mirror/openwrt
1
0
Fork 1
mirror of https://git.openwrt.org/openwrt/openwrt.git synced 2025-12-10 08:44:39 +01:00
Code Issues Projects Releases Packages Wiki Activity Actions 2

generic: add pending patch fixing resource_size()

Browse source 
Add opending patch fixing resource_size(9 if resource start and end is
set to zero.

This is to indirectly fix a helper rework in ath11k qmi code that cause
firmware load panic.

Link: https://github.com/openwrt/openwrt/pull/20964
Signed-off-by: Christian Marangi <ansuelsmth@gmail.com>
This commit is contained in:
Christian Marangi 2025-12-07 23:24:26 +01:00
parent 5c513d1f65
commit 63e2b17c01
No known key found for this signature in database
GPG key ID: AC001D09ADBFEAD7
1 changed files with 80 additions and 0 deletions
Show stats Download patch file Download diff file Expand all files Collapse all files

80
target/linux/generic/pending-6.12/210-resource-handle-wrong-resource_size-value-on-zero-st.patch Normal file
View file

@ -0,0 +1,80 @@
From 7e1241396c241f9b4fff2ff133806fef4ddd9ecc Mon Sep 17 00:00:00 2001
From: Christian Marangi <ansuelsmth@gmail.com>
Date: Sun, 7 Dec 2025 22:18:11 +0100
Subject: [PATCH] resource: handle wrong resource_size value on zero start/end
resource
Commit 900730dc4705 ("wifi: ath: Use
of_reserved_mem_region_to_resource() for "memory-region"") uncovered a
massive problem with the usage of resource_size() helper.
The reported commit caused a regression with ath11k WiFi firmware
loading and the change was just a simple replacement of duplicate code
with a new helper of_reserved_mem_region_to_resource().
On reworking this, in the commit also a check for the presence of the
node was replaced with resource_size(&res). This was done following the
logic that if the node wasn't present then it's expected that also the
resource_size is zero, mimicking the same if-else logic.
This was also the reason the regression was mostly hard to catch at
first sight as the rework is correctly done given the assumption on the
used helpers.
BUT this is actually not the case. On further inspection on
resource_size() it was found that it NEVER actually returns 0.
Even if the resource value of start and end are 0, the return value of
resource_size() will ALWAYS be 1, resulting in the broken if-else
condition ALWAYS going in the first if condition.
This was simply confirmed by reading the resource_size() logic:
return res->end - res->start + 1;
Given the confusion, also other case of such usage were searched in the
kernel and with great suprise it seems LOTS of place assume
resource_size() should return zero in the context of the resource start
and end set to 0.
Quoting for example comments in drivers/vfio/pci/vfio_pci_core.c:
/*
* The PCI core shouldn't set up a resource with a
* type but zero size. But there may be bugs that
* cause us to do that.
*/
if (!resource_size(res))
goto no_mmap;
It really seems resource_size() was tought with the assumption that
resource struct was always correctly initialized before calling it and
never set to zero.
But across the year this got lost and now there are lots of driver that
assume resource_size() returns 0 if start and end are also 0.
To better handle this and make resource_size() returns correct value in
such case, add a simple check and return 0 if both resource start and
resource end are zero.
Cc: Rob Herring (Arm) <robh@kernel.org>
Cc: stable@vger.kernel.org
Fixes: 1a4e564b7db9 ("resource: add resource_size()")
Signed-off-by: Christian Marangi <ansuelsmth@gmail.com>
---
include/linux/ioport.h | 3 +++
1 file changed, 3 insertions(+)
--- a/include/linux/ioport.h
+++ b/include/linux/ioport.h
@@ -283,6 +283,9 @@ static inline void resource_set_range(st
static inline resource_size_t resource_size(const struct resource *res)
{
+ if (!res->start && !res->end)
+ return 0;
+
return res->end - res->start + 1;
}
static inline unsigned long resource_type(const struct resource *res)