r314262 - Emit section information for extern variables.
Keane, Erich via cfe-commits
cfe-commits at lists.llvm.org
Wed Oct 4 13:34:22 PDT 2017
Elizabeth pointed out to me separately that the “isDeclarationADefinition” check ought to have prevented this from happening, so she’s taking a look I believe. The usage proposal I mentioned below might be handy as well.
My fear is that this is a case where the existing functionality in Chromium works ‘by chance’ and isn’t terribly guaranteed by the code generation of either compiler. IF this is the case, it could be a situation where this warning has found a bug.
As I’m not convinced that this is actually a clang bug, I would like to give Elizabeth a chance to poke around and make that determination first before a revert, if that is acceptable to you.
-Erich
From: thakis at google.com [mailto:thakis at google.com] On Behalf Of Nico Weber
Sent: Wednesday, October 4, 2017 1:29 PM
To: Keane, Erich <erich.keane at intel.com>
Cc: Andrews, Elizabeth <elizabeth.andrews at intel.com>; Friedman, Eli <efriedma at codeaurora.org>; cfe-commits <cfe-commits at lists.llvm.org>
Subject: Re: r314262 - Emit section information for extern variables.
All I know about this code that it used to build (and still builds with gcc) and now it doesn't, sorry :-) What that code does seems somewhat questionable, but also somewhat useful, and it feels like the behavior change from this change here for that code might have been unintentional.
Your suggestion sounds reasonable to me.
On Wed, Oct 4, 2017 at 4:18 PM, Keane, Erich <erich.keane at intel.com<mailto:erich.keane at intel.com>> wrote:
I saw that… I don’t see where it prevents the same change from being made in link.h.
As I mentioned, there is a solution to make this Warning less aggressive (in the lib/Sema/SemaDecl.cpp changes, add a condition that Old->isUsed() before the warning. I’m wondering if that solves your issue.
It would result in
extern struct r_debug _r_debug;
struct r_debug _r_debug __attribute__((nocommon, section(".r_debug")));
Compiling, but :
extern struct r_debug _r_debug;
r_debug = something();
struct r_debug _r_debug __attribute__((nocommon, section(".r_debug")));
NOT compiling (which is almost definitely a bug).
From: thakis at google.com<mailto:thakis at google.com> [mailto:thakis at google.com<mailto:thakis at google.com>] On Behalf Of Nico Weber
Sent: Wednesday, October 4, 2017 1:14 PM
To: Keane, Erich <erich.keane at intel.com<mailto:erich.keane at intel.com>>
Cc: Andrews, Elizabeth <elizabeth.andrews at intel.com<mailto:elizabeth.andrews at intel.com>>; Friedman, Eli <efriedma at codeaurora.org<mailto:efriedma at codeaurora.org>>; cfe-commits <cfe-commits at lists.llvm.org<mailto:cfe-commits at lists.llvm.org>>
Subject: Re: r314262 - Emit section information for extern variables.
For why, here's the comment from the code I linked to:
/*
* GDB looks for this symbol name when it cannot find PT_DYNAMIC->DT_DEBUG.
* We don't have a PT_DYNAMIC, so it will find this. Now all we have to do
* is arrange for this space to be filled in with the dynamic linker's
* _r_debug contents after they're initialized. That way, attaching GDB to
* this process or examining its core file will find the PIE we loaded, the
* dynamic linker, and all the shared libraries, making debugging pleasant.
*/
On Wed, Oct 4, 2017 at 4:11 PM, Keane, Erich <erich.keane at intel.com<mailto:erich.keane at intel.com>> wrote:
I’ve added Elizabeth, who is the original patch author. Hopefully she can help out here. Additionally, Eli did the original review, so hopefully he can chime in as well.
I believe the necessity for this warning came out of the discussion on fixing the section behavior. The issue here is that redeclaring with a different ‘section’ can cause some pretty nasty issues, since it isn’t terribly clear what happens if the variable is used in the meantime.
There is 1 change that I can think of that Elizabeth and I discussed, which was to only warn in the case where there was a USAGE between these two redeclarations. I’m not sure that will allow na_cl to compile, however it is my belief that if there IS a usage between link.h:64 and nacl_bootstrap.c:434, that this is a bug in nacl that is simply being uncovered thanks to this new warning.
Is there a good/reasonable reason the source of nacl wants to redeclare this with a different ‘section’?
From: thakis at google.com<mailto:thakis at google.com> [mailto:thakis at google.com<mailto:thakis at google.com>] On Behalf Of Nico Weber
Sent: Wednesday, October 4, 2017 12:59 PM
To: Keane, Erich <erich.keane at intel.com<mailto:erich.keane at intel.com>>
Cc: cfe-commits <cfe-commits at lists.llvm.org<mailto:cfe-commits at lists.llvm.org>>
Subject: Re: r314262 - Emit section information for extern variables.
Hi Erich,
this breaks existing code. NaCl does this:
#include <link.h>
struct r_debug _r_debug __attribute__((nocommon, section(".r_debug")));
(There's a lengthy-ish comment for why in https://cs.chromium.org/chromium/src/native_client/src/trusted/service_runtime/linux/nacl_bootstrap.c?q=nacl_bootstrap.c&sq=package:chromium&dr&l=424)
link.h in e.g. the debian jessie sysroot says:
extern struct r_debug _r_debug;
After this change, clang complains:
../../native_client/src/trusted/service_runtime/linux/nacl_bootstrap.c:434:16: error: section attribute is specified on redeclared variable [-Werror,-Wsection]
struct r_debug _r_debug __attribute__((nocommon, section(".r_debug")));
^
../../build/linux/debian_jessie_amd64-sysroot/usr/include/link.h:67:23: note: previous declaration is here
extern struct r_debug _r_debug;
This code used to work in clang, and continues to work in gcc. So this patch probably isn't quite the right approach. Ideas?
On Tue, Sep 26, 2017 at 7:42 PM, Erich Keane via cfe-commits <cfe-commits at lists.llvm.org<mailto:cfe-commits at lists.llvm.org>> wrote:
Author: erichkeane
Date: Tue Sep 26 16:42:34 2017
New Revision: 314262
URL: http://llvm.org/viewvc/llvm-project?rev=314262&view=rev
Log:
Emit section information for extern variables.
Currently, if _attribute_((section())) is used for extern variables,
section information is not emitted in generated IR when the variables are used.
This is expected since sections are not generated for external linkage objects.
However NiosII requires this information as it uses special GP-relative accesses
for any objects that use attribute section (.sdata). GCC keeps this attribute in
middle-end.
This change emits the section information for all targets.
Patch By: Elizabeth Andrews
Differential Revision:https://reviews.llvm.org/D36487
Modified:
cfe/trunk/include/clang/Basic/DiagnosticSemaKinds.td
cfe/trunk/lib/CodeGen/CodeGenModule.cpp
cfe/trunk/lib/Sema/SemaDecl.cpp
cfe/trunk/test/Sema/attr-section.c
Modified: cfe/trunk/include/clang/Basic/DiagnosticSemaKinds.td
URL: http://llvm.org/viewvc/llvm-project/cfe/trunk/include/clang/Basic/DiagnosticSemaKinds.td?rev=314262&r1=314261&r2=314262&view=diff
==============================================================================
--- cfe/trunk/include/clang/Basic/DiagnosticSemaKinds.td (original)
+++ cfe/trunk/include/clang/Basic/DiagnosticSemaKinds.td Tue Sep 26 16:42:34 2017
@@ -2620,6 +2620,8 @@ def err_attribute_section_invalid_for_ta
"argument to 'section' attribute is not valid for this target: %0">;
def warn_mismatched_section : Warning<
"section does not match previous declaration">, InGroup<Section>;
+def warn_attribute_section_on_redeclaration : Warning<
+ "section attribute is specified on redeclared variable">, InGroup<Section>;
def err_anonymous_property: Error<
"anonymous property is not supported">;
Modified: cfe/trunk/lib/CodeGen/CodeGenModule.cpp
URL: http://llvm.org/viewvc/llvm-project/cfe/trunk/lib/CodeGen/CodeGenModule.cpp?rev=314262&r1=314261&r2=314262&view=diff
==============================================================================
--- cfe/trunk/lib/CodeGen/CodeGenModule.cpp (original)
+++ cfe/trunk/lib/CodeGen/CodeGenModule.cpp Tue Sep 26 16:42:34 2017
@@ -2432,6 +2432,12 @@ CodeGenModule::GetOrCreateLLVMGlobal(Str
EmitGlobalVarDefinition(D);
}
+ // Emit section information for extern variables.
+ if (D->hasExternalStorage()) {
+ if (const SectionAttr *SA = D->getAttr<SectionAttr>())
+ GV->setSection(SA->getName());
+ }
+
// Handle XCore specific ABI requirements.
if (getTriple().getArch() == llvm::Triple::xcore &&
D->getLanguageLinkage() == CLanguageLinkage &&
Modified: cfe/trunk/lib/Sema/SemaDecl.cpp
URL: http://llvm.org/viewvc/llvm-project/cfe/trunk/lib/Sema/SemaDecl.cpp?rev=314262&r1=314261&r2=314262&view=diff
==============================================================================
--- cfe/trunk/lib/Sema/SemaDecl.cpp (original)
+++ cfe/trunk/lib/Sema/SemaDecl.cpp Tue Sep 26 16:42:34 2017
@@ -2607,6 +2607,16 @@ void Sema::mergeDeclAttributes(NamedDecl
}
}
+ // This redeclaration adds a section attribute.
+ if (New->hasAttr<SectionAttr>() && !Old->hasAttr<SectionAttr>()) {
+ if (auto *VD = dyn_cast<VarDecl>(New)) {
+ if (VD->isThisDeclarationADefinition() != VarDecl::Definition) {
+ Diag(New->getLocation(), diag::warn_attribute_section_on_redeclaration);
+ Diag(Old->getLocation(), diag::note_previous_declaration);
+ }
+ }
+ }
+
if (!Old->hasAttrs())
return;
Modified: cfe/trunk/test/Sema/attr-section.c
URL: http://llvm.org/viewvc/llvm-project/cfe/trunk/test/Sema/attr-section.c?rev=314262&r1=314261&r2=314262&view=diff
==============================================================================
--- cfe/trunk/test/Sema/attr-section.c (original)
+++ cfe/trunk/test/Sema/attr-section.c Tue Sep 26 16:42:34 2017
@@ -19,3 +19,7 @@ void __attribute__((section("foo,zed")))
void __attribute__((section("bar,zed"))) test2(void) {} // expected-warning {{section does not match previous declaration}}
enum __attribute__((section("NEAR,x"))) e { one }; // expected-error {{'section' attribute only applies to functions, methods, properties, and global variables}}
+
+extern int a; // expected-note {{previous declaration is here}}
+int *b = &a;
+extern int a __attribute__((section("foo,zed"))); // expected-warning {{section attribute is specified on redeclared variable}}
_______________________________________________
cfe-commits mailing list
cfe-commits at lists.llvm.org<mailto:cfe-commits at lists.llvm.org>
http://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.llvm.org/pipermail/cfe-commits/attachments/20171004/3287e201/attachment-0001.html>
More information about the cfe-commits
mailing list