During invocations of EFI functions efi_switch_mm() is used to set the
active mm to borrow a different MM for a while. This used for instance
by efi_call_virt_pointer():
efi_call_virt_pointer()
- arch_efi_call_virt_setup()
- preempt_disable()
- efi_switch_mm(&efi_mm)
- "EFI CALL"
- arch_efi_call_virt_teardown()
- efi_switch_mm(efi_scratch.prev_mm)
- preempt_enable()
efi_switch_mm() is a wrapper around switch_mm() which saves current's
->active_mm, sets the requests mm as ->active_mm and invokes
switch_mm().
I don't think that task_lock() is required during that procedure. It
protects ->mm which isn't changed here.
I doubt that there any need to set ->active_mm. It is used by the
scheduler to keep track of the "currently used mm" so it can reuse one
for the kernel thread which does not own one and take a reference on it
so it does not go away while the thread (that borrows it) is active.
The `->active_mm' that is used here needs to saved to we can use it in
the "ending" efi_switch_to() again.
Based on this, the ->active_mm assignment can go.
It needs to be mentioned that during the whole procedure (switch to
EFI's mm and back) the preemption needs to be disabled. A context switch
at this point would reset the cr3 value based on current->mm.
With this new information, also update the comment to reflect it.
The "efi_scratch.prev_mm" assignment looks racy. With two concurrent
efi.get_next_variable invocations, one would overrwrite the value from
the other. This however does not happen because "runtime-wrappers"
(virt_efi_get_next_variable()) use efi_runtime_lock during the
invocation. This is not the case for "efi_64"
(efi_thunk_get_next_variable()) which does not take any locks (or I
missed them). Maybe it would be better to let the caller save
->active_mm while the MM is borrowed.
Signed-off-by: Sebastian Andrzej Siewior <bigeasy@xxxxxxxxxxxxx>
---
arch/x86/platform/efi/efi_64.c | 10 ++++------
1 file changed, 4 insertions(+), 6 deletions(-)
diff --git a/arch/x86/platform/efi/efi_64.c b/arch/x86/platform/efi/efi_64.c
index 77873ce700ae..64bf29462348 100644
--- a/arch/x86/platform/efi/efi_64.c
+++ b/arch/x86/platform/efi/efi_64.c
@@ -619,18 +619,16 @@ void __init efi_dump_pagetable(void)
/*
* Makes the calling thread switch to/from efi_mm context. Can be used
- * for SetVirtualAddressMap() i.e. current->active_mm == init_mm as well
- * as during efi runtime calls i.e current->active_mm == current_mm.
+ * in a kernel thread and user context. Preemption needs to remain disabled
+ * while the EFI-mm is borrowed.
* We are not mm_dropping()/mm_grabbing() any mm, because we are not
- * losing/creating any references.
+ * losing/creating any references. Preemption must be disabled while the mm is
+ * switched.
*/
void efi_switch_mm(struct mm_struct *mm)
{
- task_lock(current);
efi_scratch.prev_mm = current->active_mm;
- current->active_mm = mm;
switch_mm(efi_scratch.prev_mm, mm, NULL);
- task_unlock(current);
}
#ifdef CONFIG_EFI_MIXED
--
2.18.0
--
To unsubscribe from this list: send the line "unsubscribe linux-efi" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at http://vger.kernel.org/majordomo-info.html
