Raymend Chen has probably forgotten more about programming than I'll ever know, but aren't the first two blah() function examples either missing a } or have a superfluous { after the else?
I had never considered the idea of passing too few register params so I didn't immediately think of the reuse problem. And I had no idea about Itanium's Not-a-thing bit! Always a good read from Raymond Chen.
I had fun exploiting this to detect the falling convention used by some code at runtime - there were two different options depending on OS version; one passed a jnienv* as the first param, the other did not. So if I called it with 0, I could tell which was being used based on whether the first argument was NULL or not. Only used for specific architectures with a defined ABI that behaved this way, of course.
It's not even possible to pass too few arguments to a function in C unless you go out of your way to write bad code.
You can write a function declaration that's inconsistent with its definition in another translation unit. Declaring the function in a shared header file avoids this.
You can use an old-style declaration that doesn't specify what parameters a function expects. Don't do that. Use prototypes.
You can use a cast to convert a function pointer to an incompatible type, and call through the resulting pointer. Don't do that.
You can call a function with no visible declaration if your compiler overly permissive or is operating in pre-C99 mode. Don't do that.
Read the post - not all architectures behave the same!
Itanic had variable-sized register windows, plus extra tag bits for NaT ("not a thing") placeholder values. If you didn't set one of the argument registers the callee might trap in unexpected ways when it touches the register garbage.
Sparc (not a VLIW ISA) also had rotating register windows. But ia64 had a twist on it: the register window size was dynamic and "allocated" by the callee with an alloc instruction
The only other ISA I know of that did something similar was the Am29000
The Am29000 modeled it in an interesting way though:
The register file consisted of 128 global registers but the instruction encoding allowed to specify an "indirect register index" mode where the operand register was computed from the content of gr1 plus an offset. Thus gr1 acted as a "register window stack pointer". I _think_ such a computed register index would then be used to index into a separate register file for locals (and arguments etc) but I'm not sure.
Anybody here is familiar with this quite old ISA?
(I'm really interested in the richness of the CPU design space, the history of which is fascinating)
If you have 29 arguments, I assure that you some of them are on the stack in nearly every architecture in use. Also, certain types as parameters also get passed on the stack (usually types larger than a register, or in C++ code, objects with nontrivial constructors or destructors).
Sure, but he still came up with a 2005 blog post, and attached a 2026 to it. No optimizing C compiler cares for the 2nd arg, when it's a register anyway. And if the 1st is constant, the dead branch is folded away. So the 2nd arg is dead
21 comments:
Raymend Chen has probably forgotten more about programming than I'll ever know, but aren't the first two blah() function examples either missing a } or have a superfluous { after the else?
Yes. And in the second one he has return c; when he meant return b;
Homer nods.
Until C23, you could declare a pointer to a procedure that takes an unspecified amount of any type arguments like this
https://godbolt.org/z/s6e5rnqv9If you compile with -std=c23, both gcc and clang will throw an error ( (*f)() is now the same as (*f)(void) )
You do not need the pointer at all. f() not specifying the arguments has been the case since forever. "Prototypes" (90s) are newer than C.
I had never considered the idea of passing too few register params so I didn't immediately think of the reuse problem. And I had no idea about Itanium's Not-a-thing bit! Always a good read from Raymond Chen.
I had fun exploiting this to detect the falling convention used by some code at runtime - there were two different options depending on OS version; one passed a jnienv* as the first param, the other did not. So if I called it with 0, I could tell which was being used based on whether the first argument was NULL or not. Only used for specific architectures with a defined ABI that behaved this way, of course.
It's not even possible to pass too few arguments to a function in C unless you go out of your way to write bad code.
You can write a function declaration that's inconsistent with its definition in another translation unit. Declaring the function in a shared header file avoids this.
You can use an old-style declaration that doesn't specify what parameters a function expects. Don't do that. Use prototypes.
You can use a cast to convert a function pointer to an incompatible type, and call through the resulting pointer. Don't do that.
You can call a function with no visible declaration if your compiler overly permissive or is operating in pre-C99 mode. Don't do that.
> It's not even possible to pass too few arguments to a function in C unless you go out of your way to write bad code.
This article is exclusively about undefined behaviour. "Bad code" is already baked into the assumptions of the article.
This is a site for intellectual curiosity, not pedantic dissmisal.
Seriously?
I discussed some of the technical issues behind the article. If you disagree with anything I wrote, please say so.
I'm not even saying that the issues discussed in the article aren't useful, just going into how likely they're likely to be encountered in practice.
You could also use inline assembly.
Do you really not ‘pass’ register parameters? How can anyone tell if you didn’t?
Read the post - not all architectures behave the same!
Itanic had variable-sized register windows, plus extra tag bits for NaT ("not a thing") placeholder values. If you didn't set one of the argument registers the callee might trap in unexpected ways when it touches the register garbage.
Heh, it had rotating register files too. VLIW was so weird.
Sparc (not a VLIW ISA) also had rotating register windows. But ia64 had a twist on it: the register window size was dynamic and "allocated" by the callee with an alloc instruction
The only other ISA I know of that did something similar was the Am29000
The Am29000 modeled it in an interesting way though:
The register file consisted of 128 global registers but the instruction encoding allowed to specify an "indirect register index" mode where the operand register was computed from the content of gr1 plus an offset. Thus gr1 acted as a "register window stack pointer". I _think_ such a computed register index would then be used to index into a separate register file for locals (and arguments etc) but I'm not sure.
Anybody here is familiar with this quite old ISA?
(I'm really interested in the richness of the CPU design space, the history of which is fascinating)
Rotating register windows are a distinguishing feature of the Berkeley RISC design, which predates VLIW by at least a decade.
Interesting that some CPUs have a calling convention "built-in"
I regard this yet another unintuitive Itanium quirk that makes it failed.
Of which decade is this post? I cannot think of any modern architecture which still passes args on the stack.
Itanium? Stone age
If you have 29 arguments, I assure that you some of them are on the stack in nearly every architecture in use. Also, certain types as parameters also get passed on the stack (usually types larger than a register, or in C++ code, objects with nontrivial constructors or destructors).
Sure, but he still came up with a 2005 blog post, and attached a 2026 to it. No optimizing C compiler cares for the 2nd arg, when it's a register anyway. And if the 1st is constant, the dead branch is folded away. So the 2nd arg is dead