On 12/6/21 00:05, John David Anglin wrote:
> On 2021-12-05 4:00 p.m., John David Anglin wrote:
>>> Does it boot if you remove the __init in front of map_pages?
>> I'll try. I thought of trying it but wasn't sure if map_pages() had to be an init routine or not.
> This appears to fix boot. System booted okay about six times.
Do you have huge pages enabled?
If so you could try this patch (instead of markung map_pages __init):
diff --git a/arch/parisc/mm/init.c b/arch/parisc/mm/init.c
index 1ae31db9988f..a9a510338ced 100644
--- a/arch/parisc/mm/init.c
+++ b/arch/parisc/mm/init.c
@@ -405,7 +405,7 @@ static void __init map_pages(unsigned long start_vaddr,
} else if (!kernel_set_to_readonly) {
/* still initializing, allow writing to RO memory */
prot = PAGE_KERNEL_RWX;
- huge = true;
+ huge = false;
} else if (address >= ro_start) {
/* Code (ro) and Data areas */
prot = (address < ro_end) ?
Maybe the whole kernel is initially mapped via one/multiple huge pages(s), and
then we suddenly turn huge pages partly off. Therefore maybe not all TLB entries for the code of
map_pages() is already loaded?
On the other side the problem is somewhat similar to what I see with patching
the kernel code in the fixmap code on PA1.x CPUs... it showed strange asm statements too..
Helge
