[lldb] [test/Register] Add read/write tests for x87 regs

[mgorny]

Summary

Add a partial read/write tests for x87 FPU registers. This includes
reading and writing ST registers, control registers and floating-point
exception data registers (fop, fip, fdp).

The tests assume the current (roughly incorrect) behavior of reporting
the 'abridged' 8-bit ftag state as 16-bit ftag. They also assume Linux
plugin behavior of reporting fip/fdp split into halves as (fiseg, fioff)
and (foseg, fooff).

[labath]

This is great. Thanks.

lldb/test/Shell/Register/x86-fp-read.test

8

Maybe add a # CHECK: Process {{.*}} stopped line here. That will restrict the following CHECK-DAGs to match after it, which will make FileCheck output more useful. Without it, FileCheck's "Searching from here" snippet will start at the very beginning of the lldb output, which will not be very helpful.

[mgorny]

Done as requested + improved the tests:

• switched to FDIV on memory operand, to be able to test FDP
• added disassembly-based test for FIP and print-based test for FDP (assuming Linux behavior for fiseg/fioff/foseg/fooff, this certainly breaks NetBSD and might break some more platforms)
• figured out how to make write test work while restoring exception state
• added FXSAVE-based check whether writing other registers works (this also includes potential future support for MXCSR)

[mgorny]

I'd use some help improving the assembly, as explained below.

lldb/test/Shell/Register/Inputs/x86-fp-write.cpp

24

Now, here's the problem. This works just fine when built with clang but segfaults when built with gcc (apparently tries to access 0x0).

[mgorny]

lldb/test/Shell/Register/Inputs/x86-fp-write.cpp

24

Turned out to be an alignment issue. Will update the patch shortly.

[labath]

lldb/test/Shell/Register/x86-fp-read.test

28–29

zero is pretty common value. It would be better to check that the address actually points to the address of the zero variable. I think something like this ought to do it:

print &zero
# CHECK: (uint32_t *) $0 = [[ZERO:0x[0-9a-f]*]]
print (uint32_t*)($foseg * 0x100000000 + $fooff)
# CHECK: (uint32_t *) $1 = [[ZERO]]

Something similar can be done with the disassemble command too. That way we'll see the actual register value in case it turns out to be wrong.

[mgorny]

Fixed segfault when built with GCC and fixed my poor attempt to run FXSAVE with REX.W prefix. Now it correctly gets 64-bit FIP and FDP, though it's unclear whether the apparent truncation is a bug in LLDB writing it or some other kind of limitation.

[labath]

In D88583#2305815, @mgorny wrote:

Fixed segfault when built with GCC and fixed my poor attempt to run FXSAVE with REX.W prefix. Now it correctly gets 64-bit FIP and FDP, though it's unclear whether the apparent truncation is a bug in LLDB writing it or some other kind of limitation.

I think that could be an architectural cpu limitation. IIUC, even though the addresses are nominally 64-bit, not all addresses within that space are actually usable -- only the topmost and bottommost XX GB (TB) of the address space are. It could be that the cpu does not actually store all of the bits of these registers, but rather just sign-extends the most significant usable bit.

[mgorny]

lldb/test/Shell/Register/x86-fp-read.test

28–29

Cool idea, thanks!

As for FIP, I suppose that would require getting address corresponding to the FDIV instruction somehow. I suppose we can rely on constant-length encoding on int3 and fdivs ..., so I guess $rip-3 might work.

[labath]

lldb/test/Shell/Register/x86-fp-read.test

28–29

I was thinking you could disassemble the main function and capture the address before the fdivs instruction. But yes, a fixed offset from $rip should work just fine...

[mgorny]

Update fp-read tests to verify FIP/FDP addresses, as suggested by Pavel.

[labath]

cool

[mgorny]

Use safer values for fiseg and foseg, to avoid truncation.

[mgorny]

@teemperor would you perhaps be interested in testing this on macos before I commit it?

[mgorny]

Added missing -g to x86-fp-read.cpp build, as necessary to get address of zero.

[JDevlieghere]

This is failing on GreenDragon so I've XFAILed the tests (temporarily). Can you take a look and see if they can be fixed and re-enabled?

commit 0f08a1a5b162dcd2caf1b76827b917ca69e3e48d (HEAD -> master, origin/master, origin/HEAD)
Author: Jonas Devlieghere <[e-mail]>
Date:   Sat Oct 3 22:36:28 2020 -0700

    [lldb] [test/Register] Mark new FP reg tests XFAIL on Darwin

    This is failing on GreenDragon:
    http://green.lab.llvm.org/green/view/LLDB/job/lldb-cmake/24066/

[mgorny]

In D88583#2310538, @JDevlieghere wrote:

This is failing on GreenDragon so I've XFAILed the tests (temporarily). Can you take a look and see if they can be fixed and re-enabled?

I'm afraid I don't have enough context. Would you be able to rerun them for me with FILECHECK_DUMP_INPUT_ON_FAILURE=1 envvar? Also, I'd suggest adding this to buildbot — it's really helpful if something starts failing.

[JDevlieghere]

In D88583#2310556, @mgorny wrote:

In D88583#2310538, @JDevlieghere wrote:

This is failing on GreenDragon so I've XFAILed the tests (temporarily). Can you take a look and see if they can be fixed and re-enabled?

I'm afraid I don't have enough context. Would you be able to rerun them for me with FILECHECK_DUMP_INPUT_ON_FAILURE=1 envvar? Also, I'd suggest adding this to buildbot — it's really helpful if something starts failing.

I think it recently became the default? Either way the bot has the input:

http://green.lab.llvm.org/green/view/LLDB/job/lldb-cmake/24066/testReport/junit/lldb-shell/Register/x86_64_fp_write_test/
http://green.lab.llvm.org/green/view/LLDB/job/lldb-cmake/24066/testReport/junit/lldb-shell/Register/x86_fp_read_test/

[mgorny]

Well, it's not full input but sure, it's good enough. So I suppose the problem is that the register is named stmm instead of st. I suppose we can use a regex to fix read test, and maybe teach lldb stX alias for write test.

lldb/test/Shell/Register/Inputs/x86-fp-read.cpp

+#include <cstdint>
+
+struct alignas(16) float80_raw {
+  uint64_t mantissa;
+  uint16_t sign_exp;
+};
+
+int main() {
+  float80_raw st[] = {
+    {0x8000000000000000, 0x4000},  // +2.0
+    {0x3f00000000000000, 0x0000},  // 1.654785e-4932 (denormal)
+    {0x0000000000000000, 0x0000},  // +0
+    {0x0000000000000000, 0x8000},  // -0
+    {0x8000000000000000, 0x7fff},  // +inf
+    {0x8000000000000000, 0xffff},  // -inf
+    {0xc000000000000000, 0xffff},  // nan
+    // leave st7 empty to test tag word better
+  };
+
+  // unmask invalid op exception
+  uint16_t cw = 0x037b;
+  // used as single-precision float
+  uint32_t zero = 0;
+
+  asm volatile(
+    "finit\n\t"
+    "fldcw %1\n\t"
+    // load on stack in reverse order to make the result easier to read
+    "fldt 0x60(%0)\n\t"
+    "fldt 0x50(%0)\n\t"
+    "fldt 0x40(%0)\n\t"
+    "fldt 0x30(%0)\n\t"
+    "fldt 0x20(%0)\n\t"
+    "fldt 0x10(%0)\n\t"
+    "fldt 0x00(%0)\n\t"
+    // this should trigger a divide-by-zero
+    "fdivs (%2)\n\t"
+    "int3\n\t"
+    :
+    : "a"(st), "m"(cw), "b"(&zero)
+    : "st"
+  );
+
+  return 0;
+}

lldb/test/Shell/Register/Inputs/x86-fp-write.cpp

+#include <cassert>
+#include <cinttypes>
+#include <cstdint>
+#include <cstdio>
+
+struct alignas(16) float80_raw {
+  uint8_t data[10];
+};
+
+int main() {
+  float80_raw st[8];
+  uint16_t env[14];
+  union alignas(16) {
+    uint16_t i16[256];
+    uint32_t i32[128];
+    uint64_t i64[64];
+  } fxsave;
+
+  asm volatile(
+    "finit\n\t"
+    "int3\n\t"
+#if defined(__x86_64__)
+    "fxsave64 %2\n\t"
+#else
+    "fxsave %2\n\t"
+#endif
+    "fnstenv %1\n\t"
+    "fnclex\n\t"
+    "fstpt 0x00(%0)\n\t"
+    "fstpt 0x10(%0)\n\t"
+    "fstpt 0x20(%0)\n\t"
+    "fstpt 0x30(%0)\n\t"
+    "fstpt 0x40(%0)\n\t"
+    "fstpt 0x50(%0)\n\t"
+    "fstpt 0x60(%0)\n\t"
+    "fstpt 0x70(%0)\n\t"
+    :
+    : "a"(st), "m"(env), "m"(fxsave)
+    : "st"
+  );
+
+  assert(env[0] == fxsave.i16[0]);
+  assert(env[2] == fxsave.i16[1]);
+
+  printf("fctrl = 0x%04" PRIx16 "\n", env[0]);
+  printf("fstat = 0x%04" PRIx16 "\n", env[2]);
+  printf("ftag = 0x%04" PRIx16 "\n", env[4]);
+  printf("fop = 0x%04" PRIx16 "\n", fxsave.i16[3]);
+#if defined(__x86_64__)
+  printf("fip = 0x%016" PRIx64 "\n", fxsave.i64[1]);
+  printf("fdp = 0x%016" PRIx64 "\n", fxsave.i64[2]);
+#else
+  printf("fip = 0x%08" PRIx32 "\n", fxsave.i32[2]);
+  printf("fcs = 0x%04" PRIx16 "\n", fxsave.i16[6]);
+  printf("fdp = 0x%08" PRIx32 "\n", fxsave.i32[4]);
+  printf("fds = 0x%04" PRIx16 "\n", fxsave.i16[10]);
+#endif
+  printf("mxcsr = 0x%08" PRIx32 "\n", fxsave.i32[6]);
+  printf("mxcsr_mask = 0x%08" PRIx32 "\n", fxsave.i32[7]);
+
+  for (int i = 0; i < 8; ++i) {
+    printf("st%d = { ", i);
+    for (int j = 0; j < sizeof(st->data); ++j)
+      printf("0x%02" PRIx8 " ", st[i].data[j]);
+    printf("}\n");
+  }
+
+  return 0;
+}

lldb/test/Shell/Register/x86-64-fp-write.test

+# REQUIRES: native && target-x86_64
+# RUN: %clangxx_host %p/Inputs/x86-fp-write.cpp -o %t
+# RUN: %lldb -b -s %s %t | FileCheck %s
+process launch
+
+register write fctrl 0x037b
+register write fstat 0x8884
+# note: this needs to enable all registers for writes to be effective
+# TODO: fix it to use proper ftag values instead of 'abridged'
+register write ftag 0x00ff
+register write fop 0x0033
+# the exact addresses do not matter, we want just to verify FXSAVE
+# note: fxrstor64 apparently truncates this to 48 bits, and sign extends
+#       the highest bits, so let's keep the value safely below
+register write fiseg 0x00000567
+register write fioff 0x89abcdef
+register write foseg 0x00000a98
+register write fooff 0x76543210
+
+register write st0 "{0x00 0x00 0x00 0x00 0x00 0x00 0x00 0x80 0x00 0x40}"
+register write st1 "{0x00 0x00 0x00 0x00 0x00 0x00 0x00 0x3f 0x00 0x00}"
+register write st2 "{0x00 0x00 0x00 0x00 0x00 0x00 0x00 0x00 0x00 0x00}"
+register write st3 "{0x00 0x00 0x00 0x00 0x00 0x00 0x00 0x00 0x00 0x80}"
+register write st4 "{0x00 0x00 0x00 0x00 0x00 0x00 0x00 0x80 0xff 0x7f}"
+register write st5 "{0x00 0x00 0x00 0x00 0x00 0x00 0x00 0x80 0xff 0xff}"
+register write st6 "{0x00 0x00 0x00 0x00 0x00 0x00 0x00 0xc0 0xff 0xff}"
+register write st7 "{0x00 0x00 0x00 0x00 0x00 0x00 0x00 0x00 0x00 0x00}"
+
+process continue
+
+# CHECK: process continue
+# CHECK-DAG: fctrl = 0x037b
+# CHECK-DAG: fstat = 0x8884
+# CHECK-DAG: ftag = 0xa961
+# CHECK-DAG: fop = 0x0033
+# CHECK-DAG: fip = 0x0000056789abcdef
+# CHECK-DAG: fdp = 0x00000a9876543210
+
+# CHECK-DAG: st0 = { 0x00 0x00 0x00 0x00 0x00 0x00 0x00 0x80 0x00 0x40 }
+# CHECK-DAG: st1 = { 0x00 0x00 0x00 0x00 0x00 0x00 0x00 0x3f 0x00 0x00 }
+# CHECK-DAG: st2 = { 0x00 0x00 0x00 0x00 0x00 0x00 0x00 0x00 0x00 0x00 }
+# CHECK-DAG: st3 = { 0x00 0x00 0x00 0x00 0x00 0x00 0x00 0x00 0x00 0x80 }
+# CHECK-DAG: st4 = { 0x00 0x00 0x00 0x00 0x00 0x00 0x00 0x80 0xff 0x7f }
+# CHECK-DAG: st5 = { 0x00 0x00 0x00 0x00 0x00 0x00 0x00 0x80 0xff 0xff }
+# CHECK-DAG: st6 = { 0x00 0x00 0x00 0x00 0x00 0x00 0x00 0xc0 0xff 0xff }
+# CHECK-DAG: st7 = { 0x00 0x00 0x00 0x00 0x00 0x00 0x00 0x00 0x00 0x00 }
+
+# CHECK: Process {{[0-9]+}} exited with status = 0

lldb/test/Shell/Register/x86-fp-read.test

+# REQUIRES: native && (target-x86 || target-x86_64)
+# RUN: %clangxx_host %p/Inputs/x86-fp-read.cpp -o %t
+# RUN: %lldb -b -s %s %t | FileCheck %s
+process launch
+# CHECK: Process {{.*}} stopped
+
+register read --all
+# CHECK-DAG: fctrl = 0x037b
+# CHECK-DAG: fstat = 0x8884
+# TODO: the following value is incorrect, it's a bug in the way
+# FXSAVE/XSAVE is interpreted; it should be 0xa963 once fixed
+# CHECK-DAG: ftag = 0x00fe
+# CHECK-DAG: fop = 0x0033
+
+# CHECK-DAG: st0 = {0x00 0x00 0x00 0x00 0x00 0x00 0x00 0x80 0x00 0x40}
+# CHECK-DAG: st1 = {0x00 0x00 0x00 0x00 0x00 0x00 0x00 0x3f 0x00 0x00}
+# CHECK-DAG: st2 = {0x00 0x00 0x00 0x00 0x00 0x00 0x00 0x00 0x00 0x00}
+# CHECK-DAG: st3 = {0x00 0x00 0x00 0x00 0x00 0x00 0x00 0x00 0x00 0x80}
+# CHECK-DAG: st4 = {0x00 0x00 0x00 0x00 0x00 0x00 0x00 0x80 0xff 0x7f}
+# CHECK-DAG: st5 = {0x00 0x00 0x00 0x00 0x00 0x00 0x00 0x80 0xff 0xff}
+# CHECK-DAG: st6 = {0x00 0x00 0x00 0x00 0x00 0x00 0x00 0xc0 0xff 0xff}
+# CHECK-DAG: st7 = {0x00 0x00 0x00 0x00 0x00 0x00 0x00 0x00 0x00 0x00}
+
+# fdiv (%rbx) gets encoded into 2 bytes, int3 into 1 byte
+print (void*)($pc-3)
+# CHECK: (void *) $0 = [[FDIV:0x[0-9a-f]*]]
+# TODO: we probably should not split it like this
+print (void*)($fiseg*0x100000000 + $fioff)
+# CHECK: (void *) $1 = [[FDIV]]
+print &zero
+# CHECK: (uint32_t *) $2 = [[ZERO:0x[0-9a-f]*]]
+print (uint32_t*)($foseg * 0x100000000 + $fooff)
+# CHECK: (uint32_t *) $3 = [[ZERO]]
+
+process continue
+# CHECK: Process {{[0-9]+}} exited with status = 0

lldb/test/Shell/Register/x86-fp-write.test

+# REQUIRES: native && target-x86
+# RUN: %clangxx_host %p/Inputs/x86-fp-write.cpp -o %t
+# RUN: %lldb -b -s %s %t | FileCheck %s
+process launch
+
+register write fctrl 0x037b
+register write fstat 0x8884
+# note: this needs to enable all registers for writes to be effective
+# TODO: fix it to use proper ftag values instead of 'abridged'
+register write ftag 0x00ff
+register write fop 0x0033
+# the exact addresses do not matter, we want just to verify FXSAVE
+# note: segment registers are not supported on all CPUs
+register write fioff 0x89abcdef
+register write fooff 0x76543210
+
+register write st0 "{0x00 0x00 0x00 0x00 0x00 0x00 0x00 0x80 0x00 0x40}"
+register write st1 "{0x00 0x00 0x00 0x00 0x00 0x00 0x00 0x3f 0x00 0x00}"
+register write st2 "{0x00 0x00 0x00 0x00 0x00 0x00 0x00 0x00 0x00 0x00}"
+register write st3 "{0x00 0x00 0x00 0x00 0x00 0x00 0x00 0x00 0x00 0x80}"
+register write st4 "{0x00 0x00 0x00 0x00 0x00 0x00 0x00 0x80 0xff 0x7f}"
+register write st5 "{0x00 0x00 0x00 0x00 0x00 0x00 0x00 0x80 0xff 0xff}"
+register write st6 "{0x00 0x00 0x00 0x00 0x00 0x00 0x00 0xc0 0xff 0xff}"
+register write st7 "{0x00 0x00 0x00 0x00 0x00 0x00 0x00 0x00 0x00 0x00}"
+
+process continue
+
+# CHECK: process continue
+# CHECK-DAG: fctrl = 0x037b
+# CHECK-DAG: fstat = 0x8884
+# CHECK-DAG: ftag = 0xa961
+# CHECK-DAG: fop = 0x0033
+# CHECK-DAG: fip = 0x89abcdef
+# CHECK-DAG: fdp = 0x76543210
+
+# CHECK-DAG: st0 = { 0x00 0x00 0x00 0x00 0x00 0x00 0x00 0x80 0x00 0x40 }
+# CHECK-DAG: st1 = { 0x00 0x00 0x00 0x00 0x00 0x00 0x00 0x3f 0x00 0x00 }
+# CHECK-DAG: st2 = { 0x00 0x00 0x00 0x00 0x00 0x00 0x00 0x00 0x00 0x00 }
+# CHECK-DAG: st3 = { 0x00 0x00 0x00 0x00 0x00 0x00 0x00 0x00 0x00 0x80 }
+# CHECK-DAG: st4 = { 0x00 0x00 0x00 0x00 0x00 0x00 0x00 0x80 0xff 0x7f }
+# CHECK-DAG: st5 = { 0x00 0x00 0x00 0x00 0x00 0x00 0x00 0x80 0xff 0xff }
+# CHECK-DAG: st6 = { 0x00 0x00 0x00 0x00 0x00 0x00 0x00 0xc0 0xff 0xff }
+# CHECK-DAG: st7 = { 0x00 0x00 0x00 0x00 0x00 0x00 0x00 0x00 0x00 0x00 }
+
+# CHECK: Process {{[0-9]+}} exited with status = 0