diff options
Diffstat (limited to 'tools/testing/selftests/powerpc')
13 files changed, 994 insertions, 21 deletions
diff --git a/tools/testing/selftests/powerpc/alignment/Makefile b/tools/testing/selftests/powerpc/alignment/Makefile index 16b22004e75f..083a48a008b4 100644 --- a/tools/testing/selftests/powerpc/alignment/Makefile +++ b/tools/testing/selftests/powerpc/alignment/Makefile @@ -1,4 +1,5 @@ -TEST_GEN_PROGS := copy_unaligned copy_first_unaligned paste_unaligned paste_last_unaligned +TEST_GEN_PROGS := copy_unaligned copy_first_unaligned paste_unaligned \ + paste_last_unaligned alignment_handler include ../../lib.mk diff --git a/tools/testing/selftests/powerpc/alignment/alignment_handler.c b/tools/testing/selftests/powerpc/alignment/alignment_handler.c new file mode 100644 index 000000000000..39fd362415cf --- /dev/null +++ b/tools/testing/selftests/powerpc/alignment/alignment_handler.c @@ -0,0 +1,491 @@ +/* + * Test the powerpc alignment handler on POWER8/POWER9 + * + * Copyright (C) 2017 IBM Corporation (Michael Neuling, Andrew Donnellan) + * + * This program is free software; you can redistribute it and/or + * modify it under the terms of the GNU General Public License + * as published by the Free Software Foundation; either version + * 2 of the License, or (at your option) any later version. + */ + +/* + * This selftest exercises the powerpc alignment fault handler. + * + * We create two sets of source and destination buffers, one in regular memory, + * the other cache-inhibited (we use /dev/fb0 for this). + * + * We initialise the source buffers, then use whichever set of load/store + * instructions is under test to copy bytes from the source buffers to the + * destination buffers. For the regular buffers, these instructions will + * execute normally. For the cache-inhibited buffers, these instructions + * will trap and cause an alignment fault, and the alignment fault handler + * will emulate the particular instruction under test. We then compare the + * destination buffers to ensure that the native and emulated cases give the + * same result. + * + * TODO: + * - Any FIXMEs below + * - Test VSX regs < 32 and > 32 + * - Test all loads and stores + * - Check update forms do update register + * - Test alignment faults over page boundary + * + * Some old binutils may not support all the instructions. + */ + + +#include <sys/mman.h> +#include <sys/types.h> +#include <sys/stat.h> +#include <fcntl.h> +#include <unistd.h> +#include <stdio.h> +#include <stdlib.h> +#include <string.h> +#include <assert.h> +#include <getopt.h> +#include <setjmp.h> +#include <signal.h> + +#include "utils.h" + +int bufsize; +int debug; +int testing; +volatile int gotsig; + +void sighandler(int sig, siginfo_t *info, void *ctx) +{ + struct ucontext *ucp = ctx; + + if (!testing) { + signal(sig, SIG_DFL); + kill(0, sig); + } + gotsig = sig; +#ifdef __powerpc64__ + ucp->uc_mcontext.gp_regs[PT_NIP] += 4; +#else + ucp->uc_mcontext.uc_regs->gregs[PT_NIP] += 4; +#endif +} + +#define XFORM(reg, n) " " #reg " ,%"#n",%2 ;" +#define DFORM(reg, n) " " #reg " ,0(%"#n") ;" + +#define TEST(name, ld_op, st_op, form, ld_reg, st_reg) \ + void test_##name(char *s, char *d) \ + { \ + asm volatile( \ + #ld_op form(ld_reg, 0) \ + #st_op form(st_reg, 1) \ + :: "r"(s), "r"(d), "r"(0) \ + : "memory", "vs0", "vs32", "r31"); \ + } \ + rc |= do_test(#name, test_##name) + +#define LOAD_VSX_XFORM_TEST(op) TEST(op, op, stxvd2x, XFORM, 32, 32) +#define STORE_VSX_XFORM_TEST(op) TEST(op, lxvd2x, op, XFORM, 32, 32) +#define LOAD_VSX_DFORM_TEST(op) TEST(op, op, stxv, DFORM, 32, 32) +#define STORE_VSX_DFORM_TEST(op) TEST(op, lxv, op, DFORM, 32, 32) +#define LOAD_VMX_XFORM_TEST(op) TEST(op, op, stxvd2x, XFORM, 0, 32) +#define STORE_VMX_XFORM_TEST(op) TEST(op, lxvd2x, op, XFORM, 32, 0) +#define LOAD_VMX_DFORM_TEST(op) TEST(op, op, stxv, DFORM, 0, 32) +#define STORE_VMX_DFORM_TEST(op) TEST(op, lxv, op, DFORM, 32, 0) + +#define LOAD_XFORM_TEST(op) TEST(op, op, stdx, XFORM, 31, 31) +#define STORE_XFORM_TEST(op) TEST(op, ldx, op, XFORM, 31, 31) +#define LOAD_DFORM_TEST(op) TEST(op, op, std, DFORM, 31, 31) +#define STORE_DFORM_TEST(op) TEST(op, ld, op, DFORM, 31, 31) + +#define LOAD_FLOAT_DFORM_TEST(op) TEST(op, op, stfd, DFORM, 0, 0) +#define STORE_FLOAT_DFORM_TEST(op) TEST(op, lfd, op, DFORM, 0, 0) +#define LOAD_FLOAT_XFORM_TEST(op) TEST(op, op, stfdx, XFORM, 0, 0) +#define STORE_FLOAT_XFORM_TEST(op) TEST(op, lfdx, op, XFORM, 0, 0) + + +/* FIXME: Unimplemented tests: */ +// STORE_DFORM_TEST(stq) /* FIXME: need two registers for quad */ +// STORE_DFORM_TEST(stswi) /* FIXME: string instruction */ + +// STORE_XFORM_TEST(stwat) /* AMO can't emulate or run on CI */ +// STORE_XFORM_TEST(stdat) /* ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ */ + + +/* preload byte by byte */ +void preload_data(void *dst, int offset, int width) +{ + char *c = dst; + int i; + + c += offset; + + for (i = 0 ; i < width ; i++) + c[i] = i; +} + +int test_memcpy(void *dst, void *src, int size, int offset, + void (*test_func)(char *, char *)) +{ + char *s, *d; + + s = src; + s += offset; + d = dst; + d += offset; + + assert(size == 16); + gotsig = 0; + testing = 1; + + test_func(s, d); /* run the actual test */ + + testing = 0; + if (gotsig) { + if (debug) + printf(" Got signal %i\n", gotsig); + return 1; + } + return 0; +} + +void dumpdata(char *s1, char *s2, int n, char *test_name) +{ + int i; + + printf(" %s: unexpected result:\n", test_name); + printf(" mem:"); + for (i = 0; i < n; i++) + printf(" %02x", s1[i]); + printf("\n"); + printf(" ci: "); + for (i = 0; i < n; i++) + printf(" %02x", s2[i]); + printf("\n"); +} + +int test_memcmp(void *s1, void *s2, int n, int offset, char *test_name) +{ + char *s1c, *s2c; + + s1c = s1; + s1c += offset; + s2c = s2; + s2c += offset; + + if (memcmp(s1c, s2c, n)) { + if (debug) { + printf("\n Compare failed. Offset:%i length:%i\n", + offset, n); + dumpdata(s1c, s2c, n, test_name); + } + return 1; + } + return 0; +} + +/* + * Do two memcpy tests using the same instructions. One cachable + * memory and the other doesn't. + */ +int do_test(char *test_name, void (*test_func)(char *, char *)) +{ + int offset, width, fd, rc = 0, r; + void *mem0, *mem1, *ci0, *ci1; + + printf("\tDoing %s:\t", test_name); + + fd = open("/dev/fb0", O_RDWR); + if (fd < 0) { + printf("\n"); + perror("Can't open /dev/fb0"); + SKIP_IF(1); + } + + ci0 = mmap(NULL, bufsize, PROT_WRITE, MAP_SHARED, + fd, 0x0); + ci1 = mmap(NULL, bufsize, PROT_WRITE, MAP_SHARED, + fd, bufsize); + if ((ci0 == MAP_FAILED) || (ci1 == MAP_FAILED)) { + printf("\n"); + perror("mmap failed"); + SKIP_IF(1); + } + + rc = posix_memalign(&mem0, bufsize, bufsize); + if (rc) { + printf("\n"); + return rc; + } + + rc = posix_memalign(&mem1, bufsize, bufsize); + if (rc) { + printf("\n"); + free(mem0); + return rc; + } + + /* offset = 0 no alignment fault, so skip */ + for (offset = 1; offset < 16; offset++) { + width = 16; /* vsx == 16 bytes */ + r = 0; + + /* load pattern into memory byte by byte */ + preload_data(ci0, offset, width); + preload_data(mem0, offset, width); // FIXME: remove?? + memcpy(ci0, mem0, bufsize); + memcpy(ci1, mem1, bufsize); /* initialise output to the same */ + + /* sanity check */ + test_memcmp(mem0, ci0, width, offset, test_name); + + r |= test_memcpy(ci1, ci0, width, offset, test_func); + r |= test_memcpy(mem1, mem0, width, offset, test_func); + if (r && !debug) { + printf("FAILED: Got signal"); + break; + } + + r |= test_memcmp(mem1, ci1, width, offset, test_name); + rc |= r; + if (r && !debug) { + printf("FAILED: Wrong Data"); + break; + } + } + if (!r) + printf("PASSED"); + printf("\n"); + + munmap(ci0, bufsize); + munmap(ci1, bufsize); + free(mem0); + free(mem1); + + return rc; +} + +int test_alignment_handler_vsx_206(void) +{ + int rc = 0; + + printf("VSX: 2.06B\n"); + LOAD_VSX_XFORM_TEST(lxvd2x); + LOAD_VSX_XFORM_TEST(lxvw4x); + LOAD_VSX_XFORM_TEST(lxsdx); + LOAD_VSX_XFORM_TEST(lxvdsx); + STORE_VSX_XFORM_TEST(stxvd2x); + STORE_VSX_XFORM_TEST(stxvw4x); + STORE_VSX_XFORM_TEST(stxsdx); + return rc; +} + +int test_alignment_handler_vsx_207(void) +{ + int rc = 0; + + printf("VSX: 2.07B\n"); + LOAD_VSX_XFORM_TEST(lxsspx); + LOAD_VSX_XFORM_TEST(lxsiwax); + LOAD_VSX_XFORM_TEST(lxsiwzx); + STORE_VSX_XFORM_TEST(stxsspx); + STORE_VSX_XFORM_TEST(stxsiwx); + return rc; +} + +int test_alignment_handler_vsx_300(void) +{ + int rc = 0; + + SKIP_IF(!have_hwcap2(PPC_FEATURE2_ARCH_3_00)); + printf("VSX: 3.00B\n"); + LOAD_VMX_DFORM_TEST(lxsd); + LOAD_VSX_XFORM_TEST(lxsibzx); + LOAD_VSX_XFORM_TEST(lxsihzx); + LOAD_VMX_DFORM_TEST(lxssp); + LOAD_VSX_DFORM_TEST(lxv); + LOAD_VSX_XFORM_TEST(lxvb16x); + LOAD_VSX_XFORM_TEST(lxvh8x); + LOAD_VSX_XFORM_TEST(lxvx); + LOAD_VSX_XFORM_TEST(lxvwsx); + LOAD_VSX_XFORM_TEST(lxvl); + LOAD_VSX_XFORM_TEST(lxvll); + STORE_VMX_DFORM_TEST(stxsd); + STORE_VSX_XFORM_TEST(stxsibx); + STORE_VSX_XFORM_TEST(stxsihx); + STORE_VMX_DFORM_TEST(stxssp); + STORE_VSX_DFORM_TEST(stxv); + STORE_VSX_XFORM_TEST(stxvb16x); + STORE_VSX_XFORM_TEST(stxvh8x); + STORE_VSX_XFORM_TEST(stxvx); + STORE_VSX_XFORM_TEST(stxvl); + STORE_VSX_XFORM_TEST(stxvll); + return rc; +} + +int test_alignment_handler_integer(void) +{ + int rc = 0; + + printf("Integer\n"); + LOAD_DFORM_TEST(lbz); + LOAD_DFORM_TEST(lbzu); + LOAD_XFORM_TEST(lbzx); + LOAD_XFORM_TEST(lbzux); + LOAD_DFORM_TEST(lhz); + LOAD_DFORM_TEST(lhzu); + LOAD_XFORM_TEST(lhzx); + LOAD_XFORM_TEST(lhzux); + LOAD_DFORM_TEST(lha); + LOAD_DFORM_TEST(lhau); + LOAD_XFORM_TEST(lhax); + LOAD_XFORM_TEST(lhaux); + LOAD_XFORM_TEST(lhbrx); + LOAD_DFORM_TEST(lwz); + LOAD_DFORM_TEST(lwzu); + LOAD_XFORM_TEST(lwzx); + LOAD_XFORM_TEST(lwzux); + LOAD_DFORM_TEST(lwa); + LOAD_XFORM_TEST(lwax); + LOAD_XFORM_TEST(lwaux); + LOAD_XFORM_TEST(lwbrx); + LOAD_DFORM_TEST(ld); + LOAD_DFORM_TEST(ldu); + LOAD_XFORM_TEST(ldx); + LOAD_XFORM_TEST(ldux); + LOAD_XFORM_TEST(ldbrx); + LOAD_DFORM_TEST(lmw); + STORE_DFORM_TEST(stb); + STORE_XFORM_TEST(stbx); + STORE_DFORM_TEST(stbu); + STORE_XFORM_TEST(stbux); + STORE_DFORM_TEST(sth); + STORE_XFORM_TEST(sthx); + STORE_DFORM_TEST(sthu); + STORE_XFORM_TEST(sthux); + STORE_XFORM_TEST(sthbrx); + STORE_DFORM_TEST(stw); + STORE_XFORM_TEST(stwx); + STORE_DFORM_TEST(stwu); + STORE_XFORM_TEST(stwux); + STORE_XFORM_TEST(stwbrx); + STORE_DFORM_TEST(std); + STORE_XFORM_TEST(stdx); + STORE_DFORM_TEST(stdu); + STORE_XFORM_TEST(stdux); + STORE_XFORM_TEST(stdbrx); + STORE_DFORM_TEST(stmw); + return rc; +} + +int test_alignment_handler_vmx(void) +{ + int rc = 0; + + printf("VMX\n"); + LOAD_VMX_XFORM_TEST(lvx); + + /* + * FIXME: These loads only load part of the register, so our + * testing method doesn't work. Also they don't take alignment + * faults, so it's kinda pointless anyway + * + LOAD_VMX_XFORM_TEST(lvebx) + LOAD_VMX_XFORM_TEST(lvehx) + LOAD_VMX_XFORM_TEST(lvewx) + LOAD_VMX_XFORM_TEST(lvxl) + */ + STORE_VMX_XFORM_TEST(stvx); + STORE_VMX_XFORM_TEST(stvebx); + STORE_VMX_XFORM_TEST(stvehx); + STORE_VMX_XFORM_TEST(stvewx); + STORE_VMX_XFORM_TEST(stvxl); + return rc; +} + +int test_alignment_handler_fp(void) +{ + int rc = 0; + + printf("Floating point\n"); + LOAD_FLOAT_DFORM_TEST(lfd); + LOAD_FLOAT_XFORM_TEST(lfdx); + LOAD_FLOAT_DFORM_TEST(lfdp); + LOAD_FLOAT_XFORM_TEST(lfdpx); + LOAD_FLOAT_DFORM_TEST(lfdu); + LOAD_FLOAT_XFORM_TEST(lfdux); + LOAD_FLOAT_DFORM_TEST(lfs); + LOAD_FLOAT_XFORM_TEST(lfsx); + LOAD_FLOAT_DFORM_TEST(lfsu); + LOAD_FLOAT_XFORM_TEST(lfsux); + LOAD_FLOAT_XFORM_TEST(lfiwzx); + LOAD_FLOAT_XFORM_TEST(lfiwax); + STORE_FLOAT_DFORM_TEST(stfd); + STORE_FLOAT_XFORM_TEST(stfdx); + STORE_FLOAT_DFORM_TEST(stfdp); + STORE_FLOAT_XFORM_TEST(stfdpx); + STORE_FLOAT_DFORM_TEST(stfdu); + STORE_FLOAT_XFORM_TEST(stfdux); + STORE_FLOAT_DFORM_TEST(stfs); + STORE_FLOAT_XFORM_TEST(stfsx); + STORE_FLOAT_DFORM_TEST(stfsu); + STORE_FLOAT_XFORM_TEST(stfsux); + STORE_FLOAT_XFORM_TEST(stfiwx); + + return rc; +} + +void usage(char *prog) +{ + printf("Usage: %s [options]\n", prog); + printf(" -d Enable debug error output\n"); + printf("\n"); + printf("This test requires a POWER8 or POWER9 CPU and a usable "); + printf("framebuffer at /dev/fb0.\n"); +} + +int main(int argc, char *argv[]) +{ + + struct sigaction sa; + int rc = 0; + int option = 0; + + while ((option = getopt(argc, argv, "d")) != -1) { + switch (option) { + case 'd': + debug++; + break; + default: + usage(argv[0]); + exit(1); + } + } + + bufsize = getpagesize(); + + sa.sa_sigaction = sighandler; + sigemptyset(&sa.sa_mask); + sa.sa_flags = SA_SIGINFO; + if (sigaction(SIGSEGV, &sa, NULL) == -1 + || sigaction(SIGBUS, &sa, NULL) == -1 + || sigaction(SIGILL, &sa, NULL) == -1) { + perror("sigaction"); + exit(1); + } + + rc |= test_harness(test_alignment_handler_vsx_206, + "test_alignment_handler_vsx_206"); + rc |= test_harness(test_alignment_handler_vsx_207, + "test_alignment_handler_vsx_207"); + rc |= test_harness(test_alignment_handler_vsx_300, + "test_alignment_handler_vsx_300"); + rc |= test_harness(test_alignment_handler_integer, + "test_alignment_handler_integer"); + rc |= test_harness(test_alignment_handler_vmx, + "test_alignment_handler_vmx"); + rc |= test_harness(test_alignment_handler_fp, + "test_alignment_handler_fp"); + return rc; +} diff --git a/tools/testing/selftests/powerpc/benchmarks/mmap_bench.c b/tools/testing/selftests/powerpc/benchmarks/mmap_bench.c index 8d084a2d6e74..7a0a462a2272 100644 --- a/tools/testing/selftests/powerpc/benchmarks/mmap_bench.c +++ b/tools/testing/selftests/powerpc/benchmarks/mmap_bench.c @@ -7,17 +7,34 @@ #include <stdlib.h> #include <sys/mman.h> #include <time.h> +#include <getopt.h> #include "utils.h" #define ITERATIONS 5000000 -#define MEMSIZE (128 * 1024 * 1024) +#define MEMSIZE (1UL << 27) +#define PAGE_SIZE (1UL << 16) +#define CHUNK_COUNT (MEMSIZE/PAGE_SIZE) + +static int pg_fault; +static int iterations = ITERATIONS; + +static struct option options[] = { + { "pgfault", no_argument, &pg_fault, 1 }, + { "iterations", required_argument, 0, 'i' }, + { 0, }, +}; + +static void usage(void) +{ + printf("mmap_bench <--pgfault> <--iterations count>\n"); +} int test_mmap(void) { struct timespec ts_start, ts_end; - unsigned long i = ITERATIONS; + unsigned long i = iterations; clock_gettime(CLOCK_MONOTONIC, &ts_start); @@ -25,6 +42,11 @@ int test_mmap(void) char *c = mmap(NULL, MEMSIZE, PROT_READ|PROT_WRITE, MAP_PRIVATE|MAP_ANONYMOUS, -1, 0); FAIL_IF(c == MAP_FAILED); + if (pg_fault) { + int count; + for (count = 0; count < CHUNK_COUNT; count++) + c[count << 16] = 'c'; + } munmap(c, MEMSIZE); } @@ -35,7 +57,32 @@ int test_mmap(void) return 0; } -int main(void) +int main(int argc, char *argv[]) { + signed char c; + while (1) { + int option_index = 0; + + c = getopt_long(argc, argv, "", options, &option_index); + + if (c == -1) + break; + + switch (c) { + case 0: + if (options[option_index].flag != 0) + break; + + usage(); + exit(1); + break; + case 'i': + iterations = atoi(optarg); + break; + default: + usage(); + exit(1); + } + } return test_harness(test_mmap, "mmap_bench"); } diff --git a/tools/testing/selftests/powerpc/mm/.gitignore b/tools/testing/selftests/powerpc/mm/.gitignore index e715a3f2fbf4..7d7c42ed6de9 100644 --- a/tools/testing/selftests/powerpc/mm/.gitignore +++ b/tools/testing/selftests/powerpc/mm/.gitignore @@ -1,4 +1,5 @@ hugetlb_vs_thp_test subpage_prot tempfile -prot_sao
\ No newline at end of file +prot_sao +segv_errors
\ No newline at end of file diff --git a/tools/testing/selftests/powerpc/mm/Makefile b/tools/testing/selftests/powerpc/mm/Makefile index bf315bcbe663..8ebbe96d80a8 100644 --- a/tools/testing/selftests/powerpc/mm/Makefile +++ b/tools/testing/selftests/powerpc/mm/Makefile @@ -2,7 +2,7 @@ noarg: $(MAKE) -C ../ -TEST_GEN_PROGS := hugetlb_vs_thp_test subpage_prot prot_sao +TEST_GEN_PROGS := hugetlb_vs_thp_test subpage_prot prot_sao segv_errors TEST_GEN_FILES := tempfile include ../../lib.mk diff --git a/tools/testing/selftests/powerpc/mm/segv_errors.c b/tools/testing/selftests/powerpc/mm/segv_errors.c new file mode 100644 index 000000000000..06ae76ee3ea1 --- /dev/null +++ b/tools/testing/selftests/powerpc/mm/segv_errors.c @@ -0,0 +1,78 @@ +// SPDX-License-Identifier: GPL-2.0 + +/* + * Copyright 2017 John Sperbeck + * + * Test that an access to a mapped but inaccessible area causes a SEGV and + * reports si_code == SEGV_ACCERR. + */ + +#include <stdbool.h> +#include <stdio.h> +#include <stdlib.h> +#include <string.h> +#include <unistd.h> +#include <signal.h> +#include <sys/mman.h> +#include <assert.h> +#include <ucontext.h> + +#include "utils.h" + +static bool faulted; +static int si_code; + +static void segv_handler(int n, siginfo_t *info, void *ctxt_v) +{ + ucontext_t *ctxt = (ucontext_t *)ctxt_v; + struct pt_regs *regs = ctxt->uc_mcontext.regs; + + faulted = true; + si_code = info->si_code; + regs->nip += 4; +} + +int test_segv_errors(void) +{ + struct sigaction act = { + .sa_sigaction = segv_handler, + .sa_flags = SA_SIGINFO, + }; + char c, *p = NULL; + + p = mmap(NULL, getpagesize(), 0, MAP_PRIVATE|MAP_ANONYMOUS, -1, 0); + FAIL_IF(p == MAP_FAILED); + + FAIL_IF(sigaction(SIGSEGV, &act, NULL) != 0); + + faulted = false; + si_code = 0; + + /* + * We just need a compiler barrier, but mb() works and has the nice + * property of being easy to spot in the disassembly. + */ + mb(); + c = *p; + mb(); + + FAIL_IF(!faulted); + FAIL_IF(si_code != SEGV_ACCERR); + + faulted = false; + si_code = 0; + + mb(); + *p = c; + mb(); + + FAIL_IF(!faulted); + FAIL_IF(si_code != SEGV_ACCERR); + + return 0; +} + +int main(void) +{ + return test_harness(test_segv_errors, "segv_errors"); +} diff --git a/tools/testing/selftests/powerpc/ptrace/ptrace-tm-spd-vsx.c b/tools/testing/selftests/powerpc/ptrace/ptrace-tm-spd-vsx.c index 0df3c23b7888..277dade1b382 100644 --- a/tools/testing/selftests/powerpc/ptrace/ptrace-tm-spd-vsx.c +++ b/tools/testing/selftests/powerpc/ptrace/ptrace-tm-spd-vsx.c @@ -79,8 +79,8 @@ trans: : [res] "=r" (result), [texasr] "=r" (texasr) : [fp_load] "r" (fp_load), [fp_load_ckpt] "r" (fp_load_ckpt), [sprn_texasr] "i" (SPRN_TEXASR) - : "memory", "r0", "r1", "r2", "r3", "r4", - "r8", "r9", "r10", "r11" + : "memory", "r0", "r1", "r3", "r4", + "r7", "r8", "r9", "r10", "r11" ); if (result) { diff --git a/tools/testing/selftests/powerpc/ptrace/ptrace-tm-spr.c b/tools/testing/selftests/powerpc/ptrace/ptrace-tm-spr.c index 94e57cb89769..51427a2465f6 100644 --- a/tools/testing/selftests/powerpc/ptrace/ptrace-tm-spr.c +++ b/tools/testing/selftests/powerpc/ptrace/ptrace-tm-spr.c @@ -76,8 +76,7 @@ trans: : [tfhar] "=r" (tfhar), [res] "=r" (result), [texasr] "=r" (texasr), [cptr1] "=r" (cptr1) : [sprn_texasr] "i" (SPRN_TEXASR) - : "memory", "r0", "r1", "r2", "r3", "r4", - "r8", "r9", "r10", "r11", "r31" + : "memory", "r0", "r8", "r31" ); /* There are 2 32bit instructions before tbegin. */ diff --git a/tools/testing/selftests/powerpc/ptrace/ptrace-tm-vsx.c b/tools/testing/selftests/powerpc/ptrace/ptrace-tm-vsx.c index b4081e2b22d5..17c23cabac3e 100644 --- a/tools/testing/selftests/powerpc/ptrace/ptrace-tm-vsx.c +++ b/tools/testing/selftests/powerpc/ptrace/ptrace-tm-vsx.c @@ -67,7 +67,7 @@ trans: : [res] "=r" (result), [texasr] "=r" (texasr) : [fp_load] "r" (fp_load), [fp_load_ckpt] "r" (fp_load_ckpt), [sprn_texasr] "i" (SPRN_TEXASR), [cptr1] "r" (&cptr[1]) - : "memory", "r0", "r1", "r2", "r3", "r4", + : "memory", "r0", "r1", "r3", "r4", "r7", "r8", "r9", "r10", "r11" ); diff --git a/tools/testing/selftests/powerpc/tm/.gitignore b/tools/testing/selftests/powerpc/tm/.gitignore index 241a4a4ee0e4..bb90d4b79524 100644 --- a/tools/testing/selftests/powerpc/tm/.gitignore +++ b/tools/testing/selftests/powerpc/tm/.gitignore @@ -13,3 +13,4 @@ tm-signal-context-chk-vmx tm-signal-context-chk-vsx tm-vmx-unavail tm-unavailable +tm-trap diff --git a/tools/testing/selftests/powerpc/tm/Makefile b/tools/testing/selftests/powerpc/tm/Makefile index 8ed6f8c57230..a23453943ad2 100644 --- a/tools/testing/selftests/powerpc/tm/Makefile +++ b/tools/testing/selftests/powerpc/tm/Makefile @@ -3,7 +3,7 @@ SIGNAL_CONTEXT_CHK_TESTS := tm-signal-context-chk-gpr tm-signal-context-chk-fpu tm-signal-context-chk-vmx tm-signal-context-chk-vsx TEST_GEN_PROGS := tm-resched-dscr tm-syscall tm-signal-msr-resv tm-signal-stack \ - tm-vmxcopy tm-fork tm-tar tm-tmspr tm-vmx-unavail tm-unavailable \ + tm-vmxcopy tm-fork tm-tar tm-tmspr tm-vmx-unavail tm-unavailable tm-trap \ $(SIGNAL_CONTEXT_CHK_TESTS) include ../../lib.mk @@ -18,6 +18,7 @@ $(OUTPUT)/tm-tmspr: CFLAGS += -pthread $(OUTPUT)/tm-vmx-unavail: CFLAGS += -pthread -m64 $(OUTPUT)/tm-resched-dscr: ../pmu/lib.o $(OUTPUT)/tm-unavailable: CFLAGS += -O0 -pthread -m64 -Wno-error=uninitialized -mvsx +$(OUTPUT)/tm-trap: CFLAGS += -O0 -pthread -m64 SIGNAL_CONTEXT_CHK_TESTS := $(patsubst %,$(OUTPUT)/%,$(SIGNAL_CONTEXT_CHK_TESTS)) $(SIGNAL_CONTEXT_CHK_TESTS): tm-signal.S diff --git a/tools/testing/selftests/powerpc/tm/tm-trap.c b/tools/testing/selftests/powerpc/tm/tm-trap.c new file mode 100644 index 000000000000..5d92c23ee6cb --- /dev/null +++ b/tools/testing/selftests/powerpc/tm/tm-trap.c @@ -0,0 +1,329 @@ +/* + * Copyright 2017, Gustavo Romero, IBM Corp. + * Licensed under GPLv2. + * + * Check if thread endianness is flipped inadvertently to BE on trap + * caught in TM whilst MSR.FP and MSR.VEC are zero (i.e. just after + * load_fp and load_vec overflowed). + * + * The issue can be checked on LE machines simply by zeroing load_fp + * and load_vec and then causing a trap in TM. Since the endianness + * changes to BE on return from the signal handler, 'nop' is + * thread as an illegal instruction in following sequence: + * tbegin. + * beq 1f + * trap + * tend. + * 1: nop + * + * However, although the issue is also present on BE machines, it's a + * bit trickier to check it on BE machines because MSR.LE bit is set + * to zero which determines a BE endianness that is the native + * endianness on BE machines, so nothing notably critical happens, + * i.e. no illegal instruction is observed immediately after returning + * from the signal handler (as it happens on LE machines). Thus to test + * it on BE machines LE endianness is forced after a first trap and then + * the endianness is verified on subsequent traps to determine if the + * endianness "flipped back" to the native endianness (BE). + */ + +#define _GNU_SOURCE +#include <error.h> +#include <stdio.h> +#include <stdlib.h> +#include <unistd.h> +#include <htmintrin.h> +#include <inttypes.h> +#include <pthread.h> +#include <sched.h> +#include <signal.h> +#include <stdbool.h> + +#include "tm.h" +#include "utils.h" + +#define pr_error(error_code, format, ...) \ + error_at_line(1, error_code, __FILE__, __LINE__, format, ##__VA_ARGS__) + +#define MSR_LE 1UL +#define LE 1UL + +pthread_t t0_ping; +pthread_t t1_pong; + +int exit_from_pong; + +int trap_event; +int le; + +bool success; + +void trap_signal_handler(int signo, siginfo_t *si, void *uc) +{ + ucontext_t *ucp = uc; + uint64_t thread_endianness; + + /* Get thread endianness: extract bit LE from MSR */ + thread_endianness = MSR_LE & ucp->uc_mcontext.gp_regs[PT_MSR]; + + /*** + * Little-Endian Machine + */ + + if (le) { + /* First trap event */ + if (trap_event == 0) { + /* Do nothing. Since it is returning from this trap + * event that endianness is flipped by the bug, so just + * let the process return from the signal handler and + * check on the second trap event if endianness is + * flipped or not. + */ + } + /* Second trap event */ + else if (trap_event == 1) { + /* + * Since trap was caught in TM on first trap event, if + * endianness was still LE (not flipped inadvertently) + * after returning from the signal handler instruction + * (1) is executed (basically a 'nop'), as it's located + * at address of tbegin. +4 (rollback addr). As (1) on + * LE endianness does in effect nothing, instruction (2) + * is then executed again as 'trap', generating a second + * trap event (note that in that case 'trap' is caught + * not in transacional mode). On te other hand, if after + * the return from the signal handler the endianness in- + * advertently flipped, instruction (1) is tread as a + * branch instruction, i.e. b .+8, hence instruction (3) + * and (4) are executed (tbegin.; trap;) and we get sim- + * ilaly on the trap signal handler, but now in TM mode. + * Either way, it's now possible to check the MSR LE bit + * once in the trap handler to verify if endianness was + * flipped or not after the return from the second trap + * event. If endianness is flipped, the bug is present. + * Finally, getting a trap in TM mode or not is just + * worth noting because it affects the math to determine + * the offset added to the NIP on return: the NIP for a + * trap caught in TM is the rollback address, i.e. the + * next instruction after 'tbegin.', whilst the NIP for + * a trap caught in non-transactional mode is the very + * same address of the 'trap' instruction that generated + * the trap event. + */ + + if (thread_endianness == LE) { + /* Go to 'success', i.e. instruction (6) */ + ucp->uc_mcontext.gp_regs[PT_NIP] += 16; + } else { + /* + * Thread endianness is BE, so it flipped + * inadvertently. Thus we flip back to LE and + * set NIP to go to 'failure', instruction (5). + */ + ucp->uc_mcontext.gp_regs[PT_MSR] |= 1UL; + ucp->uc_mcontext.gp_regs[PT_NIP] += 4; + } + } + } + + /*** + * Big-Endian Machine + */ + + else { + /* First trap event */ + if (trap_event == 0) { + /* + * Force thread endianness to be LE. Instructions (1), + * (3), and (4) will be executed, generating a second + * trap in TM mode. + */ + ucp->uc_mcontext.gp_regs[PT_MSR] |= 1UL; + } + /* Second trap event */ + else if (trap_event == 1) { + /* + * Do nothing. If bug is present on return from this + * second trap event endianness will flip back "automat- + * ically" to BE, otherwise thread endianness will + * continue to be LE, just as it was set above. + */ + } + /* A third trap event */ + else { + /* + * Once here it means that after returning from the sec- + * ond trap event instruction (4) (trap) was executed + * as LE, generating a third trap event. In that case + * endianness is still LE as set on return from the + * first trap event, hence no bug. Otherwise, bug + * flipped back to BE on return from the second trap + * event and instruction (4) was executed as 'tdi' (so + * basically a 'nop') and branch to 'failure' in + * instruction (5) was taken to indicate failure and we + * never get here. + */ + + /* + * Flip back to BE and go to instruction (6), i.e. go to + * 'success'. + */ + ucp->uc_mcontext.gp_regs[PT_MSR] &= ~1UL; + ucp->uc_mcontext.gp_regs[PT_NIP] += 8; + } + } + + trap_event++; +} + +void usr1_signal_handler(int signo, siginfo_t *si, void *not_used) +{ + /* Got a USR1 signal from ping(), so just tell pong() to exit */ + exit_from_pong = 1; +} + +void *ping(void *not_used) +{ + uint64_t i; + + trap_event = 0; + + /* + * Wait an amount of context switches so load_fp and load_vec overflows + * and MSR_[FP|VEC|V] is 0. + */ + for (i = 0; i < 1024*1024*512; i++) + ; + + asm goto( + /* + * [NA] means "Native Endianness", i.e. it tells how a + * instruction is executed on machine's native endianness (in + * other words, native endianness matches kernel endianness). + * [OP] means "Opposite Endianness", i.e. on a BE machine, it + * tells how a instruction is executed as a LE instruction; con- + * versely, on a LE machine, it tells how a instruction is + * executed as a BE instruction. When [NA] is omitted, it means + * that the native interpretation of a given instruction is not + * relevant for the test. Likewise when [OP] is omitted. + */ + + " tbegin. ;" /* (0) tbegin. [NA] */ + " tdi 0, 0, 0x48;" /* (1) nop [NA]; b (3) [OP] */ + " trap ;" /* (2) trap [NA] */ + ".long 0x1D05007C;" /* (3) tbegin. [OP] */ + ".long 0x0800E07F;" /* (4) trap [OP]; nop [NA] */ + " b %l[failure] ;" /* (5) b [NA]; MSR.LE flipped (bug) */ + " b %l[success] ;" /* (6) b [NA]; MSR.LE did not flip (ok)*/ + + : : : : failure, success); + +failure: + success = false; + goto exit_from_ping; + +success: + success = true; + +exit_from_ping: + /* Tell pong() to exit before leaving */ + pthread_kill(t1_pong, SIGUSR1); + return NULL; +} + +void *pong(void *not_used) +{ + while (!exit_from_pong) + /* + * Induce context switches on ping() thread + * until ping() finishes its job and signs + * to exit from this loop. + */ + sched_yield(); + + return NULL; +} + +int tm_trap_test(void) +{ + uint16_t k = 1; + + int rc; + + pthread_attr_t attr; + cpu_set_t cpuset; + + struct sigaction trap_sa; + + trap_sa.sa_flags = SA_SIGINFO; + trap_sa.sa_sigaction = trap_signal_handler; + sigaction(SIGTRAP, &trap_sa, NULL); + + struct sigaction usr1_sa; + + usr1_sa.sa_flags = SA_SIGINFO; + usr1_sa.sa_sigaction = usr1_signal_handler; + sigaction(SIGUSR1, &usr1_sa, NULL); + + /* Set only CPU 0 in the mask. Both threads will be bound to cpu 0. */ + CPU_ZERO(&cpuset); + CPU_SET(0, &cpuset); + + /* Init pthread attribute */ + rc = pthread_attr_init(&attr); + if (rc) + pr_error(rc, "pthread_attr_init()"); + + /* + * Bind thread ping() and pong() both to CPU 0 so they ping-pong and + * speed up context switches on ping() thread, speeding up the load_fp + * and load_vec overflow. + */ + rc = pthread_attr_setaffinity_np(&attr, sizeof(cpu_set_t), &cpuset); + if (rc) + pr_error(rc, "pthread_attr_setaffinity()"); + + /* Figure out the machine endianness */ + le = (int) *(uint8_t *)&k; + + printf("%s machine detected. Checking if endianness flips %s", + le ? "Little-Endian" : "Big-Endian", + "inadvertently on trap in TM... "); + + rc = fflush(0); + if (rc) + pr_error(rc, "fflush()"); + + /* Launch ping() */ + rc = pthread_create(&t0_ping, &attr, ping, NULL); + if (rc) + pr_error(rc, "pthread_create()"); + + exit_from_pong = 0; + + /* Launch pong() */ + rc = pthread_create(&t1_pong, &attr, pong, NULL); + if (rc) + pr_error(rc, "pthread_create()"); + + rc = pthread_join(t0_ping, NULL); + if (rc) + pr_error(rc, "pthread_join()"); + + rc = pthread_join(t1_pong, NULL); + if (rc) + pr_error(rc, "pthread_join()"); + + if (success) { + printf("no.\n"); /* no, endianness did not flip inadvertently */ + return EXIT_SUCCESS; + } + + printf("yes!\n"); /* yes, endianness did flip inadvertently */ + return EXIT_FAILURE; +} + +int main(int argc, char **argv) +{ + return test_harness(tm_trap_test, "tm_trap_test"); +} diff --git a/tools/testing/selftests/powerpc/tm/tm-unavailable.c b/tools/testing/selftests/powerpc/tm/tm-unavailable.c index 96c37f84ce54..e6a0fad2bfd0 100644 --- a/tools/testing/selftests/powerpc/tm/tm-unavailable.c +++ b/tools/testing/selftests/powerpc/tm/tm-unavailable.c @@ -15,6 +15,7 @@ */ #define _GNU_SOURCE +#include <error.h> #include <stdio.h> #include <stdlib.h> #include <unistd.h> @@ -33,6 +34,11 @@ #define VSX_UNA_EXCEPTION 2 #define NUM_EXCEPTIONS 3 +#define err_at_line(status, errnum, format, ...) \ + error_at_line(status, errnum, __FILE__, __LINE__, format ##__VA_ARGS__) + +#define pr_warn(code, format, ...) err_at_line(0, code, format, ##__VA_ARGS__) +#define pr_err(code, format, ...) err_at_line(1, code, format, ##__VA_ARGS__) struct Flags { int touch_fp; @@ -303,10 +309,19 @@ void test_fp_vec(int fp, int vec, pthread_attr_t *attr) * checking if the failure cause is the one we expect. */ do { + int rc; + /* Bind 'ping' to CPU 0, as specified in 'attr'. */ - pthread_create(&t0, attr, ping, (void *) &flags); - pthread_setname_np(t0, "ping"); - pthread_join(t0, &ret_value); + rc = pthread_create(&t0, attr, ping, (void *) &flags); + if (rc) + pr_err(rc, "pthread_create()"); + rc = pthread_setname_np(t0, "ping"); + if (rc) + pr_warn(rc, "pthread_setname_np"); + rc = pthread_join(t0, &ret_value); + if (rc) + pr_err(rc, "pthread_join"); + retries--; } while (ret_value != NULL && retries); @@ -320,7 +335,7 @@ void test_fp_vec(int fp, int vec, pthread_attr_t *attr) int main(int argc, char **argv) { - int exception; /* FP = 0, VEC = 1, VSX = 2 */ + int rc, exception; /* FP = 0, VEC = 1, VSX = 2 */ pthread_t t1; pthread_attr_t attr; cpu_set_t cpuset; @@ -330,13 +345,23 @@ int main(int argc, char **argv) CPU_SET(0, &cpuset); /* Init pthread attribute. */ - pthread_attr_init(&attr); + rc = pthread_attr_init(&attr); + if (rc) + pr_err(rc, "pthread_attr_init()"); /* Set CPU 0 mask into the pthread attribute. */ - pthread_attr_setaffinity_np(&attr, sizeof(cpu_set_t), &cpuset); - - pthread_create(&t1, &attr /* Bind 'pong' to CPU 0 */, pong, NULL); - pthread_setname_np(t1, "pong"); /* Name it for systemtap convenience */ + rc = pthread_attr_setaffinity_np(&attr, sizeof(cpu_set_t), &cpuset); + if (rc) + pr_err(rc, "pthread_attr_setaffinity_np()"); + + rc = pthread_create(&t1, &attr /* Bind 'pong' to CPU 0 */, pong, NULL); + if (rc) + pr_err(rc, "pthread_create()"); + + /* Name it for systemtap convenience */ + rc = pthread_setname_np(t1, "pong"); + if (rc) + pr_warn(rc, "pthread_create()"); flags.result = 0; |