Re: ppc elf_map breakage with MAP_FIXED_NOREPLACE

[Michal Hocko]

Did you have any chance to test the following hack, Anshuman?

On Thu 01-02-18 14:48:29, Michal Hocko wrote:
[...]
> Just a blind shot... Does the following make any difference?
> ---
> diff --git a/fs/binfmt_elf.c b/fs/binfmt_elf.c
> index 021fe78998ea..04b24d00c911 100644
> --- a/fs/binfmt_elf.c
> +++ b/fs/binfmt_elf.c
> @@ -895,7 +895,7 @@ static int load_elf_binary(struct linux_binprm *bprm)
>  	   the correct location in memory. */
>  	for(i = 0, elf_ppnt = elf_phdata;
>  	    i < loc->elf_ex.e_phnum; i++, elf_ppnt++) {
> -		int elf_prot = 0, elf_flags;
> +		int elf_prot = 0, elf_flags, elf_fixed = MAP_FIXED_NOREPLACE;
>  		unsigned long k, vaddr;
>  		unsigned long total_size = 0;
>  
> @@ -927,6 +927,7 @@ static int load_elf_binary(struct linux_binprm *bprm)
>  					 */
>  				}
>  			}
> +			elf_fixed = MAP_FIXED;
>  		}
>  
>  		if (elf_ppnt->p_flags & PF_R)
> @@ -944,7 +945,7 @@ static int load_elf_binary(struct linux_binprm *bprm)
>  		 * the ET_DYN load_addr calculations, proceed normally.
>  		 */
>  		if (loc->elf_ex.e_type == ET_EXEC || load_addr_set) {
> -			elf_flags |= MAP_FIXED_NOREPLACE;
> +			elf_flags |= elf_fixed;
>  		} else if (loc->elf_ex.e_type == ET_DYN) {
>  			/*
>  			 * This logic is run once for the first LOAD Program
> @@ -980,7 +981,7 @@ static int load_elf_binary(struct linux_binprm *bprm)
>  				load_bias = ELF_ET_DYN_BASE;
>  				if (current->flags & PF_RANDOMIZE)
>  					load_bias += arch_mmap_rnd();
> -				elf_flags |= MAP_FIXED_NOREPLACE;
> +				elf_flags |= elf_fixed;
>  			} else
>  				load_bias = 0;
>  
-- 
Michal Hocko
SUSE Labs