similar to: [LLVMdev] making trampolines more portable

I'm not sure whether there are any other cases where using LLVM requires knowledge of the target architecture, but needing it for trampolines is worrying me. I'm putting together things where the program that's generating the LLVM intermediate code doesn't have any way to know what the target architecture will be. Right now I'm using alloca to get a block that is expected

Duncan Sands wrote: > is it important for you to have portable bitcode (i.e. the trampoline > size as some kind of symbolic constant, maybe via a global or an > intrinsic) that works on all targets or would it be enough to have the > LLVM info for the target expose the trampoline size, so when generating > IR for a target X you would query the method for target X and if it

Eli Friedman wrote: > I didn't say we wouldn't accept a patch. It's just that up until now > nobody has particularly cared because trampolines aren't used very > much. Fair enough. If I get to the point where I understand how to implement it cleanly, I'll submit a patch. Best regards, Eric

Attached patches change how dragonegg lowers trampolines (for compatibility with Go). -- Sanjoy Das http://playingwithpointers.com -------------- next part -------------- A non-text attachment was scrubbed... Name: 0001-Fix-the-trampolines-for-Go.patch Type: text/x-diff Size: 5108 bytes Desc: not available URL:

Hi! Attached patches split init.trampoline into adjust.trampoline and init.trampoline, like in gcc. As mentioned in the previous mail, I've not made a documentation patch, since I'm not sure about what the documented semantics of llvm.adjust.trampoline should be. Thanks! -- Sanjoy Das http://playingwithpointers.com -------------- next part -------------- A non-text attachment was

Taking the address of a function inhibits optimisations for that function. Essentially any ABI changes are unavailable if we can't adjust the call site to match. The case of interest here is when a given function is called directly and indirectly, and we don't want the latter to impose a cost on the former. One approach to avoid the ABI constraint cost is to extract/outline the body of an

Hi! Attached set of patches splits llvm.init.trampoline into an "init" phase and an "adjust" phase, as discussed on the "Go on dragonegg" thread. Thanks! -- Sanjoy Das http://playingwithpointers.com -------------- next part -------------- A non-text attachment was scrubbed... Name: 0001-Split-intrinsics-and-DAG-nodes.patch Type: text/x-diff Size: 8808 bytes Desc:

Dear LLVM-Mailing list (again), I still have another beginners question in my mind - thank you for your patience. Again I was playing around, but this time with llvm::Function. In an older question I learned, that I can replace a llvm::Function directly with an address. llvm::Function *function = mainModue->getFunction("puts"); function->replaceAllUsesWith(

Hi Sanjoy, > Attached patches change how dragonegg lowers trampolines (for > compatibility with Go). I think the right approach is to add an llvm.adjust.trampoline intrinsic to LLVM (and change llvm.init.trampoline to not return a result). Then the dragonegg trampoline code will become trivial, and the Go problem will just Go away :) In fact this was how I first did it: two intrinsics,

On 04 Jan 2011, at 08:30, Jakob Stoklund Olesen wrote: > I noticed that we generate code like this for i386 PIC: > > calll L0$pb > L0$pb: > popl %eax > movl %eax, -24(%ebp) ## 4-byte Spill > > I worry that this defeats the return address prediction for returns > in the function because calls and returns no longer are matched. According to benchmarks by

Hi Sanjoy, the first and last patches look good (except that you didn't add any tests for the auto-upgrade functionality). Comments on the other two below. > Attached patches split init.trampoline into adjust.trampoline and > init.trampoline, like in gcc. > > As mentioned in the previous mail, I've not made a documentation > patch, since I'm not sure about what the

Dear Mr. Northover, Thank you for the quick reply. You are correct about the address-mode operands :) . I guess an important detail left out was that the basic block (call it A) that wants to calculate the address of the target stationary trampoline basic block (call it B) will be moved around in memory during run-time. Our earlier solution, before the feature was implemented to move around (A)

Hi Sanjoy, > Attached set of patches splits llvm.init.trampoline into an "init" > phase and an "adjust" phase, as discussed on the "Go on dragonegg" > thread. thanks for doing this. The patches look good, though the decomposition into individual patches is not that great (since things won't always work, in fact not even compile I think, with not all

Hi, > I'm parallelizing loops to be called by pthread. The thread body that I pass > to pthread_create looks like > > define i8* @loop1({ i32*, i32* }* nest %parent_frame, i8* %arg) > parent_frame is pointer to shared variables in original function > > 0x00007f0de11c41f0: mov (%r10),%rax > 0x00007f0de11c41f3: cmpl $0x63,(%rax) > 0x00007f0de11c41f6:

Hi Sam, I think an intrinsic would be ideal, but I couldn't figure out how to replace calls with an arbitrary number of arguments with a call to an intrinsic. In particular, I would like to call a trampoline instead of the original function. The trampoline performs certain actions and then jumps to the original callee. Would I have to call the intrinsic with the original arguments and then

Fellow developers, I'm parallelizing loops to be called by pthread. The thread body that I pass to pthread_create looks like define i8* @loop1({ i32*, i32* }* nest %parent_frame, i8* %arg) parent_frame is pointer to shared variables in original function 0x00007f0de11c41f0: mov (%r10),%rax 0x00007f0de11c41f3: cmpl $0x63,(%rax) 0x00007f0de11c41f6: jg 0x7f0de11c420c

Hi Hayden, Dave's answer covers this pretty well. Neither Orc nor MCJIT currently reason about replacing function bodies. They may let you add duplicate definitions, but how they'll behave if you do that isn't specified in their contracts. They definitely won't replace old definitions unless you provide a custom memory manager that's rigged to lay new definitions down on top

Hi! > I think the right approach is to add an llvm.adjust.trampoline intrinsic > to LLVM (and change llvm.init.trampoline to not return a result). Then the Won't that break previous code that relies on the current llvm.init.trampoline? I can start work on this if that is not a problem. -- Sanjoy Das http://playingwithpointers.com

Hi! Thank you for the detailed review. I've attached a new set of patches. -- Sanjoy Das http://playingwithpointers.com -------------- next part -------------- A non-text attachment was scrubbed... Name: 0001-New-trampoline-intrinsics-for-LLVM.patch Type: text/x-diff Size: 16783 bytes Desc: not available URL:

On Thu, Jun 25, 2020 at 01:04:29PM +0300, Dan Carpenter wrote: > On Wed, Jun 24, 2020 at 12:39:24PM -0700, Kees Cook wrote: > > On Wed, Jun 24, 2020 at 07:51:48PM +0300, Dan Carpenter wrote: > > > In Debian testing the initrd triggers the warning. > > > > > > [ 34.529809] process '/usr/bin/fstype' started with executable stack > > > >