hid: bpf: avoid building struct ops without JIT
PATCH 1/1 hid: bpf: avoid building struct ops without JIT
From: Arnd Bergmann <[email protected]>
The module does not do anything when the JIT is disabled, but instead causes a warning:
In file included from include/linux/bpf_verifier.h:7,
from drivers/hid/bpf/hid_bpf_struct_ops.c:10:drivers/hid/bpf/hid_bpf_struct_ops.c: In function 'hid_bpf_struct_ops_init':include/linux/bpf.h:1853:50: error: statement with no effect [-Werror=unused-value]
1853 | #define register_bpf_struct_ops(st_ops, type) ({ (void *)(st_ops); 0; })
| ^~~~~~~~~~~~~~~~
drivers/hid/bpf/hid_bpf_struct_ops.c:305:16: note: in expansion of macro 'register_bpf_struct_ops'
305 | return register_bpf_struct_ops(&bpf_hid_bpf_ops, hid_bpf_ops);
| ^~~~~~~~~~~~~~~~~~~~~~~This could be avoided by making HID-BPF just depend on JIT, but that is probably not what we want here. Checking the other users of struct_ops, I see that those just leave out the struct_ops usage, so do the same here.
--- drivers/hid/bpf/Makefile | 3 ++- drivers/hid/bpf/hid_bpf_dispatch.h | 6 ++++++ 2 files changed, 8 insertions(+), 1 deletion(-)
drivers/hid/bpf/Makefile +2 -1
@@ -8,4 +8,5 @@ LIBBPF_INCLUDE = $(srctree)/tools/lib obj-$(CONFIG_HID_BPF) += hid_bpf.o CFLAGS_hid_bpf_dispatch.o += -I$(LIBBPF_INCLUDE) CFLAGS_hid_bpf_jmp_table.o += -I$(LIBBPF_INCLUDE) -hid_bpf-objs += hid_bpf_dispatch.o hid_bpf_struct_ops.o +hid_bpf-y += hid_bpf_dispatch.o +hid_bpf-$(CONFIG_BPF_JIT) += hid_bpf_struct_ops.o
drivers/hid/bpf/hid_bpf_dispatch.h +6 -0
@@ -14,7 +14,13 @@ struct hid_bpf_ctx_kern {struct hid_device *hid_get_device (unsigned int hid_id);void hid_put_device (struct hid_device *hid);int hid_bpf_allocate_event_data (struct hid_device *hdev); +#ifdef CONFIG_BPF_JIT void __hid_bpf_ops_destroy_device (struct hid_device *hdev); +#else +static inline void __hid_bpf_ops_destroy_device (struct hid_device *hdev) +{ +} +#endif int hid_bpf_reconnect (struct hid_device *hdev);struct bpf_prog ;
Signature · 1 line
2.39.2
Benjamin Tissoires
maintainerQuoted 10 lines · Arnd Bergmann
From: Arnd Bergmann <[email protected]>
The module does not do anything when the JIT is disabled, but instead causes a warning:
In file included from
include/linux/bpf_verifier.h:7,
fromdrivers/hid/bpf/hid_bpf_struct_ops.c:10:drivers/hid/bpf/hid_bpf_struct_ops.c: In function 'hid_bpf_struct_ops_init':include/linux/bpf.h:1853:50: error: statement with no effect [-Werror=unused-value]
1853 | #defineregister_bpf_struct_ops(st_ops, type)({ (void *)(st_ops); 0; })
Quoted 8 lines
| ^~~~~~~~~~~~~~~~ drivers/hid/bpf/hid_bpf_struct_ops.c:305:16: note: in expansion of macro 'register_bpf_struct_ops' 305 | return register_bpf_struct_ops(&bpf_hid_bpf_ops, hid_bpf_ops); | ^~~~~~~~~~~~~~~~~~~~~~~This could be avoided by making HID-BPF just depend on JIT, but that is probably not what we want here. Checking the other users of
struct_ops, I see that those just leave out thestruct_opsusage, so do the same here.
Actually, if we make the struct_ops part only depend on JIT HID-BPF is kind of moot. All we could do is use HID-BPF to communicate with the device, without getting any feedback, so nothing much more than what hidraw provides.
The only "interesting" bit we could do is inject a new event on a device as if it were originated from the device itself, but I really do not see the point without the struct_ops hooks.
So I think struct_ops is now the base for HID-BPF, and if it's not available, we should not have HID-BPF at all.
Quoted 13 lines
Fixes ebc0d8093e8c ("HID: bpf: implement HID-BPF through bpf_struct_ops")Signed-off-by Arnd Bergmann <[email protected]>--- drivers/hid/bpf/Makefile | 3 ++- drivers/hid/bpf/hid_bpf_dispatch.h | 6 ++++++ 2 files changed, 8 insertions(+), 1 deletion(-)drivers/hid/bpf/Makefile +0
-0
drivers/hid/bpf/Makefile LIBBPF_INCLUDE = $(srctree)/tools/lib · line 10
Quoted 3 lines
drivers/hid/bpf/Makefile +0
-0@@ -8,4 +8,5 @@ LIBBPF_INCLUDE = $(srctree)/tools/lib obj-$(CONFIG_HID_BPF) += hid_bpf.o CFLAGS_hid_bpf_dispatch.o += -I$(LIBBPF_INCLUDE) CFLAGS_hid_bpf_jmp_table.o += -I$(LIBBPF_INCLUDE)
Unrelated to your patch, but looks like I forgot to remove that entry from the Makefile now that hid_bpf_jmp_table.c is not available :)
Cheers,
Benjamin
Quoted 24 lines
drivers/hid/bpf/Makefile +2
-1-hid_bpf-objs += hid_bpf_dispatch.o hid_bpf_struct_ops.o +hid_bpf-y += hid_bpf_dispatch.o +hid_bpf-$(CONFIG_BPF_JIT) += hid_bpf_struct_ops.odrivers/hid/bpf/hid_bpf_dispatch.h +6
-0@@ -14,7 +14,13 @@ struct hid_bpf_ctx_kern {struct hid_device *hid_get_device (unsigned int hid_id);void hid_put_device (struct hid_device *hid);int hid_bpf_allocate_event_data (struct hid_device *hdev); +#ifdef CONFIG_BPF_JIT void __hid_bpf_ops_destroy_device (struct hid_device *hdev); +#else +static inline void __hid_bpf_ops_destroy_device (struct hid_device *hdev) +{ +} +#endif int hid_bpf_reconnect (struct hid_device *hdev);struct bpf_prog ;Signature · 1 line
2.39.2
Quoted 11 lines · Benjamin Tissoires
Quoted 4 lines · Arnd Bergmann
This could be avoided by making HID-BPF just depend on JIT, but that is probably not what we want here. Checking the other users of
struct_ops, I see that those just leave out thestruct_opsusage, so do the same here.Actually, if we make the
struct_opspart only depend on JIT HID-BPF is kind of moot. All we could do is use HID-BPF to communicate with the device, without getting any feedback, so nothing much more than what hidraw provides.
Quoted 7 lines
The only "interesting" bit we could do is inject a new event on a device as if it were originated from the device itself, but I really do not see the point without the
struct_opshooks.So I think
struct_opsis now the base for HID-BPF, and if it's not available, we should not have HID-BPF at all.
Ok, got it. So my original patch was correct after all. I had tried this version and then discarded it.
Arnd
8<------
Subject: [PATCH] hid: bpf: add BPF_JIT dependency
The module does not do anything when the JIT is disabled, but instead causes a warning:
In file included from include/linux/bpf_verifier.h:7,
from drivers/hid/bpf/hid_bpf_struct_ops.c:10:drivers/hid/bpf/hid_bpf_struct_ops.c: In function 'hid_bpf_struct_ops_init':include/linux/bpf.h:1853:50: error: statement with no effect [-Werror=unused-value]
1853 | #define register_bpf_struct_ops(st_ops, type) ({ (void *)(st_ops); 0; })
| ^~~~~~~~~~~~~~~~
drivers/hid/bpf/hid_bpf_struct_ops.c:305:16: note: in expansion of macro 'register_bpf_struct_ops'
305 | return register_bpf_struct_ops(&bpf_hid_bpf_ops, hid_bpf_ops);
| ^~~~~~~~~~~~~~~~~~~~~~~Add a Kconfig dependency to only allow building the HID-BPF support when a JIT is enabled.
----
drivers/hid/bpf/Kconfig +1 -1
@@ -3,7 +3,7 @@ menu "HID-BPF support" config HID_BPF bool "HID-BPF support" - depends on BPF + depends on BPF_JIT depends on BPF_SYSCALL depends on DYNAMIC_FTRACE_WITH_DIRECT_CALLS help
Benjamin Tissoires
maintainerQuoted 52 lines · Arnd Bergmann
Quoted 11 lines · Benjamin Tissoires
Quoted 4 lines · Arnd Bergmann
This could be avoided by making HID-BPF just depend on JIT, but that is probably not what we want here. Checking the other users of
struct_ops, I see that those just leave out thestruct_opsusage, so do the same here.Actually, if we make the
struct_opspart only depend on JIT HID-BPF is kind of moot. All we could do is use HID-BPF to communicate with the device, without getting any feedback, so nothing much more than what hidraw provides.Quoted 7 lines
The only "interesting" bit we could do is inject a new event on a device as if it were originated from the device itself, but I really do not see the point without the
struct_opshooks.So I think
struct_opsis now the base for HID-BPF, and if it's not available, we should not have HID-BPF at all.Ok, got it. So my original patch was correct after all. I had tried this version and then discarded it.
Arnd
8<------
Subject: [PATCH] hid: bpf: addBPF_JITdependencyThe module does not do anything when the JIT is disabled, but instead causes a warning:
In file included from
include/linux/bpf_verifier.h:7,
fromdrivers/hid/bpf/hid_bpf_struct_ops.c:10:drivers/hid/bpf/hid_bpf_struct_ops.c: In function 'hid_bpf_struct_ops_init':include/linux/bpf.h:1853:50: error: statement with no effect [-Werror=unused-value]
1853 | #defineregister_bpf_struct_ops(st_ops, type)({ (void *)(st_ops); 0; })| ^~~~~~~~~~~~~~~~ drivers/hid/bpf/hid_bpf_struct_ops.c:305:16: note: in expansion of macro 'register_bpf_struct_ops' 305 | return register_bpf_struct_ops(&bpf_hid_bpf_ops, hid_bpf_ops); | ^~~~~~~~~~~~~~~~~~~~~~~Add a
Kconfigdependency to only allow building the HID-BPF support when a JIT is enabled.Fixes ebc0d8093e8c ("HID: bpf: implement HID-BPF through bpf_struct_ops")Signed-off-by Arnd Bergmann <[email protected]>----
drivers/hid/bpf/Kconfig +0
-0
drivers/hid/bpf/Kconfig menu "HID-BPF support" · line 9
Quoted 8 lines
drivers/hid/bpf/Kconfig +1
-1@@ -3,7 +3,7 @@ menu "HID-BPF support" config HID_BPF bool "HID-BPF support" - depends on BPF + depends on BPF_JIT depends on BPF_SYSCALL depends on DYNAMIC_FTRACE_WITH_DIRECT_CALLS help
Thanks. I've applied this patch to for-6.11/upstream-fixes in the HID tree.
Cheers,
Benjamin