[GHC] #11353: DWARF call frame information incorrect in the presence of unsafe foreign calls
#11353: DWARF call frame information incorrect in the presence of unsafe foreign calls -------------------------------------+------------------------------------- Reporter: bgamari | Owner: Type: bug | Status: new Priority: normal | Milestone: 8.2.1 Component: Compiler | Version: 7.10.3 (CodeGen) | Keywords: | Operating System: Unknown/Multiple Architecture: x86_64 | Type of failure: Debugging (amd64) | information is incorrect Test Case: | Blocked By: Blocking: | Related Tickets: Differential Rev(s): | Wiki Page: -------------------------------------+------------------------------------- Unsafe foreign calls result in adjustments to `$rsp` made in the NCG to comply with calling convention alignment requirements (see `X86.CodeGen.genCCall64'`). Unfortunately, this happens after we've generated unwind information. This results in incorrect unwinding information for `$rsp` when inside of a foreign call. The results can be quite catastrophic (e.g. segmentation faults while unwinding). Unfortunately it's really not clear what can be done about this given that these adjustments aren't present in the Cmm representation that we use to produce frame information. -- Ticket URL: http://ghc.haskell.org/trac/ghc/ticket/11353 GHC http://www.haskell.org/ghc/ The Glasgow Haskell Compiler
#11353: DWARF call frame information incorrect in the presence of unsafe foreign calls -------------------------------------+------------------------------------- Reporter: bgamari | Owner: Type: bug | Status: new Priority: normal | Milestone: 8.2.1 Component: Compiler | Version: 7.10.3 (CodeGen) | Resolution: | Keywords: Operating System: Unknown/Multiple | Architecture: x86_64 Type of failure: Debugging | (amd64) information is incorrect | Test Case: Blocked By: | Blocking: Related Tickets: | Differential Rev(s): Wiki Page: | -------------------------------------+------------------------------------- Comment (by bgamari): Perhaps we can feed the `DebugBlocks` as state into the NCG. The NCG can then make the appropriate updates before finally producing the final debug information? Ugh. -- Ticket URL: http://ghc.haskell.org/trac/ghc/ticket/11353#comment:1 GHC http://www.haskell.org/ghc/ The Glasgow Haskell Compiler
#11353: DWARF call frame information incorrect in the presence of unsafe foreign calls -------------------------------------+------------------------------------- Reporter: bgamari | Owner: Type: bug | Status: new Priority: normal | Milestone: 8.2.1 Component: Compiler | Version: 7.10.3 (CodeGen) | Resolution: | Keywords: Operating System: Unknown/Multiple | Architecture: x86_64 Type of failure: Debugging | (amd64) information is incorrect | Test Case: Blocked By: | Blocking: Related Tickets: | Differential Rev(s): Wiki Page: | -------------------------------------+------------------------------------- Comment (by bgamari): In fact, it looks like the NCG already has read-only access to the `DebugBlock`s. Perhaps this idea isn't so crazy. The tricky part is ensuring that the debug entries remain properly sorted. Perhaps you could say something like "insert this `UnwindTable` after label `c2TN`"? -- Ticket URL: http://ghc.haskell.org/trac/ghc/ticket/11353#comment:2 GHC http://www.haskell.org/ghc/ The Glasgow Haskell Compiler
#11353: DWARF call frame information incorrect in the presence of unsafe foreign calls -------------------------------------+------------------------------------- Reporter: bgamari | Owner: Type: bug | Status: new Priority: normal | Milestone: 8.2.1 Component: Compiler | Version: 7.10.3 (CodeGen) | Resolution: | Keywords: Operating System: Unknown/Multiple | Architecture: x86_64 Type of failure: Debugging | (amd64) information is incorrect | Test Case: Blocked By: | Blocking: Related Tickets: | Differential Rev(s): Wiki Page: | -------------------------------------+------------------------------------- Comment (by bgamari): Another option along these lines would be to abandon tracking label ordering at all until code generation. Then the code generator could modify the `DebugBlock` to its hearts content, so long as in the end it returned the unwinding tables in the proper order. For those playing along at home, the issue here is that DWARF requires that frame unwinding tables (FDEs) must be written in order of increasing address. -- Ticket URL: http://ghc.haskell.org/trac/ghc/ticket/11353#comment:3 GHC http://www.haskell.org/ghc/ The Glasgow Haskell Compiler
#11353: DWARF call frame information incorrect in the presence of unsafe foreign calls -------------------------------------+------------------------------------- Reporter: bgamari | Owner: Type: bug | Status: new Priority: normal | Milestone: 8.2.1 Component: Compiler | Version: 7.10.3 (CodeGen) | Resolution: | Keywords: Operating System: Unknown/Multiple | Architecture: x86_64 Type of failure: Debugging | (amd64) information is incorrect | Test Case: Blocked By: | Blocking: Related Tickets: | Differential Rev(s): Wiki Page: | -------------------------------------+------------------------------------- Comment (by simonmar): The NCG tracks the current state of %rsp during code generation (see `getDeltaNat`) and emits hints into the generated code with the pseudo- instruction `DELTA`. It seems like we should be able to generate correct directives for the assembler from the `DELTA` instructions, no? -- Ticket URL: http://ghc.haskell.org/trac/ghc/ticket/11353#comment:4 GHC http://www.haskell.org/ghc/ The Glasgow Haskell Compiler
#11353: DWARF call frame information incorrect in the presence of unsafe foreign calls -------------------------------------+------------------------------------- Reporter: bgamari | Owner: Type: bug | Status: new Priority: normal | Milestone: 8.2.1 Component: Compiler | Version: 7.10.3 (CodeGen) | Resolution: | Keywords: Operating System: Unknown/Multiple | Architecture: x86_64 Type of failure: Debugging | (amd64) information is incorrect | Test Case: Blocked By: | Blocking: Related Tickets: #11338, #11337 | Differential Rev(s): Wiki Page: | -------------------------------------+------------------------------------- Changes (by bgamari): * related: => #11338, #11337 -- Ticket URL: http://ghc.haskell.org/trac/ghc/ticket/11353#comment:5 GHC http://www.haskell.org/ghc/ The Glasgow Haskell Compiler
#11353: DWARF call frame information incorrect in the presence of unsafe foreign calls -------------------------------------+------------------------------------- Reporter: bgamari | Owner: Type: bug | Status: new Priority: normal | Milestone: 8.2.1 Component: Compiler | Version: 7.10.3 (CodeGen) | Resolution: | Keywords: Operating System: Unknown/Multiple | Architecture: x86_64 Type of failure: Debugging | (amd64) information is incorrect | Test Case: Blocked By: | Blocking: Related Tickets: #11338, #11337 | Differential Rev(s): Wiki Page: | -------------------------------------+------------------------------------- Comment (by bgamari): The problem is that you need both details from the code generator (to account for the unwind adjustments as described in this ticket) as well as the procedure's control flow graph (in order to push unwind information down to successor blocks). My current plan is, * Introduce a `UNWIND` pseudo-instruction * `CmmUnwind` nodes will be translated by the `UNWIND` instructions * Add a lazy field to `NEWBLOCK` which will capture the successors of the block * Introduce a pass after code generation which traverses the `[Instruction]` of each block and, * extract the `UNWIND` and `DELTA` instructions from a block's `[Instruction]` and use them to build a `LabelMap (UnwindTable, [SuccessorBlockId])` (the `SuccessorBlockId` being taken from the successors field of the block's `NEWBLOCK` instruction) * accumulate a `[BlockId]` to preserve the order in which we must emit unwinding tables (since they must be written in order of increasing address) * Introduce another pass of shape `LabelMap (UnwindTable, [SuccessorBlockId]) -> LabelMap UnwindTable` which pushes blocks' unwind tables "down" to their successors * Pass the resulting `UnwindTable`s to the existing unwinding table generation code -- Ticket URL: http://ghc.haskell.org/trac/ghc/ticket/11353#comment:6 GHC http://www.haskell.org/ghc/ The Glasgow Haskell Compiler
#11353: DWARF call frame information incorrect in the presence of unsafe foreign calls -------------------------------------+------------------------------------- Reporter: bgamari | Owner: Type: bug | Status: new Priority: normal | Milestone: 8.2.1 Component: Compiler | Version: 7.10.3 (CodeGen) | Resolution: | Keywords: Operating System: Unknown/Multiple | Architecture: x86_64 Type of failure: Debugging | (amd64) information is incorrect | Test Case: Blocked By: | Blocking: Related Tickets: #11338, #11337 | Differential Rev(s): Wiki Page: | -------------------------------------+------------------------------------- Comment (by scpmw): Not sure why you are worrying about the order - that gets derived from the NCG output using cmmDebugLink. This is precisely so NCG is free to do whatever it wants with blocks, even reorder or optimise them out. My first impulse here would be that you will need some sort of `UNWIND` anyway to have NCG create labels. So maybe have it carry its unwind rules, that would allow NCG to update it. In either case, you would feed that information back roughly the same way that currently `ngs_labels` gets threaded back to debug information generation. As `UNWIND` would carry unique labels, this could just be replacing the current `[Label]` block list with an an ordered `[(Label, [UnwindTable])]` debug label to unwind table map. Not sure - this is exactly the kind of complexity that I tried my best to avoid, so I haven't put a lot of thought into it. -- Ticket URL: http://ghc.haskell.org/trac/ghc/ticket/11353#comment:7 GHC http://www.haskell.org/ghc/ The Glasgow Haskell Compiler
#11353: DWARF call frame information incorrect in the presence of unsafe foreign calls -------------------------------------+------------------------------------- Reporter: bgamari | Owner: Type: bug | Status: new Priority: normal | Milestone: 8.2.1 Component: Compiler (NCG) | Version: 7.10.3 Resolution: | Keywords: Operating System: Unknown/Multiple | Architecture: x86_64 Type of failure: Debugging | (amd64) information is incorrect | Test Case: Blocked By: | Blocking: Related Tickets: #11338, #11337 | Differential Rev(s): Wiki Page: | -------------------------------------+------------------------------------- Changes (by bgamari): * component: Compiler (CodeGen) => Compiler (NCG) -- Ticket URL: http://ghc.haskell.org/trac/ghc/ticket/11353#comment:8 GHC http://www.haskell.org/ghc/ The Glasgow Haskell Compiler
#11353: DWARF call frame information incorrect in the presence of unsafe foreign calls -------------------------------------+------------------------------------- Reporter: bgamari | Owner: bgamari Type: bug | Status: new Priority: normal | Milestone: 8.2.1 Component: Compiler (NCG) | Version: 7.10.3 Resolution: | Keywords: Operating System: Unknown/Multiple | Architecture: x86_64 Type of failure: Debugging | (amd64) information is incorrect | Test Case: Blocked By: | Blocking: Related Tickets: #11338, #11337 | Differential Rev(s): Wiki Page: | -------------------------------------+------------------------------------- Changes (by bgamari): * owner: => bgamari -- Ticket URL: http://ghc.haskell.org/trac/ghc/ticket/11353#comment:9 GHC http://www.haskell.org/ghc/ The Glasgow Haskell Compiler
#11353: DWARF call frame information incorrect in the presence of unsafe foreign calls -------------------------------------+------------------------------------- Reporter: bgamari | Owner: bgamari Type: bug | Status: new Priority: high | Milestone: 8.2.1 Component: Compiler (NCG) | Version: 7.10.3 Resolution: | Keywords: Operating System: Unknown/Multiple | Architecture: x86_64 Type of failure: Debugging | (amd64) information is incorrect | Test Case: Blocked By: | Blocking: Related Tickets: #11338, #11337 | Differential Rev(s): Wiki Page: | -------------------------------------+------------------------------------- Changes (by bgamari): * priority: normal => high Comment: The patches for this are essentially done. I just need to clean them up and merge them. -- Ticket URL: http://ghc.haskell.org/trac/ghc/ticket/11353#comment:10 GHC http://www.haskell.org/ghc/ The Glasgow Haskell Compiler
#11353: DWARF call frame information incorrect in the presence of unsafe foreign
calls
-------------------------------------+-------------------------------------
Reporter: bgamari | Owner: bgamari
Type: bug | Status: new
Priority: high | Milestone: 8.2.1
Component: Compiler (NCG) | Version: 7.10.3
Resolution: | Keywords:
Operating System: Unknown/Multiple | Architecture: x86_64
Type of failure: Debugging | (amd64)
information is incorrect | Test Case:
Blocked By: | Blocking:
Related Tickets: #11338, #11337 | Differential Rev(s):
Wiki Page: |
-------------------------------------+-------------------------------------
Comment (by Ben Gamari
#11353: DWARF call frame information incorrect in the presence of unsafe foreign calls -------------------------------------+------------------------------------- Reporter: bgamari | Owner: bgamari Type: bug | Status: closed Priority: high | Milestone: 8.2.1 Component: Compiler (NCG) | Version: 7.10.3 Resolution: fixed | Keywords: Operating System: Unknown/Multiple | Architecture: x86_64 Type of failure: Debugging | (amd64) information is incorrect | Test Case: Blocked By: | Blocking: Related Tickets: #11338, #11337 | Differential Rev(s): Wiki Page: | -------------------------------------+------------------------------------- Changes (by bgamari): * status: new => closed * resolution: => fixed -- Ticket URL: http://ghc.haskell.org/trac/ghc/ticket/11353#comment:12 GHC http://www.haskell.org/ghc/ The Glasgow Haskell Compiler
#11353: DWARF call frame information incorrect in the presence of unsafe foreign calls -------------------------------------+------------------------------------- Reporter: bgamari | Owner: bgamari Type: bug | Status: closed Priority: high | Milestone: 8.2.1 Component: Compiler (NCG) | Version: 7.10.3 Resolution: fixed | Keywords: DWARF Operating System: Unknown/Multiple | Architecture: x86_64 Type of failure: Debugging | (amd64) information is incorrect | Test Case: Blocked By: | Blocking: Related Tickets: #11338, #11337 | Differential Rev(s): Wiki Page: | -------------------------------------+------------------------------------- Changes (by bgamari): * keywords: => DWARF -- Ticket URL: http://ghc.haskell.org/trac/ghc/ticket/11353#comment:13 GHC http://www.haskell.org/ghc/ The Glasgow Haskell Compiler
participants (1)
-
GHC