Re: [PATCH 01/35] lmb: prepare x86 to use lmb to replace early_res

[Benjamin Herrenschmidt <benh@kernel.crashing.org>]

On Thu, 2010-05-13 at 17:19 -0700, Yinghai Lu wrote:
> 1. expose lmb_debug
> 2. expose lmb_reserved_init_regions
> 3. expose lmb_add_region
> 4. prection for include linux/lmb.h in mm/page_alloc.c and mm/bootmem.c
> 5. lmb_find_base() should return LMB_ERROR in one failing path.
>    (this one cost me 3 hours !)
> 6. move LMB_ERROR to lmb.h

Oh well, let's start somewhere...

> Signed-off-by: Yinghai Lu <yinghai@kernel.org>
> ---
>  include/linux/lmb.h |    4 ++++
>  lib/lmb.c           |   21 +++++++++------------
>  2 files changed, 13 insertions(+), 12 deletions(-)
> 
> diff --git a/include/linux/lmb.h b/include/linux/lmb.h
> index 6f8c4bd..7987766 100644
> --- a/include/linux/lmb.h
> +++ b/include/linux/lmb.h
> @@ -19,6 +19,7 @@
>  #include <asm/lmb.h>
>  
>  #define INIT_LMB_REGIONS 128
> +#define LMB_ERROR	(~(phys_addr_t)0)

Ok so this was meant to remain internal. You seem to want to expose a
whole lot of LMB internals, I suppose for your new arch/x86/lmb.c and I
really really don't like it.

If we expose LMB_ERROR then all lmb calls that can fail should return
that. However, the API calls all return 0 instead. Changing that means
fixing all callers.

We can't just have a mix bag of result code in stuff that is exposed.

If all you need LMB_ERROR is to expose lmb_find_area() and
lmb_add_region() then make the above __ and export a public variant of
it that returns 0.

But that's not the right approach. The right thing to do I believe is to
instead change LMB to use proper errno.h values.

For things like lmb_add_region(), return then as a negative int. For
things that return a phys_addr_t as well with a proper casting macro
since I -think- we can safely consider that phys addrs in the range
-PAGE_SIZE..-1 can be error codes. Just like we do for PTR_ERR etc...

This should be a separate patch btw.

I'm also not too happy with exposing lmb_add_region(). Why would you
ever need to expose it ? Just call lmb_reserve() if you want to reserve
something. lmb_add_region() is an internal function and has no business
being used outside of the main lmb.c file.

Also:

>  	/* Calculate new doubled size */
>  	old_size = type->max * sizeof(struct lmb_region);
>  	new_size = old_size << 1;
> @@ -206,7 +199,7 @@ static int lmb_double_array(struct lmb_type *type)
>  		new_array = kmalloc(new_size, GFP_KERNEL);
>  		addr = new_array == NULL ? LMB_ERROR : __pa(new_array);
>  	} else
> -		addr = lmb_find_base(new_size, sizeof(phys_addr_t), 0, LMB_ALLOC_ACCESSIBLE);
> +		addr = lmb_find_base(new_size, sizeof(struct lmb_region), 0, LMB_ALLOC_ACCESSIBLE);

Why this change ? Does it need to be aligned to the struct size ? if you
really want that and have a good justification, make this a separate
patch and explain why you are doing that in the changeset comment.

>  	if (addr == LMB_ERROR) {
>  		pr_err("lmb: Failed to double %s array from %ld to %ld entries !\n",
>  		       lmb_type_name(type), type->max, type->max * 2);
> @@ -214,6 +207,10 @@ static int lmb_double_array(struct lmb_type *type)
>  	}
>  	new_array = __va(addr);
>  
> +	if (lmb_debug)
> +		pr_info("lmb: %s array is doubled to %ld at %llx - %llx",
> +			 lmb_type_name(type), type->max * 2, (u64)addr, (u64)addr + new_size);
> +
>  	/* Found space, we now need to move the array over before
>  	 * we add the reserved region since it may be our reserved
>  	 * array itself that is full.
> @@ -249,7 +246,7 @@ extern int __weak lmb_memory_can_coalesce(phys_addr_t addr1, phys_addr_t size1,
>  	return 1;
>  }

Cheers,
Ben.