Re: jemalloc issues on earm and earmhf

[christos%astron.com@localhost (Christos Zoulas)]

In article <qvd61b$56kh$1%blaine.gmane.org@localhost>,
Christos Zoulas <christos%astron.com@localhost> wrote:
>In article <Pine.NEB.4.64.2001110037550.6444%hlin.zia.io@localhost>,
>John Klos  <john%ziaspace.com@localhost> wrote:
>>> What do you get if you try it on an earmv6hf installation ?
>>
>>NetBSD-9 earmv6hf: hw.pagesize = 8192
>>
>>It's 8192 on NetBSD-8 earmv4, earmv6hf, earmv7hf
>>
>>It's 8192 on NetBSD-9 earmv6hf, earmv7hf
>>
>>It's 4096 on NetBSD-9 and -current aarch64
>>
>>No idea why it's different on earm (earm, earmv4, earmv5) NetBSD-9 and 
>>-current. An earm chroot on an earmv6hf system gives an endless stream of 
>>"<jemalloc>: Unsupported system page size" when trying to chroot.
>
>The code does:
>
>       os_page = os_page_detect();
>        if (os_page > PAGE) {
>                malloc_write("<jemalloc>: Unsupported system page size\n");
>                if (opt_abort) {
>                        abort();
>                }
>                return true; 
>        }
>        
>
>os_page = 8192 and I presume PAGE = 4096...
>
>From:
>
>#ifndef PGSHIFT
>#if defined(_ARM_ARCH_6)
>#define PGSHIFT         13              /* LOG2(NBPG) */ 
>#else 
>#define PGSHIFT         12              /* LOG2(NBPG) */
>#endif  
>#endif
>
>because the code was not compiled on _ARM_ARCH_6. I will take care of it.

I committed a fix and asked to pullup-9. If we don't want to support 8K
pages in any kernel we can roll it back.

christos