Re: [PATCH v2 RESEND 05/18] x86, ACPICA: Split acpi_boot_table_init() into two parts.

["Rafael J. Wysocki" <rjw@sisk.pl>]

On Monday, August 05, 2013 11:21:51 AM Tang Chen wrote:
> Hi Rafael,
> 
> On 08/02/2013 09:00 PM, Rafael J. Wysocki wrote:
> ......
> >> This patch splits acpi_boot_table_init() into two steps:
> >> 1. Parse RSDT, which cannot be overrided, and initialize
> >>     acpi_gbl_root_table_list. (step 1 + 2 above)
> >> 2. Install all ACPI tables into acpi_gbl_root_table_list.
> >>     (step 3 + 4 above)
> >>
> >> In later patches, we will do step 1 + 2 earlier.
> >
> > Please note that Linux is not the only user of the code you're modifying, so
> > you need to make it possible to use the existing functions.
> >
> > In particular, acpi_tb_parse_root_table() can't be modified the way you did it,
> > because that would require all of the users of ACPICA to be modified.
> 
> OK, I understand it. Then how about acpi_tb_install_table() ?
> 
> acpi_tb_install_table() is also an ACPICA API. But can we split the
> acpi_initrd_table_override part out ? Like the following:

I'm not sure what you mean.  acpi_tb_install_table() doesn't call
acpi_initrd_table_override() directly.

Do you want to split the acpi_tb_table_override() call out of it?

I'm afraid that still wouldn't be OK.

> 1. Initialize acpi_gbl_root_table_list earlier, and install all tables
>     provided by firmware.
> 2. Find SRAT in initrd. If no overridden SRAT, get the SRAT in 
> acpi_gbl_root_table_list
>     directly. And mark hotpluggable memory. (This the job I want to do.)
> 3. DO acpi_initrd_table_override job.
> 
> Finally it will work like the current kernel. The only difference is:
> Before the patch-set, it try to do override first, and then install 
> firmware tables.
> After the patch-set, it installs firmware tables, and then do the override.

I think I understand what you're trying to achieve and I don't have objections
agaist the goal, but the matter is *how* to do that.

Why don't you do something like this:
(1) Introduce two new functions that will each do part of
    acpi_tb_parse_root_table() such that calling them in sequence, one right
    after the other, will be exactly equivalent to the current
    acpi_tb_parse_root_table().
(2) Redefine acpi_tb_parse_root_table() as a wrapper calling those two new
    function one right after the other.
(3) Make Linux use the two new functions directly instead of calling
    acpi_tb_parse_root_table()?

Then, Linux will use your new functions and won't call acpi_tb_parse_root_table()
at all, but the other existing users of ACPICA may still call it without any
modifications.

Does this make sense to you?

Rafael


-- 
I speak only for myself.
Rafael J. Wysocki, Intel Open Source Technology Center.

--
To unsubscribe, send a message with 'unsubscribe linux-mm' in
the body to majordomo@kvack.org.  For more info on Linux MM,
see: http://www.linux-mm.org/ .
Don't email: <a href=mailto:"dont@kvack.org"> email@kvack.org </a>