summaryrefslogtreecommitdiffstats
path: root/gnu/llvm/lib/CodeGen
AgeCommit message (Collapse)AuthorFilesLines
2020-05-02Don't make an illegal adde. Avoids fatal error on PowerPC.gkoehler1-1/+3
When the DAG truncates an ISD::ADDE node, DAGCombiner may optimize it by making an adde with smaller operands. PowerPC has i1 registers, and may truncate an i32 adde to i1, but an i1 adde is not legal for PowerPC, and the legalize-ops phase can't fix it. This was causing "fatal error: error in backend: Cannot select..." cwen@ reported the error ok mortimer@ kettenis@ deraadt@
2019-11-09Move the hashed __retguard_* symbols into individual sections and markguenther1-3/+5
them as COMDATs so that the linker can individually discard them, instead of just ignoring duplicate symbols but keep the (duplicate) space. On amd64, this reduces the size of the kernel OPENBSD_RANDOM segment by 82% and the libc OPENBSD_RANDOM segment by 15%. A port that tb@ is working on experienced a 97.3% reduction...which let it actually run. ok mortimer@ deraadt@
2019-09-01Import LLVM 8.0.1 release including clang, lld and lldb.patrick1-0/+6
2019-08-12Ensure stack protector slot access is resolved by PEI.mortimer1-0/+4
On arm64, arm, and ppc it is possible that a large stack frame will cause the stack protector slot to be reallocated at the wrong end of the frame. Noticed by tj@. ok patrick@.
2019-06-23Merge LLVM 8.0.0 release.patrick12-717/+283
Prepared with help from jsg@ and mortimer@ Tested on amd64 by bcallah@, krw@, naddy@ Tested on arm64 by patrick@ Tested on macppc by kettenis@ Tested on octeon by visa@ Tested on sparc64 by claudio@
2019-06-23Import LLVM 8.0.0 release including clang, lld and lldb.patrick163-7676/+16786
2019-05-31Re-enable RETGUARD leaf function optimization for arm64.mortimer1-1/+14
It turns out MachineFrameInfo.hasCalls() is unreliable, because it is up to the backends to update this information whenever they add calls to a function, and this does not always happen. ok kettenis@
2019-03-31Do not store the retguard cookie in frame in leaf functions if possible.mortimer1-6/+45
Makes things slightly faster and also improves security in these functions, since the retguard cookie can't leak via the stack. ok deraadt@
2019-01-27Tedu files that got removed in LLVM 7.0.1.patrick3-1309/+0
2019-01-27Merge LLVM 7.0.1 release.patrick6-267/+490
With fixes from mortimer@ (thanks!) Tested by many, especially naddy@ (thanks!)
2019-01-27Import LLVM 7.0.1 release including clang, lld and lldb.patrick231-9780/+20811
2018-08-12Refactor retguard to make adding additional arches easier.mortimer2-20/+189
2018-06-22Emit trap alignment between basic blocks that are unrechable viamortimer1-3/+7
fallthrough. Avoids unnecessary jmp instructions in the middle of functions and makes disassembly nicer to read. ok guenther@ mlarkin@ deraadt@
2018-06-12Put the new retguard symbols in their own section,guenther1-1/+1
'.openbsd.randomdata.retguard', to make them easier to work with in the kernel hibernate code. ok mortimer@ deraadt@
2018-06-06Add RETGUARD to clang for amd64. This security mechanism uses per-functionmortimer4-2/+102
random cookies to protect access to function return instructions, with the effect that the integrity of the return address is protected, and function return instructions are harder to use in ROP gadgets. On function entry the return address is combined with a per-function random cookie and stored in the stack frame. The integrity of this value is verified before function return, and if this check fails, the program aborts. In this way RETGUARD is an improved stack protector, since the cookies are per-function. The verification routine is constructed such that the binary space immediately before each ret instruction is padded with int03 instructions, which makes these return instructions difficult to use in ROP gadgets. In the kernel, this has the effect of removing approximately 50% of total ROP gadgets, and 15% of unique ROP gadgets compared to the 6.3 release kernel. Function epilogues are essentially gadget free, leaving only the polymorphic gadgets that result from jumping into the instruction stream partway through other instructions. Work to remove these gadgets will continue through other mechanisms. Remaining work includes adding this mechanism to assembly routines, which must be done by hand. Many thanks to all those who helped test and provide feedback, especially deaadt, tb, espie and naddy. ok deraadt@
2018-04-06Tedu files that got removed in LLVM 6.0.0.patrick4-2500/+0
2018-04-06Merge LLVM 6.0.0 release.patrick3-631/+317
2018-04-06Import LLVM 6.0.1 release including clang, lld and lldb.patrick236-9992/+18511
"where is the kaboom?" deraadt@
2017-12-24Import LLVM 5.0.1 release including clang, lld and lldb.patrick6-15/+47
2017-10-04Tedu files that got removed in LLVM 5.0.0.patrick1-33/+0
2017-10-04Merge LLVM 5.0.0 release.patrick3-169/+442
2017-10-04Import LLVM 5.0.0 release including clang, lld and lldb.patrick220-10757/+23825
2017-08-01Use int3 trap padding between functions instead of trapsleds with a leading jump.mortimer1-1/+28
ok deraadt@
2017-03-14Import LLVM 4.0.0 release including clang and lld.patrick15-142/+213
2017-01-24Merge LLVM 4.0.0 rc1patrick3-81/+275
2017-01-24Import LLVM 4.0.0 rc1 including clang and lld to help the currentpatrick200-7540/+19533
development effort on OpenBSD/arm64.
2017-01-20Restore setting the visibility of __guard_local to hidden for betterguenther1-1/+4
code generation. Use dyn_case_or_null instead of a static cast to solve the crashes in the previous code. ok stefan@ kettenis@
2017-01-14Merge LLVM 3.9.1patrick16-1805/+90
2017-01-14Import LLVM 3.9.1 including clang and lld.patrick199-11619/+26243
2016-09-07Backport https://reviews.llvm.org/rL279449 from upstreamstefan1-4/+2
ok pascal@ Original commit message: [SSP] Do not set __guard_local to hidden for OpenBSD SSP guard_local is defined as long on OpenBSD. If the source file contains a definition of guard_local, it mismatches with the int8 pointer type used in LLVM. In that case, Module::getOrInsertGlobal() returns a cast operation instead of a GlobalVariable. Trying to set the visibility on the cast operation leads to random segfaults (seen when compiling the OpenBSD kernel, which also runs with stack protection). In the kernel, the hidden attribute does not matter. For userspace code, guard_local is defined as hidden in the startup code. If a program re-defines guard_local, the definition from the startup code will either win or the linker complains about multiple definitions (depending on whether the re-defined __guard_local is placed in the common segment or not). It also matches what gcc on OpenBSD does.
2016-09-03Use the space freed up by sparc and zaurus to import LLVM.pascal232-0/+172476
ok hackroom@
Copyright © 1996 – 2026 Jason A. Donenfeld. All Rights Reverse Engineered.