s390/kasan: move shadow mapping to decompressor

Since regular paging structs are initialized in decompressor already
move KASAN shadow mapping to decompressor as well. This helps to avoid
allocating KASAN required memory in 1 large chunk, de-duplicate paging
structs creation code and start the uncompressed kernel with KASAN
instrumentation right away. This also allows to avoid all pitfalls
accidentally calling KASAN instrumented code during KASAN initialization.
Acked-by: Heiko Carstens <hca@linux.ibm.com>
Reviewed-by: Alexander Gordeev <agordeev@linux.ibm.com>
Signed-off-by: Vasily Gorbik <gor@linux.ibm.com>
Signed-off-by: Heiko Carstens <hca@linux.ibm.com>

arch/s390/boot/boot.h

@@ -32,6 +32,13 @@ struct vmlinux_info {
 	unsigned long init_mm_off;
 	unsigned long swapper_pg_dir_off;
 	unsigned long invalid_pg_dir_off;
+#ifdef CONFIG_KASAN
+	unsigned long kasan_early_shadow_page_off;
+	unsigned long kasan_early_shadow_pte_off;
+	unsigned long kasan_early_shadow_pmd_off;
+	unsigned long kasan_early_shadow_pud_off;
+	unsigned long kasan_early_shadow_p4d_off;
+#endif
 };
 
 void startup_kernel(void);

arch/s390/boot/startup.c

@@ -266,6 +266,13 @@ static void offset_vmlinux_info(unsigned long offset)
 	vmlinux.init_mm_off += offset;
 	vmlinux.swapper_pg_dir_off += offset;
 	vmlinux.invalid_pg_dir_off += offset;
+#ifdef CONFIG_KASAN
+	vmlinux.kasan_early_shadow_page_off += offset;
+	vmlinux.kasan_early_shadow_pte_off += offset;
+	vmlinux.kasan_early_shadow_pmd_off += offset;
+	vmlinux.kasan_early_shadow_pud_off += offset;
+	vmlinux.kasan_early_shadow_p4d_off += offset;
+#endif
 }
 
 void startup_kernel(void)
@@ -307,10 +314,6 @@ void startup_kernel(void)
 	detect_physmem_online_ranges(max_physmem_end);
 	save_ipl_cert_comp_list();
 	rescue_initrd(safe_addr, ident_map_size);
-#ifdef CONFIG_KASAN
-	physmem_alloc_top_down(RR_KASAN, kasan_estimate_memory_needs(get_physmem_usable_total()),
-			       _SEGMENT_SIZE);
-#endif
 
 	if (IS_ENABLED(CONFIG_RANDOMIZE_BASE) && kaslr_enabled) {
 		random_lma = get_random_base();

arch/s390/include/asm/kasan.h

@@ -2,7 +2,7 @@
 #ifndef __ASM_KASAN_H
 #define __ASM_KASAN_H
 
-#include <asm/pgtable.h>
+#include <linux/const.h>
 
 #ifdef CONFIG_KASAN
 
@@ -13,35 +13,6 @@
 #define KASAN_SHADOW_START	KASAN_SHADOW_OFFSET
 #define KASAN_SHADOW_END	(KASAN_SHADOW_START + KASAN_SHADOW_SIZE)
 
-extern void kasan_early_init(void);
-
-/*
- * Estimate kasan memory requirements, which it will reserve
- * at the very end of available physical memory. To estimate
- * that, we take into account that kasan would require
- * 1/8 of available physical memory (for shadow memory) +
- * creating page tables for the shadow memory region.
- * To keep page tables estimates simple take the double of
- * combined ptes size.
- *
- * physmem parameter has to be already adjusted if not entire physical memory
- * would be used (e.g. due to effect of "mem=" option).
- */
-static inline unsigned long kasan_estimate_memory_needs(unsigned long physmem)
-{
-	unsigned long kasan_needs;
-	unsigned long pages;
-	/* for shadow memory */
-	kasan_needs = round_up(physmem / 8, PAGE_SIZE);
-	/* for paging structures */
-	pages = DIV_ROUND_UP(kasan_needs, PAGE_SIZE);
-	kasan_needs += DIV_ROUND_UP(pages, _PAGE_ENTRIES) * _PAGE_TABLE_SIZE * 2;
-
-	return kasan_needs;
-}
-#else
-static inline void kasan_early_init(void) { }
-static inline unsigned long kasan_estimate_memory_needs(unsigned long physmem) { return 0; }
 #endif
 
 #endif

arch/s390/include/asm/physmem_info.h

@@ -26,9 +26,6 @@ enum reserved_range_type {
 	RR_CERT_COMP_LIST,
 	RR_MEM_DETECT_EXTENDED,
 	RR_VMEM,
-#ifdef CONFIG_KASAN
-	RR_KASAN,
-#endif
 	RR_MAX
 };
 
@@ -129,9 +126,6 @@ static inline const char *get_rr_type_name(enum reserved_range_type t)
 	RR_TYPE_NAME(CERT_COMP_LIST);
 	RR_TYPE_NAME(MEM_DETECT_EXTENDED);
 	RR_TYPE_NAME(VMEM);
-#ifdef CONFIG_KASAN
-	RR_TYPE_NAME(KASAN);
-#endif
 	default:
 		return "UNKNOWN";
 	}
@@ -166,17 +160,6 @@ static inline struct reserved_range *__physmem_reserved_next(enum reserved_range
 	     range; range = __physmem_reserved_next(&t, range),			\
 	    *p_start = range ? range->start : 0, *p_end = range ? range->end : 0)
 
-static inline unsigned long get_physmem_usable_total(void)
-{
-	unsigned long start, end, total = 0;
-	int i;
-
-	for_each_physmem_usable_range(i, &start, &end)
-		total += end - start;
-
-	return total;
-}
-
 static inline unsigned long get_physmem_reserved(enum reserved_range_type type,
 						 unsigned long *addr, unsigned long *size)
 {

arch/s390/kernel/early.c

@@ -51,6 +51,14 @@ decompressor_handled_param(nokaslr);
 decompressor_handled_param(prot_virt);
 #endif
 
+static void __init kasan_early_init(void)
+{
+#ifdef CONFIG_KASAN
+	init_task.kasan_depth = 0;
+	sclp_early_printk("KernelAddressSanitizer initialized\n");
+#endif
+}
+
 static void __init reset_tod_clock(void)
 {
 	union tod_clock clk;
@@ -293,6 +301,7 @@ static void __init sort_amode31_extable(void)
 
 void __init startup_init(void)
 {
+	kasan_early_init();
 	reset_tod_clock();
 	time_early_init();
 	init_kernel_storage_key();

arch/s390/kernel/head64.S

@@ -26,9 +26,6 @@ ENTRY(startup_continue)
 	stg	%r14,__LC_CURRENT
 	larl	%r15,init_thread_union+THREAD_SIZE-STACK_FRAME_OVERHEAD-__PT_SIZE
 	brasl	%r14,sclp_early_adjust_va	# allow sclp_early_printk
-#ifdef CONFIG_KASAN
-	brasl	%r14,kasan_early_init
-#endif
 	brasl	%r14,startup_init		# s390 specific early init
 	brasl	%r14,start_kernel		# common init code
 #

arch/s390/kernel/vmlinux.lds.S

@@ -219,6 +219,13 @@ SECTIONS
 		QUAD(init_mm)
 		QUAD(swapper_pg_dir)
 		QUAD(invalid_pg_dir)
+#ifdef CONFIG_KASAN
+		QUAD(kasan_early_shadow_page)
+		QUAD(kasan_early_shadow_pte)
+		QUAD(kasan_early_shadow_pmd)
+		QUAD(kasan_early_shadow_pud)
+		QUAD(kasan_early_shadow_p4d)
+#endif
 	} :NONE
 
 	/* Debugging sections.	*/

arch/s390/mm/Makefile

@@ -10,6 +10,3 @@ obj-$(CONFIG_CMM)		+= cmm.o
 obj-$(CONFIG_HUGETLB_PAGE)	+= hugetlbpage.o
 obj-$(CONFIG_PTDUMP_CORE)	+= dump_pagetables.o
 obj-$(CONFIG_PGSTE)		+= gmap.o
-
-KASAN_SANITIZE_kasan_init.o	:= n
-obj-$(CONFIG_KASAN)		+= kasan_init.o

arch/s390/mm/kasan_init.c

-// SPDX-License-Identifier: GPL-2.0
-#include <linux/memblock.h>
-#include <linux/pgtable.h>
-#include <linux/kasan.h>
-#include <asm/physmem_info.h>
-#include <asm/processor.h>
-#include <asm/facility.h>
-#include <asm/pgalloc.h>
-#include <asm/sclp.h>
-
-static unsigned long pgalloc_pos __initdata;
-static unsigned long segment_pos __initdata;
-static bool has_edat __initdata;
-static bool has_nx __initdata;
-
-#define __sha(x) ((unsigned long)kasan_mem_to_shadow((void *)x))
-
-static void __init kasan_early_panic(const char *reason)
-{
-	sclp_early_printk("The Linux kernel failed to boot with the KernelAddressSanitizer:\n");
-	sclp_early_printk(reason);
-	disabled_wait();
-}
-
-static void * __init kasan_early_alloc_segment(void)
-{
-	unsigned long addr = segment_pos;
-
-	segment_pos += _SEGMENT_SIZE;
-	if (segment_pos > pgalloc_pos)
-		kasan_early_panic("out of memory during initialisation\n");
-
-	return __va(addr);
-}
-
-static void * __init kasan_early_alloc_pages(unsigned int order)
-{
-	pgalloc_pos -= (PAGE_SIZE << order);
-
-	if (segment_pos > pgalloc_pos)
-		kasan_early_panic("out of memory during initialisation\n");
-
-	return __va(pgalloc_pos);
-}
-
-static void * __init kasan_early_crst_alloc(unsigned long val)
-{
-	unsigned long *table;
-
-	table = kasan_early_alloc_pages(CRST_ALLOC_ORDER);
-	if (table)
-		crst_table_init(table, val);
-	return table;
-}
-
-static pte_t * __init kasan_early_pte_alloc(void)
-{
-	static void *pte_leftover;
-	pte_t *pte;
-
-	BUILD_BUG_ON(_PAGE_TABLE_SIZE * 2 != PAGE_SIZE);
-
-	if (!pte_leftover) {
-		pte_leftover = kasan_early_alloc_pages(0);
-		pte = pte_leftover + _PAGE_TABLE_SIZE;
-	} else {
-		pte = pte_leftover;
-		pte_leftover = NULL;
-	}
-	memset64((u64 *)pte, _PAGE_INVALID, PTRS_PER_PTE);
-	return pte;
-}
-
-enum populate_mode {
-	POPULATE_MAP,
-	POPULATE_ZERO_SHADOW,
-	POPULATE_SHALLOW
-};
-
-static inline pgprot_t pgprot_clear_bit(pgprot_t pgprot, unsigned long bit)
-{
-	return __pgprot(pgprot_val(pgprot) & ~bit);
-}
-
-static void __init kasan_early_pgtable_populate(unsigned long address,
-						unsigned long end,
-						enum populate_mode mode)
-{
-	pgprot_t pgt_prot_zero = PAGE_KERNEL_RO;
-	pgprot_t pgt_prot = PAGE_KERNEL;
-	pgprot_t sgt_prot = SEGMENT_KERNEL;
-	pgd_t *pg_dir;
-	p4d_t *p4_dir;
-	pud_t *pu_dir;
-	pmd_t *pm_dir;
-	pte_t *pt_dir;
-	pmd_t pmd;
-	pte_t pte;
-
-	if (!has_nx) {
-		pgt_prot_zero = pgprot_clear_bit(pgt_prot_zero, _PAGE_NOEXEC);
-		pgt_prot = pgprot_clear_bit(pgt_prot, _PAGE_NOEXEC);
-		sgt_prot = pgprot_clear_bit(sgt_prot, _SEGMENT_ENTRY_NOEXEC);
-	}
-
-	while (address < end) {
-		pg_dir = pgd_offset_k(address);
-		if (pgd_none(*pg_dir)) {
-			if (mode == POPULATE_ZERO_SHADOW &&
-			    IS_ALIGNED(address, PGDIR_SIZE) &&
-			    end - address >= PGDIR_SIZE) {
-				pgd_populate(&init_mm, pg_dir,
-						kasan_early_shadow_p4d);
-				address = (address + PGDIR_SIZE) & PGDIR_MASK;
-				continue;
-			}
-			p4_dir = kasan_early_crst_alloc(_REGION2_ENTRY_EMPTY);
-			pgd_populate(&init_mm, pg_dir, p4_dir);
-		}
-
-		if (mode == POPULATE_SHALLOW) {
-			address = (address + P4D_SIZE) & P4D_MASK;
-			continue;
-		}
-
-		p4_dir = p4d_offset(pg_dir, address);
-		if (p4d_none(*p4_dir)) {
-			if (mode == POPULATE_ZERO_SHADOW &&
-			    IS_ALIGNED(address, P4D_SIZE) &&
-			    end - address >= P4D_SIZE) {
-				p4d_populate(&init_mm, p4_dir,
-						kasan_early_shadow_pud);
-				address = (address + P4D_SIZE) & P4D_MASK;
-				continue;
-			}
-			pu_dir = kasan_early_crst_alloc(_REGION3_ENTRY_EMPTY);
-			p4d_populate(&init_mm, p4_dir, pu_dir);
-		}
-
-		pu_dir = pud_offset(p4_dir, address);
-		if (pud_none(*pu_dir)) {
-			if (mode == POPULATE_ZERO_SHADOW &&
-			    IS_ALIGNED(address, PUD_SIZE) &&
-			    end - address >= PUD_SIZE) {
-				pud_populate(&init_mm, pu_dir,
-						kasan_early_shadow_pmd);
-				address = (address + PUD_SIZE) & PUD_MASK;
-				continue;
-			}
-			pm_dir = kasan_early_crst_alloc(_SEGMENT_ENTRY_EMPTY);
-			pud_populate(&init_mm, pu_dir, pm_dir);
-		}
-
-		pm_dir = pmd_offset(pu_dir, address);
-		if (pmd_none(*pm_dir)) {
-			if (IS_ALIGNED(address, PMD_SIZE) &&
-			    end - address >= PMD_SIZE) {
-				if (mode == POPULATE_ZERO_SHADOW) {
-					pmd_populate(&init_mm, pm_dir, kasan_early_shadow_pte);
-					address = (address + PMD_SIZE) & PMD_MASK;
-					continue;
-				} else if (has_edat) {
-					void *page = kasan_early_alloc_segment();
-
-					memset(page, 0, _SEGMENT_SIZE);
-					pmd = __pmd(__pa(page));
-					pmd = set_pmd_bit(pmd, sgt_prot);
-					set_pmd(pm_dir, pmd);
-					address = (address + PMD_SIZE) & PMD_MASK;
-					continue;
-				}
-			}
-			pt_dir = kasan_early_pte_alloc();
-			pmd_populate(&init_mm, pm_dir, pt_dir);
-		} else if (pmd_large(*pm_dir)) {
-			address = (address + PMD_SIZE) & PMD_MASK;
-			continue;
-		}
-
-		pt_dir = pte_offset_kernel(pm_dir, address);
-		if (pte_none(*pt_dir)) {
-			void *page;
-
-			switch (mode) {
-			case POPULATE_MAP:
-				page = kasan_early_alloc_pages(0);
-				memset(page, 0, PAGE_SIZE);
-				pte = __pte(__pa(page));
-				pte = set_pte_bit(pte, pgt_prot);
-				set_pte(pt_dir, pte);
-				break;
-			case POPULATE_ZERO_SHADOW:
-				page = kasan_early_shadow_page;
-				pte = __pte(__pa(page));
-				pte = set_pte_bit(pte, pgt_prot_zero);
-				set_pte(pt_dir, pte);
-				break;
-			case POPULATE_SHALLOW:
-				/* should never happen */
-				break;
-			}
-		}
-		address += PAGE_SIZE;
-	}
-}
-
-static void __init kasan_early_detect_facilities(void)
-{
-	if (test_facility(8)) {
-		has_edat = true;
-		__ctl_set_bit(0, 23);
-	}
-	if (!noexec_disabled && test_facility(130)) {
-		has_nx = true;
-		__ctl_set_bit(0, 20);
-	}
-}
-
-void __init kasan_early_init(void)
-{
-	pte_t pte_z = __pte(__pa(kasan_early_shadow_page) | pgprot_val(PAGE_KERNEL_RO));
-	pmd_t pmd_z = __pmd(__pa(kasan_early_shadow_pte) | _SEGMENT_ENTRY);
-	pud_t pud_z = __pud(__pa(kasan_early_shadow_pmd) | _REGION3_ENTRY);
-	p4d_t p4d_z = __p4d(__pa(kasan_early_shadow_pud) | _REGION2_ENTRY);
-	unsigned long pgalloc_pos_initial, segment_pos_initial;
-	unsigned long untracked_end = MODULES_VADDR;
-	unsigned long start, end;
-	int i;
-
-	kasan_early_detect_facilities();
-	if (!has_nx)
-		pte_z = clear_pte_bit(pte_z, __pgprot(_PAGE_NOEXEC));
-
-	BUILD_BUG_ON(!IS_ALIGNED(KASAN_SHADOW_START, P4D_SIZE));
-	BUILD_BUG_ON(!IS_ALIGNED(KASAN_SHADOW_END, P4D_SIZE));
-
-	/* init kasan zero shadow */
-	crst_table_init((unsigned long *)kasan_early_shadow_p4d, p4d_val(p4d_z));
-	crst_table_init((unsigned long *)kasan_early_shadow_pud, pud_val(pud_z));
-	crst_table_init((unsigned long *)kasan_early_shadow_pmd, pmd_val(pmd_z));
-	memset64((u64 *)kasan_early_shadow_pte, pte_val(pte_z), PTRS_PER_PTE);
-
-	/* segment allocations go bottom up -> <- pgalloc go top down */
-	segment_pos_initial = physmem_info.reserved[RR_KASAN].start;
-	segment_pos = segment_pos_initial;
-	pgalloc_pos_initial = physmem_info.reserved[RR_KASAN].end;
-	pgalloc_pos = pgalloc_pos_initial;
-	/*
-	 * Current memory layout:
-	 * +- 0 -------------+	       +- shadow start -+
-	 * |1:1 ident mapping|	      /|1/8 of ident map|
-	 * |		     |	     / |		|
-	 * +-end of ident map+	    /  +----------------+
-	 * | ... gap ...     |	   /   |    kasan	|
-	 * |		     |	  /    |  zero page	|
-	 * +- vmalloc area  -+	 /     |   mapping	|
-	 * | vmalloc_size    |	/      | (untracked)	|
-	 * +- modules vaddr -+ /       +----------------+
-	 * | 2Gb	     |/        |    unmapped	| allocated per module
-	 * +- shadow start  -+	       +----------------+
-	 * | 1/8 addr space  |	       | zero pg mapping| (untracked)
-	 * +- shadow end ----+---------+- shadow end ---+
-	 *
-	 * Current memory layout (KASAN_VMALLOC):
-	 * +- 0 -------------+	       +- shadow start -+
-	 * |1:1 ident mapping|	      /|1/8 of ident map|
-	 * |		     |	     / |		|
-	 * +-end of ident map+	    /  +----------------+
-	 * | ... gap ...     |	   /   | kasan zero page| (untracked)
-	 * |		     |	  /    | mapping	|
-	 * +- vmalloc area  -+	 /     +----------------+
-	 * | vmalloc_size    |	/      |shallow populate|
-	 * +- modules vaddr -+ /       +----------------+
-	 * | 2Gb	     |/        |shallow populate|
-	 * +- shadow start  -+	       +----------------+
-	 * | 1/8 addr space  |	       | zero pg mapping| (untracked)
-	 * +- shadow end ----+---------+- shadow end ---+
-	 */
-	/* populate kasan shadow (for identity mapping and zero page mapping) */
-	for_each_physmem_usable_range(i, &start, &end)
-		kasan_early_pgtable_populate(__sha(start), __sha(end), POPULATE_MAP);
-	if (IS_ENABLED(CONFIG_KASAN_VMALLOC)) {
-		untracked_end = VMALLOC_START;
-		/* shallowly populate kasan shadow for vmalloc and modules */
-		kasan_early_pgtable_populate(__sha(VMALLOC_START), __sha(MODULES_END),
-					     POPULATE_SHALLOW);
-	}
-	/* populate kasan shadow for untracked memory */
-	kasan_early_pgtable_populate(__sha(ident_map_size), __sha(untracked_end),
-				     POPULATE_ZERO_SHADOW);
-	kasan_early_pgtable_populate(__sha(MODULES_END), __sha(_REGION1_SIZE),
-				     POPULATE_ZERO_SHADOW);
-	/* enable kasan */
-	init_task.kasan_depth = 0;
-	sclp_early_printk("KernelAddressSanitizer initialized\n");
-	memblock_reserve(segment_pos_initial, segment_pos - segment_pos_initial);
-	memblock_reserve(pgalloc_pos, pgalloc_pos_initial - pgalloc_pos);
-}

arch/s390/mm/vmem.c

@@ -5,6 +5,7 @@
 
 #include <linux/memory_hotplug.h>
 #include <linux/memblock.h>
+#include <linux/kasan.h>
 #include <linux/pfn.h>
 #include <linux/mm.h>
 #include <linux/init.h>
@@ -664,6 +665,9 @@ static void __init memblock_region_swap(void *a, void *b, int size)
 	swap(*(struct memblock_region *)a, *(struct memblock_region *)b);
 }
 
+#ifdef CONFIG_KASAN
+#define __sha(x)	((unsigned long)kasan_mem_to_shadow((void *)x))
+#endif
 /*
  * map whole physical memory to virtual memory (identity mapping)
  * we reserve enough space in the vmalloc area for vmemmap to hotplug
@@ -733,6 +737,13 @@ void __init vmem_map_init(void)
 			     SET_MEMORY_RW | SET_MEMORY_NX);
 	}
 
+#ifdef CONFIG_KASAN
+	for_each_mem_range(i, &base, &end)
+		__set_memory(__sha(base),
+			     (__sha(end) - __sha(base)) >> PAGE_SHIFT,
+			     SET_MEMORY_RW | SET_MEMORY_NX);
+#endif
+
 	__set_memory((unsigned long)_stext,
 		     (unsigned long)(_etext - _stext) >> PAGE_SHIFT,
 		     SET_MEMORY_RO | SET_MEMORY_X);