Merge remote-tracking branch 'stable/linux-4.4.y' into lts-4.4.y-hikey4.4.106-hikey

1 files changed, 25 insertions, 14 deletions

diff --git a/arch/x86/platform/efi/efi.c b/arch/x86/platform/efi/efi.c
index 3c1f3cd7b2ba..ad285404ea7f 100644
--- a/arch/x86/platform/efi/efi.c
+++ b/arch/x86/platform/efi/efi.c
@@ -869,7 +869,7 @@ static void __init kexec_enter_virtual_mode(void)
  * This function will switch the EFI runtime services to virtual mode.
  * Essentially, we look through the EFI memmap and map every region that
  * has the runtime attribute bit set in its memory descriptor into the
- * efi_pgd page table.
+ * ->trampoline_pgd page table using a top-down VA allocation scheme.
  *
  * The old method which used to update that memory descriptor with the
  * virtual address obtained from ioremap() is still supported when the
@@ -879,8 +879,8 @@ static void __init kexec_enter_virtual_mode(void)
  *
  * The new method does a pagetable switch in a preemption-safe manner
  * so that we're in a different address space when calling a runtime
- * function. For function arguments passing we do copy the PUDs of the
- * kernel page table into efi_pgd prior to each call.
+ * function. For function arguments passing we do copy the PGDs of the
+ * kernel page table into ->trampoline_pgd prior to each call.
  *
  * Specially for kexec boot, efi runtime maps in previous kernel should
  * be passed in via setup_data. In that case runtime ranges will be mapped
@@ -895,12 +895,6 @@ static void __init __efi_enter_virtual_mode(void)
 
 	efi.systab = NULL;
 
-	if (efi_alloc_page_tables()) {
-		pr_err("Failed to allocate EFI page tables\n");
-		clear_bit(EFI_RUNTIME_SERVICES, &efi.flags);
-		return;
-	}
-
 	efi_merge_regions();
 	new_memmap = efi_map_regions(&count, &pg_shift);
 	if (!new_memmap) {
@@ -960,11 +954,28 @@ static void __init __efi_enter_virtual_mode(void)
 	efi_runtime_mkexec();
 
 	/*
-	 * We mapped the descriptor array into the EFI pagetable above
-	 * but we're not unmapping it here because if we're running in
-	 * EFI mixed mode we need all of memory to be accessible when
-	 * we pass parameters to the EFI runtime services in the
-	 * thunking code.
+	 * We mapped the descriptor array into the EFI pagetable above but we're
+	 * not unmapping it here. Here's why:
+	 *
+	 * We're copying select PGDs from the kernel page table to the EFI page
+	 * table and when we do so and make changes to those PGDs like unmapping
+	 * stuff from them, those changes appear in the kernel page table and we
+	 * go boom.
+	 *
+	 * From setup_real_mode():
+	 *
+	 * ...
+	 * trampoline_pgd[0] = init_level4_pgt[pgd_index(__PAGE_OFFSET)].pgd;
+	 *
+	 * In this particular case, our allocation is in PGD 0 of the EFI page
+	 * table but we've copied that PGD from PGD[272] of the EFI page table:
+	 *
+	 *	pgd_index(__PAGE_OFFSET = 0xffff880000000000) = 272
+	 *
+	 * where the direct memory mapping in kernel space is.
+	 *
+	 * new_memmap's VA comes from that direct mapping and thus clearing it,
+	 * it would get cleared in the kernel page table too.
 	 *
 	 * efi_cleanup_page_tables(__pa(new_memmap), 1 << pg_shift);
 	 */