arm64: Implement PR_{GET,SET}_MEM_MODEL for always-TSO CPUs

Some ARM64 implementations are known to always use the TSO memory model.
Add trivial support for the PR_{GET,SET}_MEM_MODEL prctl, which allows
userspace to learn this fact.

Known TSO implementations:
- Nvidia Denver
- Nvidia Carmel
- Fujitsu A64FX

Signed-off-by: Hector Martin <[email protected]>

Author: marcan

arch/arm64/Kconfig

@@ -2176,6 +2176,15 @@ config ARM64_DEBUG_PRIORITY_MASKING
 	  If unsure, say N
 endif # ARM64_PSEUDO_NMI
 
+config ARM64_MEMORY_MODEL_CONTROL
+	bool "Runtime memory model control"
+	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/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 */
@@ -953,7 +955,7 @@ 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++) {
@@ -1053,6 +1055,7 @@ 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,
@@ -1421,7 +1424,7 @@ 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 +1,39 @@
+// SPDX-License-Identifier: GPL-2.0-only
+/*
+ * Contains implementation-defined CPU feature definitions.
+ */
+
+#include <asm/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_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 (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

@@ -540,6 +540,25 @@ static inline void actlr_thread_switch(struct task_struct *next)
 }
 #endif
 
+#ifdef CONFIG_ARM64_MEMORY_MODEL_CONTROL
+int arch_prctl_mem_model_get(struct task_struct *t)
+{
+	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 (val == PR_SET_MEM_MODEL_DEFAULT)
+		return 0;
+
+	return -EINVAL;
+}
+#endif
+
 /*
  * Thread switching.
  */
@@ -679,6 +698,10 @@ void arch_setup_new_exec(void)
 		arch_prctl_spec_ctrl_set(current, PR_SPEC_STORE_BYPASS,
 					 PR_SPEC_ENABLE);
 	}
+
+#ifdef CONFIG_ARM64_MEMORY_MODEL_CONTROL
+	arch_prctl_mem_model_set(current, PR_SET_MEM_MODEL_DEFAULT);
+#endif
 }
 
 #ifdef CONFIG_ARM64_TAGGED_ADDR_ABI

arch/arm64/tools/cpucaps

@@ -44,6 +44,7 @@ HAS_SB
 HAS_STAGE2_FWB
 HAS_TIDCP1
 HAS_TLB_RANGE
+HAS_TSO_FIXED
 HAS_VIRT_HOST_EXTN
 HAS_WFXT
 HW_DBM