ARM: fix stack frame management

[myungjoo]

Fix #4638

There have been a lot of ASM functions that may break stack unwinding.

Tested along with @parjong 's #4503 (merged)
and modified #4581 (appended in included in this PR)

Signed-off-by: MyungJoo Ham myungjoo.ham@samsung.com

[myungjoo]

I'll continue working on from Line 799. there are more functions to be fixed.

[myungjoo]

Now, it seems that I've reviewed and corrected every single ASM/ARM function that is going to be unwinded.

I assume that JIT_Stelem_Ref_NotExactMatch is not going to be unwinded.

[myungjoo]

TEST RESULTS:

bash-3.2# ./corerun csc.exe /lib:. /r:mscorlib.dll throw.cs
Microsoft (R) Visual C# Compiler version 42.42.42.42
Copyright (C) Microsoft Corporation. All rights reserved.

throw.cs(6,42): warning CS7095: Filter expression is a constant, consider removing the filter
throw.cs(6,31): warning CS0168: The variable 'e' is declared but never used
bash-3.2# ./corerun throw.exe
Hello World!
bash-3.2# cat throw.cs
namespace ConsoleApplication { public class Program {

  public static void Main(string[] args) {
    try {
      throw new System.Exception("Throw");
    } catch (System.Exception e)  when ( true ) {
      System.Console.WriteLine("Hello World!");
    }
  }

} }
bash-3.2# ./corerun seh_tests.exe 
Testcase successful
bash-3.2# uname -r -o
3.10.65 GNU/Linux
bash-3.2# cat /proc/cpuinfo
Processor   : ARMv7 Processor rev 5 (v7l)
processor   : 0
model name  : ARMv7 Processor rev 5 (v7l)
BogoMIPS    : 2002.62
Features    : swp half thumb fastmult vfp edsp neon vfpv3 tls vfpv4 idiva idivt vfpd32 
CPU implementer : 0x41
CPU architecture: 7
CPU variant : 0x0
CPU part    : 0xc07
CPU revision    : 5

Hardware    : Z3
Revision    : 0000
Serial      : 0000c6bf00009200
bash-3.2#

Based: dfcb7f0 + #4503 + Modified #4581.

Modified #4581 is:

diff --git a/src/vm/arm/ehhelpers.S b/src/vm/arm/ehhelpers.S
index dfe1b11..c32a06b 100644
--- a/src/vm/arm/ehhelpers.S
+++ b/src/vm/arm/ehhelpers.S
@@ -47,6 +47,8 @@ OFFSET_OF_FRAME=(4 + SIZEOF__GSCookie)
         //
         // Runtime check for 8-byte alignment. 
         PROLOG_STACK_SAVE r7
+        // We lose stack unwindability here by configuring fp(r7) incorrectely
+        // here.
         and r0, r7, #4
         sub sp, sp, r0

@@ -99,6 +101,7 @@ OFFSET_OF_FRAME=(4 + SIZEOF__GSCookie)
         NESTED_ENTRY CallEHFunclet, _TEXT, NoHandler

         PROLOG_PUSH  "{r4-r11, lr}"
+        PROLOG_STACK_SAVE_OFFSET  r7, #12
         alloc_stack  4

         // On entry:
@@ -124,7 +127,8 @@ OFFSET_OF_FRAME=(4 + SIZEOF__GSCookie)
         // frame pointer for accessing the locals in the parent method.
         NESTED_ENTRY CallEHFilterFunclet, _TEXT, NoHandler

-        PROLOG_PUSH  "{lr}"
+        PROLOG_PUSH  "{r7, lr}"
+        PROLOG_STACK_SAVE r7
         alloc_stack  4

         // On entry:
@@ -140,6 +144,6 @@ OFFSET_OF_FRAME=(4 + SIZEOF__GSCookie)
         blx r2

         free_stack   4
-        EPILOG_POP   "{pc}"
+        EPILOG_POP   "{r7, pc}"

         NESTED_END CallEHFilterFunclet, _TEXT

[myungjoo]

@dotnet-bot test OSX x64 Checked Build Please

[myungjoo]

As @parjong's #4581 looks better if based on this PR, I'm importing and rebasing his patch and putting here. (will be appended to this PR soon)

[janvorli]

Can we use PROLOG_PUSH followed by PROLOG_STACK_SAVE_OFFSET here and at several other places below? We have a convention to not to use the unwinder pseudoinstructions directly in the .S files.

[myungjoo]

Sure we can. The patchset is updated to use the macro.

[myungjoo]

@dotnet-bot Test OSX x64 Checked Build and Test Please

[janvorli]

LGTM, thanks!

[parjong]

@janvorli OSX x64 Checked Build and Test failed with "No space left on device" error as follows:

BEGIN EXECUTION /Users/dotnet-bot/j/workspace/dotnet_coreclr/master/checked_osx_tst_prtest/bin/tests/Windows_NT.x64.Checked/Tests/coreoverlay/corerun Collect_Default_1.exe 0 {0x5943147-0x10cc56480} ASSERT [PROCESS] at /Users/dotnet-bot/j/workspace/dotnet_coreclr/master/checked_osx_prtest/src/pal/src/thread/process.cpp.1848: sem_open(/clrct000144800000000057289b62) failed: 28 (No space left on device) ./Collect_Default_1.sh: line 62: 83072 Trace/BPT trap: 5 $_DebuggerFullPath "$CORE_ROOT/corerun" Collect_Default_1.exe $CLRTestExecutionArguments $Host_Args Expected: 100 Actual: 133 END EXECUTION - FAILED

Is there any issue with test VM?

[janvorli]

@parjong yes, we have issues with OSX CI machines at the moment, most of the test runs are failing. The issue is not a disk space problem though, it is a semaphore leak that was accumulating over time and now it has hit.
The issue is being worked on.

[parjong]

@janvorli Thanks you for comment 👍

[janvorli]

I am going to merge this in since the change is completely unrelated to OSX and doesn't touch any OSX related files.