[OS-BUILD PATCHv3] mm/sparsemem: Fix 'mem_section' will never be NULL gcc 12 warning

[Justin M. Forbes (via Email Bridge)]

From: Waiman Long <longman(a)redhat.com&gt;

mm/sparsemem: Fix 'mem_section' will never be NULL gcc 12 warning

The gcc 12 compiler reports a "'mem_section' will never be NULL"
warning on the following code:

    static inline struct mem_section *__nr_to_section(unsigned long nr)
    {
    #ifdef CONFIG_SPARSEMEM_EXTREME
        if (!mem_section)
                return NULL;
    #endif
        if (!mem_section[SECTION_NR_TO_ROOT(nr)])
                return NULL;
       :

It happens with both CONFIG_SPARSEMEM_EXTREME on and off. The mem_section
definition is

    #ifdef CONFIG_SPARSEMEM_EXTREME
    extern struct mem_section **mem_section;
    #else
    extern struct mem_section mem_section[NR_SECTION_ROOTS][SECTIONS_PER_ROOT];
    #endif

In the CONFIG_SPARSEMEM_EXTREME case, mem_section obviously cannot
be NULL, but *mem_section can be if memory hasn't been allocated for
the dynamic mem_section[] array yet. In the !CONFIG_SPARSEMEM_EXTREME
case, mem_section is a static 2-dimensional array and so the check
"!mem_section[SECTION_NR_TO_ROOT(nr)]" doesn't make sense.

Fix this warning by checking for "!*mem_section" instead of
"!mem_section" and moving the "!mem_section[SECTION_NR_TO_ROOT(nr)]"
check up inside the CONFIG_SPARSEMEM_EXTREME block.

Fixes: 83e3c48729d9 ("mm/sparsemem: Allocate mem_section at runtime for
CONFIG_SPARSEMEM_EXTREME=y")
Fixes: 3e347261a80b ("sparsemem extreme implementation")
Reported-by: Justin Forbes <jforbes(a)redhat.com&gt;
Signed-off-by: Waiman Long <longman(a)redhat.com&gt;

diff --git a/include/linux/mmzone.h b/include/linux/mmzone.h
index blahblah..blahblah 100644
--- a/include/linux/mmzone.h
+++ b/include/linux/mmzone.h
@@ -1390,11 +1390,9 @@ static inline unsigned long *section_to_usemap(struct mem_section
*ms)
 static inline struct mem_section *__nr_to_section(unsigned long nr)
 {
 #ifdef CONFIG_SPARSEMEM_EXTREME
-	if (!mem_section)
+	if (!*mem_section || !mem_section[SECTION_NR_TO_ROOT(nr)])
 		return NULL;
 #endif
-	if (!mem_section[SECTION_NR_TO_ROOT(nr)])
-		return NULL;
 	return &mem_section[SECTION_NR_TO_ROOT(nr)][nr & SECTION_ROOT_MASK];
 }
 extern size_t mem_section_usage_size(void);

--
https://gitlab.com/cki-project/kernel-ark/-/merge_requests/1598

[Mark Salter (via Email Bridge)]

From: Mark Salter on gitlab.com
https://gitlab.com/cki-project/kernel-ark/-/merge_requests/1598#note_8660...

The "if (!*mem_section")" is causing a panic on aarch64 systems
(CONFIG_SPARSEMEM_EXTREME=y) which have no memory in the first mem_section
root. The original "if (!mem_section)" is the correct test.

See [bz2053214](https://bugzilla.redhat.com/show_bug.cgi?id=2053214)

[Justin M. Forbes (via Email Bridge)]

From: Justin M. Forbes on gitlab.com
https://gitlab.com/cki-project/kernel-ark/-/merge_requests/1598#note_8670...

Yes, it was backed out from the queue upstream because of a similar problem on
m1 macs (support not upstreamed yet).  @llong1 is there an updated version in
the works?