Re: Program that segfaults with -Ofast

[Christer Solskogen <christer dot solskogen at gmail dot com>]

On 05.01.2018 11:57, Mason wrote:
> On 04/01/2018 17:51, Christer Solskogen wrote:
>
> > Probably because it's about amiberry, which is a uae4arm-rpi fork.
> > And because O3/Ofast probably does some magic.
>
> (IMO, -Ofast is an "unsafe" optimization level: "It enables optimizations
> that are not valid for all standard-compliant programs.")

I'm not sure why the developer chose that. It was not my decision.
> Could you run the crashing binary under valgrind?
> Do you get interesting results?

==7821== Conditional jump or move depends on uninitialised value(s)
==7821==    at 0x29478C: read_kickstart (memory.cpp:581)
==7821==    by 0x29478C: read_kickstart_version(uae_prefs*) 
(memory.cpp:1151)
==7821==    by 0x28E243: fixup_prefs(uae_prefs*, bool) (main.cpp:195)
==7821==    by 0x1AB34F: cfgfile_load(uae_prefs*, char const*, int*, 
int, int) (cfgfile.cpp:3969)
==7821==    by 0x2F593B: target_cfgfile_load(uae_prefs*, char const*, 
int, int) (amiberry.cpp:527)
==7821==    by 0x28F927: parse_cmdline_and_init_file (main.cpp:608)
==7821==    by 0x28F927: real_main2 (main.cpp:686)
==7821==    by 0x28F927: real_main(int, char**) (main.cpp:784)
==7821==    by 0x151D97: main (amiberry.cpp:920)
==7821==
==7821== Conditional jump or move depends on uninitialised value(s)
==7821==    at 0x29479C: read_kickstart (memory.cpp:584)
==7821==    by 0x29479C: read_kickstart_version(uae_prefs*) 
(memory.cpp:1151)
==7821==    by 0x28E243: fixup_prefs(uae_prefs*, bool) (main.cpp:195)
==7821==    by 0x1AB34F: cfgfile_load(uae_prefs*, char const*, int*, 
int, int) (cfgfile.cpp:3969)
==7821==    by 0x2F593B: target_cfgfile_load(uae_prefs*, char const*, 
int, int) (amiberry.cpp:527)
==7821==    by 0x28F927: parse_cmdline_and_init_file (main.cpp:608)
==7821==    by 0x28F927: real_main2 (main.cpp:686)
==7821==    by 0x28F927: real_main(int, char**) (main.cpp:784)
==7821==    by 0x151D97: main (amiberry.cpp:920)
==7821==
==7821== Invalid read of size 4
==7821==    at 0x28E2AC: fixup_prefs(uae_prefs*, bool) (main.cpp:207)
==7821==    by 0x1AB34F: cfgfile_load(uae_prefs*, char const*, int*, 
int, int) (cfgfile.cpp:3969)
==7821==    by 0x2F593B: target_cfgfile_load(uae_prefs*, char const*, 
int, int) (amiberry.cpp:527)
==7821==    by 0x28F927: parse_cmdline_and_init_file (main.cpp:608)
==7821==    by 0x28F927: real_main2 (main.cpp:686)
==7821==    by 0x28F927: real_main(int, char**) (main.cpp:784)
==7821==    by 0x151D97: main (amiberry.cpp:920)
==7821==  Address 0x29bb8 is not stack'd, malloc'd or (recently) free'd



> I cloned the repo, and looked at src/memory.cpp
> read_kickstart() is static, and is never called through a function pointer.

Ok. I really don't know what that means, and if it's a good or a bad 
thing. I have some knowledge about C, and very little (close to nothing) 
about C++.


> $ git grep -n read_kickstart
> src/memory.cpp:424:static int read_kickstart (struct zfile *f, uae_u8 *mem, int size, int dochecksum, int noalias)
> src/memory.cpp:540:      read_kickstart (f, extendedkickmem_bank.baseaddr, extendedkickmem_bank.allocated, 0, 1);
> src/memory.cpp:613:     read_kickstart (f, extendedkickmem_bank.baseaddr, ROM_SIZE_512, 0, 1);
> src/memory.cpp:617:     read_kickstart (f, kickmem_bank.baseaddr, ROM_SIZE_512, 1, 0);
> src/memory.cpp:679:       size = read_kickstart (f, kickmem_bank.baseaddr, maxsize, 1, 0);
> src/memory.cpp:690:         read_kickstart (f, extendedkickmem_bank.baseaddr, extendedkickmem_bank.allocated, 0, 1);
> src/memory.cpp:697:         read_kickstart (f, extendedkickmem2_bank.baseaddr, ROM_SIZE_512, 0, 1);
> src/memory.cpp:699:         read_kickstart (f, extendedkickmem2_bank.baseaddr + ROM_SIZE_512, ROM_SIZE_512, 0, 1);
>
> What is weird is that every call to read_kickstart() has the 'size' argument
> either ROM_SIZE_512, or ROM_SIZE_256, or ROM_SIZE_512 * 2. It's never 32 like
> in the sanitizer's stack dump...

Is that something that O3/Ofast can do?

--
chs