Hi Tyler,

On 20/01/17 20:58, Baicar, Tyler wrote:

On 1/19/2017 10:57 AM, James Morse wrote:

quoted

On 18/01/17 23:51, Baicar, Tyler wrote:

quoted

On 1/18/2017 7:50 AM, James Morse wrote:

quoted

On 12/01/17 18:15, Tyler Baicar wrote:

quoted

diff --git a/drivers/acpi/apei/ghes.c b/drivers/acpi/apei/ghes.c

quoted

There are two other things that need changing to make the in_nmi() code path
work on arm64.
Always reserve the virtual-address-space forcing GHES_IOREMAP_PAGES to be 2
regardless of CONFIG_HAVE_ACPI_APEI_NMI. This is almost revert of
594c7255dce7a13cac50cf2470cc56e2c3b0494e (but that did a few other things too).

Looks simple enough, should I force it to 2 in all cases, or add a check for
CONFIG_HAVE_ACPI_APEI_SEA
similar to the check for CONFIG_HAVE_ACPI_APEI_NMI?

Its just address space not actual memory it is reserving right? I think just
reserve two pages all the time to save eye-sore #ifdefs!

quoted

We also need to fix ghes_ioremap_pfn_nmi() to use arch_apei_get_mem_attribute()
and not assume PAGE_KERNEL.

So just change the call to ioremap_page_range to:

ioremap_page_range(vaddr, vaddr + PAGE_SIZE, pfn << PAGE_SHIFT,
arch_apei_get_mem_attribute());

(you need to give arch_apei_get_mem_attribute() the address...) copying whatever
ghes_ioremap_pfn_irq() does a few lines down is probably best.


Thanks,

James