[llvm-dev] Range lists, zero-length functions, linker gc (original) (raw)

Eric Christopher via llvm-dev llvm-dev at lists.llvm.org
Thu May 28 17:39:23 PDT 2020

(FWIW I'm following this, but also haven't disagreed with anything Dave has said yet :)

On Thu, May 28, 2020 at 3:55 PM David Blaikie <dblaikie at gmail.com> wrote:

On Thu, May 28, 2020 at 6:03 AM Alexey Lapshin <alapshin at accesssoftek.com> wrote:

Hi David,

>So there have been several recent discussions about the issues around >DWARF-agnostic linking and gc-sections, linkonce function definitions being >dropped, etc - and just how much DWARF-awareness would be suitable >in a linker to help with this situation. > I'd like to discuss a narrower instance of this issue: Zero length gc'd/deduplicated functions. > LLVM seems to at least produce zero length functions in a few cases: > * non-void function without a return statement > * function definition containing only llvmunreachable > (both of these trap at -O0, but at higher optimization levels even the trap > instruction is removed & you get the full power UB of control flowing off > the end of the function into whatever other bytes are after that function) > So, for context, debugranges (this whole issue doesn't exist in DWARFv5, > FWIW) is a list of address pairs, terminated by a pair of zeros. > With function sections, or even just with normal C++ inline functions, > the CU will have a range entry for that function that consists of two relocations > - to the start and end of the function. Generally the start of the function is the > start of the section, and the end is "start of function + length of function (aka addend)". > Usually any relocation to the section would keep that section "alive" during linking - > but that would cause debug info to defeat linker GC and deduplication. So there's > special rules for how linkers handle these relocations in debug info to allow the > sections to be dropped - what do you write in the bytes that requested the relocation? > Binutils ld: Special cases only debugranges, resolving all relocations to dead > code to 1. In other debug sections, these values are all resolved to zero. > Gold and lld: Special cases all debug info sections - resolving all relocations > to "addend" (so begin usually goes to zero, end goes to "size of function") > These special rules are designed to ensure omitted/gc'd/deduplicated functions > don't cause the range list to terminate prematurely (which would happen if begin/end > were both resolved to zero). >But with an empty function, gold and lld's strategy here fails to avoid terminating a >range list by accident. > What should we do about it? > 1) Ensure no zero-length functions exist? (doesn't address backwards > compatibility/existing functions/other compilers) > 2) adopt the binutils approach to this (at least in debugranges - maybe in all > debug sections? (doing it in other sections could break ) > 3) Revisit the discussion about using an even more 'blessed' value, > like int max-1? ( https://reviews.llvm.org/D59553 ) > (I don't have links to all the recent threads about this discussion - I think D59553 > might've spawned a separate broader discussion/non-review - oh, Alexey wrote a > good summary with links to other discussions here: > http://lists.llvm.org/pipermail/llvm-dev/2019-September/135068.html ) > Thoughts? I think for the problem of "zero length functions and .debugranges" binutils approach looks good: >Special cases only debugranges, resolving all relocations to >dead code to 1. In other debug sections, these values are all resolved to >zero. But, this would not completely solve the problem from https://reviews.llvm.org/D59553 - Overlapped address ranges. Binutils approach will solve the problem if the address range specified as startaddress:endaddress. While resolving relocations, it would replace such a range with 1:1. However, It would not work if address ranges were specified as startaddress:length since the length is not relocated. This case could be additionally fixed by fast scan debuginfo for HighPC defined as length and changing it to 1. Something which you suggested here: http://lists.llvm.org/pipermail/llvm-dev/2020-May/141599.html. Hmm, I don't /think/ I intended to suggest anything that would have to parse all the debuginfo, even if just to fixup highpc. I meant that debugrnglist for the CU at least (rnglist has fewer problems - you can't accidentally terminate it early, but still has the "large functions in programs that use relatively low code addresses can't just be resolved to "addend" because then [0, length) of the large function might overlap into_ _that code address range") could be modified by a DWARF-aware linker to_ _remove the unused chunks. The DWARF that describes a specific function_ _using lowpc/highpc - it may be split into a .dwo file and unreachable by_ _the linker - so it /needs/ a magic value for the address referenced by the_ _lowpc to indicate that it is invalid._ _Which all comes back to "we probably need to pick a value that's_ _explicitly invalid" and -2 (max - 1) seems to be about the right thing._ _So it looks like following solution could fix both problems and be_ _relatively fast:_ _"Resolve all relocations from debug sections into dead code to 1. Parse_ _debug sections and replace HighPc of an address range pointing to dead code_ _and specified as length to 1"._ _That second part seems pretty expensive compared to anything else the_ _linker is doing with debug info. I'd try to avoid it if at all possible._ _As the result all address ranges pointing into dead code would be marked_ _as zero length._ _There still exist another problem:_ _DWARF4: "A range list entry (but not a base address selection or end of_ _list entry) whose beginning and_ _ending addresses are equal has no effect because the size of the range_ _covered by such an_ _entry is zero."_ _DWARF5: "A bounded range entry whose beginning and ending address offsets_ _are equal_ _(including zero) indicates an empty range and may be ignored."_ _These rules allow us to ignore zero-length address ranges. I.e., some_ _tool reading DWARF is permitted to ignore related DWARF entries._ _I agree it allows consumers to ignore that entry in the range list because_ _that entry is zero-length/equivalent to not being present at all - I don't_ _think that means consumers can ignore the DIE that refers to this range_ _list. I think it's valid DWARF to have a CU that only describes types,_ _without any code attached to it at all. Or for a subprogram that's been_ _eliminated to still be used by a consumer for name lookup purposes - so the_ _consumer can tell the user there is a function called "f1" and tell the_ _user what parameter types, return type it has, etc - not ignore it entirely._ _In that case, there could be ignored essential descriptions. That problem_ _could happen with -flto=thin example_ _https://reviews.llvm.org/D54747#1503720 . In this example, all type definitions except one were replaced with declarations by thinlto. The definition, which was left, is in a piece of debug info related to deleted code. According to zero-length rule, that definition could be ignored, and finally, incomplete debug info could be used. Yeah, I think the bug there is the linker dropping object files just because they have no exxecutable code in them - I think the patch that did that was reverted, if I'm remembering correctly. So, it probably should be forbidden to generate debuginfo, which could become incomplete after removing pieces related to zero length address ranges. Otherwise, creating zero-length address ranges could lead to incomplete debug info. Thank you, Alexey.

-------------- next part -------------- An HTML attachment was scrubbed... URL: <http://lists.llvm.org/pipermail/llvm-dev/attachments/20200528/6e94e509/attachment.html>

More information about the llvm-dev mailing list