTROL
+	bool "Runtime memory model control"
+	default ARCH_APPLE
+	select ARM64_ACTLR_STATE
+	help
+	  Some ARM64 CPUs support runtime switching of the CPU memory
+	  model, which can be useful to emulate other CPU architectures
+	  which have different memory models. Say Y to enable support
+	  for the PR_SET_MEM_MODEL/PR_GET_MEM_MODEL prctl() calls on
+	  CPUs with this feature.
+
 config RELOCATABLE
 	bool "Build a relocatable kernel image" if EXPERT
 	select ARCH_HAS_RELR

arch/arm64/include/asm/apple_cpufeature.h

@@ -0,0 +1,15 @@
+// SPDX-License-Identifier: GPL-2.0
+
+#ifndef __ASM_APPLE_CPUFEATURES_H
+#define __ASM_APPLE_CPUFEATURES_H
+
+#include <linux/bits.h>
+#include <asm/sysreg.h>
+
+#define AIDR_APPLE_TSO_SHIFT	9
+#define AIDR_APPLE_TSO		BIT(9)
+
+#define ACTLR_APPLE_TSO_SHIFT	1
+#define ACTLR_APPLE_TSO		BIT(1)
+
+#endif

arch/arm64/include/asm/cpufeature.h

@@ -915,6 +915,12 @@ static inline unsigned int get_vmid_bits(u64 mmfr1)
 	return 8;
 }
 
+static __always_inline bool system_has_actlr_state(void)
+{
+	return IS_ENABLED(CONFIG_ARM64_ACTLR_STATE) &&
+		cpus_have_const_cap(ARM64_HAS_TSO_APPLE);
+}
+
 struct arm64_ftr_reg *get_arm64_ftr_reg(u32 sys_id);
 
 extern struct arm64_ftr_override id_aa64mmfr1_override;

arch/arm64/include/asm/processor.h

@@ -179,6 +179,9 @@ struct thread_struct {
 	u64			sctlr_user;
 	u64			svcr;
 	u64			tpidr2_el0;
+#ifdef CONFIG_ARM64_ACTLR_STATE
+	u64			actlr;
+#endif
 };
 
 static inline unsigned int thread_get_vl(struct thread_struct *thread,

arch/arm64/kernel/Makefile

@@ -34,7 +34,7 @@ obj-y			:= debug-monitors.o entry.o irq.o fpsimd.o		\
 			   cpufeature.o alternative.o cacheinfo.o		\
 			   smp.o smp_spin_table.o topology.o smccc-call.o	\
 			   syscall.o proton-pack.o idreg-override.o idle.o	\
-			   patching.o
+			   patching.o cpufeature_impdef.o
 
 obj-$(CONFIG_COMPAT)			+= sys32.o signal32.o			\
 					   sys_compat.o

arch/arm64/kernel/cpufeature.c

@@ -134,6 +134,8 @@ DEFINE_STATIC_KEY_FALSE(arm64_mismatched_32bit_el0);
  */
 static cpumask_var_t cpu_32bit_el0_mask __cpumask_var_read_mostly;
 
+void __init init_cpu_hwcaps_indirect_list_impdef(void);
+
 void dump_cpu_features(void)
 {
 	/* file-wide pr_fmt adds "CPU features: " prefix */
@@ -946,7 +948,7 @@ static void init_cpu_ftr_reg(u32 sys_reg, u64 new)
 extern const struct arm64_cpu_capabilities arm64_errata[];
 static const struct arm64_cpu_capabilities arm64_features[];
 
-static void __init
+void __init
 init_cpu_hwcaps_indirect_list_from_array(const struct arm64_cpu_capabilities *caps)
 {
 	for (; caps->matches; caps++) {
@@ -1046,6 +1048,7 @@ void __init init_cpu_features(struct cpuinfo_arm64 *info)
 	 * before we handle the boot CPU below.
 	 */
 	init_cpu_hwcaps_indirect_list();
+	init_cpu_hwcaps_indirect_list_impdef();
 
 	/*
 	 * Detect and enable early CPU capabilities based on the boot CPU,
@@ -1414,7 +1417,7 @@ has_always(const struct arm64_cpu_capabilities *entry, int scope)
 	return true;
 }
 
-static bool
+bool
 feature_matches(u64 reg, const struct arm64_cpu_capabilities *entry)
 {
 	int val = cpuid_feature_extract_field_width(reg, entry->field_pos,

arch/arm64/kernel/cpufeature_impdef.c

@@ -0,0 +1,62 @@
+// SPDX-License-Identifier: GPL-2.0-only
+/*
+ * Contains implementation-defined CPU feature definitions.
+ */
+
+#include <asm/cpufeature.h>
+#include <asm/apple_cpufeature.h>
+
+void __init init_cpu_hwcaps_indirect_list_from_array(const struct arm64_cpu_capabilities *caps);
+bool feature_matches(u64 reg, const struct arm64_cpu_capabilities *entry);
+
+bool has_apple_feature(const struct arm64_cpu_capabilities *entry, int scope)
+{
+	u64 val;
+	WARN_ON(scope != SCOPE_SYSTEM);
+
+	if (read_cpuid_implementor() != ARM_CPU_IMP_APPLE)
+		return false;
+
+	val = read_sysreg(aidr_el1);
+	return feature_matches(val, entry);
+}
+
+bool has_tso_fixed(const struct arm64_cpu_capabilities *entry, int scope)
+{
+	/* List of CPUs that always use the TSO memory model */
+	static const struct midr_range fixed_tso_list[] = {
+		MIDR_ALL_VERSIONS(MIDR_NVIDIA_DENVER),
+		MIDR_ALL_VERSIONS(MIDR_NVIDIA_CARMEL),
+		MIDR_ALL_VERSIONS(MIDR_FUJITSU_A64FX),
+		{ /* sentinel */ }
+	};
+
+	return is_midr_in_range_list(read_cpuid_id(), fixed_tso_list);
+}
+
+static const struct arm64_cpu_capabilities arm64_impdef_features[] = {
+#ifdef CONFIG_ARM64_MEMORY_MODEL_CONTROL
+	{
+		.desc = "TSO memory model (Apple)",
+		.capability = ARM64_HAS_TSO_APPLE,
+		.type = ARM64_CPUCAP_SYSTEM_FEATURE,
+		.matches = has_apple_feature,
+		.field_pos = AIDR_APPLE_TSO_SHIFT,
+		.field_width = 1,
+		.sign = FTR_UNSIGNED,
+		.min_field_value = 1,
+	},
+	{
+		.desc = "TSO memory model (Fixed)",
+		.capability = ARM64_HAS_TSO_FIXED,
+		.type = ARM64_CPUCAP_SYSTEM_FEATURE,
+		.matches = has_tso_fixed,
+	},
+#endif
+	{},
+};
+
+void __init init_cpu_hwcaps_indirect_list_impdef(void)
+{
+	init_cpu_hwcaps_indirect_list_from_array(arm64_impdef_features);
+}

arch/arm64/kernel/process.c

@@ -43,6 +43,7 @@
 #include <linux/stacktrace.h>
 
 #include <asm/alternative.h>
+#include <asm/apple_cpufeature.h>
 #include <asm/compat.h>
 #include <asm/cpufeature.h>
 #include <asm/cacheflush.h>
@@ -374,6 +375,9 @@ int copy_thread(struct task_struct *p, const struct kernel_clone_args *args)
 		if (system_supports_tpidr2())
 			p->thread.tpidr2_el0 = read_sysreg_s(SYS_TPIDR2_EL0);
 
+		if (system_has_actlr_state())
+			p->thread.actlr = read_sysreg(actlr_el1);
+
 		if (stack_start) {
 			if (is_compat_thread(task_thread_info(p)))
 				childregs->compat_sp = stack_start;
@@ -516,6 +520,58 @@ void update_sctlr_el1(u64 sctlr)
 	isb();
 }
 
+/*
+ * IMPDEF control register ACTLR_EL1 handling. Some CPUs use this to
+ * expose features that can be controlled by userspace.
+ */
+static void actlr_thread_switch(struct task_struct *next)
+{
+	if (!system_has_actlr_state())
+		return;
+
+	current->thread.actlr = read_sysreg(actlr_el1);
+	write_sysreg(next->thread.actlr, actlr_el1);
+}
+
+#ifdef CONFIG_ARM64_MEMORY_MODEL_CONTROL
+int arch_prctl_mem_model_get(struct task_struct *t)
+{
+	if (cpus_have_const_cap(ARM64_HAS_TSO_APPLE) &&
+		t->thread.actlr & ACTLR_APPLE_TSO)
+		return PR_SET_MEM_MODEL_TSO;
+
+	return PR_SET_MEM_MODEL_DEFAULT;
+}
+
+int arch_prctl_mem_model_set(struct task_struct *t, unsigned long val)
+{
+	if (cpus_have_const_cap(ARM64_HAS_TSO_FIXED) && val == PR_SET_MEM_MODEL_TSO)
+		return 0;
+
+	if (cpus_have_const_cap(ARM64_HAS_TSO_APPLE)) {
+		WARN_ON(!system_has_actlr_state());
+
+		switch (val) {
+		case PR_SET_MEM_MODEL_TSO:
+			t->thread.actlr |= ACTLR_APPLE_TSO;
+			break;
+		case PR_SET_MEM_MODEL_DEFAULT:
+			t->thread.actlr &= ~ACTLR_APPLE_TSO;
+			break;
+		default:
+			return -EINVAL;
+		}
+		write_sysreg(t->thread.actlr, actlr_el1);
+		return 0;
+	}
+
+	if (val == PR_SET_MEM_MODEL_DEFAULT)
+		return 0;
+
+	return -EINVAL;
+}
+#endif
+
 /*
  * Thread switching.
  */
@@ -533,6 +589,7 @@ struct task_struct *__switch_to(struct task_struct *prev,
 	ssbs_thread_switch(next);
 	erratum_1418040_thread_switch(next);
 	ptrauth_thread_switch_user(next);
+	actlr_thread_switch(next);
 
 	/*
 	 * Complete any pending TLB or cache maintenance on this CPU in case
@@ -654,6 +711,10 @@ void arch_setup_new_exec(void)
 		arch_prctl_spec_ctrl_set(current, PR_SPEC_STORE_BYPASS,
 					 PR_SPEC_ENABLE);
 	}
+
+	if (IS_ENABLED(CONFIG_ARM64_MEMORY_MODEL_CONTROL)) {
+		arch_prctl_mem_model_set(current, PR_SET_MEM_MODEL_DEFAULT);
+	}
 }
 
 #ifdef CONFIG_ARM64_TAGGED_ADDR_ABI

arch/arm64/kernel/setup.c

@@ -380,6 +380,14 @@ void __init __no_sanitize_address setup_arch(char **cmdline_p)
 	 */
 	init_task.thread_info.ttbr0 = phys_to_ttbr(__pa_symbol(reserved_pg_dir));
 #endif
+#ifdef CONFIG_ARM64_ACTLR_STATE
+	/* Store the boot CPU ACTLR_EL1 value as the default. This will only
+	 * be actually restored during context switching iff the platform is
+	 * known to use ACTLR_EL1 for exposable features and its layout is
+	 * known to be the same on all CPUs.
+	 */
+	init_task.thread.actlr = read_sysreg(actlr_el1);
+#endif
 
 	if (boot_args[1] || boot_args[2] || boot_args[3]) {
 		pr_err("WARNING: x1-x3 nonzero in violation of boot protocol:\n"

arch/arm64/tools/cpucaps

@@ -43,6 +43,8 @@ HAS_SB
 HAS_STAGE2_FWB
 HAS_TIDCP1
 HAS_TLB_RANGE
+HAS_TSO_APPLE
+HAS_TSO_FIXED
 HAS_VIRT_HOST_EXTN
 HAS_WFXT
 HW_DBM