Merge patch series "riscv: Use Kconfig to set unaligned access speed"

Charlie Jenkins <charlie@rivosinc.com> says:

If the hardware unaligned access speed is known at compile time, it is
possible to avoid running the unaligned access speed probe to speedup
boot-time.

* b4-shazam-merge:
  riscv: Set unaligned access speed at compile time
  riscv: Decouple emulated unaligned accesses from access speed
  riscv: Only check online cpus for emulated accesses
  riscv: lib: Introduce has_fast_unaligned_access()

Link: https://lore.kernel.org/r/20240308-disable_misaligned_probe_config-v9-0-a388770ba0ce@rivosinc.com
Signed-off-by: Palmer Dabbelt <palmer@rivosinc.com>

8 files changed, 374 insertions, 293 deletions

diff --git a/arch/riscv/Kconfig b/arch/riscv/Kconfig
index 0bfcfec67ed5..8ebafe337eac 100644
--- a/arch/riscv/Kconfig
+++ b/arch/riscv/Kconfig
@@ -704,27 +704,61 @@ config THREAD_SIZE_ORDER
 	  affects irq stack size, which is equal to thread stack size.
 
 config RISCV_MISALIGNED
-	bool "Support misaligned load/store traps for kernel and userspace"
+	bool
 	select SYSCTL_ARCH_UNALIGN_ALLOW
-	default y
 	help
-	  Say Y here if you want the kernel to embed support for misaligned
-	  load/store for both kernel and userspace. When disable, misaligned
-	  accesses will generate SIGBUS in userspace and panic in kernel.
+	  Embed support for emulating misaligned loads and stores.
+
+choice
+	prompt "Unaligned Accesses Support"
+	default RISCV_PROBE_UNALIGNED_ACCESS
+	help
+	  This determines the level of support for unaligned accesses. This
+	  information is used by the kernel to perform optimizations. It is also
+	  exposed to user space via the hwprobe syscall. The hardware will be
+	  probed at boot by default.
+
+config RISCV_PROBE_UNALIGNED_ACCESS
+	bool "Probe for hardware unaligned access support"
+	select RISCV_MISALIGNED
+	help
+	  During boot, the kernel will run a series of tests to determine the
+	  speed of unaligned accesses. This probing will dynamically determine
+	  the speed of unaligned accesses on the underlying system. If unaligned
+	  memory accesses trap into the kernel as they are not supported by the
+	  system, the kernel will emulate the unaligned accesses to preserve the
+	  UABI.
+
+config RISCV_EMULATED_UNALIGNED_ACCESS
+	bool "Emulate unaligned access where system support is missing"
+	select RISCV_MISALIGNED
+	help
+	  If unaligned memory accesses trap into the kernel as they are not
+	  supported by the system, the kernel will emulate the unaligned
+	  accesses to preserve the UABI. When the underlying system does support
+	  unaligned accesses, the unaligned accesses are assumed to be slow.
+
+config RISCV_SLOW_UNALIGNED_ACCESS
+	bool "Assume the system supports slow unaligned memory accesses"
+	depends on NONPORTABLE
+	help
+	  Assume that the system supports slow unaligned memory accesses. The
+	  kernel and userspace programs may not be able to run at all on systems
+	  that do not support unaligned memory accesses.
 
 config RISCV_EFFICIENT_UNALIGNED_ACCESS
-	bool "Assume the CPU supports fast unaligned memory accesses"
+	bool "Assume the system supports fast unaligned memory accesses"
 	depends on NONPORTABLE
 	select DCACHE_WORD_ACCESS if MMU
 	select HAVE_EFFICIENT_UNALIGNED_ACCESS
 	help
-	  Say Y here if you want the kernel to assume that the CPU supports
-	  efficient unaligned memory accesses.  When enabled, this option
-	  improves the performance of the kernel on such CPUs.  However, the
-	  kernel will run much more slowly, or will not be able to run at all,
-	  on CPUs that do not support efficient unaligned memory accesses.
+	  Assume that the system supports fast unaligned memory accesses. When
+	  enabled, this option improves the performance of the kernel on such
+	  systems. However, the kernel and userspace programs will run much more
+	  slowly, or will not be able to run at all, on systems that do not
+	  support efficient unaligned memory accesses.
 
-	  If unsure what to do here, say N.
+endchoice
 
 endmenu # "Platform type"
 
diff --git a/arch/riscv/include/asm/cpufeature.h b/arch/riscv/include/asm/cpufeature.h
index 5a626ed2c47a..46061f5e9764 100644
--- a/arch/riscv/include/asm/cpufeature.h
+++ b/arch/riscv/include/asm/cpufeature.h
@@ -1,6 +1,6 @@
 /* SPDX-License-Identifier: GPL-2.0-only */
 /*
- * Copyright 2022-2023 Rivos, Inc
+ * Copyright 2022-2024 Rivos, Inc
  */
 
 #ifndef _ASM_CPUFEATURE_H
@@ -28,29 +28,38 @@ struct riscv_isainfo {
 
 DECLARE_PER_CPU(struct riscv_cpuinfo, riscv_cpuinfo);
 
-DECLARE_PER_CPU(long, misaligned_access_speed);
-
 /* Per-cpu ISA extensions. */
 extern struct riscv_isainfo hart_isa[NR_CPUS];
 
 void riscv_user_isa_enable(void);
 
-#ifdef CONFIG_RISCV_MISALIGNED
-bool unaligned_ctl_available(void);
-bool check_unaligned_access_emulated(int cpu);
+#if defined(CONFIG_RISCV_MISALIGNED)
+bool check_unaligned_access_emulated_all_cpus(void);
 void unaligned_emulation_finish(void);
+bool unaligned_ctl_available(void);
+DECLARE_PER_CPU(long, misaligned_access_speed);
 #else
 static inline bool unaligned_ctl_available(void)
 {
 	return false;
 }
+#endif
+
+#if defined(CONFIG_RISCV_PROBE_UNALIGNED_ACCESS)
+DECLARE_STATIC_KEY_FALSE(fast_unaligned_access_speed_key);
 
-static inline bool check_unaligned_access_emulated(int cpu)
+static __always_inline bool has_fast_unaligned_accesses(void)
 {
-	return false;
+	return static_branch_likely(&fast_unaligned_access_speed_key);
+}
+#else
+static __always_inline bool has_fast_unaligned_accesses(void)
+{
+	if (IS_ENABLED(CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS))
+		return true;
+	else
+		return false;
 }
-
-static inline void unaligned_emulation_finish(void) {}
 #endif
 
 unsigned long riscv_get_elf_hwcap(void);
@@ -135,6 +144,4 @@ static __always_inline bool riscv_cpu_has_extension_unlikely(int cpu, const unsi
 	return __riscv_isa_extension_available(hart_isa[cpu].isa, ext);
 }
 
-DECLARE_STATIC_KEY_FALSE(fast_misaligned_access_speed_key);
-
 #endif
diff --git a/arch/riscv/kernel/Makefile b/arch/riscv/kernel/Makefile
index f71910718053..c8085126a6f9 100644
--- a/arch/riscv/kernel/Makefile
+++ b/arch/riscv/kernel/Makefile
@@ -38,7 +38,6 @@ extra-y += vmlinux.lds
 obj-y	+= head.o
 obj-y	+= soc.o
 obj-$(CONFIG_RISCV_ALTERNATIVE) += alternative.o
-obj-y	+= copy-unaligned.o
 obj-y	+= cpu.o
 obj-y	+= cpufeature.o
 obj-y	+= entry.o
@@ -62,6 +61,9 @@ obj-y	+= tests/
 obj-$(CONFIG_MMU) += vdso.o vdso/
 
 obj-$(CONFIG_RISCV_MISALIGNED)	+= traps_misaligned.o
+obj-$(CONFIG_RISCV_MISALIGNED)	+= unaligned_access_speed.o
+obj-$(CONFIG_RISCV_PROBE_UNALIGNED_ACCESS)	+= copy-unaligned.o
+
 obj-$(CONFIG_FPU)		+= fpu.o
 obj-$(CONFIG_RISCV_ISA_V)	+= vector.o
 obj-$(CONFIG_RISCV_ISA_V)	+= kernel_mode_vector.o
diff --git a/arch/riscv/kernel/cpufeature.c b/arch/riscv/kernel/cpufeature.c
index 0c7688fa8376..afeae3ff43dc 100644
--- a/arch/riscv/kernel/cpufeature.c
+++ b/arch/riscv/kernel/cpufeature.c
@@ -11,7 +11,6 @@
 #include <linux/cpu.h>
 #include <linux/cpuhotplug.h>
 #include <linux/ctype.h>
-#include <linux/jump_label.h>
 #include <linux/log2.h>
 #include <linux/memory.h>
 #include <linux/module.h>
@@ -21,20 +20,12 @@
 #include <asm/cacheflush.h>
 #include <asm/cpufeature.h>
 #include <asm/hwcap.h>
-#include <asm/hwprobe.h>
 #include <asm/patch.h>
 #include <asm/processor.h>
 #include <asm/vector.h>
 
-#include "copy-unaligned.h"
-
 #define NUM_ALPHA_EXTS ('z' - 'a' + 1)
 
-#define MISALIGNED_ACCESS_JIFFIES_LG2 1
-#define MISALIGNED_BUFFER_SIZE 0x4000
-#define MISALIGNED_BUFFER_ORDER get_order(MISALIGNED_BUFFER_SIZE)
-#define MISALIGNED_COPY_SIZE ((MISALIGNED_BUFFER_SIZE / 2) - 0x80)
-
 unsigned long elf_hwcap __read_mostly;
 
 /* Host ISA bitmap */
@@ -43,11 +34,6 @@ static DECLARE_BITMAP(riscv_isa, RISCV_ISA_EXT_MAX) __read_mostly;
 /* Per-cpu ISA extensions. */
 struct riscv_isainfo hart_isa[NR_CPUS];
 
-/* Performance information */
-DEFINE_PER_CPU(long, misaligned_access_speed);
-
-static cpumask_t fast_misaligned_access;
-
 /**
  * riscv_isa_extension_base() - Get base extension word
  *
@@ -707,247 +693,6 @@ unsigned long riscv_get_elf_hwcap(void)
 	return hwcap;
 }
 
-static int check_unaligned_access(void *param)
-{
-	int cpu = smp_processor_id();
-	u64 start_cycles, end_cycles;
-	u64 word_cycles;
-	u64 byte_cycles;
-	int ratio;
-	unsigned long start_jiffies, now;
-	struct page *page = param;
-	void *dst;
-	void *src;
-	long speed = RISCV_HWPROBE_MISALIGNED_SLOW;
-
-	if (check_unaligned_access_emulated(cpu))
-		return 0;
-
-	/* Make an unaligned destination buffer. */
-	dst = (void *)((unsigned long)page_address(page) | 0x1);
-	/* Unalign src as well, but differently (off by 1 + 2 = 3). */
-	src = dst + (MISALIGNED_BUFFER_SIZE / 2);
-	src += 2;
-	word_cycles = -1ULL;
-	/* Do a warmup. */
-	__riscv_copy_words_unaligned(dst, src, MISALIGNED_COPY_SIZE);
-	preempt_disable();
-	start_jiffies = jiffies;
-	while ((now = jiffies) == start_jiffies)
-		cpu_relax();
-
-	/*
-	 * For a fixed amount of time, repeatedly try the function, and take
-	 * the best time in cycles as the measurement.
-	 */
-	while (time_before(jiffies, now + (1 << MISALIGNED_ACCESS_JIFFIES_LG2))) {
-		start_cycles = get_cycles64();
-		/* Ensure the CSR read can't reorder WRT to the copy. */
-		mb();
-		__riscv_copy_words_unaligned(dst, src, MISALIGNED_COPY_SIZE);
-		/* Ensure the copy ends before the end time is snapped. */
-		mb();
-		end_cycles = get_cycles64();
-		if ((end_cycles - start_cycles) < word_cycles)
-			word_cycles = end_cycles - start_cycles;
-	}
-
-	byte_cycles = -1ULL;
-	__riscv_copy_bytes_unaligned(dst, src, MISALIGNED_COPY_SIZE);
-	start_jiffies = jiffies;
-	while ((now = jiffies) == start_jiffies)
-		cpu_relax();
-
-	while (time_before(jiffies, now + (1 << MISALIGNED_ACCESS_JIFFIES_LG2))) {
-		start_cycles = get_cycles64();
-		mb();
-		__riscv_copy_bytes_unaligned(dst, src, MISALIGNED_COPY_SIZE);
-		mb();
-		end_cycles = get_cycles64();
-		if ((end_cycles - start_cycles) < byte_cycles)
-			byte_cycles = end_cycles - start_cycles;
-	}
-
-	preempt_enable();
-
-	/* Don't divide by zero. */
-	if (!word_cycles || !byte_cycles) {
-		pr_warn("cpu%d: rdtime lacks granularity needed to measure unaligned access speed\n",
-			cpu);
-
-		return 0;
-	}
-
-	if (word_cycles < byte_cycles)
-		speed = RISCV_HWPROBE_MISALIGNED_FAST;
-
-	ratio = div_u64((byte_cycles * 100), word_cycles);
-	pr_info("cpu%d: Ratio of byte access time to unaligned word access is %d.%02d, unaligned accesses are %s\n",
-		cpu,
-		ratio / 100,
-		ratio % 100,
-		(speed == RISCV_HWPROBE_MISALIGNED_FAST) ? "fast" : "slow");
-
-	per_cpu(misaligned_access_speed, cpu) = speed;
-
-	/*
-	 * Set the value of fast_misaligned_access of a CPU. These operations
-	 * are atomic to avoid race conditions.
-	 */
-	if (speed == RISCV_HWPROBE_MISALIGNED_FAST)
-		cpumask_set_cpu(cpu, &fast_misaligned_access);
-	else
-		cpumask_clear_cpu(cpu, &fast_misaligned_access);
-
-	return 0;
-}
-
-static void check_unaligned_access_nonboot_cpu(void *param)
-{
-	unsigned int cpu = smp_processor_id();
-	struct page **pages = param;
-
-	if (smp_processor_id() != 0)
-		check_unaligned_access(pages[cpu]);
-}
-
-DEFINE_STATIC_KEY_FALSE(fast_misaligned_access_speed_key);
-
-static void modify_unaligned_access_branches(cpumask_t *mask, int weight)
-{
-	if (cpumask_weight(mask) == weight)
-		static_branch_enable_cpuslocked(&fast_misaligned_access_speed_key);
-	else
-		static_branch_disable_cpuslocked(&fast_misaligned_access_speed_key);
-}
-
-static void set_unaligned_access_static_branches_except_cpu(int cpu)
-{
-	/*
-	 * Same as set_unaligned_access_static_branches, except excludes the
-	 * given CPU from the result. When a CPU is hotplugged into an offline
-	 * state, this function is called before the CPU is set to offline in
-	 * the cpumask, and thus the CPU needs to be explicitly excluded.
-	 */
-
-	cpumask_t fast_except_me;
-
-	cpumask_and(&fast_except_me, &fast_misaligned_access, cpu_online_mask);
-	cpumask_clear_cpu(cpu, &fast_except_me);
-
-	modify_unaligned_access_branches(&fast_except_me, num_online_cpus() - 1);
-}
-
-static void set_unaligned_access_static_branches(void)
-{
-	/*
-	 * This will be called after check_unaligned_access_all_cpus so the
-	 * result of unaligned access speed for all CPUs will be available.
-	 *
-	 * To avoid the number of online cpus changing between reading
-	 * cpu_online_mask and calling num_online_cpus, cpus_read_lock must be
-	 * held before calling this function.
-	 */
-
-	cpumask_t fast_and_online;
-
-	cpumask_and(&fast_and_online, &fast_misaligned_access, cpu_online_mask);
-
-	modify_unaligned_access_branches(&fast_and_online, num_online_cpus());
-}
-
-static int lock_and_set_unaligned_access_static_branch(void)
-{
-	cpus_read_lock();
-	set_unaligned_access_static_branches();
-	cpus_read_unlock();
-
-	return 0;
-}
-
-arch_initcall_sync(lock_and_set_unaligned_access_static_branch);
-
-static int riscv_online_cpu(unsigned int cpu)
-{
-	static struct page *buf;
-
-	/* We are already set since the last check */
-	if (per_cpu(misaligned_access_speed, cpu) != RISCV_HWPROBE_MISALIGNED_UNKNOWN)
-		goto exit;
-
-	buf = alloc_pages(GFP_KERNEL, MISALIGNED_BUFFER_ORDER);
-	if (!buf) {
-		pr_warn("Allocation failure, not measuring misaligned performance\n");
-		return -ENOMEM;
-	}
-
-	check_unaligned_access(buf);
-	__free_pages(buf, MISALIGNED_BUFFER_ORDER);
-
-exit:
-	set_unaligned_access_static_branches();
-
-	return 0;
-}
-
-static int riscv_offline_cpu(unsigned int cpu)
-{
-	set_unaligned_access_static_branches_except_cpu(cpu);
-
-	return 0;
-}
-
-/* Measure unaligned access on all CPUs present at boot in parallel. */
-static int check_unaligned_access_all_cpus(void)
-{
-	unsigned int cpu;
-	unsigned int cpu_count = num_possible_cpus();
-	struct page **bufs = kzalloc(cpu_count * sizeof(struct page *),
-				     GFP_KERNEL);
-
-	if (!bufs) {
-		pr_warn("Allocation failure, not measuring misaligned performance\n");
-		return 0;
-	}
-
-	/*
-	 * Allocate separate buffers for each CPU so there's no fighting over
-	 * cache lines.
-	 */
-	for_each_cpu(cpu, cpu_online_mask) {
-		bufs[cpu] = alloc_pages(GFP_KERNEL, MISALIGNED_BUFFER_ORDER);
-		if (!bufs[cpu]) {
-			pr_warn("Allocation failure, not measuring misaligned performance\n");
-			goto out;
-		}
-	}
-
-	/* Check everybody except 0, who stays behind to tend jiffies. */
-	on_each_cpu(check_unaligned_access_nonboot_cpu, bufs, 1);
-
-	/* Check core 0. */
-	smp_call_on_cpu(0, check_unaligned_access, bufs[0], true);
-
-	/*
-	 * Setup hotplug callbacks for any new CPUs that come online or go
-	 * offline.
-	 */
-	cpuhp_setup_state_nocalls(CPUHP_AP_ONLINE_DYN, "riscv:online",
-				  riscv_online_cpu, riscv_offline_cpu);
-
-out:
-	unaligned_emulation_finish();
-	for_each_cpu(cpu, cpu_online_mask) {
-		if (bufs[cpu])
-			__free_pages(bufs[cpu], MISALIGNED_BUFFER_ORDER);
-	}
-
-	kfree(bufs);
-	return 0;
-}
-
-arch_initcall(check_unaligned_access_all_cpus);
-
 void riscv_user_isa_enable(void)
 {
 	if (riscv_cpu_has_extension_unlikely(smp_processor_id(), RISCV_ISA_EXT_ZICBOZ))
diff --git a/arch/riscv/kernel/sys_hwprobe.c b/arch/riscv/kernel/sys_hwprobe.c
index a7c56b41efd2..8cae41a502dd 100644
--- a/arch/riscv/kernel/sys_hwprobe.c
+++ b/arch/riscv/kernel/sys_hwprobe.c
@@ -147,6 +147,7 @@ static bool hwprobe_ext0_has(const struct cpumask *cpus, unsigned long ext)
 	return (pair.value & ext);
 }
 
+#if defined(CONFIG_RISCV_PROBE_UNALIGNED_ACCESS)
 static u64 hwprobe_misaligned(const struct cpumask *cpus)
 {
 	int cpu;
@@ -169,6 +170,18 @@ static u64 hwprobe_misaligned(const struct cpumask *cpus)
 
 	return perf;
 }
+#else
+static u64 hwprobe_misaligned(const struct cpumask *cpus)
+{
+	if (IS_ENABLED(CONFIG_RISCV_EFFICIENT_UNALIGNED_ACCESS))
+		return RISCV_HWPROBE_MISALIGNED_FAST;
+
+	if (IS_ENABLED(CONFIG_RISCV_EMULATED_UNALIGNED_ACCESS) && unaligned_ctl_available())
+		return RISCV_HWPROBE_MISALIGNED_EMULATED;
+
+	return RISCV_HWPROBE_MISALIGNED_SLOW;
+}
+#endif
 
 static void hwprobe_one_pair(struct riscv_hwprobe *pair,
 			     const struct cpumask *cpus)
diff --git a/arch/riscv/kernel/traps_misaligned.c b/arch/riscv/kernel/traps_misaligned.c
index 8ded225e8c5b..2adb7c3e4dd5 100644
--- a/arch/riscv/kernel/traps_misaligned.c
+++ b/arch/riscv/kernel/traps_misaligned.c
@@ -413,7 +413,9 @@ int handle_misaligned_load(struct pt_regs *regs)
 
 	perf_sw_event(PERF_COUNT_SW_ALIGNMENT_FAULTS, 1, regs, addr);
 
+#ifdef CONFIG_RISCV_PROBE_UNALIGNED_ACCESS
 	*this_cpu_ptr(&misaligned_access_speed) = RISCV_HWPROBE_MISALIGNED_EMULATED;
+#endif
 
 	if (!unaligned_enabled)
 		return -1;
@@ -596,7 +598,7 @@ int handle_misaligned_store(struct pt_regs *regs)
 	return 0;
 }
 
-bool check_unaligned_access_emulated(int cpu)
+static bool check_unaligned_access_emulated(int cpu)
 {
 	long *mas_ptr = per_cpu_ptr(&misaligned_access_speed, cpu);
 	unsigned long tmp_var, tmp_val;
@@ -623,7 +625,7 @@ bool check_unaligned_access_emulated(int cpu)
 	return misaligned_emu_detected;
 }
 
-void unaligned_emulation_finish(void)
+bool check_unaligned_access_emulated_all_cpus(void)
 {
 	int cpu;
 
@@ -632,13 +634,12 @@ void unaligned_emulation_finish(void)
 	 * accesses emulated since tasks requesting such control can run on any
 	 * CPU.
 	 */
-	for_each_present_cpu(cpu) {
-		if (per_cpu(misaligned_access_speed, cpu) !=
-					RISCV_HWPROBE_MISALIGNED_EMULATED) {
-			return;
-		}
-	}
+	for_each_online_cpu(cpu)
+		if (!check_unaligned_access_emulated(cpu))
+			return false;
+
 	unaligned_ctl = true;
+	return true;
 }
 
 bool unaligned_ctl_available(void)
diff --git a/arch/riscv/kernel/unaligned_access_speed.c b/arch/riscv/kernel/unaligned_access_speed.c
new file mode 100644
index 000000000000..52264ea4f0bd
--- /dev/null
+++ b/arch/riscv/kernel/unaligned_access_speed.c
@@ -0,0 +1,282 @@
+// SPDX-License-Identifier: GPL-2.0-only
+/*
+ * Copyright 2024 Rivos Inc.
+ */
+
+#include <linux/cpu.h>
+#include <linux/cpumask.h>
+#include <linux/jump_label.h>
+#include <linux/mm.h>
+#include <linux/smp.h>
+#include <linux/types.h>
+#include <asm/cpufeature.h>
+#include <asm/hwprobe.h>
+
+#include "copy-unaligned.h"
+
+#define MISALIGNED_ACCESS_JIFFIES_LG2 1
+#define MISALIGNED_BUFFER_SIZE 0x4000
+#define MISALIGNED_BUFFER_ORDER get_order(MISALIGNED_BUFFER_SIZE)
+#define MISALIGNED_COPY_SIZE ((MISALIGNED_BUFFER_SIZE / 2) - 0x80)
+
+DEFINE_PER_CPU(long, misaligned_access_speed);
+
+#ifdef CONFIG_RISCV_PROBE_UNALIGNED_ACCESS
+static cpumask_t fast_misaligned_access;
+static int check_unaligned_access(void *param)
+{
+	int cpu = smp_processor_id();
+	u64 start_cycles, end_cycles;
+	u64 word_cycles;
+	u64 byte_cycles;
+	int ratio;
+	unsigned long start_jiffies, now;
+	struct page *page = param;
+	void *dst;
+	void *src;
+	long speed = RISCV_HWPROBE_MISALIGNED_SLOW;
+
+	if (per_cpu(misaligned_access_speed, cpu) != RISCV_HWPROBE_MISALIGNED_UNKNOWN)
+		return 0;
+
+	/* Make an unaligned destination buffer. */
+	dst = (void *)((unsigned long)page_address(page) | 0x1);
+	/* Unalign src as well, but differently (off by 1 + 2 = 3). */
+	src = dst + (MISALIGNED_BUFFER_SIZE / 2);
+	src += 2;
+	word_cycles = -1ULL;
+	/* Do a warmup. */
+	__riscv_copy_words_unaligned(dst, src, MISALIGNED_COPY_SIZE);
+	preempt_disable();
+	start_jiffies = jiffies;
+	while ((now = jiffies) == start_jiffies)
+		cpu_relax();
+
+	/*
+	 * For a fixed amount of time, repeatedly try the function, and take
+	 * the best time in cycles as the measurement.
+	 */
+	while (time_before(jiffies, now + (1 << MISALIGNED_ACCESS_JIFFIES_LG2))) {
+		start_cycles = get_cycles64();
+		/* Ensure the CSR read can't reorder WRT to the copy. */
+		mb();
+		__riscv_copy_words_unaligned(dst, src, MISALIGNED_COPY_SIZE);
+		/* Ensure the copy ends before the end time is snapped. */
+		mb();
+		end_cycles = get_cycles64();
+		if ((end_cycles - start_cycles) < word_cycles)
+			word_cycles = end_cycles - start_cycles;
+	}
+
+	byte_cycles = -1ULL;
+	__riscv_copy_bytes_unaligned(dst, src, MISALIGNED_COPY_SIZE);
+	start_jiffies = jiffies;
+	while ((now = jiffies) == start_jiffies)
+		cpu_relax();
+
+	while (time_before(jiffies, now + (1 << MISALIGNED_ACCESS_JIFFIES_LG2))) {
+		start_cycles = get_cycles64();
+		mb();
+		__riscv_copy_bytes_unaligned(dst, src, MISALIGNED_COPY_SIZE);
+		mb();
+		end_cycles = get_cycles64();
+		if ((end_cycles - start_cycles) < byte_cycles)
+			byte_cycles = end_cycles - start_cycles;
+	}
+
+	preempt_enable();
+
+	/* Don't divide by zero. */
+	if (!word_cycles || !byte_cycles) {
+		pr_warn("cpu%d: rdtime lacks granularity needed to measure unaligned access speed\n",
+			cpu);
+
+		return 0;
+	}
+
+	if (word_cycles < byte_cycles)
+		speed = RISCV_HWPROBE_MISALIGNED_FAST;
+
+	ratio = div_u64((byte_cycles * 100), word_cycles);
+	pr_info("cpu%d: Ratio of byte access time to unaligned word access is %d.%02d, unaligned accesses are %s\n",
+		cpu,
+		ratio / 100,
+		ratio % 100,
+		(speed == RISCV_HWPROBE_MISALIGNED_FAST) ? "fast" : "slow");
+
+	per_cpu(misaligned_access_speed, cpu) = speed;
+
+	/*
+	 * Set the value of fast_misaligned_access of a CPU. These operations
+	 * are atomic to avoid race conditions.
+	 */
+	if (speed == RISCV_HWPROBE_MISALIGNED_FAST)
+		cpumask_set_cpu(cpu, &fast_misaligned_access);
+	else
+		cpumask_clear_cpu(cpu, &fast_misaligned_access);
+
+	return 0;
+}
+
+static void check_unaligned_access_nonboot_cpu(void *param)
+{
+	unsigned int cpu = smp_processor_id();
+	struct page **pages = param;
+
+	if (smp_processor_id() != 0)
+		check_unaligned_access(pages[cpu]);
+}
+
+DEFINE_STATIC_KEY_FALSE(fast_unaligned_access_speed_key);
+
+static void modify_unaligned_access_branches(cpumask_t *mask, int weight)
+{
+	if (cpumask_weight(mask) == weight)
+		static_branch_enable_cpuslocked(&fast_unaligned_access_speed_key);
+	else
+		static_branch_disable_cpuslocked(&fast_unaligned_access_speed_key);
+}
+
+static void set_unaligned_access_static_branches_except_cpu(int cpu)
+{
+	/*
+	 * Same as set_unaligned_access_static_branches, except excludes the
+	 * given CPU from the result. When a CPU is hotplugged into an offline
+	 * state, this function is called before the CPU is set to offline in
+	 * the cpumask, and thus the CPU needs to be explicitly excluded.
+	 */
+
+	cpumask_t fast_except_me;
+
+	cpumask_and(&fast_except_me, &fast_misaligned_access, cpu_online_mask);
+	cpumask_clear_cpu(cpu, &fast_except_me);
+
+	modify_unaligned_access_branches(&fast_except_me, num_online_cpus() - 1);
+}
+
+static void set_unaligned_access_static_branches(void)
+{
+	/*
+	 * This will be called after check_unaligned_access_all_cpus so the
+	 * result of unaligned access speed for all CPUs will be available.
+	 *
+	 * To avoid the number of online cpus changing between reading
+	 * cpu_online_mask and calling num_online_cpus, cpus_read_lock must be
+	 * held before calling this function.
+	 */
+
+	cpumask_t fast_and_online;
+
+	cpumask_and(&fast_and_online, &fast_misaligned_access, cpu_online_mask);
+
+	modify_unaligned_access_branches(&fast_and_online, num_online_cpus());
+}
+
+static int lock_and_set_unaligned_access_static_branch(void)
+{
+	cpus_read_lock();
+	set_unaligned_access_static_branches();
+	cpus_read_unlock();
+
+	return 0;
+}
+
+arch_initcall_sync(lock_and_set_unaligned_access_static_branch);
+
+static int riscv_online_cpu(unsigned int cpu)
+{
+	static struct page *buf;
+
+	/* We are already set since the last check */
+	if (per_cpu(misaligned_access_speed, cpu) != RISCV_HWPROBE_MISALIGNED_UNKNOWN)
+		goto exit;
+
+	buf = alloc_pages(GFP_KERNEL, MISALIGNED_BUFFER_ORDER);
+	if (!buf) {
+		pr_warn("Allocation failure, not measuring misaligned performance\n");
+		return -ENOMEM;
+	}
+
+	check_unaligned_access(buf);
+	__free_pages(buf, MISALIGNED_BUFFER_ORDER);
+
+exit:
+	set_unaligned_access_static_branches();
+
+	return 0;
+}
+
+static int riscv_offline_cpu(unsigned int cpu)
+{
+	set_unaligned_access_static_branches_except_cpu(cpu);
+
+	return 0;
+}
+
+/* Measure unaligned access speed on all CPUs present at boot in parallel. */
+static int check_unaligned_access_speed_all_cpus(void)
+{
+	unsigned int cpu;
+	unsigned int cpu_count = num_possible_cpus();
+	struct page **bufs = kzalloc(cpu_count * sizeof(struct page *),
+				     GFP_KERNEL);
+
+	if (!bufs) {
+		pr_warn("Allocation failure, not measuring misaligned performance\n");
+		return 0;
+	}
+
+	/*
+	 * Allocate separate buffers for each CPU so there's no fighting over
+	 * cache lines.
+	 */
+	for_each_cpu(cpu, cpu_online_mask) {
+		bufs[cpu] = alloc_pages(GFP_KERNEL, MISALIGNED_BUFFER_ORDER);
+		if (!bufs[cpu]) {
+			pr_warn("Allocation failure, not measuring misaligned performance\n");
+			goto out;
+		}
+	}
+
+	/* Check everybody except 0, who stays behind to tend jiffies. */
+	on_each_cpu(check_unaligned_access_nonboot_cpu, bufs, 1);
+
+	/* Check core 0. */
+	smp_call_on_cpu(0, check_unaligned_access, bufs[0], true);
+
+	/*
+	 * Setup hotplug callbacks for any new CPUs that come online or go
+	 * offline.
+	 */
+	cpuhp_setup_state_nocalls(CPUHP_AP_ONLINE_DYN, "riscv:online",
+				  riscv_online_cpu, riscv_offline_cpu);
+
+out:
+	for_each_cpu(cpu, cpu_online_mask) {
+		if (bufs[cpu])
+			__free_pages(bufs[cpu], MISALIGNED_BUFFER_ORDER);
+	}
+
+	kfree(bufs);
+	return 0;
+}
+
+static int check_unaligned_access_all_cpus(void)
+{
+	bool all_cpus_emulated = check_unaligned_access_emulated_all_cpus();
+
+	if (!all_cpus_emulated)
+		return check_unaligned_access_speed_all_cpus();
+
+	return 0;
+}
+#else /* CONFIG_RISCV_PROBE_UNALIGNED_ACCESS */
+static int check_unaligned_access_all_cpus(void)
+{
+	check_unaligned_access_emulated_all_cpus();
+
+	return 0;
+}
+#endif
+
+arch_initcall(check_unaligned_access_all_cpus);
diff --git a/arch/riscv/lib/csum.c b/arch/riscv/lib/csum.c
index af3df5274ccb..7178e0acfa22 100644
--- a/arch/riscv/lib/csum.c
+++ b/arch/riscv/lib/csum.c
@@ -3,7 +3,7 @@
  * Checksum library
  *
  * Influenced by arch/arm64/lib/csum.c
- * Copyright (C) 2023 Rivos Inc.
+ * Copyright (C) 2023-2024 Rivos Inc.
  */
 #include <linux/bitops.h>
 #include <linux/compiler.h>
@@ -318,10 +318,7 @@ unsigned int do_csum(const unsigned char *buff, int len)
 	 * branches. The largest chunk of overlap was delegated into the
 	 * do_csum_common function.
 	 */
-	if (static_branch_likely(&fast_misaligned_access_speed_key))
-		return do_csum_no_alignment(buff, len);
-
-	if (((unsigned long)buff & OFFSET_MASK) == 0)
+	if (has_fast_unaligned_accesses() || (((unsigned long)buff & OFFSET_MASK) == 0))
 		return do_csum_no_alignment(buff, len);
 
 	return do_csum_with_alignment(buff, len);