On Sun, Dec 9, 2018 at 7:25 AM Andy Shevchenko
<andy.shevchenko@xxxxxxxxx> wrote:
>
> Cc: Dan (Dan, it seems in the file mentioned in the bug the only
> change between affected and non-affected verions is yours)
>
> Thanks for the report!
Thanks for forwarding, can you give the attached patch a try to get
more information?
We should remove the BUG_ON() regardless of the outcome,
reserve_memtype() failures should not bring down the system.
From 5be0556e589c21b3b931f95c1bf3d6a378c07caf Mon Sep 17 00:00:00 2001
From: Dan Williams <dan.j.williams@xxxxxxxxx>
Date: Sun, 9 Dec 2018 09:09:50 -0800
Subject: [PATCH] Debug physical address sanitization conflict.
---
arch/x86/mm/pat.c | 14 +++++++++++---
1 file changed, 11 insertions(+), 3 deletions(-)
diff --git a/arch/x86/mm/pat.c b/arch/x86/mm/pat.c
index 08013524fba1..225d89e24e38 100644
--- a/arch/x86/mm/pat.c
+++ b/arch/x86/mm/pat.c
@@ -540,13 +540,21 @@ int reserve_memtype(u64 start, u64 end, enum page_cache_mode req_type,
enum page_cache_mode *new_type)
{
struct memtype *new;
+ u64 s_start, s_end;
enum page_cache_mode actual_type;
int is_range_ram;
int err = 0;
- start = sanitize_phys(start);
- end = sanitize_phys(end);
- BUG_ON(start >= end); /* end is exclusive */
+ s_start = sanitize_phys(start);
+ s_end = sanitize_phys(end);
+ if (s_start >= s_end) {
+ pr_err("%s failed [mem %#010Lx-%#010Lx], [s_mem %#010Lx-%#010Lx], type: %s\n",
+ __func__, start, end - 1, s_start, s_end - 1,
+ cattr_name(req_type));
+ return -EINVAL;
+ }
+ start = s_start;
+ end = s_end;
if (!pat_enabled()) {
/* This is identical to page table setting without PAT */
--
2.17.2
