Revert "x86/efi: Build our own page table structures"

This reverts commit 36e0f05a which is
commit 67a9108e upstream.

Turns there was too many other issues with this patch to make it viable
for the stable tree.
Reported-by: Ben Hutchings <ben.hutchings@codethink.co.uk>
Cc: Matt Fleming <matt@codeblueprint.co.uk>
Cc: Borislav Petkov <bp@suse.de>
Cc: Andrew Morton <akpm@linux-foundation.org>
Cc: Andy Lutomirski <luto@amacapital.net>
Cc: Andy Lutomirski <luto@kernel.org>
Cc: Ard Biesheuvel <ard.biesheuvel@linaro.org>
Cc: Borislav Petkov <bp@alien8.de>
Cc: Brian Gerst <brgerst@gmail.com>
Cc: Dave Jones <davej@codemonkey.org.uk>
Cc: Denys Vlasenko <dvlasenk@redhat.com>
Cc: H. Peter Anvin <hpa@zytor.com>
Cc: Linus Torvalds <torvalds@linux-foundation.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Sai Praneeth Prakhya <sai.praneeth.prakhya@intel.com>
Cc: Stephen Smalley <sds@tycho.nsa.gov>
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: Toshi Kani <toshi.kani@hp.com>
Cc: linux-efi@vger.kernel.org
Cc: Ingo Molnar <mingo@kernel.org>
Cc: "Ghannam, Yazen" <Yazen.Ghannam@amd.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>

arch/x86/include/asm/efi.h

@@ -136,7 +136,6 @@ extern void __init efi_memory_uc(u64 addr, unsigned long size);
 extern void __init efi_map_region(efi_memory_desc_t *md);
 extern void __init efi_map_region_fixed(efi_memory_desc_t *md);
 extern void efi_sync_low_kernel_mappings(void);
-extern int __init efi_alloc_page_tables(void);
 extern int __init efi_setup_page_tables(unsigned long pa_memmap, unsigned num_pages);
 extern void __init efi_cleanup_page_tables(unsigned long pa_memmap, unsigned num_pages);
 extern void __init old_map_region(efi_memory_desc_t *md);

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);
 	 */

arch/x86/platform/efi/efi_32.c

@@ -38,11 +38,6 @@
  * say 0 - 3G.
  */
 
-int __init efi_alloc_page_tables(void)
-{
-	return 0;
-}
-
 void efi_sync_low_kernel_mappings(void) {}
 void __init efi_dump_pagetable(void) {}
 int __init efi_setup_page_tables(unsigned long pa_memmap, unsigned num_pages)

arch/x86/platform/efi/efi_64.c

@@ -40,7 +40,6 @@
 #include <asm/fixmap.h>
 #include <asm/realmode.h>
 #include <asm/time.h>
-#include <asm/pgalloc.h>
 
 /*
  * We allocate runtime services regions bottom-up, starting from -4G, i.e.
@@ -122,92 +121,22 @@ void __init efi_call_phys_epilog(pgd_t *save_pgd)
 	early_code_mapping_set_exec(0);
 }
 
-static pgd_t *efi_pgd;
-
-/*
- * We need our own copy of the higher levels of the page tables
- * because we want to avoid inserting EFI region mappings (EFI_VA_END
- * to EFI_VA_START) into the standard kernel page tables. Everything
- * else can be shared, see efi_sync_low_kernel_mappings().
- */
-int __init efi_alloc_page_tables(void)
-{
-	pgd_t *pgd;
-	pud_t *pud;
-	gfp_t gfp_mask;
-
-	if (efi_enabled(EFI_OLD_MEMMAP))
-		return 0;
-
-	gfp_mask = GFP_KERNEL | __GFP_NOTRACK | __GFP_REPEAT | __GFP_ZERO;
-	efi_pgd = (pgd_t *)__get_free_page(gfp_mask);
-	if (!efi_pgd)
-		return -ENOMEM;
-
-	pgd = efi_pgd + pgd_index(EFI_VA_END);
-
-	pud = pud_alloc_one(NULL, 0);
-	if (!pud) {
-		free_page((unsigned long)efi_pgd);
-		return -ENOMEM;
-	}
-
-	pgd_populate(NULL, pgd, pud);
-
-	return 0;
-}
-
 /*
  * Add low kernel mappings for passing arguments to EFI functions.
  */
 void efi_sync_low_kernel_mappings(void)
 {
-	unsigned num_entries;
-	pgd_t *pgd_k, *pgd_efi;
-	pud_t *pud_k, *pud_efi;
+	unsigned num_pgds;
+	pgd_t *pgd = (pgd_t *)__va(real_mode_header->trampoline_pgd);
 
 	if (efi_enabled(EFI_OLD_MEMMAP))
 		return;
 
-	/*
-	 * We can share all PGD entries apart from the one entry that
-	 * covers the EFI runtime mapping space.
-	 *
-	 * Make sure the EFI runtime region mappings are guaranteed to
-	 * only span a single PGD entry and that the entry also maps
-	 * other important kernel regions.
-	 */
-	BUILD_BUG_ON(pgd_index(EFI_VA_END) != pgd_index(MODULES_END));
-	BUILD_BUG_ON((EFI_VA_START & PGDIR_MASK) !=
-			(EFI_VA_END & PGDIR_MASK));
-
-	pgd_efi = efi_pgd + pgd_index(PAGE_OFFSET);
-	pgd_k = pgd_offset_k(PAGE_OFFSET);
-
-	num_entries = pgd_index(EFI_VA_END) - pgd_index(PAGE_OFFSET);
-	memcpy(pgd_efi, pgd_k, sizeof(pgd_t) * num_entries);
+	num_pgds = pgd_index(MODULES_END - 1) - pgd_index(PAGE_OFFSET);
 
-	/*
-	 * We share all the PUD entries apart from those that map the
-	 * EFI regions. Copy around them.
-	 */
-	BUILD_BUG_ON((EFI_VA_START & ~PUD_MASK) != 0);
-	BUILD_BUG_ON((EFI_VA_END & ~PUD_MASK) != 0);
-
-	pgd_efi = efi_pgd + pgd_index(EFI_VA_END);
-	pud_efi = pud_offset(pgd_efi, 0);
-
-	pgd_k = pgd_offset_k(EFI_VA_END);
-	pud_k = pud_offset(pgd_k, 0);
-
-	num_entries = pud_index(EFI_VA_END);
-	memcpy(pud_efi, pud_k, sizeof(pud_t) * num_entries);
-
-	pud_efi = pud_offset(pgd_efi, EFI_VA_START);
-	pud_k = pud_offset(pgd_k, EFI_VA_START);
-
-	num_entries = PTRS_PER_PUD - pud_index(EFI_VA_START);
-	memcpy(pud_efi, pud_k, sizeof(pud_t) * num_entries);
+	memcpy(pgd + pgd_index(PAGE_OFFSET),
+		init_mm.pgd + pgd_index(PAGE_OFFSET),
+		sizeof(pgd_t) * num_pgds);
 }
 
 int __init efi_setup_page_tables(unsigned long pa_memmap, unsigned num_pages)
@@ -220,8 +149,8 @@ int __init efi_setup_page_tables(unsigned long pa_memmap, unsigned num_pages)
 	if (efi_enabled(EFI_OLD_MEMMAP))
 		return 0;
 
-	efi_scratch.efi_pgt = (pgd_t *)__pa(efi_pgd);
-	pgd = efi_pgd;
+	efi_scratch.efi_pgt = (pgd_t *)(unsigned long)real_mode_header->trampoline_pgd;
+	pgd = __va(efi_scratch.efi_pgt);
 
 	/*
 	 * It can happen that the physical address of new_memmap lands in memory
@@ -267,14 +196,16 @@ int __init efi_setup_page_tables(unsigned long pa_memmap, unsigned num_pages)
 
 void __init efi_cleanup_page_tables(unsigned long pa_memmap, unsigned num_pages)
 {
-	kernel_unmap_pages_in_pgd(efi_pgd, pa_memmap, num_pages);
+	pgd_t *pgd = (pgd_t *)__va(real_mode_header->trampoline_pgd);
+
+	kernel_unmap_pages_in_pgd(pgd, pa_memmap, num_pages);
 }
 
 static void __init __map_region(efi_memory_desc_t *md, u64 va)
 {
+	pgd_t *pgd = (pgd_t *)__va(real_mode_header->trampoline_pgd);
 	unsigned long flags = 0;
 	unsigned long pfn;
-	pgd_t *pgd = efi_pgd;
 
 	if (!(md->attribute & EFI_MEMORY_WB))
 		flags |= _PAGE_PCD;
@@ -383,7 +314,9 @@ void __init efi_runtime_mkexec(void)
 void __init efi_dump_pagetable(void)
 {
 #ifdef CONFIG_EFI_PGT_DUMP
-	ptdump_walk_pgd_level(NULL, efi_pgd);
+	pgd_t *pgd = (pgd_t *)__va(real_mode_header->trampoline_pgd);
+
+	ptdump_walk_pgd_level(NULL, pgd);
 #endif
 }