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 <marcan@marcan.st>
Reviewed-by: Neal Gompa <neal@gompa.dev>

Author: marcan

arch/arm64/Kconfig

@@ -2294,6 +2294,15 @@ 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/include/asm/cpufeature.h

@@ -1078,6 +1078,10 @@ static inline bool cpu_has_lpa2(void)
 #endif
 }
 
+void __init init_cpucap_indirect_list_impdef(void);
+void __init init_cpucap_indirect_list_from_array(const struct arm64_cpu_capabilities *caps);
+bool cpufeature_matches(u64 reg, const struct arm64_cpu_capabilities *entry);
+
 #endif /* __ASSEMBLER__ */
 
 #endif

arch/arm64/kernel/Makefile

@@ -34,6 +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 idle.o patching.o pi/	\
+			   cpufeature_impdef.o					\
 			   rsi.o jump_label.o
 
 obj-$(CONFIG_COMPAT)			+= sys32.o signal32.o			\

arch/arm64/kernel/cpufeature.c

@@ -1080,7 +1080,7 @@ 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_cpucap_indirect_list_from_array(const struct arm64_cpu_capabilities *caps)
 {
 	for (; caps->matches; caps++) {
@@ -1592,8 +1592,8 @@ has_always(const struct arm64_cpu_capabilities *entry, int scope)
 	return true;
 }
 
-static bool
-feature_matches(u64 reg, const struct arm64_cpu_capabilities *entry)
+bool
+cpufeature_matches(u64 reg, const struct arm64_cpu_capabilities *entry)
 {
 	int val, min, max;
 	u64 tmp;
@@ -1646,14 +1646,14 @@ has_user_cpuid_feature(const struct arm64_cpu_capabilities *entry, int scope)
 	if (!mask)
 		return false;
 
-	return feature_matches(val, entry);
+	return cpufeature_matches(val, entry);
 }
 
 static bool
 has_cpuid_feature(const struct arm64_cpu_capabilities *entry, int scope)
 {
 	u64 val = read_scoped_sysreg(entry, scope);
-	return feature_matches(val, entry);
+	return cpufeature_matches(val, entry);
 }
 
 const struct cpumask *system_32bit_el0_cpumask(void)
@@ -3888,6 +3888,7 @@ void __init setup_boot_cpu_features(void)
 	 * handle the boot CPU.
 	 */
 	init_cpucap_indirect_list();
+	init_cpucap_indirect_list_impdef();
 
 	/*
 	 * Detect broken pseudo-NMI. Must be called _before_ the call to

arch/arm64/kernel/cpufeature_impdef.c

@@ -0,0 +1,38 @@
+// SPDX-License-Identifier: GPL-2.0-only
+/*
+ * Contains implementation-defined CPU feature definitions.
+ */
+
+#include <asm/cpufeature.h>
+
+#ifdef CONFIG_ARM64_MEMORY_MODEL_CONTROL
+static 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(fixed_tso_list);
+}
+#endif
+
+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_EARLY_LOCAL_CPU_FEATURE,
+		.matches = has_tso_fixed,
+	},
+#endif
+	{},
+};
+
+void __init init_cpucap_indirect_list_impdef(void)
+{
+	init_cpucap_indirect_list_from_array(arm64_impdef_features);
+}

arch/arm64/kernel/process.c

@@ -41,6 +41,7 @@
 #include <linux/thread_info.h>
 #include <linux/prctl.h>
 #include <linux/stacktrace.h>
+#include <linux/memory_ordering_model.h>
 
 #include <asm/alternative.h>
 #include <asm/arch_timer.h>
@@ -699,6 +700,25 @@ void update_sctlr_el1(u64 sctlr)
 	isb();
 }
 
+#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 (alternative_has_cap_unlikely(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.
  */
@@ -840,6 +860,10 @@ 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

@@ -62,6 +62,7 @@ HAS_STAGE2_FWB
 HAS_TCR2
 HAS_TIDCP1
 HAS_TLB_RANGE
+HAS_TSO_FIXED
 HAS_VA52
 HAS_VIRT_HOST_EXTN
 HAS_WFXT