[PATCH] Fix SPARSEMEM_EXTREME support on Linux v4.15 when passing vmlinux

[Omar Sandoval]

Since kernel commit 83e3c48729d9 ("mm/sparsemem: Allocate mem_section at
runtime for CONFIG_SPARSEMEM_EXTREME=y"), mem_section is a dynamically
allocated array of pointers to mem_section instead of a static one
(i.e., struct mem_section ** instead of struct mem_section * []). This
adds an extra layer of indirection that breaks makedumpfile, which will
end up with a bunch of bogus mem_maps.

Since kernel commit a0b1280368d1 ("kdump: write correct address of
mem_section into vmcoreinfo"), the mem_section symbol in vmcoreinfo
contains the address of the actual struct mem_section * array instead of
the address of the pointer in .bss, which gets rid of the extra
indirection. However, makedumpfile still uses the debugging symbol from
the vmlinux image. Fix this by allowing symbols from the vmcore to
override symbols from the vmlinux image. As the comment in initial()
says, "vmcoreinfo in /proc/vmcore is more reliable than -x/-i option".

Signed-off-by: Omar Sandoval <osandov@xxxxxx>
---
 makedumpfile.h | 6 ++++--
 1 file changed, 4 insertions(+), 2 deletions(-)

diff --git a/makedumpfile.h b/makedumpfile.h
index 57cf4d9..d68c798 100644
--- a/makedumpfile.h
+++ b/makedumpfile.h
@@ -274,8 +274,10 @@ do { \
 } while (0)
 #define READ_SYMBOL(str_symbol, symbol) \
 do { \
-	if (SYMBOL(symbol) == NOT_FOUND_SYMBOL) { \
-		SYMBOL(symbol) = read_vmcoreinfo_symbol(STR_SYMBOL(str_symbol)); \
+	unsigned long _tmp_symbol; \
+	_tmp_symbol = read_vmcoreinfo_symbol(STR_SYMBOL(str_symbol)); \
+	if (_tmp_symbol != NOT_FOUND_SYMBOL) { \
+		SYMBOL(symbol) = _tmp_symbol; \
 		if (SYMBOL(symbol) == INVALID_SYMBOL_DATA) \
 			return FALSE; \
 	} \
-- 
2.9.5