Changes since v1:
Addressed comments from Bhupesh Sharma, Thomas Gleixner, and Ard Biesheuvel:
* made __uv_bios_call() static * moved the efi_enabled() cleanup to its own patchlet * explained the reason for renaming the efi_runtime_lock semaphore * dropped the reviewed-bys as they should be given on the mailing list * Cc'ng stable@vger.kernel.org given the nature of the problem addressed by the patches
---
Calls into UV BIOS were not being serialised which is wrong as it violates EFI runtime rules, and bad as it does result in all sorts of potentially hard to track down hangs and panics when efi_scratch.prev_mm gets clobbered whenever efi_switch_mm() gets called concurrently from two different CPUs.
Patch #1 makes the efi_runtime_lock semaphore visible to EFI runtime callers defined outside drivers/firmware/efi/runtime-wrappers.c in preparation for using it to serialise calls into UV BIOS; the lock is also renamed to efi_runtime_sem so that it can coexist with the efi_runtime_lock spinlock defined in arch/x86/platform/efi/efi_64.c when CONFIG_EFI_MIXED is enabled.
Patch #2 removes uv_bios_call_reentrant() because it's dead code.
Patch #3 is a cleanup that drops test_bit() in favour of the ad hoc efi_enabled().
Patch #4 makes uv_bios_call() variants use efi_runtime_sem to protect against concurrency.
Cc: Russ Anderson rja@hpe.com Cc: Mike Travis mike.travis@hpe.com Cc: Dimitri Sivanich sivanich@hpe.com Cc: Steve Wahl steve.wahl@hpe.com Cc: stable@vger.kernel.org
Hedi Berriche (4): efi/x86: turn EFI runtime semaphore into a global lock x86/platform/UV: kill uv_bios_call_reentrant() as it has no callers x86/platform/UV: use efi_enabled() instead of test_bit() x86/platform/UV: use efi_runtime_sem to serialise BIOS calls
arch/x86/include/asm/uv/bios.h | 4 +- arch/x86/platform/uv/bios_uv.c | 33 ++++++++------ drivers/firmware/efi/runtime-wrappers.c | 60 ++++++++++++------------- include/linux/efi.h | 3 ++ 4 files changed, 55 insertions(+), 45 deletions(-)