Re: Address space limit?

[email protected] (Christos Zoulas) Thu, 6 Jun 2019 20:56:51 -0000 (UTC)

Newsgroups	gmane.os.netbsd.ports.m68k,gmane.os.netbsd.ports.hp300
Message-ID	<[email protected]>

In article <[email protected]>,
Izumi Tsutsui  <[email protected]> wrote:
>> I've been told that this might be an issue with the m68k pmap code
>> allocating only one page for the page table, and assuming the page
>> table is contiguous. Any suggestions on a workaround?
>
>One trivial workaround is to use 8KB/page settings:
>
>Index: include/param.h
>===================================================================
>RCS file: /cvsroot/src/sys/arch/hp300/include/param.h,v
>retrieving revision 1.53
>diff -u -p -d -r1.53 param.h
>--- include/param.h	10 Feb 2012 17:35:49 -0000	1.53
>+++ include/param.h	10 Jan 2019 15:46:06 -0000
>@@ -47,7 +47,7 @@
> #define	_MACHINE	hp300
> #define	MACHINE		"hp300"
> 
>-#define	PGSHIFT		12		/* LOG2(NBPG) */
>+#define	PGSHIFT		13		/* LOG2(NBPG) */
> #define	KERNBASE	0x00000000	/* start of kernel virtual */
> 
> #define	UPAGES		2		/* pages of u-area */
>
>

1. I think this might break existing binaries that use jemalloc because:

bool
pages_boot(void) {
        os_page = os_page_detect();
        if (os_page > PAGE) {
		boom.

2. There is a comment in vmparam.h about 4K pages..

christos