add arm64 support for clock frequency helpers

homa_clock() calls get_cycles(), whose backing source is
arch-specific: RDTSC on x86 (frequency tsc_khz) and the
generic timer on arm64 (frequency arch_timer_get_cntfrq()).

Make homa_clock_khz() return the correct value per arch:
 - x86:   tsc_khz  (not cpu_khz -- the kernel tracks them
          separately and they can diverge)
 - arm64: arch_timer_get_cntfrq() / 1000
 - other: 1000000 (fallback)

Guard the tsc_khz-dependent scaling in homa_metrics_print()
behind CONFIG_X86 and emit raw cycle counts on other arches
where Linux and homa_clock() share the same counter.

Replace bare tsc_khz references in timetrace.c with
homa_clock_khz() so the emitted cpu_khz header is correct
on all architectures.

Signed-off-by: Serapheim Dimitropoulos <sdimitropoulos@coreweave.com>

Author: sdimitro

homa_impl.h

@@ -46,6 +46,9 @@
 #ifndef __UPSTREAM__ /* See strip.py */
 #include "homa.h"
 #include <linux/version.h>
+#ifdef CONFIG_ARM64
+#include <clocksource/arm_arch_timer.h>
+#endif
 #include "homa_devel.h"
 #else /* See strip.py */
 #include <linux/homa.h>
@@ -842,7 +845,13 @@ static inline u64 homa_clock_khz(void)
 	return 1000000;
 #else /* __UNIT_TEST__ */
 #ifndef __UPSTREAM__ /* See strip.py */
-	return cpu_khz;
+#ifdef CONFIG_X86
+	return tsc_khz;
+#elif defined(CONFIG_ARM64)
+	return arch_timer_get_cntfrq() / 1000;
+#else
+	return 1000000;
+#endif
 #else /* See strip.py */
 	return 1000000;
 #endif /* See strip.py */

homa_metrics.c

@@ -241,13 +241,26 @@ char *homa_metrics_print(void)
 		M("bypass_softirq_cycles", m->bypass_softirq_cycles,
 		  "Time spent in homa_softirq during bypass from GRO\n");
 
-		/* Adjust stats gathered in Linux that use rdtsc. */
+		/* Adjust stats gathered in Linux via get_cycles (RDTSC on
+		 * x86, arch timer on arm64). On x86 both Linux and
+		 * homa_clock() read the TSC, so the ratio is 1; the
+		 * explicit conversion is kept in case the sources ever
+		 * diverge. On non-x86 the counters are identical, so raw
+		 * values are emitted directly.
+		 */
+#ifdef CONFIG_X86
 		M("linux_softirq_cycles", m->linux_softirq_cycles *
 		  (homa_clock_khz() / 1000) / (tsc_khz / 1000),
 		  "Time spent in all Linux SoftIRQ\n");
 		M("napi_cycles", m->napi_cycles * (homa_clock_khz() / 1000) /
 		  (tsc_khz / 1000),
 		  "Time spent in NAPI-level packet handling\n");
+#else
+		M("linux_softirq_cycles", m->linux_softirq_cycles,
+		  "Time spent in all Linux SoftIRQ\n");
+		M("napi_cycles", m->napi_cycles,
+		  "Time spent in NAPI-level packet handling\n");
+#endif
 		M("linux_softirqd_actions", m->linux_softirqd_actions,
 		  "SoftIRQ actions taken in the background softirqd thread\n");
 		M("send_cycles", m->send_cycles,

timetrace.c

@@ -390,7 +390,8 @@ int tt_proc_open(struct inode *inode, struct file *file)
 
 	if (!tt_test_no_khz) {
 		pf->bytes_available = snprintf(pf->msg_storage, TT_PF_BUF_SIZE,
-					       "cpu_khz: %u\n", tsc_khz);
+					       "cpu_khz: %llu\n",
+					       homa_clock_khz());
 	}
 
 done:
@@ -631,7 +632,7 @@ void tt_print_file(char *path)
 
 	bytes_used += snprintf(buffer + bytes_used,
 			sizeof(buffer) - bytes_used,
-			"cpu_khz: %u\n", tsc_khz);
+			"cpu_khz: %llu\n", homa_clock_khz());
 
 	/* Each iteration of this loop printk's one event. */
 	while (true) {
@@ -758,7 +759,7 @@ void tt_printk(void)
 	}
 #endif
 
-	pr_err("cpu_khz: %u, start: %llu\n", tsc_khz, start_time);
+	pr_err("cpu_khz: %llu, start: %llu\n", homa_clock_khz(), start_time);
 
 	/* Each iteration of this loop printk's one event. */
 	while (true) {