The commit ca29a0bf122145 ("tracing: gfp: Remove duplication of recording
GFP flags") caused the following regression in printf_test selftest:
[ 46.208199] test_printf: kvasprintf(..., "%pGg", ...) returned 'none|0xfc000000', expected '0xfc000000'
[ 46.208209] test_printf: kvasprintf(..., "%pGg", ...) returned '__GFP_HIGH|none|0xfc000000', expected '__GFP_HIGH|0xfc000000'
The problem is the new '{ 0, "none" }' entry in __def_gfpflag_names macro
and the following code:
char *format_flags(char *buf, char *end, unsigned long flags,
const struct trace_print_flags *names)
{
[...]
if ((flags & mask) != mask)
continue;
[...]
}
The purpose of the code is to print the name of a mask instead of bits,
for example, printk "GFP_ZONEMASK", instead of
"__GFP_DMA|__GFP_HIGHMEM|__GFP_DMA32|__GFP_MOVABLE".
Unfortunately, the mask "0" pass this check and "none" is always
printed.
A solution would be to move TRACE_GFP_FLAGS up so that it is not
the last entry. But it breaks the rule that named masks must
be defined before names of single bytes. Otherwise, it would
print the names of the bytes instead of the mask.
Instead, replace '{ 0, "none" }' with '{ 0, NULL }'. It works because
__def_gfpflag_names defines a standalone array and this is the standard
trailing entry. The code processing these arrays always ends the cycle
when flag->name == NULL.
Fixes: ca29a0bf122145 ("tracing: gfp: Remove duplication of recording GFP flags")
Signed-off-by: Petr Mladek <pmladek@xxxxxxxx>
---
include/trace/events/mmflags.h | 2 +-
1 file changed, 1 insertion(+), 1 deletion(-)
diff --git a/include/trace/events/mmflags.h b/include/trace/events/mmflags.h
index 82371177ef79..15aae955a10b 100644
--- a/include/trace/events/mmflags.h
+++ b/include/trace/events/mmflags.h
@@ -101,7 +101,7 @@ TRACE_DEFINE_ENUM(___GFP_LAST_BIT);
gfpflag_string(GFP_DMA32), \
gfpflag_string(__GFP_RECLAIM), \
TRACE_GFP_FLAGS \
- { 0, "none" }
+ { 0, NULL }
#define show_gfp_flags(flags) \
(flags) ? __print_flags(flags, "|", __def_gfpflag_names \
--
2.48.1
