[9fans] compiling xen3 with more recent pc kernel

[9fans] compiling xen3 with more recent pc kernel

[Skip Tavakkolian]

i had to make the following changes to the kernel for xen3 to match up
to recent labs pc kernel and to compile successfully.  can anyone verify
the changes are correct?  (i could to a smoke test on EC2, but don't
feel adventurous at the moment)

thanks,

% yesterday -D mem.h fns.h trap.c
diff -n /n/dump/2015/0621/sys/src/9/xen3/mem.h /sys/src/9/xen3/mem.h
/n/dump/2015/0621/sys/src/9/xen3/mem.h:15,16 c /sys/src/9/xen3/mem.h:15,16
< #define	ROUND(s, sz)	(((s)+((sz)-1))&~((sz)-1))
< #define	PGROUND(s)	ROUND(s, BY2PG)
---
> /* #define	ROUND(s, sz)	(((s)+((sz)-1))&~((sz)-1))	*/
> /* #define	PGROUND(s)	ROUND(s, BY2PG)		*/
/n/dump/2015/0621/sys/src/9/xen3/mem.h:51 a /sys/src/9/xen3/mem.h:52
> #define	UTROUND(t)	ROUNDUP((t), BY2PG)
diff -n /n/dump/2015/0621/sys/src/9/xen3/fns.h /sys/src/9/xen3/fns.h
/n/dump/2015/0621/sys/src/9/xen3/fns.h:100 a /sys/src/9/xen3/fns.h:101
> void	validalign(uintptr, unsigned);
diff -n /n/dump/2015/0621/sys/src/9/xen3/trap.c /sys/src/9/xen3/trap.c
/n/dump/2015/0621/sys/src/9/xen3/trap.c:949 a /sys/src/9/xen3/trap.c:950,954
> void
> validalign(uintptr addr, unsigned align)
> {
> }
>

Re: [9fans] compiling xen3 with more recent pc kernel

[David du Colombier]

> i had to make the following changes to the kernel for xen3 to match up
> to recent labs pc kernel and to compile successfully.  can anyone
> verify the changes are correct?  (i could to a smoke test on EC2, but
> don't feel adventurous at the moment)

The evenaddr function in trap.c should be replaced by validalign:

void
validalign(uintptr addr, unsigned align)
{
	/*
	 * Plan 9 is a 32-bit O/S, and the hardware it runs on
	 * does not usually have instructions which move 64-bit
	 * quantities directly, synthesizing the operations
	 * with 32-bit move instructions. Therefore, the compiler
	 * (and hardware) usually only enforce 32-bit alignment,
	 * if at all.
	 *
	 * Take this out if the architecture warrants it.
	 */
	if(align == sizeof(vlong))
		align = sizeof(long);

	/*
	 * Check align is a power of 2, then addr alignment.
	 */
	if((align != 0 && !(align & (align-1))) && !(addr & (align-1)))
		return;
	postnote(up, 1, "sys: odd address", NDebug);
	error(Ebadarg);
	/*NOTREACHED*/
}

This is a change done by Jim McKie on 2014-05-15.

--
David du Colombier

Re: [9fans] compiling xen3 with more recent pc kernel

[Skip Tavakkolian]

thanks.

i saw the note and was wondering if for xen3 it could be
stubbed out.

>> i had to make the following changes to the kernel for xen3 to match up
>> to recent labs pc kernel and to compile successfully.  can anyone
>> verify the changes are correct?  (i could to a smoke test on EC2, but
>> don't feel adventurous at the moment)
>
> The evenaddr function in trap.c should be replaced by validalign:
>
> void
> validalign(uintptr addr, unsigned align)
> {
> 	/*
> 	 * Plan 9 is a 32-bit O/S, and the hardware it runs on
> 	 * does not usually have instructions which move 64-bit
> 	 * quantities directly, synthesizing the operations
> 	 * with 32-bit move instructions. Therefore, the compiler
> 	 * (and hardware) usually only enforce 32-bit alignment,
> 	 * if at all.
> 	 *
> 	 * Take this out if the architecture warrants it.
> 	 */
> 	if(align == sizeof(vlong))
> 		align = sizeof(long);
>
> 	/*
> 	 * Check align is a power of 2, then addr alignment.
> 	 */
> 	if((align != 0 && !(align & (align-1))) && !(addr & (align-1)))
> 		return;
> 	postnote(up, 1, "sys: odd address", NDebug);
> 	error(Ebadarg);
> 	/*NOTREACHED*/
> }
>
> This is a change done by Jim McKie on 2014-05-15.
>
> --
> David du Colombier