<div dir="ltr"><div class="gmail_extra"><div class="gmail_quote">Thank you George for writing this up!</div><div class="gmail_quote"><br></div><div class="gmail_quote">On Mon, Dec 4, 2017 at 7:11 AM, George Rimar <span dir="ltr"><<a href="mailto:grimar@accesssoftek.com" target="_blank">grimar@accesssoftek.com</a>></span> wrote:<br><blockquote class="gmail_quote" style="margin:0 0 0 .8ex;border-left:1px #ccc solid;padding-left:1ex">
<div dir="ltr" style="font-size:12pt;color:#000000;background-color:#ffffff;font-family:Calibri,Arial,Helvetica,sans-serif">
<p></p>
<div>Hi all !</div>
<div><br>
</div>
<div>We have an issue with LLD, it is "relocation R_X86_64_32 out of range" (PR31109)</div>
<div>which occurs during resolving relocations in debug sections. It looks happens</div>
<div>because .debug_info section can be too large sometimes and 32x relocation is not enough</div>
<div>to represent the value. One of possible solutions looks to be to deduplicate information</div>
<div>to reduce .debug_info size.</div>
<div>The rest of mail contains information about experiments I did, the obtained results and</div>
<div>some questions and suggestions as well.</div>
<div><br>
</div>
<div>I was investigating idea to deduplicate debug types information. Idea is described at</div>
<div>p276 of DWARF4 specification (<a href="http://www.dwarfstd.org/doc/DWARF4.pdf" target="_blank">http://www.dwarfstd.org/doc/<wbr>DWARF4.pdf</a>). It suggests</div>
<div>to split types information out of .debug_info and emit multiple .debug_types sections</div>
<div>with use of COMDATs. Both clang and gcc I tested implements -fdebug-types-section flag for that:</div>
<div><br>
</div>
<div>-fdebug-types-section, -fno-debug-types-section</div>
<div>Place debug types in their own section (ELF Only)</div>
<div>gcc's description is here: <a href="https://gcc.gnu.org/onlinedocs/gcc-6.4.0/gcc/Debugging-Options.html#Debugging-Options" target="_blank">https://gcc.gnu.org/<wbr>onlinedocs/gcc-6.4.0/gcc/<wbr>Debugging-Options.html#<wbr>Debugging-Options</a>.</div>
<div><br>
</div>
<div>This flag is disabled by default. I compared clang binaries to see the difference<br>
</div>
<div>with and without the linker side optimisation.</div>
<div>1) Clang built with -g has size of 1.7 GB, .debug_info section size is 894.5 Mb.</div>
<div>2) Clang built with -g -fdebug-types-section has size of 1.0 GB.</div>
<div> .debug_types size is 26.267 MB, .debug_info size is 227.7 MB.</div>
<div><br>
</div>
<div>Difference is huge and I believe shows (though probably for most of readers here it was</div>
<div>already obvious) that optimization can be useful. Though <span style="font-family:Calibri,Arial,Helvetica,sans-serif;font-size:16px;background-color:rgb(255,255,255)">
-fdebug-types-section</span> is disabled by default.</div>
<div>Looks it was initially disabled because not all of DWARF consumers were aware of .debug_types section.</div>
<div><br>
</div>
<div>Now in 2017 situation is different. I think most of DWARF consumers knows about .debug_types, but:</div>
<div>1) DWARF5 specification explicitly eliminates the .debug_types section introduced in DWARF4:</div>
<div> p8, "1.4 Changes from Version 4 to Version 5" <a href="http://dwarfstd.org/doc/DWARF5.pdf" target="_blank">http://dwarfstd.org/doc/<wbr>DWARF5.pdf</a></div>
<div>2) Instead of emiting multiple .debug_types it suggests to emit multiple .debug_info COMDAT</div>
<div> sections. (p375, p376).</div>
<div><br>
</div>
<div>And it seems currently there is no way to make clang to emit multiple .debug_info with type information</div>
<div>like DWARF5 suggests. I tried command line below:</div>
<div>-g -fdebug-types-section -gdwarf-5</div>
<div>It still emits .debug_types and does not look there is a flag for emiting multiple .debug_info.</div>
<div>Looking at whole LLVM code (lib/mc, lib/CodeGen) actually it seems it is just always assumed .debug_info is</div>
<div>a unique section in object.</div>
<div>(also not sure why clang emits .debug_types when -gdwarf-5 flag is set, as this section is incompatible with v5,</div>
<div>probably it is a bug).</div>
<div><br>
</div>
<div>So my questions are following:</div>
<div>1) Do we want to try to implement multiple .debug_info approach ? As it seems can be very useful sometimes.</div>
<div>2) For now in LLD may be we may want to extend our error message from "relocation X out of range" to something</div>
<div> suggesting to use -fdebug-types-section (only for relocations in debug sections) ?</div></div></blockquote><div><br></div><div>What we ideally should do is to print out a hint message to add a flag to force the compiler to emit DWARF64 debug info (I don't know the flag name) and -fdebug-type-sections, along with a brief message describing why we can't satisfy the user request (e.g. relocation X is too large for 32-bit DWARF info). That said, looks like LLVM's DWARF64 support is incomplete yet, so it may make sense to print out a hint message as you suggested.</div><div><br></div><blockquote class="gmail_quote" style="margin:0 0 0 .8ex;border-left:1px #ccc solid;padding-left:1ex"><div dir="ltr" style="font-size:12pt;color:#000000;background-color:#ffffff;font-family:Calibri,Arial,Helvetica,sans-serif">
<div>3) Why -fdebug-types-section is disabled by default ?<br>
</div>
<p><br>
</p>
<div id="m_1404985757845891049Signature">
<div name="divtagdefaultwrapper">
<div class="m_1404985757845891049BodyFragment"><font size="2">
<div class="m_1404985757845891049PlainText">Best regards,<br>
George | Developer | <span style="background-color:rgb(255,255,255);color:rgb(33,33,33);font-family:Calibri,sans-serif;font-size:13.3333px">Access Softek, Inc</span></div>
</font></div>
</div>
</div>
</div>
</blockquote></div><br></div></div>