Documentation/srso: Document a method for checking safe RET operates properly

Add a method to quickly verify whether safe RET operates properly on
a given system using perf tool.

Also, add a selftest which does the same thing.

Signed-off-by: Borislav Petkov (AMD) <[e-mail]>
Link: https://lore.kernel.org/r/[e-mail]

Documentation/admin-guide/hw-vuln/srso.rst

···
 In older Zen1 and Zen2, this is accomplished using a reinterpretation
 technique similar to Retbleed one: srso_untrain_ret() and
 srso_safe_ret().
+
+Checking the safe RET mitigation actually works
+-----------------------------------------------
+
+In case one wants to validate whether the SRSO safe RET mitigation works
+on a kernel, one could use two performance counters
+
+* PMC_0xc8 - Count of RET/RET lw retired
+* PMC_0xc9 - Count of RET/RET lw retired mispredicted
+
+and compare the number of RETs retired properly vs those retired
+mispredicted, in kernel mode. Another way of specifying those events
+is::
+
+        # perf list ex_ret_near_ret
+
+        List of pre-defined events (to be used in -e or -M):
+
+        core:
+          ex_ret_near_ret
+               [Retired Near Returns]
+          ex_ret_near_ret_mispred
+               [Retired Near Returns Mispredicted]
+
+Either the command using the event mnemonics::
+
+        # perf stat -e ex_ret_near_ret:k -e ex_ret_near_ret_mispred:k sleep 10s
+
+or using the raw PMC numbers::
+
+        # perf stat -e cpu/event=0xc8,umask=0/k -e cpu/event=0xc9,umask=0/k sleep 10s
+
+should give the same amount. I.e., every RET retired should be
+mispredicted::
+
+        [root@brent: ~/kernel/linux/tools/perf> ./perf stat -e cpu/event=0xc8,umask=0/k -e cpu/event=0xc9,umask=0/k sleep 10s
+
+         Performance counter stats for 'sleep 10s':
+
+                   137,167      cpu/event=0xc8,umask=0/k
+                   137,173      cpu/event=0xc9,umask=0/k
+
+              10.004110303 seconds time elapsed
+
+               0.000000000 seconds user
+               0.004462000 seconds sys
+
+vs the case when the mitigation is disabled (spec_rstack_overflow=off)
+or not functioning properly, showing usually a lot smaller number of
+mispredicted retired RETs vs the overall count of retired RETs during
+a workload::
+
+       [root@brent: ~/kernel/linux/tools/perf> ./perf stat -e cpu/event=0xc8,umask=0/k -e cpu/event=0xc9,umask=0/k sleep 10s
+
+        Performance counter stats for 'sleep 10s':
+
+                  201,627      cpu/event=0xc8,umask=0/k
+                    4,074      cpu/event=0xc9,umask=0/k
+
+             10.003267252 seconds time elapsed
+
+              0.002729000 seconds user
+              0.000000000 seconds sys
+
+Also, there is a selftest which performs the above, go to
+tools/testing/selftests/x86/ and do::
+
+        make srso
+        ./srso

tools/testing/selftests/x86/Makefile

···
 
 all_64: $(BINARIES_64)
 
-EXTRA_CLEAN := $(BINARIES_32) $(BINARIES_64)
+EXTRA_CLEAN := $(BINARIES_32) $(BINARIES_64) srso
 
 $(BINARIES_32): $(OUTPUT)/%_32: %.c helpers.h
 	$(CC) -m32 -o $@ $(CFLAGS) $(EXTRA_CFLAGS) $< $(EXTRA_FILES) -lrt -ldl -lm

tools/testing/selftests/x86/srso.c

···
+// SPDX-License-Identifier: GPL-2.0
+#include <linux/perf_event.h>
+#include <cpuid.h>
+#include <errno.h>
+#include <stdio.h>
+#include <stdlib.h>
+#include <string.h>
+#include <sys/ioctl.h>
+#include <sys/syscall.h>
+#include <unistd.h>
+
+int main(void)
+{
+	struct perf_event_attr ret_attr, mret_attr;
+	long long count_rets, count_rets_mispred;
+	int rrets_fd, mrrets_fd;
+	unsigned int cpuid1_eax, b, c, d;
+
+	__cpuid(1, cpuid1_eax, b, c, d);
+
+	if (cpuid1_eax < 0x00800f00 ||
+	    cpuid1_eax > 0x00afffff) {
+		fprintf(stderr, "This needs to run on a Zen[1-4] machine (CPUID(1).EAX: 0x%x). Exiting...\n", cpuid1_eax);
+		exit(EXIT_FAILURE);
+	}
+
+	memset(&ret_attr, 0, sizeof(struct perf_event_attr));
+	memset(&mret_attr, 0, sizeof(struct perf_event_attr));
+
+	ret_attr.type = mret_attr.type = PERF_TYPE_RAW;
+	ret_attr.size = mret_attr.size = sizeof(struct perf_event_attr);
+	ret_attr.config = 0xc8;
+	mret_attr.config = 0xc9;
+	ret_attr.disabled = mret_attr.disabled = 1;
+	ret_attr.exclude_user = mret_attr.exclude_user = 1;
+	ret_attr.exclude_hv = mret_attr.exclude_hv = 1;
+
+	rrets_fd = syscall(SYS_perf_event_open, &ret_attr, 0, -1, -1, 0);
+	if (rrets_fd == -1) {
+		perror("opening retired RETs fd");
+		exit(EXIT_FAILURE);
+	}
+
+	mrrets_fd = syscall(SYS_perf_event_open, &mret_attr, 0, -1, -1, 0);
+	if (mrrets_fd == -1) {
+		perror("opening retired mispredicted RETs fd");
+		exit(EXIT_FAILURE);
+	}
+
+	ioctl(rrets_fd, PERF_EVENT_IOC_RESET, 0);
+	ioctl(mrrets_fd, PERF_EVENT_IOC_RESET, 0);
+
+	ioctl(rrets_fd, PERF_EVENT_IOC_ENABLE, 0);
+	ioctl(mrrets_fd, PERF_EVENT_IOC_ENABLE, 0);
+
+	printf("Sleeping for 10 seconds\n");
+	sleep(10);
+
+	ioctl(rrets_fd, PERF_EVENT_IOC_DISABLE, 0);
+	ioctl(mrrets_fd, PERF_EVENT_IOC_DISABLE, 0);
+
+	read(rrets_fd, &count_rets, sizeof(long long));
+	read(mrrets_fd, &count_rets_mispred, sizeof(long long));
+
+	printf("RETs: (%lld retired <-> %lld mispredicted)\n",
+		count_rets, count_rets_mispred);
+	printf("SRSO Safe-RET mitigation works correctly if both counts are almost equal.\n");
+
+	return 0;
+}