Skip to contentSkip to conversation

efi/runtime-wrappers: fix linker warnings

PATCH Applied Breno Leitao 2 participants 4 messages Jul 13 – Jul 14

Subsystems (2)
lore
Share & tools
lore
#0 lore raw

cover efi/runtime-wrappers: fix linker warnings

When I've submitted my last patchset titled "efi/runtime-wrappers: bound the wait for EFI runtime service calls", I've got two reports

1) efi_rts_park_worker() loops forever and never returns, but wasn't marked __noreturn, so objtool warns about the unreachable code after the call.
https://lore.kernel.org/all/[email protected]/

2) efi_runtime_lock_owner is optimised away under CONFIG_BUG=n, leaving a dangling debug-info reference that breaks the link.
https://lore.kernel.org/all/[email protected]/

Patch 1 fixes the objtool warning; patch 2 fixes the link error.

---
Breno Leitao (2):
efi/runtime-wrappers: mark efi_rts_park_worker() as __noreturn efi/runtime-wrappers: mark efi_runtime_lock_owner as __used

 drivers/firmware/efi/runtime-wrappers.c | 4 ++--
 include/linux/efi.h                     | 2 +-
 tools/objtool/noreturns.h               | 1 +
 3 files changed, 4 insertions(+), 3 deletions(-)

---
base-commit: cf790fdb851867532b9a0f88588dcce648dbffe9
change-id: 20260713-efi_fixes-b7193dd37bcf

Best regards,

Signature · 1 line
Breno Leitao <[email protected]>
#1 lore raw

PATCH 1/2 efi/runtime-wrappers: mark efi_rts_park_worker() as __noreturn

efi_rts_park_worker() loops in schedule() forever and never returns, but it is not annotated as such. When efi_crash_gracefully_on_page_fault() calls it, objtool cannot tell the call is a dead end and warns about the unreachable code that follows it:

vmlinux.o: warning: objtool: efi_crash_gracefully_on_page_fault+0xc3: efi_rts_park_worker() missing __noreturn in .c/.h or NORETURN() in noreturns.h

Mark both the declaration and the definition __noreturn, and add the function to objtool's noreturn list.

Fixes 3e5ba97c181e ("efi/runtime-wrappers: factor out efi_rts_park_worker()")
---
 drivers/firmware/efi/runtime-wrappers.c | 2 +-
 include/linux/efi.h                     | 2 +-
 tools/objtool/noreturns.h               | 1 +
 3 files changed, 3 insertions(+), 2 deletions(-)
drivers/firmware/efi/runtime-wrappers.c +1 -1
index 591a725b10613..6f42da11a9178 100644
@@ -224,7 +224,7 @@ extern struct semaphore __efi_uv_runtime_lock __alias(efi_runtime_lock);
  * have been disabled and its efi_rts_work is abandoned. Loop in schedule()
  * so a spurious wakeup cannot resume it.
  */
-void efi_rts_park_worker(void)
+void __noreturn efi_rts_park_worker(void)
 {
 	for (;;) {
 		set_current_state(TASK_IDLE);
include/linux/efi.h +1 -1
index 015505423277e..b3c83516593d1 100644
@@ -1256,7 +1256,7 @@ extern struct efi_runtime_work efi_rts_work;
 /* Workqueue to queue EFI Runtime Services */
 extern struct workqueue_struct *efi_rts_wq;
 
-void efi_rts_park_worker(void);
+void __noreturn efi_rts_park_worker(void);
 
 struct linux_efi_memreserve {
 	int		size;			// allocated size of the array
tools/objtool/noreturns.h +1 -0
index 14f8ab653449c..0e9dfd0a2446c 100644
@@ -26,6 +26,7 @@ NORETURN(cpu_startup_entry)
 NORETURN(do_exit)
 NORETURN(do_group_exit)
 NORETURN(do_task_dead)
+NORETURN(efi_rts_park_worker)
 NORETURN(ex_handler_msr_mce)
 NORETURN(hlt_play_dead)
 NORETURN(hv_ghcb_terminate)
Signature · 1 line
2.53.0-Meta
#2 lore raw

PATCH 2/2 efi/runtime-wrappers: mark efi_runtime_lock_owner as __used

In reply to Breno Leitao

efi_runtime_lock_owner is only ever read by efi_runtime_assert_lock_held(). When CONFIG_BUG=n, WARN_ON() collapses to a statement that discards its condition, so the compiler drops that read, treats the variable as write-only and eliminates it. Debug info still references the symbol, so a build carrying it (e.g. CONFIG_DEBUG_INFO_SPLIT) fails to link:

ld: drivers/firmware/efi/runtime-wrappers.o:(.debug_addr+0x104): undefined reference to `efi_runtime_lock_owner'

Mark the variable __used so its storage is always emitted. This seems to be better than alternatives (READ_ONCE/__maybe_unused).

Fixes a2860501203c ("efi/runtime-wrappers: Keep track of the efi_runtime_lock owner")
---
 drivers/firmware/efi/runtime-wrappers.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)
drivers/firmware/efi/runtime-wrappers.c +1 -1
index 6f42da11a9178..2344b9d1e81f8 100644
@@ -210,7 +210,7 @@ void efi_call_virt_check_flags(unsigned long flags, const void *caller)
  */
 static DEFINE_SEMAPHORE(efi_runtime_lock, 1);
 
-static struct task_struct *efi_runtime_lock_owner;
+static struct task_struct *efi_runtime_lock_owner __used;
 
 /*
  * Expose the EFI runtime lock to the UV platform
Signature · 1 line
2.53.0-Meta
maintainer Applied
#3 lore raw
In reply to Breno Leitao
Quoted 12 lines · Breno Leitao

When I've submitted my last patchset titled "efi/runtime-wrappers: bound the wait for EFI runtime service calls", I've got two reports

1) efi_rts_park_worker() loops forever and never returns, but wasn't marked __noreturn, so objtool warns about the unreachable code after the call.
https://lore.kernel.org/all/[email protected]/

2) efi_runtime_lock_owner is optimised away under CONFIG_BUG=n, leaving a dangling debug-info reference that breaks the link.
https://lore.kernel.org/all/[email protected]/

Quoted 8 lines

Patch 1 fixes the objtool warning; patch 2 fixes the link error.

---
Breno Leitao (2):
efi/runtime-wrappers: mark efi_rts_park_worker() as __noreturn efi/runtime-wrappers: mark efi_runtime_lock_owner as __used

Queued up now, thanks.