<div dir="ltr"><div dir="ltr">On Thu, 7 Nov 2019 at 10:30, Y Song via cfe-dev <<a href="mailto:cfe-dev@lists.llvm.org">cfe-dev@lists.llvm.org</a>> wrote:<br></div><div class="gmail_quote"><blockquote class="gmail_quote" style="margin:0px 0px 0px 0.8ex;border-left:1px solid rgb(204,204,204);padding-left:1ex">Hi,<br>
<br>
Currently, I am thinking to add address_space attribute support for<br>
x86_64 and BPF to compile linux kernel. The main goal is to get<br>
address_space information into debug_info and then later to dwarf/BTF.<br>
But address_space enforcement itself can also help with better code<br>
for the kernel.<br>
<br>
The RFC patch is here:<br>
<a href="https://reviews.llvm.org/D69393" rel="noreferrer" target="_blank">https://reviews.llvm.org/D69393</a><br>
<br>
During experiments, we found two issues which I would like to get<br>
expert opinion:<br>
<br>
issue 1: address_space attributes on function pointers<br>
=========================================<br>
<br>
clang does not allow address_space attribute for function pointers,<br>
specifically, in clang/lib/Sema/SemaType.cpp,<br>
<br>
// ISO/IEC TR 18037 S5.3 (amending C99 6.7.3): "A function type shall not be<br>
// qualified by an address-space qualifier."<br>
if (Type->isFunctionType()) {<br>
S.Diag(Attr.getLoc(), diag::err_attribute_address_function_type);<br>
Attr.setInvalid();<br>
return;<br>
}<br>
<br>
But linux kernel tries to annotate signal handling function pointer<br>
with address space to indicate it is accessing user space.<br>
<br>
typedef __signalfn_t __user *__sighandler_t;<br>
typedef __restorefn_t __user *__sigrestore_t;<br>
<br>
Such attribute makes sense for linux since indeed the signal handler<br>
code resides in user space and the kernel pointer<br>
merely points to user memory here.<br>
<br>
What is the rationale for this restriction in standard? How we could<br>
make clang tolerate such restrictions for X86/BPF/...?<br>
<br>
issue 2: typeof(*ptr) where ptr has address_space attribute.<br>
<br>
-bash-4.4$ cat t2.c<br>
int test(int __attribute__((address_space(1))) *p) {<br>
#ifdef NO_TYPEOF<br>
int t = *p;<br>
#else<br>
typeof(*p) t = *p;<br>
#endif<br>
return t;<br>
}<br>
-bash-4.4$ clang -c t2.c<br>
t2.c:5:14: error: automatic variable qualified with an address space<br>
typeof(*p) t = *p;<br>
^<br>
1 error generated.<br>
-bash-4.4$ clang -DNO_TYPEOF -c t2.c<br>
-bash-4.4$<br>
<br>
The IR generated without NO_TYPEOF macro:<br>
%p.addr = alloca i32 addrspace(1)*, align 8<br>
%t = alloca i32, align 4<br>
store i32 addrspace(1)* %p, i32 addrspace(1)** %p.addr, align 8<br>
%0 = load i32 addrspace(1)*, i32 addrspace(1)** %p.addr, align 8<br>
%1 = load i32, i32 addrspace(1)* %0, align 4<br>
store i32 %1, i32* %t, align 4<br>
%2 = load i32, i32* %t, align 4<br>
ret i32 %2<br>
<br>
So we can directly load a i32 addrspacee(1)* to an i32. Maybe<br>
typeof(*p) should just i32 and we should not emit the error at all?<br></blockquote><div><br></div><div>I don't think this suggestion has been fully addressed yet. It's true that we could make typeof(*p) remove the address space qualifier, but that would be inconsistent with how typeof(expr) generally behaves (it doesn't remove type qualifiers from the type of its argument), and it can be important to preserve the address space when the result is used to form another type. For example:</div><div><br></div><div>int test(int __attribute__((address_space(1))) *p) {<br></div><div> const typeof(*p) *const_p = p; // should preserve address_space here</div><div><br></div><div>'typeof' is a GNU extension, and generally we want to follow GCC's behavior for GNU extensions, but GCC doesn't implement this attribute (at least not in builds that I've tested), so we have on explicit guidance there. Additionally, it seems wise to keep open the option of specifying an address space on a local variable (even if we don't support it today) for future extensions (such as <a href="https://www.springerprofessional.de/en/scads/6859296">https://www.springerprofessional.de/en/scads/6859296</a>) rather than ignoring an address space attribute on a local variable.</div><div><br></div><div>Can you give an example of code suffering from issue 2? It would help to see how reasonable or unreasonable such code seems to be, and whether this would be best addressed in the compiler or in the code.</div></div></div>