Re: [PATCH] - Mapping ACPI tables as CACHED

[Milton Miller <miltonm@bga.com>]

On Tue, 14 Dec 2010 at about 23:35:28 -0500 (EST), Len Brown wrote:
> 
> [ 1.444475] LENB: free_initmem() NOT!
> is where free_initmem(); would have been called.
> 
> So the calls to acpi_os_map_memory() at 2.586160, kill the machine.
> 
> My guess is that they are related to the opregion stuff used by i915;
> but there is no reason that acpi_os_map_memory can't be called
> as a result of AML at run-time any-time later.
> 
> BTW. what, exactly does this notation do for us?
> 
> void __iomem *__init_refok acpi_os_map_memory(...) {

A function that returns a void pointer for dereference via iomem
accessors that is not init but may conditionally call init text.

__init_refok says that the code was audited to conditionally
use init vs non-init text and data.

> 
> Also, I see no calls within E820_RAM.
> 
> I see a call at 0.534537 which is not in E820_RAM, ACPI, or NVS,
> but is in a reserved region.
> 
> I'm not sure the concept of checking against E820
> is better than simply calling ioremap_cache() always.
> 
> thanks,

after digging back through the thread for the patch, it would seem
that both __acpi_map_table_permanent and e820_all_mapped would
need to be moved from init to normal text, as this is replacing
the non-init path.  I'll let the maintainers decide if that is the
right solution (vs unconditonally mapping cached, or something else).

milton