summaryrefslogtreecommitdiffstats
path: root/kernel/tools/testing/selftests/x86
diff options
context:
space:
mode:
authorYunhong Jiang <yunhong.jiang@intel.com>2015-08-04 12:17:53 -0700
committerYunhong Jiang <yunhong.jiang@intel.com>2015-08-04 15:44:42 -0700
commit9ca8dbcc65cfc63d6f5ef3312a33184e1d726e00 (patch)
tree1c9cafbcd35f783a87880a10f85d1a060db1a563 /kernel/tools/testing/selftests/x86
parent98260f3884f4a202f9ca5eabed40b1354c489b29 (diff)
Add the rt linux 4.1.3-rt3 as base
Import the rt linux 4.1.3-rt3 as OPNFV kvm base. It's from git://git.kernel.org/pub/scm/linux/kernel/git/rt/linux-rt-devel.git linux-4.1.y-rt and the base is: commit 0917f823c59692d751951bf5ea699a2d1e2f26a2 Author: Sebastian Andrzej Siewior <bigeasy@linutronix.de> Date: Sat Jul 25 12:13:34 2015 +0200 Prepare v4.1.3-rt3 Signed-off-by: Sebastian Andrzej Siewior <bigeasy@linutronix.de> We lose all the git history this way and it's not good. We should apply another opnfv project repo in future. Change-Id: I87543d81c9df70d99c5001fbdf646b202c19f423 Signed-off-by: Yunhong Jiang <yunhong.jiang@intel.com>
Diffstat (limited to 'kernel/tools/testing/selftests/x86')
-rw-r--r--kernel/tools/testing/selftests/x86/.gitignore2
-rw-r--r--kernel/tools/testing/selftests/x86/Makefile59
-rwxr-xr-xkernel/tools/testing/selftests/x86/check_cc.sh16
-rw-r--r--kernel/tools/testing/selftests/x86/entry_from_vm86.c114
-rw-r--r--kernel/tools/testing/selftests/x86/sigreturn.c684
-rw-r--r--kernel/tools/testing/selftests/x86/single_step_syscall.c181
-rw-r--r--kernel/tools/testing/selftests/x86/trivial_32bit_program.c18
-rw-r--r--kernel/tools/testing/selftests/x86/trivial_64bit_program.c18
8 files changed, 1092 insertions, 0 deletions
diff --git a/kernel/tools/testing/selftests/x86/.gitignore b/kernel/tools/testing/selftests/x86/.gitignore
new file mode 100644
index 000000000..15034fef9
--- /dev/null
+++ b/kernel/tools/testing/selftests/x86/.gitignore
@@ -0,0 +1,2 @@
+*_32
+*_64
diff --git a/kernel/tools/testing/selftests/x86/Makefile b/kernel/tools/testing/selftests/x86/Makefile
new file mode 100644
index 000000000..9b0d8baf2
--- /dev/null
+++ b/kernel/tools/testing/selftests/x86/Makefile
@@ -0,0 +1,59 @@
+all:
+
+include ../lib.mk
+
+.PHONY: all all_32 all_64 warn_32bit_failure clean
+
+TARGETS_C_BOTHBITS := sigreturn single_step_syscall
+TARGETS_C_32BIT_ONLY := entry_from_vm86
+
+TARGETS_C_32BIT_ALL := $(TARGETS_C_BOTHBITS) $(TARGETS_C_32BIT_ONLY)
+BINARIES_32 := $(TARGETS_C_32BIT_ALL:%=%_32)
+BINARIES_64 := $(TARGETS_C_BOTHBITS:%=%_64)
+
+CFLAGS := -O2 -g -std=gnu99 -pthread -Wall
+
+UNAME_M := $(shell uname -m)
+CAN_BUILD_I386 := $(shell ./check_cc.sh $(CC) trivial_32bit_program.c -m32)
+CAN_BUILD_X86_64 := $(shell ./check_cc.sh $(CC) trivial_64bit_program.c)
+
+ifeq ($(CAN_BUILD_I386),1)
+all: all_32
+TEST_PROGS += $(BINARIES_32)
+endif
+
+ifeq ($(CAN_BUILD_X86_64),1)
+all: all_64
+TEST_PROGS += $(BINARIES_64)
+endif
+
+all_32: $(BINARIES_32)
+
+all_64: $(BINARIES_64)
+
+clean:
+ $(RM) $(BINARIES_32) $(BINARIES_64)
+
+$(TARGETS_C_32BIT_ALL:%=%_32): %_32: %.c
+ $(CC) -m32 -o $@ $(CFLAGS) $(EXTRA_CFLAGS) $^ -lrt -ldl
+
+$(TARGETS_C_BOTHBITS:%=%_64): %_64: %.c
+ $(CC) -m64 -o $@ $(CFLAGS) $(EXTRA_CFLAGS) $^ -lrt -ldl
+
+# x86_64 users should be encouraged to install 32-bit libraries
+ifeq ($(CAN_BUILD_I386)$(CAN_BUILD_X86_64),01)
+all: warn_32bit_failure
+
+warn_32bit_failure:
+ @echo "Warning: you seem to have a broken 32-bit build" 2>&1; \
+ echo "environment. This will reduce test coverage of 64-bit" 2>&1; \
+ echo "kernels. If you are using a Debian-like distribution," 2>&1; \
+ echo "try:"; 2>&1; \
+ echo ""; \
+ echo " apt-get install gcc-multilib libc6-i386 libc6-dev-i386"; \
+ echo ""; \
+ echo "If you are using a Fedora-like distribution, try:"; \
+ echo ""; \
+ echo " yum install glibc-devel.*i686"; \
+ exit 0;
+endif
diff --git a/kernel/tools/testing/selftests/x86/check_cc.sh b/kernel/tools/testing/selftests/x86/check_cc.sh
new file mode 100755
index 000000000..172d3293f
--- /dev/null
+++ b/kernel/tools/testing/selftests/x86/check_cc.sh
@@ -0,0 +1,16 @@
+#!/bin/sh
+# check_cc.sh - Helper to test userspace compilation support
+# Copyright (c) 2015 Andrew Lutomirski
+# GPL v2
+
+CC="$1"
+TESTPROG="$2"
+shift 2
+
+if "$CC" -o /dev/null "$TESTPROG" -O0 "$@" 2>/dev/null; then
+ echo 1
+else
+ echo 0
+fi
+
+exit 0
diff --git a/kernel/tools/testing/selftests/x86/entry_from_vm86.c b/kernel/tools/testing/selftests/x86/entry_from_vm86.c
new file mode 100644
index 000000000..5c38a1876
--- /dev/null
+++ b/kernel/tools/testing/selftests/x86/entry_from_vm86.c
@@ -0,0 +1,114 @@
+/*
+ * entry_from_vm86.c - tests kernel entries from vm86 mode
+ * Copyright (c) 2014-2015 Andrew Lutomirski
+ *
+ * This exercises a few paths that need to special-case vm86 mode.
+ *
+ * GPL v2.
+ */
+
+#define _GNU_SOURCE
+
+#include <assert.h>
+#include <stdlib.h>
+#include <sys/syscall.h>
+#include <sys/signal.h>
+#include <sys/ucontext.h>
+#include <unistd.h>
+#include <stdio.h>
+#include <string.h>
+#include <inttypes.h>
+#include <sys/mman.h>
+#include <err.h>
+#include <stddef.h>
+#include <stdbool.h>
+#include <errno.h>
+#include <sys/vm86.h>
+
+static unsigned long load_addr = 0x10000;
+static int nerrs = 0;
+
+asm (
+ ".pushsection .rodata\n\t"
+ ".type vmcode_bound, @object\n\t"
+ "vmcode:\n\t"
+ "vmcode_bound:\n\t"
+ ".code16\n\t"
+ "bound %ax, (2048)\n\t"
+ "int3\n\t"
+ "vmcode_sysenter:\n\t"
+ "sysenter\n\t"
+ ".size vmcode, . - vmcode\n\t"
+ "end_vmcode:\n\t"
+ ".code32\n\t"
+ ".popsection"
+ );
+
+extern unsigned char vmcode[], end_vmcode[];
+extern unsigned char vmcode_bound[], vmcode_sysenter[];
+
+static void do_test(struct vm86plus_struct *v86, unsigned long eip,
+ const char *text)
+{
+ long ret;
+
+ printf("[RUN]\t%s from vm86 mode\n", text);
+ v86->regs.eip = eip;
+ ret = vm86(VM86_ENTER, v86);
+
+ if (ret == -1 && errno == ENOSYS) {
+ printf("[SKIP]\tvm86 not supported\n");
+ return;
+ }
+
+ if (VM86_TYPE(ret) == VM86_INTx) {
+ char trapname[32];
+ int trapno = VM86_ARG(ret);
+ if (trapno == 13)
+ strcpy(trapname, "GP");
+ else if (trapno == 5)
+ strcpy(trapname, "BR");
+ else if (trapno == 14)
+ strcpy(trapname, "PF");
+ else
+ sprintf(trapname, "%d", trapno);
+
+ printf("[OK]\tExited vm86 mode due to #%s\n", trapname);
+ } else if (VM86_TYPE(ret) == VM86_UNKNOWN) {
+ printf("[OK]\tExited vm86 mode due to unhandled GP fault\n");
+ } else {
+ printf("[OK]\tExited vm86 mode due to type %ld, arg %ld\n",
+ VM86_TYPE(ret), VM86_ARG(ret));
+ }
+}
+
+int main(void)
+{
+ struct vm86plus_struct v86;
+ unsigned char *addr = mmap((void *)load_addr, 4096,
+ PROT_READ | PROT_WRITE | PROT_EXEC,
+ MAP_ANONYMOUS | MAP_PRIVATE, -1,0);
+ if (addr != (unsigned char *)load_addr)
+ err(1, "mmap");
+
+ memcpy(addr, vmcode, end_vmcode - vmcode);
+ addr[2048] = 2;
+ addr[2050] = 3;
+
+ memset(&v86, 0, sizeof(v86));
+
+ v86.regs.cs = load_addr / 16;
+ v86.regs.ss = load_addr / 16;
+ v86.regs.ds = load_addr / 16;
+ v86.regs.es = load_addr / 16;
+
+ assert((v86.regs.cs & 3) == 0); /* Looks like RPL = 0 */
+
+ /* #BR -- should deliver SIG??? */
+ do_test(&v86, vmcode_bound - vmcode, "#BR");
+
+ /* SYSENTER -- should cause #GP or #UD depending on CPU */
+ do_test(&v86, vmcode_sysenter - vmcode, "SYSENTER");
+
+ return (nerrs == 0 ? 0 : 1);
+}
diff --git a/kernel/tools/testing/selftests/x86/sigreturn.c b/kernel/tools/testing/selftests/x86/sigreturn.c
new file mode 100644
index 000000000..b5aa1bab7
--- /dev/null
+++ b/kernel/tools/testing/selftests/x86/sigreturn.c
@@ -0,0 +1,684 @@
+/*
+ * sigreturn.c - tests for x86 sigreturn(2) and exit-to-userspace
+ * Copyright (c) 2014-2015 Andrew Lutomirski
+ *
+ * This program is free software; you can redistribute it and/or modify
+ * it under the terms and conditions of the GNU General Public License,
+ * version 2, as published by the Free Software Foundation.
+ *
+ * This program is distributed in the hope it will be useful, but
+ * WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
+ * General Public License for more details.
+ *
+ * This is a series of tests that exercises the sigreturn(2) syscall and
+ * the IRET / SYSRET paths in the kernel.
+ *
+ * For now, this focuses on the effects of unusual CS and SS values,
+ * and it has a bunch of tests to make sure that ESP/RSP is restored
+ * properly.
+ *
+ * The basic idea behind these tests is to raise(SIGUSR1) to create a
+ * sigcontext frame, plug in the values to be tested, and then return,
+ * which implicitly invokes sigreturn(2) and programs the user context
+ * as desired.
+ *
+ * For tests for which we expect sigreturn and the subsequent return to
+ * user mode to succeed, we return to a short trampoline that generates
+ * SIGTRAP so that the meat of the tests can be ordinary C code in a
+ * SIGTRAP handler.
+ *
+ * The inner workings of each test is documented below.
+ *
+ * Do not run on outdated, unpatched kernels at risk of nasty crashes.
+ */
+
+#define _GNU_SOURCE
+
+#include <sys/time.h>
+#include <time.h>
+#include <stdlib.h>
+#include <sys/syscall.h>
+#include <unistd.h>
+#include <stdio.h>
+#include <string.h>
+#include <inttypes.h>
+#include <sys/mman.h>
+#include <sys/signal.h>
+#include <sys/ucontext.h>
+#include <asm/ldt.h>
+#include <err.h>
+#include <setjmp.h>
+#include <stddef.h>
+#include <stdbool.h>
+#include <sys/ptrace.h>
+#include <sys/user.h>
+
+/*
+ * In principle, this test can run on Linux emulation layers (e.g.
+ * Illumos "LX branded zones"). Solaris-based kernels reserve LDT
+ * entries 0-5 for their own internal purposes, so start our LDT
+ * allocations above that reservation. (The tests don't pass on LX
+ * branded zones, but at least this lets them run.)
+ */
+#define LDT_OFFSET 6
+
+/* An aligned stack accessible through some of our segments. */
+static unsigned char stack16[65536] __attribute__((aligned(4096)));
+
+/*
+ * An aligned int3 instruction used as a trampoline. Some of the tests
+ * want to fish out their ss values, so this trampoline copies ss to eax
+ * before the int3.
+ */
+asm (".pushsection .text\n\t"
+ ".type int3, @function\n\t"
+ ".align 4096\n\t"
+ "int3:\n\t"
+ "mov %ss,%eax\n\t"
+ "int3\n\t"
+ ".size int3, . - int3\n\t"
+ ".align 4096, 0xcc\n\t"
+ ".popsection");
+extern char int3[4096];
+
+/*
+ * At startup, we prepapre:
+ *
+ * - ldt_nonexistent_sel: An LDT entry that doesn't exist (all-zero
+ * descriptor or out of bounds).
+ * - code16_sel: A 16-bit LDT code segment pointing to int3.
+ * - data16_sel: A 16-bit LDT data segment pointing to stack16.
+ * - npcode32_sel: A 32-bit not-present LDT code segment pointing to int3.
+ * - npdata32_sel: A 32-bit not-present LDT data segment pointing to stack16.
+ * - gdt_data16_idx: A 16-bit GDT data segment pointing to stack16.
+ * - gdt_npdata32_idx: A 32-bit not-present GDT data segment pointing to
+ * stack16.
+ *
+ * For no particularly good reason, xyz_sel is a selector value with the
+ * RPL and LDT bits filled in, whereas xyz_idx is just an index into the
+ * descriptor table. These variables will be zero if their respective
+ * segments could not be allocated.
+ */
+static unsigned short ldt_nonexistent_sel;
+static unsigned short code16_sel, data16_sel, npcode32_sel, npdata32_sel;
+
+static unsigned short gdt_data16_idx, gdt_npdata32_idx;
+
+static unsigned short GDT3(int idx)
+{
+ return (idx << 3) | 3;
+}
+
+static unsigned short LDT3(int idx)
+{
+ return (idx << 3) | 7;
+}
+
+/* Our sigaltstack scratch space. */
+static char altstack_data[SIGSTKSZ];
+
+static void sethandler(int sig, void (*handler)(int, siginfo_t *, void *),
+ int flags)
+{
+ struct sigaction sa;
+ memset(&sa, 0, sizeof(sa));
+ sa.sa_sigaction = handler;
+ sa.sa_flags = SA_SIGINFO | flags;
+ sigemptyset(&sa.sa_mask);
+ if (sigaction(sig, &sa, 0))
+ err(1, "sigaction");
+}
+
+static void clearhandler(int sig)
+{
+ struct sigaction sa;
+ memset(&sa, 0, sizeof(sa));
+ sa.sa_handler = SIG_DFL;
+ sigemptyset(&sa.sa_mask);
+ if (sigaction(sig, &sa, 0))
+ err(1, "sigaction");
+}
+
+static void add_ldt(const struct user_desc *desc, unsigned short *var,
+ const char *name)
+{
+ if (syscall(SYS_modify_ldt, 1, desc, sizeof(*desc)) == 0) {
+ *var = LDT3(desc->entry_number);
+ } else {
+ printf("[NOTE]\tFailed to create %s segment\n", name);
+ *var = 0;
+ }
+}
+
+static void setup_ldt(void)
+{
+ if ((unsigned long)stack16 > (1ULL << 32) - sizeof(stack16))
+ errx(1, "stack16 is too high\n");
+ if ((unsigned long)int3 > (1ULL << 32) - sizeof(int3))
+ errx(1, "int3 is too high\n");
+
+ ldt_nonexistent_sel = LDT3(LDT_OFFSET + 2);
+
+ const struct user_desc code16_desc = {
+ .entry_number = LDT_OFFSET + 0,
+ .base_addr = (unsigned long)int3,
+ .limit = 4095,
+ .seg_32bit = 0,
+ .contents = 2, /* Code, not conforming */
+ .read_exec_only = 0,
+ .limit_in_pages = 0,
+ .seg_not_present = 0,
+ .useable = 0
+ };
+ add_ldt(&code16_desc, &code16_sel, "code16");
+
+ const struct user_desc data16_desc = {
+ .entry_number = LDT_OFFSET + 1,
+ .base_addr = (unsigned long)stack16,
+ .limit = 0xffff,
+ .seg_32bit = 0,
+ .contents = 0, /* Data, grow-up */
+ .read_exec_only = 0,
+ .limit_in_pages = 0,
+ .seg_not_present = 0,
+ .useable = 0
+ };
+ add_ldt(&data16_desc, &data16_sel, "data16");
+
+ const struct user_desc npcode32_desc = {
+ .entry_number = LDT_OFFSET + 3,
+ .base_addr = (unsigned long)int3,
+ .limit = 4095,
+ .seg_32bit = 1,
+ .contents = 2, /* Code, not conforming */
+ .read_exec_only = 0,
+ .limit_in_pages = 0,
+ .seg_not_present = 1,
+ .useable = 0
+ };
+ add_ldt(&npcode32_desc, &npcode32_sel, "npcode32");
+
+ const struct user_desc npdata32_desc = {
+ .entry_number = LDT_OFFSET + 4,
+ .base_addr = (unsigned long)stack16,
+ .limit = 0xffff,
+ .seg_32bit = 1,
+ .contents = 0, /* Data, grow-up */
+ .read_exec_only = 0,
+ .limit_in_pages = 0,
+ .seg_not_present = 1,
+ .useable = 0
+ };
+ add_ldt(&npdata32_desc, &npdata32_sel, "npdata32");
+
+ struct user_desc gdt_data16_desc = {
+ .entry_number = -1,
+ .base_addr = (unsigned long)stack16,
+ .limit = 0xffff,
+ .seg_32bit = 0,
+ .contents = 0, /* Data, grow-up */
+ .read_exec_only = 0,
+ .limit_in_pages = 0,
+ .seg_not_present = 0,
+ .useable = 0
+ };
+
+ if (syscall(SYS_set_thread_area, &gdt_data16_desc) == 0) {
+ /*
+ * This probably indicates vulnerability to CVE-2014-8133.
+ * Merely getting here isn't definitive, though, and we'll
+ * diagnose the problem for real later on.
+ */
+ printf("[WARN]\tset_thread_area allocated data16 at index %d\n",
+ gdt_data16_desc.entry_number);
+ gdt_data16_idx = gdt_data16_desc.entry_number;
+ } else {
+ printf("[OK]\tset_thread_area refused 16-bit data\n");
+ }
+
+ struct user_desc gdt_npdata32_desc = {
+ .entry_number = -1,
+ .base_addr = (unsigned long)stack16,
+ .limit = 0xffff,
+ .seg_32bit = 1,
+ .contents = 0, /* Data, grow-up */
+ .read_exec_only = 0,
+ .limit_in_pages = 0,
+ .seg_not_present = 1,
+ .useable = 0
+ };
+
+ if (syscall(SYS_set_thread_area, &gdt_npdata32_desc) == 0) {
+ /*
+ * As a hardening measure, newer kernels don't allow this.
+ */
+ printf("[WARN]\tset_thread_area allocated npdata32 at index %d\n",
+ gdt_npdata32_desc.entry_number);
+ gdt_npdata32_idx = gdt_npdata32_desc.entry_number;
+ } else {
+ printf("[OK]\tset_thread_area refused 16-bit data\n");
+ }
+}
+
+/* State used by our signal handlers. */
+static gregset_t initial_regs, requested_regs, resulting_regs;
+
+/* Instructions for the SIGUSR1 handler. */
+static volatile unsigned short sig_cs, sig_ss;
+static volatile sig_atomic_t sig_trapped, sig_err, sig_trapno;
+
+/* Abstractions for some 32-bit vs 64-bit differences. */
+#ifdef __x86_64__
+# define REG_IP REG_RIP
+# define REG_SP REG_RSP
+# define REG_AX REG_RAX
+
+struct selectors {
+ unsigned short cs, gs, fs, ss;
+};
+
+static unsigned short *ssptr(ucontext_t *ctx)
+{
+ struct selectors *sels = (void *)&ctx->uc_mcontext.gregs[REG_CSGSFS];
+ return &sels->ss;
+}
+
+static unsigned short *csptr(ucontext_t *ctx)
+{
+ struct selectors *sels = (void *)&ctx->uc_mcontext.gregs[REG_CSGSFS];
+ return &sels->cs;
+}
+#else
+# define REG_IP REG_EIP
+# define REG_SP REG_ESP
+# define REG_AX REG_EAX
+
+static greg_t *ssptr(ucontext_t *ctx)
+{
+ return &ctx->uc_mcontext.gregs[REG_SS];
+}
+
+static greg_t *csptr(ucontext_t *ctx)
+{
+ return &ctx->uc_mcontext.gregs[REG_CS];
+}
+#endif
+
+/* Number of errors in the current test case. */
+static volatile sig_atomic_t nerrs;
+
+/*
+ * SIGUSR1 handler. Sets CS and SS as requested and points IP to the
+ * int3 trampoline. Sets SP to a large known value so that we can see
+ * whether the value round-trips back to user mode correctly.
+ */
+static void sigusr1(int sig, siginfo_t *info, void *ctx_void)
+{
+ ucontext_t *ctx = (ucontext_t*)ctx_void;
+
+ memcpy(&initial_regs, &ctx->uc_mcontext.gregs, sizeof(gregset_t));
+
+ *csptr(ctx) = sig_cs;
+ *ssptr(ctx) = sig_ss;
+
+ ctx->uc_mcontext.gregs[REG_IP] =
+ sig_cs == code16_sel ? 0 : (unsigned long)&int3;
+ ctx->uc_mcontext.gregs[REG_SP] = (unsigned long)0x8badf00d5aadc0deULL;
+ ctx->uc_mcontext.gregs[REG_AX] = 0;
+
+ memcpy(&requested_regs, &ctx->uc_mcontext.gregs, sizeof(gregset_t));
+ requested_regs[REG_AX] = *ssptr(ctx); /* The asm code does this. */
+
+ return;
+}
+
+/*
+ * Called after a successful sigreturn. Restores our state so that
+ * the original raise(SIGUSR1) returns.
+ */
+static void sigtrap(int sig, siginfo_t *info, void *ctx_void)
+{
+ ucontext_t *ctx = (ucontext_t*)ctx_void;
+
+ sig_err = ctx->uc_mcontext.gregs[REG_ERR];
+ sig_trapno = ctx->uc_mcontext.gregs[REG_TRAPNO];
+
+ unsigned short ss;
+ asm ("mov %%ss,%0" : "=r" (ss));
+
+ greg_t asm_ss = ctx->uc_mcontext.gregs[REG_AX];
+ if (asm_ss != sig_ss && sig == SIGTRAP) {
+ /* Sanity check failure. */
+ printf("[FAIL]\tSIGTRAP: ss = %hx, frame ss = %hx, ax = %llx\n",
+ ss, *ssptr(ctx), (unsigned long long)asm_ss);
+ nerrs++;
+ }
+
+ memcpy(&resulting_regs, &ctx->uc_mcontext.gregs, sizeof(gregset_t));
+ memcpy(&ctx->uc_mcontext.gregs, &initial_regs, sizeof(gregset_t));
+
+ sig_trapped = sig;
+}
+
+/*
+ * Checks a given selector for its code bitness or returns -1 if it's not
+ * a usable code segment selector.
+ */
+int cs_bitness(unsigned short cs)
+{
+ uint32_t valid = 0, ar;
+ asm ("lar %[cs], %[ar]\n\t"
+ "jnz 1f\n\t"
+ "mov $1, %[valid]\n\t"
+ "1:"
+ : [ar] "=r" (ar), [valid] "+rm" (valid)
+ : [cs] "r" (cs));
+
+ if (!valid)
+ return -1;
+
+ bool db = (ar & (1 << 22));
+ bool l = (ar & (1 << 21));
+
+ if (!(ar & (1<<11)))
+ return -1; /* Not code. */
+
+ if (l && !db)
+ return 64;
+ else if (!l && db)
+ return 32;
+ else if (!l && !db)
+ return 16;
+ else
+ return -1; /* Unknown bitness. */
+}
+
+/* Finds a usable code segment of the requested bitness. */
+int find_cs(int bitness)
+{
+ unsigned short my_cs;
+
+ asm ("mov %%cs,%0" : "=r" (my_cs));
+
+ if (cs_bitness(my_cs) == bitness)
+ return my_cs;
+ if (cs_bitness(my_cs + (2 << 3)) == bitness)
+ return my_cs + (2 << 3);
+ if (my_cs > (2<<3) && cs_bitness(my_cs - (2 << 3)) == bitness)
+ return my_cs - (2 << 3);
+ if (cs_bitness(code16_sel) == bitness)
+ return code16_sel;
+
+ printf("[WARN]\tCould not find %d-bit CS\n", bitness);
+ return -1;
+}
+
+static int test_valid_sigreturn(int cs_bits, bool use_16bit_ss, int force_ss)
+{
+ int cs = find_cs(cs_bits);
+ if (cs == -1) {
+ printf("[SKIP]\tCode segment unavailable for %d-bit CS, %d-bit SS\n",
+ cs_bits, use_16bit_ss ? 16 : 32);
+ return 0;
+ }
+
+ if (force_ss != -1) {
+ sig_ss = force_ss;
+ } else {
+ if (use_16bit_ss) {
+ if (!data16_sel) {
+ printf("[SKIP]\tData segment unavailable for %d-bit CS, 16-bit SS\n",
+ cs_bits);
+ return 0;
+ }
+ sig_ss = data16_sel;
+ } else {
+ asm volatile ("mov %%ss,%0" : "=r" (sig_ss));
+ }
+ }
+
+ sig_cs = cs;
+
+ printf("[RUN]\tValid sigreturn: %d-bit CS (%hx), %d-bit SS (%hx%s)\n",
+ cs_bits, sig_cs, use_16bit_ss ? 16 : 32, sig_ss,
+ (sig_ss & 4) ? "" : ", GDT");
+
+ raise(SIGUSR1);
+
+ nerrs = 0;
+
+ /*
+ * Check that each register had an acceptable value when the
+ * int3 trampoline was invoked.
+ */
+ for (int i = 0; i < NGREG; i++) {
+ greg_t req = requested_regs[i], res = resulting_regs[i];
+ if (i == REG_TRAPNO || i == REG_IP)
+ continue; /* don't care */
+ if (i == REG_SP) {
+ printf("\tSP: %llx -> %llx\n", (unsigned long long)req,
+ (unsigned long long)res);
+
+ /*
+ * In many circumstances, the high 32 bits of rsp
+ * are zeroed. For example, we could be a real
+ * 32-bit program, or we could hit any of a number
+ * of poorly-documented IRET or segmented ESP
+ * oddities. If this happens, it's okay.
+ */
+ if (res == (req & 0xFFFFFFFF))
+ continue; /* OK; not expected to work */
+ }
+
+ bool ignore_reg = false;
+#if __i386__
+ if (i == REG_UESP)
+ ignore_reg = true;
+#else
+ if (i == REG_CSGSFS) {
+ struct selectors *req_sels =
+ (void *)&requested_regs[REG_CSGSFS];
+ struct selectors *res_sels =
+ (void *)&resulting_regs[REG_CSGSFS];
+ if (req_sels->cs != res_sels->cs) {
+ printf("[FAIL]\tCS mismatch: requested 0x%hx; got 0x%hx\n",
+ req_sels->cs, res_sels->cs);
+ nerrs++;
+ }
+
+ if (req_sels->ss != res_sels->ss) {
+ printf("[FAIL]\tSS mismatch: requested 0x%hx; got 0x%hx\n",
+ req_sels->ss, res_sels->ss);
+ nerrs++;
+ }
+
+ continue;
+ }
+#endif
+
+ /* Sanity check on the kernel */
+ if (i == REG_AX && requested_regs[i] != resulting_regs[i]) {
+ printf("[FAIL]\tAX (saved SP) mismatch: requested 0x%llx; got 0x%llx\n",
+ (unsigned long long)requested_regs[i],
+ (unsigned long long)resulting_regs[i]);
+ nerrs++;
+ continue;
+ }
+
+ if (requested_regs[i] != resulting_regs[i] && !ignore_reg) {
+ /*
+ * SP is particularly interesting here. The
+ * usual cause of failures is that we hit the
+ * nasty IRET case of returning to a 16-bit SS,
+ * in which case bits 16:31 of the *kernel*
+ * stack pointer persist in ESP.
+ */
+ printf("[FAIL]\tReg %d mismatch: requested 0x%llx; got 0x%llx\n",
+ i, (unsigned long long)requested_regs[i],
+ (unsigned long long)resulting_regs[i]);
+ nerrs++;
+ }
+ }
+
+ if (nerrs == 0)
+ printf("[OK]\tall registers okay\n");
+
+ return nerrs;
+}
+
+static int test_bad_iret(int cs_bits, unsigned short ss, int force_cs)
+{
+ int cs = force_cs == -1 ? find_cs(cs_bits) : force_cs;
+ if (cs == -1)
+ return 0;
+
+ sig_cs = cs;
+ sig_ss = ss;
+
+ printf("[RUN]\t%d-bit CS (%hx), bogus SS (%hx)\n",
+ cs_bits, sig_cs, sig_ss);
+
+ sig_trapped = 0;
+ raise(SIGUSR1);
+ if (sig_trapped) {
+ char errdesc[32] = "";
+ if (sig_err) {
+ const char *src = (sig_err & 1) ? " EXT" : "";
+ const char *table;
+ if ((sig_err & 0x6) == 0x0)
+ table = "GDT";
+ else if ((sig_err & 0x6) == 0x4)
+ table = "LDT";
+ else if ((sig_err & 0x6) == 0x2)
+ table = "IDT";
+ else
+ table = "???";
+
+ sprintf(errdesc, "%s%s index %d, ",
+ table, src, sig_err >> 3);
+ }
+
+ char trapname[32];
+ if (sig_trapno == 13)
+ strcpy(trapname, "GP");
+ else if (sig_trapno == 11)
+ strcpy(trapname, "NP");
+ else if (sig_trapno == 12)
+ strcpy(trapname, "SS");
+ else if (sig_trapno == 32)
+ strcpy(trapname, "IRET"); /* X86_TRAP_IRET */
+ else
+ sprintf(trapname, "%d", sig_trapno);
+
+ printf("[OK]\tGot #%s(0x%lx) (i.e. %s%s)\n",
+ trapname, (unsigned long)sig_err,
+ errdesc, strsignal(sig_trapped));
+ return 0;
+ } else {
+ printf("[FAIL]\tDid not get SIGSEGV\n");
+ return 1;
+ }
+}
+
+int main()
+{
+ int total_nerrs = 0;
+ unsigned short my_cs, my_ss;
+
+ asm volatile ("mov %%cs,%0" : "=r" (my_cs));
+ asm volatile ("mov %%ss,%0" : "=r" (my_ss));
+ setup_ldt();
+
+ stack_t stack = {
+ .ss_sp = altstack_data,
+ .ss_size = SIGSTKSZ,
+ };
+ if (sigaltstack(&stack, NULL) != 0)
+ err(1, "sigaltstack");
+
+ sethandler(SIGUSR1, sigusr1, 0);
+ sethandler(SIGTRAP, sigtrap, SA_ONSTACK);
+
+ /* Easy cases: return to a 32-bit SS in each possible CS bitness. */
+ total_nerrs += test_valid_sigreturn(64, false, -1);
+ total_nerrs += test_valid_sigreturn(32, false, -1);
+ total_nerrs += test_valid_sigreturn(16, false, -1);
+
+ /*
+ * Test easy espfix cases: return to a 16-bit LDT SS in each possible
+ * CS bitness. NB: with a long mode CS, the SS bitness is irrelevant.
+ *
+ * This catches the original missing-espfix-on-64-bit-kernels issue
+ * as well as CVE-2014-8134.
+ */
+ total_nerrs += test_valid_sigreturn(64, true, -1);
+ total_nerrs += test_valid_sigreturn(32, true, -1);
+ total_nerrs += test_valid_sigreturn(16, true, -1);
+
+ if (gdt_data16_idx) {
+ /*
+ * For performance reasons, Linux skips espfix if SS points
+ * to the GDT. If we were able to allocate a 16-bit SS in
+ * the GDT, see if it leaks parts of the kernel stack pointer.
+ *
+ * This tests for CVE-2014-8133.
+ */
+ total_nerrs += test_valid_sigreturn(64, true,
+ GDT3(gdt_data16_idx));
+ total_nerrs += test_valid_sigreturn(32, true,
+ GDT3(gdt_data16_idx));
+ total_nerrs += test_valid_sigreturn(16, true,
+ GDT3(gdt_data16_idx));
+ }
+
+ /*
+ * We're done testing valid sigreturn cases. Now we test states
+ * for which sigreturn itself will succeed but the subsequent
+ * entry to user mode will fail.
+ *
+ * Depending on the failure mode and the kernel bitness, these
+ * entry failures can generate SIGSEGV, SIGBUS, or SIGILL.
+ */
+ clearhandler(SIGTRAP);
+ sethandler(SIGSEGV, sigtrap, SA_ONSTACK);
+ sethandler(SIGBUS, sigtrap, SA_ONSTACK);
+ sethandler(SIGILL, sigtrap, SA_ONSTACK); /* 32-bit kernels do this */
+
+ /* Easy failures: invalid SS, resulting in #GP(0) */
+ test_bad_iret(64, ldt_nonexistent_sel, -1);
+ test_bad_iret(32, ldt_nonexistent_sel, -1);
+ test_bad_iret(16, ldt_nonexistent_sel, -1);
+
+ /* These fail because SS isn't a data segment, resulting in #GP(SS) */
+ test_bad_iret(64, my_cs, -1);
+ test_bad_iret(32, my_cs, -1);
+ test_bad_iret(16, my_cs, -1);
+
+ /* Try to return to a not-present code segment, triggering #NP(SS). */
+ test_bad_iret(32, my_ss, npcode32_sel);
+
+ /*
+ * Try to return to a not-present but otherwise valid data segment.
+ * This will cause IRET to fail with #SS on the espfix stack. This
+ * exercises CVE-2014-9322.
+ *
+ * Note that, if espfix is enabled, 64-bit Linux will lose track
+ * of the actual cause of failure and report #GP(0) instead.
+ * This would be very difficult for Linux to avoid, because
+ * espfix64 causes IRET failures to be promoted to #DF, so the
+ * original exception frame is never pushed onto the stack.
+ */
+ test_bad_iret(32, npdata32_sel, -1);
+
+ /*
+ * Try to return to a not-present but otherwise valid data
+ * segment without invoking espfix. Newer kernels don't allow
+ * this to happen in the first place. On older kernels, though,
+ * this can trigger CVE-2014-9322.
+ */
+ if (gdt_npdata32_idx)
+ test_bad_iret(32, GDT3(gdt_npdata32_idx), -1);
+
+ return total_nerrs ? 1 : 0;
+}
diff --git a/kernel/tools/testing/selftests/x86/single_step_syscall.c b/kernel/tools/testing/selftests/x86/single_step_syscall.c
new file mode 100644
index 000000000..50c26358e
--- /dev/null
+++ b/kernel/tools/testing/selftests/x86/single_step_syscall.c
@@ -0,0 +1,181 @@
+/*
+ * single_step_syscall.c - single-steps various x86 syscalls
+ * Copyright (c) 2014-2015 Andrew Lutomirski
+ *
+ * This program is free software; you can redistribute it and/or modify
+ * it under the terms and conditions of the GNU General Public License,
+ * version 2, as published by the Free Software Foundation.
+ *
+ * This program is distributed in the hope it will be useful, but
+ * WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
+ * General Public License for more details.
+ *
+ * This is a very simple series of tests that makes system calls with
+ * the TF flag set. This exercises some nasty kernel code in the
+ * SYSENTER case: SYSENTER does not clear TF, so SYSENTER with TF set
+ * immediately issues #DB from CPL 0. This requires special handling in
+ * the kernel.
+ */
+
+#define _GNU_SOURCE
+
+#include <sys/time.h>
+#include <time.h>
+#include <stdlib.h>
+#include <sys/syscall.h>
+#include <unistd.h>
+#include <stdio.h>
+#include <string.h>
+#include <inttypes.h>
+#include <sys/mman.h>
+#include <sys/signal.h>
+#include <sys/ucontext.h>
+#include <asm/ldt.h>
+#include <err.h>
+#include <setjmp.h>
+#include <stddef.h>
+#include <stdbool.h>
+#include <sys/ptrace.h>
+#include <sys/user.h>
+
+static void sethandler(int sig, void (*handler)(int, siginfo_t *, void *),
+ int flags)
+{
+ struct sigaction sa;
+ memset(&sa, 0, sizeof(sa));
+ sa.sa_sigaction = handler;
+ sa.sa_flags = SA_SIGINFO | flags;
+ sigemptyset(&sa.sa_mask);
+ if (sigaction(sig, &sa, 0))
+ err(1, "sigaction");
+}
+
+static volatile sig_atomic_t sig_traps;
+
+#ifdef __x86_64__
+# define REG_IP REG_RIP
+# define WIDTH "q"
+#else
+# define REG_IP REG_EIP
+# define WIDTH "l"
+#endif
+
+static unsigned long get_eflags(void)
+{
+ unsigned long eflags;
+ asm volatile ("pushf" WIDTH "\n\tpop" WIDTH " %0" : "=rm" (eflags));
+ return eflags;
+}
+
+static void set_eflags(unsigned long eflags)
+{
+ asm volatile ("push" WIDTH " %0\n\tpopf" WIDTH
+ : : "rm" (eflags) : "flags");
+}
+
+#define X86_EFLAGS_TF (1UL << 8)
+
+static void sigtrap(int sig, siginfo_t *info, void *ctx_void)
+{
+ ucontext_t *ctx = (ucontext_t*)ctx_void;
+
+ if (get_eflags() & X86_EFLAGS_TF) {
+ set_eflags(get_eflags() & ~X86_EFLAGS_TF);
+ printf("[WARN]\tSIGTRAP handler had TF set\n");
+ _exit(1);
+ }
+
+ sig_traps++;
+
+ if (sig_traps == 10000 || sig_traps == 10001) {
+ printf("[WARN]\tHit %d SIGTRAPs with si_addr 0x%lx, ip 0x%lx\n",
+ (int)sig_traps,
+ (unsigned long)info->si_addr,
+ (unsigned long)ctx->uc_mcontext.gregs[REG_IP]);
+ }
+}
+
+static void check_result(void)
+{
+ unsigned long new_eflags = get_eflags();
+ set_eflags(new_eflags & ~X86_EFLAGS_TF);
+
+ if (!sig_traps) {
+ printf("[FAIL]\tNo SIGTRAP\n");
+ exit(1);
+ }
+
+ if (!(new_eflags & X86_EFLAGS_TF)) {
+ printf("[FAIL]\tTF was cleared\n");
+ exit(1);
+ }
+
+ printf("[OK]\tSurvived with TF set and %d traps\n", (int)sig_traps);
+ sig_traps = 0;
+}
+
+int main()
+{
+ int tmp;
+
+ sethandler(SIGTRAP, sigtrap, 0);
+
+ printf("[RUN]\tSet TF and check nop\n");
+ set_eflags(get_eflags() | X86_EFLAGS_TF);
+ asm volatile ("nop");
+ check_result();
+
+#ifdef __x86_64__
+ printf("[RUN]\tSet TF and check syscall-less opportunistic sysret\n");
+ set_eflags(get_eflags() | X86_EFLAGS_TF);
+ extern unsigned char post_nop[];
+ asm volatile ("pushf" WIDTH "\n\t"
+ "pop" WIDTH " %%r11\n\t"
+ "nop\n\t"
+ "post_nop:"
+ : : "c" (post_nop) : "r11");
+ check_result();
+#endif
+
+ printf("[RUN]\tSet TF and check int80\n");
+ set_eflags(get_eflags() | X86_EFLAGS_TF);
+ asm volatile ("int $0x80" : "=a" (tmp) : "a" (SYS_getpid));
+ check_result();
+
+ /*
+ * This test is particularly interesting if fast syscalls use
+ * SYSENTER: it triggers a nasty design flaw in SYSENTER.
+ * Specifically, SYSENTER does not clear TF, so either SYSENTER
+ * or the next instruction traps at CPL0. (Of course, Intel
+ * mostly forgot to document exactly what happens here.) So we
+ * get a CPL0 fault with usergs (on 64-bit kernels) and possibly
+ * no stack. The only sane way the kernel can possibly handle
+ * it is to clear TF on return from the #DB handler, but this
+ * happens way too early to set TF in the saved pt_regs, so the
+ * kernel has to do something clever to avoid losing track of
+ * the TF bit.
+ *
+ * Needless to say, we've had bugs in this area.
+ */
+ syscall(SYS_getpid); /* Force symbol binding without TF set. */
+ printf("[RUN]\tSet TF and check a fast syscall\n");
+ set_eflags(get_eflags() | X86_EFLAGS_TF);
+ syscall(SYS_getpid);
+ check_result();
+
+ /* Now make sure that another fast syscall doesn't set TF again. */
+ printf("[RUN]\tFast syscall with TF cleared\n");
+ fflush(stdout); /* Force a syscall */
+ if (get_eflags() & X86_EFLAGS_TF) {
+ printf("[FAIL]\tTF is now set\n");
+ exit(1);
+ }
+ if (sig_traps) {
+ printf("[FAIL]\tGot SIGTRAP\n");
+ exit(1);
+ }
+ printf("[OK]\tNothing unexpected happened\n");
+
+ return 0;
+}
diff --git a/kernel/tools/testing/selftests/x86/trivial_32bit_program.c b/kernel/tools/testing/selftests/x86/trivial_32bit_program.c
new file mode 100644
index 000000000..fabdf0f51
--- /dev/null
+++ b/kernel/tools/testing/selftests/x86/trivial_32bit_program.c
@@ -0,0 +1,18 @@
+/*
+ * Trivial program to check that we have a valid 32-bit build environment.
+ * Copyright (c) 2015 Andy Lutomirski
+ * GPL v2
+ */
+
+#ifndef __i386__
+# error wrong architecture
+#endif
+
+#include <stdio.h>
+
+int main()
+{
+ printf("\n");
+
+ return 0;
+}
diff --git a/kernel/tools/testing/selftests/x86/trivial_64bit_program.c b/kernel/tools/testing/selftests/x86/trivial_64bit_program.c
new file mode 100644
index 000000000..b994946c4
--- /dev/null
+++ b/kernel/tools/testing/selftests/x86/trivial_64bit_program.c
@@ -0,0 +1,18 @@
+/*
+ * Trivial program to check that we have a valid 32-bit build environment.
+ * Copyright (c) 2015 Andy Lutomirski
+ * GPL v2
+ */
+
+#ifndef __x86_64__
+# error wrong architecture
+#endif
+
+#include <stdio.h>
+
+int main()
+{
+ printf("\n");
+
+ return 0;
+}