Re: [PATCH] Fix acpi_dmar_zap/reinstate() (fixes S3 regression)

[Tomasz Wroblewski <tomasz.wroblewski@citrix.com>]

On 22/01/13 13:58, Jan Beulich wrote:

Jan, thanks for your comments

> I recognize the need of fixing this, but not this way. We have
> ioremap() now, and hence the patch could be using this,
> without re-running the whole acpi_get_table(), but just using
> the stored physical address of the table (retrieving of which
> would be the only real code addition needed here).
>
> For the older trees with non-functional ioremap(), I'd prefer
> simply adding the table range to the 1:1 mapping (thus making
> ioremap() work for that range, should use of that be needed;
> if not needed, that's certainly worth considering this even for
> -unstable).
>
> Also, with your change not even attempting to fix the underlying
> issue of the ACPI code storing a pointer to the mapped table in
> acpi_gbl_root_table_list.tables[].pointer, I can't even see how
> your patch is supposed to work. I'd expect the stored pointer to
> get re-used by acpi_get_table()/acpi_tb_verify_table(), and hence
> this shouldn't win you anything.
>    
It works because the existing code in acpi_get_table actually sets 
acpi_gbl_root_table_list.tables[i].pointer = NULL every call anyway 
(after setting *out_table pointer).

I'll have a go at trying this with ioremap() and retrievable table's 
physical pointer instead.