[PATCH v3 0/5] Increased address space for 64 bit

[PATCH v3 0/5] Increased address space for 64 bit

[benjamin]

From: Benjamin Berg <benjamin.berg@intel.com>

This patchset fixes a few bugs, adds a new method of discovering the
host task size and finally adds four level page table support. All of
this means the userspace TASK_SIZE is much larger and in turns permits
userspace applications that need a lot of virtual addresses to work
fine.

One such application is ASAN which uses a fixed address in memory that
would otherwise not be addressable.

v2:
* Improved double clone logic using CLONE_VFORK
* Kconfig fixes pointed out by Tiwei Bie

v3:
* Undo incorrect change in child wait loop

Benjamin Berg (5):
  um: Fix stub_start address calculation
  um: Limit TASK_SIZE to the addressable range
  um: Do a double clone to disable rseq
  um: Discover host_task_size from envp
  um: Add 4 level page table support

 arch/um/Kconfig                      |   1 +
 arch/um/include/asm/page.h           |  14 +++-
 arch/um/include/asm/pgalloc.h        |  11 ++-
 arch/um/include/asm/pgtable-4level.h | 119 +++++++++++++++++++++++++++
 arch/um/include/asm/pgtable.h        |   6 +-
 arch/um/include/shared/as-layout.h   |   2 +-
 arch/um/include/shared/os.h          |   2 +-
 arch/um/kernel/mem.c                 |  17 +++-
 arch/um/kernel/um_arch.c             |  14 +++-
 arch/um/os-Linux/main.c              |   9 +-
 arch/um/os-Linux/skas/process.c      |  53 +++++++++++-
 arch/x86/um/Kconfig                  |  38 ++++++---
 arch/x86/um/os-Linux/task_size.c     |  19 ++++-
 13 files changed, 273 insertions(+), 32 deletions(-)
 create mode 100644 arch/um/include/asm/pgtable-4level.h

-- 
2.45.1

[PATCH v3 1/5] um: Fix stub_start address calculation

[benjamin]

From: Benjamin Berg <benjamin.berg@intel.com>

The calculation was wrong as it only subtracted one and then rounded
down for alignment. However, this is incorrect if host_task_size is not
already aligned.

This probably worked fine because on 64 bit the host_task_size is bigger
than returned by os_get_top_address.

Signed-off-by: Benjamin Berg <benjamin.berg@intel.com>
---
 arch/um/kernel/um_arch.c | 3 ++-
 1 file changed, 2 insertions(+), 1 deletion(-)

diff --git a/arch/um/kernel/um_arch.c b/arch/um/kernel/um_arch.c
index e95f805e5004..0d8b1a73cd5b 100644
--- a/arch/um/kernel/um_arch.c
+++ b/arch/um/kernel/um_arch.c
@@ -331,7 +331,8 @@ int __init linux_main(int argc, char **argv)
 	/* reserve a few pages for the stubs (taking care of data alignment) */
 	/* align the data portion */
 	BUILD_BUG_ON(!is_power_of_2(STUB_DATA_PAGES));
-	stub_start = (host_task_size - 1) & ~(STUB_DATA_PAGES * PAGE_SIZE - 1);
+	stub_start = (host_task_size - STUB_DATA_PAGES * PAGE_SIZE) &
+		     ~(STUB_DATA_PAGES * PAGE_SIZE - 1);
 	/* another page for the code portion */
 	stub_start -= PAGE_SIZE;
 	host_task_size = stub_start;
-- 
2.45.1

[PATCH v3 2/5] um: Limit TASK_SIZE to the addressable range

[benjamin]

From: Benjamin Berg <benjamin.berg@intel.com>

We may have a TASK_SIZE from the host that is bigger than UML is able to
address with a three-level pagetable. Guard against that by clipping the
maximum TASK_SIZE to the maximum addressable area.

Signed-off-by: Benjamin Berg <benjamin.berg@intel.com>
---
 arch/um/kernel/um_arch.c | 7 ++++++-
 1 file changed, 6 insertions(+), 1 deletion(-)

diff --git a/arch/um/kernel/um_arch.c b/arch/um/kernel/um_arch.c
index 0d8b1a73cd5b..5ab1a92b6bf7 100644
--- a/arch/um/kernel/um_arch.c
+++ b/arch/um/kernel/um_arch.c
@@ -337,11 +337,16 @@ int __init linux_main(int argc, char **argv)
 	stub_start -= PAGE_SIZE;
 	host_task_size = stub_start;
 
+	/* Limit TASK_SIZE to what is addressable by the page table */
+	task_size = host_task_size;
+	if (task_size > PTRS_PER_PGD * PGDIR_SIZE)
+		task_size = PTRS_PER_PGD * PGDIR_SIZE;
+
 	/*
 	 * TASK_SIZE needs to be PGDIR_SIZE aligned or else exit_mmap craps
 	 * out
 	 */
-	task_size = host_task_size & PGDIR_MASK;
+	task_size = task_size & PGDIR_MASK;
 
 	/* OS sanity checks that need to happen before the kernel runs */
 	os_early_checks();
-- 
2.45.1

[PATCH v3 3/5] um: Do a double clone to disable rseq

[benjamin]

From: Benjamin Berg <benjamin.berg@intel.com>

Newer glibc versions are enabling rseq support by default. This remains
enabled in the cloned child process, potentially causing the host kernel
to write/read memory in the child.

It appears that this was purely not an issue because the used memory
area happened to be above TASK_SIZE and remains mapped.

Note that a better approach would be to exec a small static binary that
does not link with other libraries. Using a memfd and execveat the
binary could be embedded into UML itself and it would result in an
entirely clean execution environment for userspace.

Signed-off-by: Benjamin Berg <benjamin.berg@intel.com>

---

v2: Improved clone logic using CLONE_VFORK
v3: Undo incorrect change in child wait loop
---
 arch/um/os-Linux/skas/process.c | 53 ++++++++++++++++++++++++++++++---
 1 file changed, 49 insertions(+), 4 deletions(-)

diff --git a/arch/um/os-Linux/skas/process.c b/arch/um/os-Linux/skas/process.c
index 41a288dcfc34..93d3d44337cd 100644
--- a/arch/um/os-Linux/skas/process.c
+++ b/arch/um/os-Linux/skas/process.c
@@ -255,6 +255,32 @@ static int userspace_tramp(void *stack)
 int userspace_pid[NR_CPUS];
 int kill_userspace_mm[NR_CPUS];
 
+struct tramp_data {
+	int pid;
+	void *clone_sp;
+	void *stack;
+};
+
+static int userspace_tramp_clone_vm(void *data)
+{
+	struct tramp_data *tramp_data = data;
+
+	/*
+	 * At this point we are still in the same VM as the parent, but rseq
+	 * has been disabled for this process.
+	 * Continue with the clone into the new userspace process, the kernel
+	 * continues as soon as this process quits (CLONE_VFORK).
+	 */
+
+	tramp_data->pid = clone(userspace_tramp, tramp_data->clone_sp,
+				CLONE_PARENT | CLONE_FILES | SIGCHLD,
+				tramp_data->stack);
+	if (tramp_data->pid < 0)
+		tramp_data->pid = -errno;
+
+	exit(0);
+}
+
 /**
  * start_userspace() - prepare a new userspace process
  * @stub_stack:	pointer to the stub stack.
@@ -268,9 +294,10 @@ int kill_userspace_mm[NR_CPUS];
  */
 int start_userspace(unsigned long stub_stack)
 {
+	struct tramp_data tramp_data;
 	void *stack;
 	unsigned long sp;
-	int pid, status, n, flags, err;
+	int pid, status, n, err;
 
 	/* setup a temporary stack page */
 	stack = mmap(NULL, UM_KERN_PAGE_SIZE,
@@ -286,10 +313,13 @@ int start_userspace(unsigned long stub_stack)
 	/* set stack pointer to the end of the stack page, so it can grow downwards */
 	sp = (unsigned long)stack + UM_KERN_PAGE_SIZE;
 
-	flags = CLONE_FILES | SIGCHLD;
+	tramp_data.stack = (void *) stub_stack;
+	tramp_data.clone_sp = (void *) sp;
+	tramp_data.pid = -EINVAL;
 
-	/* clone into new userspace process */
-	pid = clone(userspace_tramp, (void *) sp, flags, (void *) stub_stack);
+	/* first stage CLONE_VM clone using VFORK and no signal notification */
+	pid = clone(userspace_tramp_clone_vm, (void *) sp,
+		    CLONE_VM | CLONE_FILES | CLONE_VFORK, &tramp_data);
 	if (pid < 0) {
 		err = -errno;
 		printk(UM_KERN_ERR "%s : clone failed, errno = %d\n",
@@ -297,6 +327,21 @@ int start_userspace(unsigned long stub_stack)
 		return err;
 	}
 
+	n = waitpid(pid, &status, WUNTRACED | WNOHANG | __WCLONE);
+	if (n < 0 || !WIFEXITED(status) || WEXITSTATUS(status)) {
+		err = -errno;
+		printk(UM_KERN_ERR "%s : wait failed, errno = %d, status = %d\n",
+		       __func__, n < 0 ? errno : 0, status);
+		goto out_kill;
+	}
+
+	pid = tramp_data.pid;
+	if (pid < 0) {
+		printk(UM_KERN_ERR "%s : second clone failed, errno = %d\n",
+		       __func__, -pid);
+		return pid;
+	}
+
 	do {
 		CATCH_EINTR(n = waitpid(pid, &status, WUNTRACED | __WALL));
 		if (n < 0) {
-- 
2.45.1

[PATCH v3 4/5] um: Discover host_task_size from envp

[benjamin]

From: Benjamin Berg <benjamin.berg@intel.com>

When loading the UML binary, the host kernel will place the stack at the
highest possible address. It will then map the program name and
environment variables onto the start of the stack.

As such, an easy way to figure out the host_task_size is to use the
highest pointer to an environment variable as a reference.

Ensure that this works by disabling address layout randomization and
re-executing UML in case it was enabled.

This increases the available TASK_SIZE for 64 bit UML considerably.

Signed-off-by: Benjamin Berg <benjamin.berg@intel.com>
---
 arch/um/include/shared/as-layout.h |  2 +-
 arch/um/include/shared/os.h        |  2 +-
 arch/um/kernel/um_arch.c           |  4 ++--
 arch/um/os-Linux/main.c            |  9 ++++++++-
 arch/x86/um/os-Linux/task_size.c   | 19 +++++++++++++++----
 5 files changed, 27 insertions(+), 9 deletions(-)

diff --git a/arch/um/include/shared/as-layout.h b/arch/um/include/shared/as-layout.h
index c22f46a757dc..480bb44ea1f2 100644
--- a/arch/um/include/shared/as-layout.h
+++ b/arch/um/include/shared/as-layout.h
@@ -48,7 +48,7 @@ extern unsigned long brk_start;
 extern unsigned long host_task_size;
 extern unsigned long stub_start;
 
-extern int linux_main(int argc, char **argv);
+extern int linux_main(int argc, char **argv, char **envp);
 extern void uml_finishsetup(void);
 
 struct siginfo;
diff --git a/arch/um/include/shared/os.h b/arch/um/include/shared/os.h
index aff8906304ea..db644fc67069 100644
--- a/arch/um/include/shared/os.h
+++ b/arch/um/include/shared/os.h
@@ -327,7 +327,7 @@ extern int __ignore_sigio_fd(int fd);
 extern int get_pty(void);
 
 /* sys-$ARCH/task_size.c */
-extern unsigned long os_get_top_address(void);
+extern unsigned long os_get_top_address(char **envp);
 
 long syscall(long number, ...);
 
diff --git a/arch/um/kernel/um_arch.c b/arch/um/kernel/um_arch.c
index 5ab1a92b6bf7..046eaf356b28 100644
--- a/arch/um/kernel/um_arch.c
+++ b/arch/um/kernel/um_arch.c
@@ -305,7 +305,7 @@ static void parse_cache_line(char *line)
 	}
 }
 
-int __init linux_main(int argc, char **argv)
+int __init linux_main(int argc, char **argv, char **envp)
 {
 	unsigned long avail, diff;
 	unsigned long virtmem_size, max_physmem;
@@ -327,7 +327,7 @@ int __init linux_main(int argc, char **argv)
 	if (have_console == 0)
 		add_arg(DEFAULT_COMMAND_LINE_CONSOLE);
 
-	host_task_size = os_get_top_address();
+	host_task_size = os_get_top_address(envp);
 	/* reserve a few pages for the stubs (taking care of data alignment) */
 	/* align the data portion */
 	BUILD_BUG_ON(!is_power_of_2(STUB_DATA_PAGES));
diff --git a/arch/um/os-Linux/main.c b/arch/um/os-Linux/main.c
index f98ff79cdbf7..9a61b1767795 100644
--- a/arch/um/os-Linux/main.c
+++ b/arch/um/os-Linux/main.c
@@ -11,6 +11,7 @@
 #include <signal.h>
 #include <string.h>
 #include <sys/resource.h>
+#include <sys/personality.h>
 #include <as-layout.h>
 #include <init.h>
 #include <kern_util.h>
@@ -108,6 +109,12 @@ int __init main(int argc, char **argv, char **envp)
 	char **new_argv;
 	int ret, i, err;
 
+	/* Disable randomization and re-exec if it was changed successfully */
+	ret = personality(PER_LINUX | ADDR_NO_RANDOMIZE);
+	if (ret >= 0 && (ret & (PER_LINUX | ADDR_NO_RANDOMIZE)) !=
+			 (PER_LINUX | ADDR_NO_RANDOMIZE))
+		execve("/proc/self/exe", argv, envp);
+
 	set_stklim();
 
 	setup_env_path();
@@ -140,7 +147,7 @@ int __init main(int argc, char **argv, char **envp)
 #endif
 
 	change_sig(SIGPIPE, 0);
-	ret = linux_main(argc, argv);
+	ret = linux_main(argc, argv, envp);
 
 	/*
 	 * Disable SIGPROF - I have no idea why libc doesn't do this or turn
diff --git a/arch/x86/um/os-Linux/task_size.c b/arch/x86/um/os-Linux/task_size.c
index 1dc9adc20b1c..33c26291545a 100644
--- a/arch/x86/um/os-Linux/task_size.c
+++ b/arch/x86/um/os-Linux/task_size.c
@@ -65,7 +65,7 @@ static int page_ok(unsigned long page)
 	return ok;
 }
 
-unsigned long os_get_top_address(void)
+unsigned long os_get_top_address(char **envp)
 {
 	struct sigaction sa, old;
 	unsigned long bottom = 0;
@@ -142,10 +142,21 @@ unsigned long os_get_top_address(void)
 
 #else
 
-unsigned long os_get_top_address(void)
+unsigned long os_get_top_address(char **envp)
 {
-	/* The old value of CONFIG_TOP_ADDR */
-	return 0x7fc0002000;
+	unsigned long top_addr = (unsigned long) &top_addr;
+	int i;
+
+	/* The earliest variable should be after the program name in ELF */
+	for (i = 0; envp[i]; i++) {
+		if ((unsigned long) envp[i] > top_addr)
+			top_addr = (unsigned long) envp[i];
+	}
+
+	top_addr &= ~(UM_KERN_PAGE_SIZE - 1);
+	top_addr += UM_KERN_PAGE_SIZE;
+
+	return top_addr;
 }
 
 #endif
-- 
2.45.1

[PATCH v3 5/5] um: Add 4 level page table support

[benjamin]

From: Benjamin Berg <benjamin.berg@intel.com>

The larger memory space is useful to support more applications inside
UML. One example for this is ASAN instrumentation of userspace
applications which requires addresses that would otherwise not be
available.

Signed-off-by: Benjamin Berg <benjamin.berg@intel.com>

---

v2:
- Do not hide option behind the EXPERT flag
- Fix typo in new "Two-level pagetables" option
---
 arch/um/Kconfig                      |   1 +
 arch/um/include/asm/page.h           |  14 +++-
 arch/um/include/asm/pgalloc.h        |  11 ++-
 arch/um/include/asm/pgtable-4level.h | 119 +++++++++++++++++++++++++++
 arch/um/include/asm/pgtable.h        |   6 +-
 arch/um/kernel/mem.c                 |  17 +++-
 arch/x86/um/Kconfig                  |  38 ++++++---
 7 files changed, 189 insertions(+), 17 deletions(-)
 create mode 100644 arch/um/include/asm/pgtable-4level.h

diff --git a/arch/um/Kconfig b/arch/um/Kconfig
index 93a5a8999b07..5d111fc8ccb7 100644
--- a/arch/um/Kconfig
+++ b/arch/um/Kconfig
@@ -208,6 +208,7 @@ config MMAPPER
 
 config PGTABLE_LEVELS
 	int
+	default 4 if 4_LEVEL_PGTABLES
 	default 3 if 3_LEVEL_PGTABLES
 	default 2
 
diff --git a/arch/um/include/asm/page.h b/arch/um/include/asm/page.h
index 9ef9a8aedfa6..c3b2ae03b60c 100644
--- a/arch/um/include/asm/page.h
+++ b/arch/um/include/asm/page.h
@@ -57,14 +57,22 @@ typedef unsigned long long phys_t;
 typedef struct { unsigned long pte; } pte_t;
 typedef struct { unsigned long pgd; } pgd_t;
 
-#ifdef CONFIG_3_LEVEL_PGTABLES
+#if CONFIG_PGTABLE_LEVELS > 2
+
 typedef struct { unsigned long pmd; } pmd_t;
 #define pmd_val(x)	((x).pmd)
 #define __pmd(x) ((pmd_t) { (x) } )
-#endif
 
-#define pte_val(x)	((x).pte)
+#if CONFIG_PGTABLE_LEVELS > 3
 
+typedef struct { unsigned long pud; } pud_t;
+#define pud_val(x)	((x).pud)
+#define __pud(x) ((pud_t) { (x) } )
+
+#endif /* CONFIG_PGTABLE_LEVELS > 3 */
+#endif /* CONFIG_PGTABLE_LEVELS > 2 */
+
+#define pte_val(x)	((x).pte)
 
 #define pte_get_bits(p, bits) ((p).pte & (bits))
 #define pte_set_bits(p, bits) ((p).pte |= (bits))
diff --git a/arch/um/include/asm/pgalloc.h b/arch/um/include/asm/pgalloc.h
index de5e31c64793..04fb4e6969a4 100644
--- a/arch/um/include/asm/pgalloc.h
+++ b/arch/um/include/asm/pgalloc.h
@@ -31,7 +31,7 @@ do {								\
 	tlb_remove_page_ptdesc((tlb), (page_ptdesc(pte)));	\
 } while (0)
 
-#ifdef CONFIG_3_LEVEL_PGTABLES
+#if CONFIG_PGTABLE_LEVELS > 2
 
 #define __pmd_free_tlb(tlb, pmd, address)			\
 do {								\
@@ -39,6 +39,15 @@ do {								\
 	tlb_remove_page_ptdesc((tlb), virt_to_ptdesc(pmd));	\
 } while (0)
 
+#if CONFIG_PGTABLE_LEVELS > 3
+
+#define __pud_free_tlb(tlb, pud, address)			\
+do {								\
+	pagetable_pud_dtor(virt_to_ptdesc(pud));		\
+	tlb_remove_page_ptdesc((tlb), virt_to_ptdesc(pud));	\
+} while (0)
+
+#endif
 #endif
 
 #endif
diff --git a/arch/um/include/asm/pgtable-4level.h b/arch/um/include/asm/pgtable-4level.h
new file mode 100644
index 000000000000..f912fcc16b7a
--- /dev/null
+++ b/arch/um/include/asm/pgtable-4level.h
@@ -0,0 +1,119 @@
+/* SPDX-License-Identifier: GPL-2.0 */
+/*
+ * Copyright 2003 PathScale Inc
+ * Derived from include/asm-i386/pgtable.h
+ */
+
+#ifndef __UM_PGTABLE_4LEVEL_H
+#define __UM_PGTABLE_4LEVEL_H
+
+#include <asm-generic/pgtable-nop4d.h>
+
+/* PGDIR_SHIFT determines what a fourth-level page table entry can map */
+
+#define PGDIR_SHIFT	39
+#define PGDIR_SIZE	(1UL << PGDIR_SHIFT)
+#define PGDIR_MASK	(~(PGDIR_SIZE-1))
+
+/* PUD_SHIFT determines the size of the area a third-level page table can
+ * map
+ */
+
+#define PUD_SHIFT	30
+#define PUD_SIZE	(1UL << PUD_SHIFT)
+#define PUD_MASK	(~(PUD_SIZE-1))
+
+/* PMD_SHIFT determines the size of the area a second-level page table can
+ * map
+ */
+
+#define PMD_SHIFT	21
+#define PMD_SIZE	(1UL << PMD_SHIFT)
+#define PMD_MASK	(~(PMD_SIZE-1))
+
+/*
+ * entries per page directory level
+ */
+
+#define PTRS_PER_PTE 512
+#define PTRS_PER_PMD 512
+#define PTRS_PER_PUD 512
+#define PTRS_PER_PGD 512
+
+#define USER_PTRS_PER_PGD ((TASK_SIZE + (PGDIR_SIZE - 1)) / PGDIR_SIZE)
+
+#define pte_ERROR(e) \
+        printk("%s:%d: bad pte %p(%016lx).\n", __FILE__, __LINE__, &(e), \
+	       pte_val(e))
+#define pmd_ERROR(e) \
+        printk("%s:%d: bad pmd %p(%016lx).\n", __FILE__, __LINE__, &(e), \
+	       pmd_val(e))
+#define pud_ERROR(e) \
+        printk("%s:%d: bad pud %p(%016lx).\n", __FILE__, __LINE__, &(e), \
+	       pud_val(e))
+#define pgd_ERROR(e) \
+        printk("%s:%d: bad pgd %p(%016lx).\n", __FILE__, __LINE__, &(e), \
+	       pgd_val(e))
+
+#define pud_none(x)	(!(pud_val(x) & ~_PAGE_NEWPAGE))
+#define	pud_bad(x)	((pud_val(x) & (~PAGE_MASK & ~_PAGE_USER)) != _KERNPG_TABLE)
+#define pud_present(x)	(pud_val(x) & _PAGE_PRESENT)
+#define pud_populate(mm, pud, pmd) \
+	set_pud(pud, __pud(_PAGE_TABLE + __pa(pmd)))
+
+#define set_pud(pudptr, pudval) (*(pudptr) = (pudval))
+
+#define p4d_none(x)	(!(p4d_val(x) & ~_PAGE_NEWPAGE))
+#define	p4d_bad(x)	((p4d_val(x) & (~PAGE_MASK & ~_PAGE_USER)) != _KERNPG_TABLE)
+#define p4d_present(x)	(p4d_val(x) & _PAGE_PRESENT)
+#define p4d_populate(mm, p4d, pud) \
+	set_p4d(p4d, __p4d(_PAGE_TABLE + __pa(pud)))
+
+#define set_p4d(p4dptr, p4dval) (*(p4dptr) = (p4dval))
+
+
+static inline int pgd_newpage(pgd_t pgd)
+{
+	return(pgd_val(pgd) & _PAGE_NEWPAGE);
+}
+
+static inline void pgd_mkuptodate(pgd_t pgd) { pgd_val(pgd) &= ~_PAGE_NEWPAGE; }
+
+#define set_pmd(pmdptr, pmdval) (*(pmdptr) = (pmdval))
+
+static inline void pud_clear (pud_t *pud)
+{
+	set_pud(pud, __pud(_PAGE_NEWPAGE));
+}
+
+static inline void p4d_clear (p4d_t *p4d)
+{
+	set_p4d(p4d, __p4d(_PAGE_NEWPAGE));
+}
+
+#define pud_page(pud) phys_to_page(pud_val(pud) & PAGE_MASK)
+#define pud_pgtable(pud) ((pmd_t *) __va(pud_val(pud) & PAGE_MASK))
+
+#define p4d_page(p4d) phys_to_page(p4d_val(p4d) & PAGE_MASK)
+#define p4d_pgtable(p4d) ((pud_t *) __va(p4d_val(p4d) & PAGE_MASK))
+
+static inline unsigned long pte_pfn(pte_t pte)
+{
+	return phys_to_pfn(pte_val(pte));
+}
+
+static inline pte_t pfn_pte(unsigned long page_nr, pgprot_t pgprot)
+{
+	pte_t pte;
+	phys_t phys = pfn_to_phys(page_nr);
+
+	pte_set_val(pte, phys, pgprot);
+	return pte;
+}
+
+static inline pmd_t pfn_pmd(unsigned long page_nr, pgprot_t pgprot)
+{
+	return __pmd((page_nr << PAGE_SHIFT) | pgprot_val(pgprot));
+}
+
+#endif
diff --git a/arch/um/include/asm/pgtable.h b/arch/um/include/asm/pgtable.h
index e1ece21dbe3f..71a7651e2db7 100644
--- a/arch/um/include/asm/pgtable.h
+++ b/arch/um/include/asm/pgtable.h
@@ -24,9 +24,11 @@
 /* We borrow bit 10 to store the exclusive marker in swap PTEs. */
 #define _PAGE_SWP_EXCLUSIVE	0x400
 
-#ifdef CONFIG_3_LEVEL_PGTABLES
+#if CONFIG_PGTABLE_LEVELS == 4
+#include <asm/pgtable-4level.h>
+#elif CONFIG_PGTABLE_LEVELS == 3
 #include <asm/pgtable-3level.h>
-#else
+#elif CONFIG_PGTABLE_LEVELS == 2
 #include <asm/pgtable-2level.h>
 #endif
 
diff --git a/arch/um/kernel/mem.c b/arch/um/kernel/mem.c
index ca91accd64fc..2dc0d90c0550 100644
--- a/arch/um/kernel/mem.c
+++ b/arch/um/kernel/mem.c
@@ -99,7 +99,7 @@ static void __init one_page_table_init(pmd_t *pmd)
 
 static void __init one_md_table_init(pud_t *pud)
 {
-#ifdef CONFIG_3_LEVEL_PGTABLES
+#if CONFIG_PGTABLE_LEVELS > 2
 	pmd_t *pmd_table = (pmd_t *) memblock_alloc_low(PAGE_SIZE, PAGE_SIZE);
 	if (!pmd_table)
 		panic("%s: Failed to allocate %lu bytes align=%lx\n",
@@ -110,6 +110,19 @@ static void __init one_md_table_init(pud_t *pud)
 #endif
 }
 
+static void __init one_ud_table_init(p4d_t *p4d)
+{
+#if CONFIG_PGTABLE_LEVELS > 3
+	pud_t *pud_table = (pud_t *) memblock_alloc_low(PAGE_SIZE, PAGE_SIZE);
+	if (!pud_table)
+		panic("%s: Failed to allocate %lu bytes align=%lx\n",
+		      __func__, PAGE_SIZE, PAGE_SIZE);
+
+	set_p4d(p4d, __p4d(_KERNPG_TABLE + (unsigned long) __pa(pud_table)));
+	BUG_ON(pud_table != pud_offset(p4d, 0));
+#endif
+}
+
 static void __init fixrange_init(unsigned long start, unsigned long end,
 				 pgd_t *pgd_base)
 {
@@ -127,6 +140,8 @@ static void __init fixrange_init(unsigned long start, unsigned long end,
 
 	for ( ; (i < PTRS_PER_PGD) && (vaddr < end); pgd++, i++) {
 		p4d = p4d_offset(pgd, vaddr);
+		if (p4d_none(*p4d))
+			one_ud_table_init(p4d);
 		pud = pud_offset(p4d, vaddr);
 		if (pud_none(*pud))
 			one_md_table_init(pud);
diff --git a/arch/x86/um/Kconfig b/arch/x86/um/Kconfig
index 186f13268401..454ad560f627 100644
--- a/arch/x86/um/Kconfig
+++ b/arch/x86/um/Kconfig
@@ -28,16 +28,34 @@ config X86_64
 	def_bool 64BIT
 	select MODULES_USE_ELF_RELA
 
-config 3_LEVEL_PGTABLES
-	bool "Three-level pagetables" if !64BIT
-	default 64BIT
-	help
-	  Three-level pagetables will let UML have more than 4G of physical
-	  memory.  All the memory that can't be mapped directly will be treated
-	  as high memory.
-
-	  However, this it experimental on 32-bit architectures, so if unsure say
-	  N (on x86-64 it's automatically enabled, instead, as it's safe there).
+choice
+	prompt "Pagetable levels"
+	default 2_LEVEL_PGTABLES if !64BIT
+	default 4_LEVEL_PGTABLES if 64BIT
+
+	config 2_LEVEL_PGTABLES
+		bool "Two-level pagetables" if !64BIT
+		depends on !64BIT
+		help
+		  Two-level page table for 32-bit architectures.
+
+	config 3_LEVEL_PGTABLES
+		bool "Three-level pagetables" if 64BIT
+		help
+		  Three-level pagetables will let UML have more than 4G of physical
+		  memory.  All the memory that can't be mapped directly will be treated
+		  as high memory.
+
+		  However, this it experimental on 32-bit architectures, so if unsure say
+		  N (on x86-64 it's automatically enabled, instead, as it's safe there).
+
+	config 4_LEVEL_PGTABLES
+		bool "Four-level pagetables" if 64BIT
+		depends on 64BIT
+		help
+		  Four-level pagetables, gives a bigger address space which can be
+		  useful for some applications (e.g. ASAN).
+endchoice
 
 config ARCH_HAS_SC_SIGNALS
 	def_bool !64BIT
-- 
2.45.1