On 02/03/17 15:08, Vijay Kilari wrote:
On Thu, Mar 2, 2017 at 8:18 PM, Julien Grall <julien.gr...@arm.com> wrote:
Hello Vijay,
On 02/03/17 12:25, Vijay Kilari wrote:
On Mon, Feb 20, 2017 at 11:35 PM, Julien Grall <julien.gr...@arm.com>
wrote:
On 09/02/17 15:56, vijay.kil...@gmail.com wrote:
[...]
Rather than reimplementing the wheel, it might be better to hook the
parsing
in bootfdt.c. This would avoid an extra parsing of the device-tree,
duplicate the code and expose primitive for early DT parsing.
The process_memory_node() is called only if EFI_BOOT is not enabled. So
hooking might not work.
This series adds this change (see patch #5) and the code is not set in
stone.
I have not added process_memory_node() in the patch #5, I have only
introduced the check.
That's exactly what I said...
We should rework the code when it makes sense rather than trying to find a
more convolute way.
I see two restrictions.
1) If we make a hook from process_memory_node(). This is called much
earlier than
numa_init().
2) We cannot move numa_init() before setup_mm().
I didn't ask to move numa_init earlier but moving the parsing earlier.
AFAICT, it does not require to allocate memory nor rely on anything
initialized big in numa_init but nodes_clear(...).
Regards,
--
Julien Grall
_______________________________________________
Xen-devel mailing list
Xen-devel@lists.xen.org
https://lists.xen.org/xen-devel
