Merge branch 'x86-fpu-for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/tip/tip

Pull x86 fpu changes from Ingo Molnar:
 "There are two main areas of changes:

   - Rework of the extended FPU state code to robustify the kernel's
     usage of cpuid provided xstate sizes - and related changes (Dave
     Hansen)"

   - math emulation enhancements: new modern FPU instructions support,
     with testcases, plus cleanups (Denys Vlasnko)"

* 'x86-fpu-for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/tip/tip: (23 commits)
  x86/fpu: Fixup uninitialized feature_name warning
  x86/fpu/math-emu: Add support for FISTTP instructions
  x86/fpu/math-emu, selftests: Add test for FISTTP instructions
  x86/fpu/math-emu: Add support for FCMOVcc insns
  x86/fpu/math-emu: Add support for F[U]COMI[P] insns
  x86/fpu/math-emu: Remove define layer for undocumented opcodes
  x86/fpu/math-emu, selftests: Add tests for FCMOV and FCOMI insns
  x86/fpu/math-emu: Remove !NO_UNDOC_CODE
  x86/fpu: Check CPU-provided sizes against struct declarations
  x86/fpu: Check to ensure increasing-offset xstate offsets
  x86/fpu: Correct and check XSAVE xstate size calculations
  x86/fpu: Add C structures for AVX-512 state components
  x86/fpu: Rework YMM definition
  x86/fpu/mpx: Rework MPX 'xstate' types
  x86/fpu: Add xfeature_enabled() helper instead of test_bit()
  x86/fpu: Remove 'xfeature_nr'
  x86/fpu: Rework XSTATE_* macros to remove magic '2'
  x86/fpu: Rename XFEATURES_NR_MAX
  x86/fpu: Rename XSAVE macros
  x86/fpu: Remove partial LWP support definitions
  ...

4 files changed, 564 insertions, 2 deletions

diff --git a/tools/testing/selftests/x86/Makefile b/tools/testing/selftests/x86/Makefile
index 389701f59940..eabcff411984 100644
--- a/tools/testing/selftests/x86/Makefile
+++ b/tools/testing/selftests/x86/Makefile
@@ -5,7 +5,8 @@ include ../lib.mk
 .PHONY: all all_32 all_64 warn_32bit_failure clean
 
 TARGETS_C_BOTHBITS := single_step_syscall sysret_ss_attrs ldt_gdt syscall_nt ptrace_syscall
-TARGETS_C_32BIT_ONLY := entry_from_vm86 syscall_arg_fault sigreturn test_syscall_vdso unwind_vdso
+TARGETS_C_32BIT_ONLY := entry_from_vm86 syscall_arg_fault sigreturn test_syscall_vdso unwind_vdso \
+			test_FCMOV test_FCOMI test_FISTTP
 
 TARGETS_C_32BIT_ALL := $(TARGETS_C_BOTHBITS) $(TARGETS_C_32BIT_ONLY)
 BINARIES_32 := $(TARGETS_C_32BIT_ALL:%=%_32)
@@ -35,7 +36,7 @@ clean:
 	$(RM) $(BINARIES_32) $(BINARIES_64)
 
 $(TARGETS_C_32BIT_ALL:%=%_32): %_32: %.c
-	$(CC) -m32 -o $@ $(CFLAGS) $(EXTRA_CFLAGS) $^ -lrt -ldl
+	$(CC) -m32 -o $@ $(CFLAGS) $(EXTRA_CFLAGS) $^ -lrt -ldl -lm
 
 $(TARGETS_C_BOTHBITS:%=%_64): %_64: %.c
 	$(CC) -m64 -o $@ $(CFLAGS) $(EXTRA_CFLAGS) $^ -lrt -ldl
diff --git a/tools/testing/selftests/x86/test_FCMOV.c b/tools/testing/selftests/x86/test_FCMOV.c
new file mode 100644
index 000000000000..4adcca0c80c4
--- /dev/null
+++ b/tools/testing/selftests/x86/test_FCMOV.c
@@ -0,0 +1,93 @@
+#undef _GNU_SOURCE
+#define _GNU_SOURCE 1
+#undef __USE_GNU
+#define __USE_GNU 1
+#include <unistd.h>
+#include <stdlib.h>
+#include <string.h>
+#include <stdio.h>
+#include <signal.h>
+#include <sys/types.h>
+#include <sys/select.h>
+#include <sys/time.h>
+#include <sys/wait.h>
+
+#define TEST(insn) \
+long double __attribute__((noinline)) insn(long flags) \
+{						\
+	long double out;			\
+	asm ("\n"				\
+	"	push	%1""\n"			\
+	"	popf""\n"			\
+	"	fldpi""\n"			\
+	"	fld1""\n"			\
+	"	" #insn " %%st(1), %%st" "\n"	\
+	"	ffree	%%st(1)" "\n"		\
+	: "=t" (out)				\
+	: "r" (flags)				\
+	);					\
+	return out;				\
+}
+
+TEST(fcmovb)
+TEST(fcmove)
+TEST(fcmovbe)
+TEST(fcmovu)
+TEST(fcmovnb)
+TEST(fcmovne)
+TEST(fcmovnbe)
+TEST(fcmovnu)
+
+enum {
+	CF = 1 << 0,
+	PF = 1 << 2,
+	ZF = 1 << 6,
+};
+
+void sighandler(int sig)
+{
+	printf("[FAIL]\tGot signal %d, exiting\n", sig);
+	exit(1);
+}
+
+int main(int argc, char **argv, char **envp)
+{
+	int err = 0;
+
+	/* SIGILL triggers on 32-bit kernels w/o fcomi emulation
+	 * when run with "no387 nofxsr". Other signals are caught
+	 * just in case.
+	 */
+	signal(SIGILL, sighandler);
+	signal(SIGFPE, sighandler);
+	signal(SIGSEGV, sighandler);
+
+	printf("[RUN]\tTesting fcmovCC instructions\n");
+	/* If fcmovCC() returns 1.0, the move wasn't done */
+	err |= !(fcmovb(0)   == 1.0); err |= !(fcmovnb(0)  != 1.0);
+	err |= !(fcmove(0)   == 1.0); err |= !(fcmovne(0)  != 1.0);
+	err |= !(fcmovbe(0)  == 1.0); err |= !(fcmovnbe(0) != 1.0);
+	err |= !(fcmovu(0)   == 1.0); err |= !(fcmovnu(0)  != 1.0);
+
+	err |= !(fcmovb(CF)  != 1.0); err |= !(fcmovnb(CF)  == 1.0);
+	err |= !(fcmove(CF)  == 1.0); err |= !(fcmovne(CF)  != 1.0);
+	err |= !(fcmovbe(CF) != 1.0); err |= !(fcmovnbe(CF) == 1.0);
+	err |= !(fcmovu(CF)  == 1.0); err |= !(fcmovnu(CF)  != 1.0);
+
+	err |= !(fcmovb(ZF)  == 1.0); err |= !(fcmovnb(ZF)  != 1.0);
+	err |= !(fcmove(ZF)  != 1.0); err |= !(fcmovne(ZF)  == 1.0);
+	err |= !(fcmovbe(ZF) != 1.0); err |= !(fcmovnbe(ZF) == 1.0);
+	err |= !(fcmovu(ZF)  == 1.0); err |= !(fcmovnu(ZF)  != 1.0);
+
+	err |= !(fcmovb(PF)  == 1.0); err |= !(fcmovnb(PF)  != 1.0);
+	err |= !(fcmove(PF)  == 1.0); err |= !(fcmovne(PF)  != 1.0);
+	err |= !(fcmovbe(PF) == 1.0); err |= !(fcmovnbe(PF) != 1.0);
+	err |= !(fcmovu(PF)  != 1.0); err |= !(fcmovnu(PF)  == 1.0);
+
+        if (!err)
+                printf("[OK]\tfcmovCC\n");
+	else
+		printf("[FAIL]\tfcmovCC errors: %d\n", err);
+
+	return err;
+}
diff --git a/tools/testing/selftests/x86/test_FCOMI.c b/tools/testing/selftests/x86/test_FCOMI.c
new file mode 100644
index 000000000000..db4933e31af9
--- /dev/null
+++ b/tools/testing/selftests/x86/test_FCOMI.c
@@ -0,0 +1,331 @@
+#undef _GNU_SOURCE
+#define _GNU_SOURCE 1
+#undef __USE_GNU
+#define __USE_GNU 1
+#include <unistd.h>
+#include <stdlib.h>
+#include <string.h>
+#include <stdio.h>
+#include <signal.h>
+#include <sys/types.h>
+#include <sys/select.h>
+#include <sys/time.h>
+#include <sys/wait.h>
+#include <fenv.h>
+
+enum {
+	CF = 1 << 0,
+	PF = 1 << 2,
+	ZF = 1 << 6,
+	ARITH = CF | PF | ZF,
+};
+
+long res_fcomi_pi_1;
+long res_fcomi_1_pi;
+long res_fcomi_1_1;
+long res_fcomi_nan_1;
+/* sNaN is s|111 1111 1|1xx xxxx xxxx xxxx xxxx xxxx */
+/* qNaN is s|111 1111 1|0xx xxxx xxxx xxxx xxxx xxxx (some x must be nonzero) */
+int snan = 0x7fc11111;
+int qnan = 0x7f811111;
+unsigned short snan1[5];
+/* sNaN80 is s|111 1111 1111 1111 |10xx xx...xx (some x must be nonzero) */
+unsigned short snan80[5] = { 0x1111, 0x1111, 0x1111, 0x8111, 0x7fff };
+
+int test(long flags)
+{
+	feclearexcept(FE_DIVBYZERO|FE_INEXACT|FE_INVALID|FE_OVERFLOW|FE_UNDERFLOW);
+
+	asm ("\n"
+
+	"	push	%0""\n"
+	"	popf""\n"
+	"	fld1""\n"
+	"	fldpi""\n"
+	"	fcomi	%%st(1), %%st" "\n"
+	"	ffree	%%st(0)" "\n"
+	"	ffree	%%st(1)" "\n"
+	"	pushf""\n"
+	"	pop	res_fcomi_1_pi""\n"
+
+	"	push	%0""\n"
+	"	popf""\n"
+	"	fldpi""\n"
+	"	fld1""\n"
+	"	fcomi	%%st(1), %%st" "\n"
+	"	ffree	%%st(0)" "\n"
+	"	ffree	%%st(1)" "\n"
+	"	pushf""\n"
+	"	pop	res_fcomi_pi_1""\n"
+
+	"	push	%0""\n"
+	"	popf""\n"
+	"	fld1""\n"
+	"	fld1""\n"
+	"	fcomi	%%st(1), %%st" "\n"
+	"	ffree	%%st(0)" "\n"
+	"	ffree	%%st(1)" "\n"
+	"	pushf""\n"
+	"	pop	res_fcomi_1_1""\n"
+	:
+	: "r" (flags)
+	);
+	if ((res_fcomi_1_pi & ARITH) != (0)) {
+		printf("[BAD]\tfcomi_1_pi with flags:%lx\n", flags);
+		return 1;
+	}
+	if ((res_fcomi_pi_1 & ARITH) != (CF)) {
+		printf("[BAD]\tfcomi_pi_1 with flags:%lx->%lx\n", flags, res_fcomi_pi_1 & ARITH);
+		return 1;
+	}
+	if ((res_fcomi_1_1 & ARITH) != (ZF)) {
+		printf("[BAD]\tfcomi_1_1 with flags:%lx\n", flags);
+		return 1;
+	}
+	if (fetestexcept(FE_INVALID) != 0) {
+		printf("[BAD]\tFE_INVALID is set in %s\n", __func__);
+		return 1;
+	}
+	return 0;
+}
+
+int test_qnan(long flags)
+{
+	feclearexcept(FE_DIVBYZERO|FE_INEXACT|FE_INVALID|FE_OVERFLOW|FE_UNDERFLOW);
+
+	asm ("\n"
+	"	push	%0""\n"
+	"	popf""\n"
+	"	flds	qnan""\n"
+	"	fld1""\n"
+	"	fnclex""\n"		// fld of a qnan raised FE_INVALID, clear it
+	"	fcomi	%%st(1), %%st" "\n"
+	"	ffree	%%st(0)" "\n"
+	"	ffree	%%st(1)" "\n"
+	"	pushf""\n"
+	"	pop	res_fcomi_nan_1""\n"
+	:
+	: "r" (flags)
+	);
+	if ((res_fcomi_nan_1 & ARITH) != (ZF|CF|PF)) {
+		printf("[BAD]\tfcomi_qnan_1 with flags:%lx\n", flags);
+		return 1;
+	}
+	if (fetestexcept(FE_INVALID) != FE_INVALID) {
+		printf("[BAD]\tFE_INVALID is not set in %s\n", __func__);
+		return 1;
+	}
+	return 0;
+}
+
+int testu_qnan(long flags)
+{
+	feclearexcept(FE_DIVBYZERO|FE_INEXACT|FE_INVALID|FE_OVERFLOW|FE_UNDERFLOW);
+
+	asm ("\n"
+	"	push	%0""\n"
+	"	popf""\n"
+	"	flds	qnan""\n"
+	"	fld1""\n"
+	"	fnclex""\n"		// fld of a qnan raised FE_INVALID, clear it
+	"	fucomi	%%st(1), %%st" "\n"
+	"	ffree	%%st(0)" "\n"
+	"	ffree	%%st(1)" "\n"
+	"	pushf""\n"
+	"	pop	res_fcomi_nan_1""\n"
+	:
+	: "r" (flags)
+	);
+	if ((res_fcomi_nan_1 & ARITH) != (ZF|CF|PF)) {
+		printf("[BAD]\tfcomi_qnan_1 with flags:%lx\n", flags);
+		return 1;
+	}
+	if (fetestexcept(FE_INVALID) != 0) {
+		printf("[BAD]\tFE_INVALID is set in %s\n", __func__);
+		return 1;
+	}
+	return 0;
+}
+
+int testu_snan(long flags)
+{
+	feclearexcept(FE_DIVBYZERO|FE_INEXACT|FE_INVALID|FE_OVERFLOW|FE_UNDERFLOW);
+
+	asm ("\n"
+	"	push	%0""\n"
+	"	popf""\n"
+//	"	flds	snan""\n"	// WRONG, this will convert 32-bit fp snan to a *qnan* in 80-bit fp register!
+//	"	fstpt	snan1""\n"	// if uncommented, it prints "snan1:7fff c111 1100 0000 0000" - c111, not 8111!
+//	"	fnclex""\n"		// flds of a snan raised FE_INVALID, clear it
+	"	fldt	snan80""\n"	// fldt never raise FE_INVALID
+	"	fld1""\n"
+	"	fucomi	%%st(1), %%st" "\n"
+	"	ffree	%%st(0)" "\n"
+	"	ffree	%%st(1)" "\n"
+	"	pushf""\n"
+	"	pop	res_fcomi_nan_1""\n"
+	:
+	: "r" (flags)
+	);
+	if ((res_fcomi_nan_1 & ARITH) != (ZF|CF|PF)) {
+		printf("[BAD]\tfcomi_qnan_1 with flags:%lx\n", flags);
+		return 1;
+	}
+//	printf("snan:%x snan1:%04x %04x %04x %04x %04x\n", snan, snan1[4], snan1[3], snan1[2], snan1[1], snan1[0]);
+	if (fetestexcept(FE_INVALID) != FE_INVALID) {
+		printf("[BAD]\tFE_INVALID is not set in %s\n", __func__);
+		return 1;
+	}
+	return 0;
+}
+
+int testp(long flags)
+{
+	feclearexcept(FE_DIVBYZERO|FE_INEXACT|FE_INVALID|FE_OVERFLOW|FE_UNDERFLOW);
+
+	asm ("\n"
+
+	"	push	%0""\n"
+	"	popf""\n"
+	"	fld1""\n"
+	"	fldpi""\n"
+	"	fcomip	%%st(1), %%st" "\n"
+	"	ffree	%%st(0)" "\n"
+	"	pushf""\n"
+	"	pop	res_fcomi_1_pi""\n"
+
+	"	push	%0""\n"
+	"	popf""\n"
+	"	fldpi""\n"
+	"	fld1""\n"
+	"	fcomip	%%st(1), %%st" "\n"
+	"	ffree	%%st(0)" "\n"
+	"	pushf""\n"
+	"	pop	res_fcomi_pi_1""\n"
+
+	"	push	%0""\n"
+	"	popf""\n"
+	"	fld1""\n"
+	"	fld1""\n"
+	"	fcomip	%%st(1), %%st" "\n"
+	"	ffree	%%st(0)" "\n"
+	"	pushf""\n"
+	"	pop	res_fcomi_1_1""\n"
+	:
+	: "r" (flags)
+	);
+	if ((res_fcomi_1_pi & ARITH) != (0)) {
+		printf("[BAD]\tfcomi_1_pi with flags:%lx\n", flags);
+		return 1;
+	}
+	if ((res_fcomi_pi_1 & ARITH) != (CF)) {
+		printf("[BAD]\tfcomi_pi_1 with flags:%lx->%lx\n", flags, res_fcomi_pi_1 & ARITH);
+		return 1;
+	}
+	if ((res_fcomi_1_1 & ARITH) != (ZF)) {
+		printf("[BAD]\tfcomi_1_1 with flags:%lx\n", flags);
+		return 1;
+	}
+	if (fetestexcept(FE_INVALID) != 0) {
+		printf("[BAD]\tFE_INVALID is set in %s\n", __func__);
+		return 1;
+	}
+	return 0;
+}
+
+int testp_qnan(long flags)
+{
+	feclearexcept(FE_DIVBYZERO|FE_INEXACT|FE_INVALID|FE_OVERFLOW|FE_UNDERFLOW);
+
+	asm ("\n"
+	"	push	%0""\n"
+	"	popf""\n"
+	"	flds	qnan""\n"
+	"	fld1""\n"
+	"	fnclex""\n"		// fld of a qnan raised FE_INVALID, clear it
+	"	fcomip	%%st(1), %%st" "\n"
+	"	ffree	%%st(0)" "\n"
+	"	pushf""\n"
+	"	pop	res_fcomi_nan_1""\n"
+	:
+	: "r" (flags)
+	);
+	if ((res_fcomi_nan_1 & ARITH) != (ZF|CF|PF)) {
+		printf("[BAD]\tfcomi_qnan_1 with flags:%lx\n", flags);
+		return 1;
+	}
+	if (fetestexcept(FE_INVALID) != FE_INVALID) {
+		printf("[BAD]\tFE_INVALID is not set in %s\n", __func__);
+		return 1;
+	}
+	return 0;
+}
+
+int testup_qnan(long flags)
+{
+	feclearexcept(FE_DIVBYZERO|FE_INEXACT|FE_INVALID|FE_OVERFLOW|FE_UNDERFLOW);
+
+	asm ("\n"
+	"	push	%0""\n"
+	"	popf""\n"
+	"	flds	qnan""\n"
+	"	fld1""\n"
+	"	fnclex""\n"		// fld of a qnan raised FE_INVALID, clear it
+	"	fucomip	%%st(1), %%st" "\n"
+	"	ffree	%%st(0)" "\n"
+	"	pushf""\n"
+	"	pop	res_fcomi_nan_1""\n"
+	:
+	: "r" (flags)
+	);
+	if ((res_fcomi_nan_1 & ARITH) != (ZF|CF|PF)) {
+		printf("[BAD]\tfcomi_qnan_1 with flags:%lx\n", flags);
+		return 1;
+	}
+	if (fetestexcept(FE_INVALID) != 0) {
+		printf("[BAD]\tFE_INVALID is set in %s\n", __func__);
+		return 1;
+	}
+	return 0;
+}
+
+void sighandler(int sig)
+{
+	printf("[FAIL]\tGot signal %d, exiting\n", sig);
+	exit(1);
+}
+
+int main(int argc, char **argv, char **envp)
+{
+	int err = 0;
+
+	/* SIGILL triggers on 32-bit kernels w/o fcomi emulation
+	 * when run with "no387 nofxsr". Other signals are caught
+	 * just in case.
+	 */
+	signal(SIGILL, sighandler);
+	signal(SIGFPE, sighandler);
+	signal(SIGSEGV, sighandler);
+
+	printf("[RUN]\tTesting f[u]comi[p] instructions\n");
+	err |= test(0);
+	err |= test_qnan(0);
+	err |= testu_qnan(0);
+	err |= testu_snan(0);
+	err |= test(CF|ZF|PF);
+	err |= test_qnan(CF|ZF|PF);
+	err |= testu_qnan(CF|ZF|PF);
+	err |= testu_snan(CF|ZF|PF);
+	err |= testp(0);
+	err |= testp_qnan(0);
+	err |= testup_qnan(0);
+	err |= testp(CF|ZF|PF);
+	err |= testp_qnan(CF|ZF|PF);
+	err |= testup_qnan(CF|ZF|PF);
+	if (!err)
+		printf("[OK]\tf[u]comi[p]\n");
+	else
+		printf("[FAIL]\tf[u]comi[p] errors: %d\n", err);
+
+	return err;
+}
diff --git a/tools/testing/selftests/x86/test_FISTTP.c b/tools/testing/selftests/x86/test_FISTTP.c
new file mode 100644
index 000000000000..b8e61a047f6b
--- /dev/null
+++ b/tools/testing/selftests/x86/test_FISTTP.c
@@ -0,0 +1,137 @@
+#undef _GNU_SOURCE
+#define _GNU_SOURCE 1
+#undef __USE_GNU
+#define __USE_GNU 1
+#include <unistd.h>
+#include <stdlib.h>
+#include <string.h>
+#include <stdio.h>
+#include <signal.h>
+#include <sys/types.h>
+#include <sys/select.h>
+#include <sys/time.h>
+#include <sys/wait.h>
+#include <fenv.h>
+
+unsigned long long res64 = -1;
+unsigned int res32 = -1;
+unsigned short res16 = -1;
+
+int test(void)
+{
+	int ex;
+
+	feclearexcept(FE_DIVBYZERO|FE_INEXACT|FE_INVALID|FE_OVERFLOW|FE_UNDERFLOW);
+	asm volatile ("\n"
+	"	fld1""\n"
+	"	fisttp	res16""\n"
+	"	fld1""\n"
+	"	fisttpl	res32""\n"
+	"	fld1""\n"
+	"	fisttpll res64""\n"
+	: : : "memory"
+	);
+	if (res16 != 1 || res32 != 1 || res64 != 1) {
+		printf("[BAD]\tfisttp 1\n");
+		return 1;
+	}
+	ex = fetestexcept(FE_DIVBYZERO|FE_INEXACT|FE_INVALID|FE_OVERFLOW|FE_UNDERFLOW);
+	if (ex != 0) {
+		printf("[BAD]\tfisttp 1: wrong exception state\n");
+		return 1;
+	}
+
+	feclearexcept(FE_DIVBYZERO|FE_INEXACT|FE_INVALID|FE_OVERFLOW|FE_UNDERFLOW);
+	asm volatile ("\n"
+	"	fldpi""\n"
+	"	fisttp	res16""\n"
+	"	fldpi""\n"
+	"	fisttpl	res32""\n"
+	"	fldpi""\n"
+	"	fisttpll res64""\n"
+	: : : "memory"
+	);
+	if (res16 != 3 || res32 != 3 || res64 != 3) {
+		printf("[BAD]\tfisttp pi\n");
+		return 1;
+	}
+	ex = fetestexcept(FE_DIVBYZERO|FE_INEXACT|FE_INVALID|FE_OVERFLOW|FE_UNDERFLOW);
+	if (ex != FE_INEXACT) {
+		printf("[BAD]\tfisttp pi: wrong exception state\n");
+		return 1;
+	}
+
+	feclearexcept(FE_DIVBYZERO|FE_INEXACT|FE_INVALID|FE_OVERFLOW|FE_UNDERFLOW);
+	asm volatile ("\n"
+	"	fldpi""\n"
+	"	fchs""\n"
+	"	fisttp	res16""\n"
+	"	fldpi""\n"
+	"	fchs""\n"
+	"	fisttpl	res32""\n"
+	"	fldpi""\n"
+	"	fchs""\n"
+	"	fisttpll res64""\n"
+	: : : "memory"
+	);
+	if (res16 != 0xfffd || res32 != 0xfffffffd || res64 != 0xfffffffffffffffdULL) {
+		printf("[BAD]\tfisttp -pi\n");
+		return 1;
+	}
+	ex = fetestexcept(FE_DIVBYZERO|FE_INEXACT|FE_INVALID|FE_OVERFLOW|FE_UNDERFLOW);
+	if (ex != FE_INEXACT) {
+		printf("[BAD]\tfisttp -pi: wrong exception state\n");
+		return 1;
+	}
+
+	feclearexcept(FE_DIVBYZERO|FE_INEXACT|FE_INVALID|FE_OVERFLOW|FE_UNDERFLOW);
+	asm volatile ("\n"
+	"	fldln2""\n"
+	"	fisttp	res16""\n"
+	"	fldln2""\n"
+	"	fisttpl	res32""\n"
+	"	fldln2""\n"
+	"	fisttpll res64""\n"
+	: : : "memory"
+	);
+	/* Test truncation to zero (round-to-nearest would give 1 here) */
+	if (res16 != 0 || res32 != 0 || res64 != 0) {
+		printf("[BAD]\tfisttp ln2\n");
+		return 1;
+	}
+	ex = fetestexcept(FE_DIVBYZERO|FE_INEXACT|FE_INVALID|FE_OVERFLOW|FE_UNDERFLOW);
+	if (ex != FE_INEXACT) {
+		printf("[BAD]\tfisttp ln2: wrong exception state\n");
+		return 1;
+	}
+
+	return 0;
+}
+
+void sighandler(int sig)
+{
+	printf("[FAIL]\tGot signal %d, exiting\n", sig);
+	exit(1);
+}
+
+int main(int argc, char **argv, char **envp)
+{
+	int err = 0;
+
+	/* SIGILL triggers on 32-bit kernels w/o fisttp emulation
+	 * when run with "no387 nofxsr". Other signals are caught
+	 * just in case.
+	 */
+	signal(SIGILL, sighandler);
+	signal(SIGFPE, sighandler);
+	signal(SIGSEGV, sighandler);
+
+	printf("[RUN]\tTesting fisttp instructions\n");
+	err |= test();
+	if (!err)
+		printf("[OK]\tfisttp\n");
+	else
+		printf("[FAIL]\tfisttp errors: %d\n", err);
+
+	return err;
+}