linux-kernel - Re: [PATCH 07/31] lmb: Add reserve_lmb/free

lists.openwall.net		lists / announce owl-users owl-dev john-users john-dev passwdqc-users yescrypt popa3d-users / oss-security kernel-hardening musl sabotage tlsify passwords / crypt-dev xvendor / Bugtraq Full-Disclosure linux-kernel linux-netdev linux-ext4 linux-hardening linux-cve-announce PHC
Open Source and information security mailing list archives
Hash Suite: Windows password security audit tool. GUI, reports in PDF.
[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]
Message-ID: <1269901252.2286.11.camel@concordia>
Date:	Tue, 30 Mar 2010 09:20:52 +1100
From:	Michael Ellerman <michael@...erman.id.au>
To:	Yinghai Lu <yinghai@...nel.org>
Cc:	Ingo Molnar <mingo@...e.hu>, Thomas Gleixner <tglx@...utronix.de>,
	"H. Peter Anvin" <hpa@...or.com>,
	Andrew Morton <akpm@...ux-foundation.org>,
	David Miller <davem@...emloft.net>,
	Benjamin Herrenschmidt <benh@...nel.crashing.org>,
	Linus Torvalds <torvalds@...ux-foundation.org>,
	Johannes Weiner <hannes@...xchg.org>,
	linux-kernel@...r.kernel.org, linux-arch@...r.kernel.org
Subject: Re: [PATCH 07/31] lmb: Add reserve_lmb/free_lmb

On Mon, 2010-03-29 at 09:45 -0700, Yinghai Lu wrote:
> On 03/29/2010 05:22 AM, Michael Ellerman wrote:
> > On Sun, 2010-03-28 at 19:43 -0700, Yinghai Lu wrote:
> >> They will check if the region array is big enough.
> >>
> >> __check_and_double_region_array will try to double the region if that array spare
> >> slots if not big enough.
> >> find_lmb_area() is used to find good postion for new region array.
> >> Old array will be copied to new array.
> >>
> >> Arch code should provide to get_max_mapped, so the new array have accessiable
> >> address
> > ..
> >> diff --git a/mm/lmb.c b/mm/lmb.c
> >> index d5d5dc4..9798458 100644
> >> --- a/mm/lmb.c
> >> +++ b/mm/lmb.c
> >> @@ -551,6 +551,95 @@ int lmb_find(struct lmb_property *res)
> >>  	return -1;
> >>  }
> >>  
> >> +u64 __weak __init get_max_mapped(void)
> >> +{
> >> +	u64 end = max_low_pfn;
> >> +
> >> +	end <<= PAGE_SHIFT;
> >> +
> >> +	return end;
> >> +}
> > 
> > ^ This is (sort of) what lmb.rmo_size represents. So maybe instead of
> > adding this function, we could just say that the arch code needs to set
> > rmo_size up with an appropriate value, and then use that below. Though
> > maybe that's conflating things.
> 
> ok
> 
> will have another patch following this patchset. to use rmo_size replace get_max_mapped()

No don't, Benh's idea was better. Leave rmo_size for now, we can clean
that up later.

We just need a lmb.alloc_limit and a lmb_set_alloc_limit() which arch
code calls when it knows what the alloc limit is (and can call multiple
times during boot). Or maybe it should be called "default_alloc_limit",
but that's getting a bit long winded.

> 
> long __init_lmb lmb_add(u64 base, u64 size)
> {
>         struct lmb_region *_rgn = &lmb.memory;
> 
>         /* On pSeries LPAR systems, the first LMB is our RMO region. */
>         if (base == 0)
>                 lmb.rmo_size = size;
> 
>         return lmb_add_region(_rgn, base, size);
> 
> }
> 
> looks scary.
> maybe later powerpc could used lmb_find and set_lmb_rmo_size in their arch code.

It's not really scary, and it gives you a hint where the code came from
originally :)

We can remove that later though, with some powerpc code to detect the
first memory region before we put it into lmb.

> > ...
> >> +
> >> +void __init add_lmb_memory(u64 start, u64 end)
> >> +{
> >> +	__check_and_double_region_array(&lmb.memory, &lmb_memory_region[0], start, end);
> >> +	lmb_add(start, end - start);
> >> +}
> >> +
> >> +void __init reserve_lmb(u64 start, u64 end, char *name)
> >> +{
> >> +	if (start == end)
> >> +		return;
> >> +
> >> +	if (WARN_ONCE(start > end, "reserve_lmb: wrong range [%#llx, %#llx]\n", start, end))
> >> +		return;
> >> +
> >> +	__check_and_double_region_array(&lmb.reserved, &lmb_reserved_region[0], start, end);
> >> +	lmb_reserve(start, end - start);
> >> +}
> >> +
> >> +void __init free_lmb(u64 start, u64 end)
> >> +{
> >> +	if (start == end)
> >> +		return;
> >> +
> >> +	if (WARN_ONCE(start > end, "free_lmb: wrong range [%#llx, %#llx]\n", start, end))
> >> +		return;
> >> +
> >> +	/* keep punching hole, could run out of slots too */
> >> +	__check_and_double_region_array(&lmb.reserved, &lmb_reserved_region[0], start, end);
> >> +	lmb_free(start, end - start);
> >> +}
> > 
> > Doesn't this mean that if I call lmb_alloc() or lmb_free() too many
> > times then I'll potentially run out of space? So doesn't that
> > essentially break the existing API?
> 
> No, I didn't touch existing API, arches other than x86 should have little change about 
> lmb.memory.region
> lmb.reserved.region
> become pointer from array.

But that's my point. You shouldn't need to touch the existing API, and
you shouldn't need to add a new parallel API. You should just be able to
add the logic for doubling the array in the lmb core, and then everyone
gets dynamically expandable lmb. I don't see any reason why we want to
have two APIs.

> > It seems to me that rather than adding these "special" routines that
> > check for enough space on the way in, instead you should be checking in
> > lmb_add_region() - which is where AFAICS all allocs/frees/reserves
> > eventually end up if they need to insert a new region.
> 
> later i prefer to replace lmb_alloc with find_lmb_area + reserve_lmb.

Why? The existing code has been working for years and is well tested?

cheers



Download attachment "signature.asc" of type "application/pgp-signature" (198 bytes)