Skip to contentSkip to conversation

Revert "mm/page_alloc: make should_fail_alloc_page() static"

PATCH In review Marco Elver 3 participants 4 messages Jul 5, 2021

Subsystems (3)
lore
Share & tools
lore
#0 lore raw

PATCH 1/1 Revert "mm/page_alloc: make should_fail_alloc_page() static"

This reverts commit f7173090033c70886d925995e9dfdfb76dbb2441.

Commit 76cd61739fd1 ("mm/error_inject: Fix allow_error_inject function signatures") explicitly made should_fail_alloc_page() non-static, due to worries of remaining compiler optimizations in the absence of function side-effects while being noinline.

Furthermore, kernel/bpf/verifier.c pushes should_fail_alloc_page onto the btf_non_sleepable_error_inject BTF IDs set, which when enabling CONFIG_DEBUG_INFO_BTF results in an error at the BTFIDS stage:

FAILED unresolved symbol should_fail_alloc_page

To avoid the W=1 warning, add a function declaration right above the function itself, with a comment it is required in a BTF IDs set.

Fixes f7173090033c ("mm/page_alloc: make should_fail_alloc_page() static")
Cc Mel Gorman <[email protected]>
Cc Alexei Starovoitov <[email protected]>
---
 mm/page_alloc.c | 8 +++++++-
 1 file changed, 7 insertions(+), 1 deletion(-)
mm/page_alloc.c +7 -1
index d6e94cc8066c..16e71d48d84e 100644
@@ -3831,7 +3831,13 @@ static inline bool __should_fail_alloc_page(gfp_t gfp_mask, unsigned int order)
 
 #endif /* CONFIG_FAIL_PAGE_ALLOC */
 
-static noinline bool should_fail_alloc_page(gfp_t gfp_mask, unsigned int order)
+/*
+ * should_fail_alloc_page() is only called by page_alloc.c, however, is also
+ * included in a BTF IDs set and must remain non-static. Declare it to avoid a
+ * "missing prototypes" warning, and make it clear this is intentional.
+ */
+bool should_fail_alloc_page(gfp_t gfp_mask, unsigned int order);
+noinline bool should_fail_alloc_page(gfp_t gfp_mask, unsigned int order)
 {
 	return __should_fail_alloc_page(gfp_mask, order);
 }
Signature · 1 line
2.32.0.93.g670b81a890-goog
Acked-by on v1 1/1
#1 lore raw
Quoted 13 lines · Marco Elver

This reverts commit f7173090033c70886d925995e9dfdfb76dbb2441.

Commit 76cd61739fd1 ("mm/error_inject: Fix allow_error_inject function signatures") explicitly made should_fail_alloc_page() non-static, due to worries of remaining compiler optimizations in the absence of function side-effects while being noinline.

Furthermore, kernel/bpf/verifier.c pushes should_fail_alloc_page onto the btf_non_sleepable_error_inject BTF IDs set, which when enabling CONFIG_DEBUG_INFO_BTF results in an error at the BTFIDS stage:

FAILED unresolved symbol should_fail_alloc_page

Quoted 7 lines

To avoid the W=1 warning, add a function declaration right above the function itself, with a comment it is required in a BTF IDs set.

Fixes f7173090033c ("mm/page_alloc: make should_fail_alloc_page() static")
Cc Mel Gorman <[email protected]>
Cc Alexei Starovoitov <[email protected]>

Out of curiousity though, why does block/blk-core.c not require something similar for should_fail_bio?

Signature · 2 lines
Mel Gorman
SUSE Labs
#2 lore raw
Quoted 22 lines · Mel Gorman
Quoted 13 lines · Marco Elver

This reverts commit f7173090033c70886d925995e9dfdfb76dbb2441.

Commit 76cd61739fd1 ("mm/error_inject: Fix allow_error_inject function signatures") explicitly made should_fail_alloc_page() non-static, due to worries of remaining compiler optimizations in the absence of function side-effects while being noinline.

Furthermore, kernel/bpf/verifier.c pushes should_fail_alloc_page onto the btf_non_sleepable_error_inject BTF IDs set, which when enabling CONFIG_DEBUG_INFO_BTF results in an error at the BTFIDS stage:

FAILED unresolved symbol should_fail_alloc_page

Quoted 7 lines

To avoid the W=1 warning, add a function declaration right above the function itself, with a comment it is required in a BTF IDs set.

Fixes f7173090033c ("mm/page_alloc: make should_fail_alloc_page() static")
Cc Mel Gorman <[email protected]>
Cc Alexei Starovoitov <[email protected]>
Quoted 4 lines

Out of curiousity though, why does block/blk-core.c not require something similar for should_fail_bio?

It seems kernel/bpf/verifier.c doesn't refer to it in an BTF IDs set. Looks like should_fail_alloc_page is special for BPF purposes. I'm not a BPF maintainer, so hopefully someone can explain why should_fail_alloc_page is special for BPF.

Thanks,
-- Marco

#3 lore raw
In reply to Marco Elver
Quoted 7 lines · Marco Elver

This reverts commit f7173090033c70886d925995e9dfdfb76dbb2441.

Commit 76cd61739fd1 ("mm/error_inject: Fix allow_error_inject function signatures") explicitly made should_fail_alloc_page() non-static, due to worries of remaining compiler optimizations in the absence of function side-effects while being noinline.

Quoted 8 lines

Furthermore, kernel/bpf/verifier.c pushes should_fail_alloc_page onto the btf_non_sleepable_error_inject BTF IDs set, which when enabling CONFIG_DEBUG_INFO_BTF results in an error at the BTFIDS stage:

FAILED unresolved symbol should_fail_alloc_page

To avoid the W=1 warning, add a function declaration right above the function itself, with a comment it is required in a BTF IDs set.

NAK. We're not going to make symbols pointlessly global for broken instrumentation coe. Someone needs to fixthis eBPF mess as we had the same kind of issue before already.