[arm32] throw exception does not set CONTEXT_UNWOUND_TO_CALL flag

[chunseoklee]

dotnet/coreclr#6117 PR need more discussion.

One strange thing I found is that when we throw exception in csharp source as in coreclr/tests/src/baseservices/exceptions/generics/try-finally-struct01.cs, frameContext->ContextFlags is set to neither CONTEXT_EXCEPTION_ACTIVE nor CONTEXT_UNWOUND_TO_CALL at the beginning of UnwindManagedExceptionPass1. In this situation, PC is out of range of proper EH table range, which implies wrong EH found.
For example below,
The range of exception handler for our exception throw is [18 to 8C). EH#0: try [0018..008C) handled by [00B0..00DA) (finally)) But, when we start to look for EH for exception throw(8A), Pc indicates to address of IL 8C. Since both CONTEXT_EXCEPTION_ACTIVE and CONTEXT_UNWOUND_TO_CALL is off, IP decrement step(in ExceptionTracker::InitializeCrawlFrame) is skipped and cannot find proper EH.

In summary, I have 2 questions :
(1) Is it OK that CONTEXT_UNWOUND_TO_CALL is off? (IMHO, it should be ON here)
(2) Is it OK that CONTEXT_EXCEPTION_ACTIVE is off? (IMHO, it should be OFF here. Thus, the patch should be modified.)

I will update comment if have any progress.

    public void InternalExceptionTest(bool throwException)
    {
        hit = false; 
        try
        {
            if (throwException)
            {
                throw new GenException<T>();
            }
            Test.Eval(!throwException);
        }
        finally
        {
            hit = true;
        }
        Test.Eval(!throwException);
    }



; Assembly listing for method Gen`1[Int32][System.Int32]:InternalExceptionTest(bool):this
; Emitting BLENDED_CODE for generic ARM CPU
; optimized code
; r11 based frame
; fully interruptible
; Final local variable assignments
;
;  V00 this         [V00,T00] (  4,   4  )   byref  ->  [sp+0x04]   do-not-enreg[H] this
;  V01 arg1         [V01    ] (  5,   5  )    bool  ->  [sp+0x14]   do-not-enreg[X] addr-exposed
;  V02 tmp0         [V02,T01] (  3,   0  )     ref  ->  [sp+0x00]  
;# V03 OutArgs      [V03    ] (  1,   1  )  lclBlk ( 0) [sp+0x00]  
;  V04 PSPSym       [V04    ] (  1,   1  )     int  ->  [sp+0x08]   do-not-enreg[X] addr-exposed
;
; Lcl frame size = 12

G_M27436_IG01:
000000  B402           push    {r1}
000002  E92D 4800      push    {r11,lr}
000006  B083           sub     sp, 12
000008  F10D 0B0C      add     r11, sp, 12
00000C  A906           add     r1, sp, 24
00000E  9102           str     r1, [sp+0x08]    // [V04 PSPSym]
000010  9001           str     r0, [sp+0x04]    // [V00 this]

G_M27436_IG02:
000012  9B01           ldr     r3, [sp+0x04]    // [V00 this]
000014  2200           movs    r2, 0
000016  701A           strb    r2, [r3]

G_M27436_IG03:
000018  F89D 3014      ldrb    r3, [sp+0x14]    // [V01 arg1]
00001C  2B00           cmp     r3, 0
00001E  D110           bne     SHORT G_M27436_IG07

G_M27436_IG04:
000020  F89D 3014      ldrb    r3, [sp+0x14]    // [V01 arg1]
000024  2B00           cmp     r3, 0
000026  D001           beq     SHORT G_M27436_IG05
000028  2000           movs    r0, 0
00002A  E000           b       SHORT G_M27436_IG06

G_M27436_IG05:
00002C  2001           movs    r0, 1

G_M27436_IG06:
00002E  F24E 0339      movw    r3, 0xe039
000032  F2CB 13DE      movt    r3, 0xb1de
000036  4798           blx     r3       // Test:Eval(bool)
000038  F64E 2E03      movw    lr, LOW ADDRESS G_M27436_IG09
00003C  F2CB 1EDE      movt    lr, HIGH ADDRESS G_M27436_IG09
000040  E036           b       SHORT G_M27436_IG13

G_M27436_IG07:
000042  2001           movs    r0, 1
000044  F64D 21C0      movw    r1, 0xdac0
000048  F2CB 5198      movt    r1, 0xb598
00004C  F642 23E5      movw    r3, 0x2ae5
000050  F2CB 634C      movt    r3, 0xb64c
000054  4798           blx     r3       // CORINFO_HELP_STRCNS
000056  F24E 2341      movw    r3, 0xe241
00005A  F2CB 13DE      movt    r3, 0xb1de
00005E  4798           blx     r3       // System.Console:WriteLine(ref)
000060  F241 1024      movw    r0, 0x1124
000064  F2CB 5099      movt    r0, 0xb599
000068  F242 0305      movw    r3, 0x2005
00006C  F2CB 634C      movt    r3, 0xb64c
000070  4798           blx     r3       // CORINFO_HELP_NEWSFAST
000072  9000           str     r0, [sp] // [V02 tmp0]
000074  9800           ldr     r0, [sp]
000076  F64F 7351      movw    r3, 0xff51
00007A  F2CB 4390      movt    r3, 0xb490
00007E  4798           blx     r3       // System.Exception:.ctor():this
000080  9800           ldr     r0, [sp] // [V02 tmp0]
000082  F24C 53ED      movw    r3, 0xc5ed
000086  F2CB 634C      movt    r3, 0xb64c
00008A  4798           blx     r3       // CORINFO_HELP_THROW

G_M27436_IG08:
00008C  BF00           nop     

G_M27436_IG09:
00008E  F89D 3014      ldrb    r3, [sp+0x14]    // [V01 arg1]
000092  2B00           cmp     r3, 0
000094  D001           beq     SHORT G_M27436_IG10
000096  2000           movs    r0, 0
000098  E000           b       SHORT G_M27436_IG11

G_M27436_IG10:
00009A  2001           movs    r0, 1

G_M27436_IG11:
00009C  F24E 0339      movw    r3, 0xe039
0000A0  F2CB 13DE      movt    r3, 0xb1de
0000A4  4798           blx     r3       // Test:Eval(bool)

G_M27436_IG12:
0000A6  B003           add     sp, 12
0000A8  E8BD 4800      pop     {r11,lr}
0000AC  B001           add     sp, 4
0000AE  4770           bx      lr

G_M27436_IG13:
0000B0  E92D 480C      push    {r2,r3,r11,lr}
0000B4  F10B 030C      add     r3, r11, 12
0000B8  9300           str     r3, [sp]

G_M27436_IG14:
0000BA  F242 3338      movw    r3, 0x2338
0000BE  F2CB 3380      movt    r3, 0xb380
0000C2  6818           ldr     r0, [r3]
0000C4  F24E 2341      movw    r3, 0xe241
0000C8  F2CB 13DE      movt    r3, 0xb1de
0000CC  4798           blx     r3       // System.Console:WriteLine(ref)
0000CE  F85B 3C08      ldr     r3, [r11-0x08]   // [V00 this]
0000D2  2201           movs    r2, 1
0000D4  701A           strb    r2, [r3]

G_M27436_IG15:
0000D6  E8BD 880C      pop     {r2,r3,r11,pc}

; Total bytes of code 218, prolog size 18 for method Gen`1[Int32][System.Int32]:InternalExceptionTest(bool):this
; ============================================================
*************** EH table for Gen`1[Int32][System.Int32]:InternalExceptionTest(bool):this
1 EH table entries, 0 duplicate clauses, 1 total EH entries reported to VM
EH#0: try [0018..008C) handled by [00B0..00DA) (finally)

[janvorli]

CONTEXT_UNWOUND_TO_CALL is set by the managed code unwinder when we unwind (on Windows, it is also cleared by the unwinder when we unwind to a trap frame, but that doesn't apply to Unix).
CONTEXT_EXCEPTION_ACTIVE is set only when we unwind to the frame where a hardware exception occured. That means that the PC is at the failing instruction and should not be augmented.
The fact that frameContext->ContextFlags is set to neither CONTEXT_EXCEPTION_ACTIVE nor CONTEXT_UNWOUND_TO_CALL at the beginning of UnwindManagedExceptionPass1 means there is a problem with setting the CONTEXT_UNWOUND_TO_CALL flag. I think that we need to set that flag in the native unwinder (for ARM / ARM64 only) when we don't set the CONTEXT_EXCEPTION_ACTIVE and vice versa. That means in the code below:
https://github.com/dotnet/coreclr/blob/master/src/pal/src/exception/seh-unwind.cpp#L289-L296.
It seems we can change it to the code below to fix the problem:

    if (unw_is_signal_frame(&cursor) > 0)
    {
        context->ContextFlags |= CONTEXT_EXCEPTION_ACTIVE;
#if defined(_ARM_) || defined(_ARM64_)
        context->ContextFlags &= ~CONTEXT_UNWOUND_TO_CALL;
#endif // _ARM_ || _ARM64_
    }
    else
    {
        context->ContextFlags &= ~CONTEXT_EXCEPTION_ACTIVE;
#if defined(_ARM_) || defined(_ARM64_)
        context->ContextFlags |= CONTEXT_UNWOUND_TO_CALL;
#endif // _ARM_ || _ARM64_
    }

[chunseoklee]

@janvorli Thank you for kind description.
Then, only one of two enums should be set in each unwind step ? It means that my PR is ridiculous and that it is sufficient to check only one of two flags. Am I right?

[janvorli]

@chunseoklee right, if you make the change I've suggested (and provided I am not mistaken), then the change in your PR should not be needed.

[chunseoklee]

@janvorli Thank you for help.
I will update PR as you suggested.