Bug 490509

Summary: GCC register allocation wrongly using ebp
Product: [Fedora] Fedora Reporter: Glauber Costa <gcosta>
Component: gccAssignee: Jakub Jelinek <jakub>
Status: CLOSED RAWHIDE QA Contact: Fedora Extras Quality Assurance <extras-qa>
Severity: medium Docs Contact:
Priority: low    
Version: rawhideCC: aoliva, ehabkost, jakub, virt-maint
Target Milestone: ---   
Target Release: ---   
Hardware: All   
OS: Linux   
Whiteboard:
Fixed In Version: Doc Type: Bug Fix
Doc Text:
Story Points: ---
Clone Of: Environment:
Last Closed: 2009-04-02 09:20:39 UTC Type: ---
Regression: --- Mount Type: ---
Documentation: --- CRM:
Verified Versions: Category: ---
oVirt Team: --- RHEL 7.3 requirements from Atomic Host:
Cloudforms Team: --- Target Upstream Version:
Embargoed:
Bug Depends On:    
Bug Blocks: 490512    
Attachments:
Description Flags
Preprocessed gcc output
none
Patch that fixes the bug none

Description Glauber Costa 2009-03-16 18:49:03 UTC
Created attachment 335390 [details]
Preprocessed gcc output

Description of problem:


Version-Release number of selected component (if applicable):

gcc version 4.4.0 20090313 (Red Hat 4.4.0-0.26) (GCC) 


How reproducible:
Always

Steps to Reproduce:
1. Compile qemu on an i386 host (./configure --target-list=i386-softmmu; make)
2. Run
3. See a segfault
  
Actual results:
QEMU segfaults.

Expected results:
QEMU runs. Sorry for not having a smaller test case for this

Additional info:
This works with older versions of gcc, including 4.3. I'm attaching to this BZ a copy of the preprocessed version of the culprit file (op_helper.c)

If you look closely, you'll see:

register struct CPUX86State *env asm("ebp");

Meaning gcc should not touch it, since it is a global var (or touch it and save it later). The generated object code has something like this:

    b1a3:       09 e8                   or     %ebp,%eax
    b1a5:       89 d5                   mov    %edx,%ebp

which seems invalid to me.

Comment 1 Alexandre Oliva 2009-03-16 20:10:17 UTC
Created attachment 335419 [details]
Patch that fixes the bug

Here's a patch that seems to fix the problem.  Still requires further testing.

Comment 2 Jakub Jelinek 2009-03-17 10:14:22 UTC
When tested, are you going to submit it to gcc-patches together with a testcase (if a small testcase is distillable)?  Thanks.

Comment 3 Alexandre Oliva 2009-03-17 18:48:59 UTC
Yup.  I'm not sure about a testcase, though.  Register allocation testcases are too fragile, even for glaring errors like this.

Comment 4 Alexandre Oliva 2009-03-19 01:23:11 UTC
Patch proposed upstream:
http://gcc.gnu.org/ml/gcc-patches/2009-03/msg00891.html

Comment 5 Jakub Jelinek 2009-04-02 09:20:39 UTC
This should be fixed in 4.4.0-0.29 and above.