seq_buf: always inline seq_buf_terminate() and the helpers around it
PATCH v2 1/1 seq_buf: always inline seq_buf_terminate() and the helpers around it
Commit 7b8cd26b6aab8 ("seq_buf: Mark a seq_buf overflowed when terminating it loses a byte") made __seq_buf_terminate() bigger, and clang no longer inlines seq_buf_terminate() into its two callers in kernel/trace/trace_events.c. Both pass bootup_event_seq, which is __initdata, so the out-of-line copy sits in .text with that address in it and modpost warns:
WARNING: modpost: vmlinux: section mismatch in reference: seq_buf_terminate+0x12 (section: .text) -> bootup_event_seq (section: .init.data)
Both callers are __init, so nothing uses the buffer after it is freed. But the warning shows up in allmodconfig builds, and it is an error with CONFIG_SECTION_MISMATCH_WARN_ONLY=n.
Always inlining only seq_buf_terminate() and __seq_buf_terminate() moves the problem: clang then leaves seq_buf_set_overflow() out of line with the same address in it. So mark every inline helper that trace_events.c reaches with bootup_event_seq __always_inline: seq_buf_terminate(), __seq_buf_terminate(), seq_buf_set_overflow(), seq_buf_buffer_left(), seq_buf_has_overflowed() and seq_buf_used().
---7b8cd26b6aab8 is on both for-next/kspp and for-linus/hardening.
Changes in v2:
- v1 moved the warning to seq_buf_set_overflow() instead of removing it, as the kernel test robot found [1]. Mark all six helpers.
- Checked with modpost itself this time, run on trace_events.o: x86 and arm64 with clang 21 (allmodconfig, defconfig, defconfig with CC_OPTIMIZE_FOR_SIZE, and the robot's config) and with gcc 16 (x86 allmodconfig and the size-optimized configs). No mismatch is left.
v1: https://lore.kernel.org/r/[email protected]
[1] https://lore.kernel.org/oe-kbuild-all/[email protected]/
include/linux/seq_buf.h | 12 ++++++------ 1 file changed, 6 insertions(+), 6 deletions(-)
include/linux/seq_buf.h +6 -6
@@ -62,7 +62,7 @@ seq_buf_init(struct seq_buf *s, char *buf, unsigned int size)* seq_buf have a buffer that might overflow. When this happens * len is set to be greater than size. */ -static inline bool +static __always_inline boolseq_buf_has_overflowed (struct seq_buf *s) {return s->len > s->size;
@@ -74,7 +74,7 @@ seq_buf_has_overflowed(struct seq_buf *s)* seq_buf_str() terminates it, so clear whatever was not written: a writer * sets len to how much it filled, and anything past that was never adopted. */ -static inline void +static __always_inline voidseq_buf_set_overflow (struct seq_buf *s) {if (s->len < s->size)
@@ -111,7 +111,7 @@ seq_buf_init_append(struct seq_buf *s, char *buf, unsigned int size)/* * How much buffer is left on the seq_buf? */ -static inline unsigned int +static __always_inline unsigned intseq_buf_buffer_left (struct seq_buf *s) {if (seq_buf_has_overflowed (s))
@@ -121,7 +121,7 @@ seq_buf_buffer_left(struct seq_buf *s) }/* How much buffer was written? */ -static inline unsigned int seq_buf_used (struct seq_buf *s) +static __always_inline unsigned int seq_buf_used(struct seq_buf *s) {return min (s->len, s->size); }
@@ -134,7 +134,7 @@ static inline unsigned int seq_buf_used(struct seq_buf *s)* * Returns: the offset of the NUL. */ -static inline size_t __seq_buf_terminate (struct seq_buf *s) +static __always_inline size_t __seq_buf_terminate(struct seq_buf *s) {size_t end;
@@ -240,7 +240,7 @@ static inline size_t seq_buf_strlen(struct seq_buf *s)* After this function is called, s->buffer is safe to use * in string operations. */ -static inline void seq_buf_terminate (struct seq_buf *s) +static __always_inline void seq_buf_terminate(struct seq_buf *s) {if (s->size ==0 )return ;
base-commit: 7b8cd26b6aab8cf147165610d13964f79200a5a7
Signature · 1 line
2.53.0
Quoted 20 lines · Sasha Levin
Commit
7b8cd26b6aab8("seq_buf: Mark aseq_bufoverflowed when terminating it loses a byte") made__seq_buf_terminate()bigger, and clang no longer inlinesseq_buf_terminate()into its two callers inkernel/trace/trace_events.c. Both passbootup_event_seq, which is__initdata, so the out-of-line copy sits in .text with that address in it and modpost warns:WARNING: modpost: vmlinux: section mismatch in reference:
seq_buf_terminate+0x12(section: .text) ->bootup_event_seq(section: .init.data)Both callers are
__init, so nothing uses the buffer after it is freed. But the warning shows up in allmodconfig builds, and it is an error withCONFIG_SECTION_MISMATCH_WARN_ONLY=n.Always inlining only
seq_buf_terminate()and__seq_buf_terminate()moves the problem: clang then leavesseq_buf_set_overflow()out of line with the same address in it. So mark every inline helper thattrace_events.creaches withbootup_event_seq__always_inline:seq_buf_terminate(),__seq_buf_terminate(),seq_buf_set_overflow(),seq_buf_buffer_left(),seq_buf_has_overflowed()andseq_buf_used().
Quoted 8 lines
Fixes 7b8cd26b6aab8 ("seq_buf: Mark a seq_buf overflowed when terminating it loses a byte")Reported-by kernel test robot <[email protected]>Suggested-by Nathan Chancellor <[email protected]>Signed-off-by Sasha Levin <[email protected]>
Signature · 2 lines
Cheers, Nathan