RISC-V: Probe misaligned access speed in parallel

Probing for misaligned access speed takes about 0.06 seconds. On a
system with 64 cores, doing this in smp_callin() means it's done
serially, extending boot time by 3.8 seconds. That's a lot of boot time.

Instead of measuring each CPU serially, let's do the measurements on
all CPUs in parallel. If we disable preemption on all CPUs, the
jiffies stop ticking, so we can do this in stages of 1) everybody
except core 0, then 2) core 0. The allocations are all done outside of
on_each_cpu() to avoid calling alloc_pages() with interrupts disabled.

For hotplugged CPUs that come in after the boot time measurement,
register CPU hotplug callbacks, and do the measurement there. Interrupts
are enabled in those callbacks, so they're fine to do alloc_pages() in.
Reported-by: Jisheng Zhang <jszhang@kernel.org>
Closes: https://lore.kernel.org/all/mhng-9359993d-6872-4134-83ce-c97debe1cf9a@palmer-ri-x1c9/T/#mae9b8f40016f9df428829d33360144dc5026bcbf
Fixes: 584ea656 ("RISC-V: Probe for unaligned access speed")
Signed-off-by: Evan Green <evan@rivosinc.com>
Link: https://lore.kernel.org/r/20231106225855.3121724-1-evan@rivosinc.comSigned-off-by: Palmer Dabbelt <palmer@rivosinc.com>

arch/riscv/include/asm/cpufeature.h

@@ -30,7 +30,6 @@ DECLARE_PER_CPU(long, misaligned_access_speed);
 /* Per-cpu ISA extensions. */
 extern struct riscv_isainfo hart_isa[NR_CPUS];
 
-void check_unaligned_access(int cpu);
 void riscv_user_isa_enable(void);
 
 #ifdef CONFIG_RISCV_MISALIGNED

arch/riscv/kernel/cpufeature.c

@@ -8,6 +8,7 @@
 
 #include <linux/acpi.h>
 #include <linux/bitmap.h>
+#include <linux/cpuhotplug.h>
 #include <linux/ctype.h>
 #include <linux/log2.h>
 #include <linux/memory.h>
@@ -29,6 +30,7 @@
 
 #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;
@@ -557,30 +559,21 @@ unsigned long riscv_get_elf_hwcap(void)
 	return hwcap;
 }
 
-void check_unaligned_access(int cpu)
+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;
+	struct page *page = param;
 	void *dst;
 	void *src;
 	long speed = RISCV_HWPROBE_MISALIGNED_SLOW;
 
 	if (check_unaligned_access_emulated(cpu))
-		return;
-
-	/* We are already set since the last check */
-	if (per_cpu(misaligned_access_speed, cpu) != RISCV_HWPROBE_MISALIGNED_UNKNOWN)
-		return;
-
-	page = alloc_pages(GFP_NOWAIT, get_order(MISALIGNED_BUFFER_SIZE));
-	if (!page) {
-		pr_warn("Can't alloc pages to measure memcpy performance");
-		return;
-	}
+		return 0;
 
 	/* Make an unaligned destination buffer. */
 	dst = (void *)((unsigned long)page_address(page) | 0x1);
@@ -634,7 +627,7 @@ void check_unaligned_access(int cpu)
 		pr_warn("cpu%d: rdtime lacks granularity needed to measure unaligned access speed\n",
 			cpu);
 
-		goto out;
+		return 0;
 	}
 
 	if (word_cycles < byte_cycles)
@@ -648,19 +641,84 @@ void check_unaligned_access(int cpu)
 		(speed == RISCV_HWPROBE_MISALIGNED_FAST) ? "fast" : "slow");
 
 	per_cpu(misaligned_access_speed, cpu) = speed;
+	return 0;
+}
 
-out:
-	__free_pages(page, get_order(MISALIGNED_BUFFER_SIZE));
+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]);
+}
+
+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)
+		return 0;
+
+	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);
+	return 0;
 }
 
-static int __init check_unaligned_access_boot_cpu(void)
+/* Measure unaligned access on all CPUs present at boot in parallel. */
+static int check_unaligned_access_all_cpus(void)
 {
-	check_unaligned_access(0);
+	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 callback for any new CPUs that come online. */
+	cpuhp_setup_state_nocalls(CPUHP_AP_ONLINE_DYN, "riscv:online",
+				  riscv_online_cpu, NULL);
+
+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_boot_cpu);
+arch_initcall(check_unaligned_access_all_cpus);
 
 void riscv_user_isa_enable(void)
 {

arch/riscv/kernel/smpboot.c

@@ -247,7 +247,6 @@ asmlinkage __visible void smp_callin(void)
 	riscv_ipi_enable();
 
 	numa_add_cpu(curr_cpuid);
-	check_unaligned_access(curr_cpuid);
 	set_cpu_online(curr_cpuid, 1);
 
 	if (has_vector()) {