[RFC][PATCH -mm 0/4] Hibernation: Arbitrary boot kernel support on x86_64

[RFC][PATCH -mm 0/4] Hibernation: Arbitrary boot kernel support on x86_64

[Rafael J. Wysocki]

Hi,

The following patches make it possible to restore a hibernation image with the
help of a kernel different from the image one.

The first patch adds the generic, platform independent code needed for that.
The second patch implements the idea on x86_64 and the two remaining patches
are generalizations.

Details in the changelogs, comments welcome.

Greetings,
Rafael


-- 
"Premature optimization is the root of all evil." - Donald Knuth

[RFC][PATCH -mm 1/4] Hibernation: Arbitrary boot kernel support - generic code

[Rafael J. Wysocki]

From: Rafael J. Wysocki <rjw@sisk.pl>

Add the bits needed for supporting arbitrary boot kernels to the common
hibernation code.

To support arbitrary boot kernels, make it possible to replace the 'struct
new_utsname' and the kernel version in the hibernation image header by some
architecture specific data that will be used to verify if the image is valid
and to restore the image.

Signed-off-by: Rafael J. Wysocki <rjw@sisk.pl>
---
 kernel/power/power.h    |    6 +++-
 kernel/power/snapshot.c |   60 ++++++++++++++++++++++++++++++++++++------------
 2 files changed, 49 insertions(+), 17 deletions(-)

Index: linux-2.6.23-rc3/kernel/power/power.h
===================================================================
--- linux-2.6.23-rc3.orig/kernel/power/power.h
+++ linux-2.6.23-rc3/kernel/power/power.h
@@ -11,14 +11,16 @@ struct swsusp_info {
 	unsigned long		size;
 } __attribute__((aligned(PAGE_SIZE)));
 
-
-
 #ifdef CONFIG_HIBERNATION
+/* Maximum size of architecture specific data in a hibernation header */
+#define MAX_ARCH_HEADER_SIZE	(sizeof(struct new_utsname) + 4)
+
 /*
  * Keep some memory free so that I/O operations can succeed without paging
  * [Might this be more than 4 MB?]
  */
 #define PAGES_FOR_IO	((4096 * 1024) >> PAGE_SHIFT)
+
 /*
  * Keep 1 MB of memory free so that device drivers can allocate some pages in
  * their .suspend() routines without breaking the suspend to disk.
Index: linux-2.6.23-rc3/kernel/power/snapshot.c
===================================================================
--- linux-2.6.23-rc3.orig/kernel/power/snapshot.c
+++ linux-2.6.23-rc3/kernel/power/snapshot.c
@@ -1239,17 +1239,29 @@ asmlinkage int swsusp_save(void)
 	return 0;
 }
 
-static void init_header(struct swsusp_info *info)
+#ifdef ARCH_HAS_HIBERNATION_HEADER
+static int init_header_complete(struct swsusp_info *info)
 {
-	memset(info, 0, sizeof(struct swsusp_info));
+	return arch_hibernation_header_save(info, MAX_ARCH_HEADER_SIZE);
+}
+#else /* !ARCH_HAS_HIBERNATION_HEADER */
+static int init_header_complete(struct swsusp_info *info)
+{
+	memcpy(&info->uts, init_utsname(), sizeof(struct new_utsname));
 	info->version_code = LINUX_VERSION_CODE;
+	return 0;
+}
+#endif /* !ARCH_HAS_HIBERNATION_HEADER */
+
+static int init_header(struct swsusp_info *info)
+{
+	memset(info, 0, sizeof(struct swsusp_info));
 	info->num_physpages = num_physpages;
-	memcpy(&info->uts, init_utsname(), sizeof(struct new_utsname));
-	info->cpus = num_online_cpus();
 	info->image_pages = nr_copy_pages;
 	info->pages = nr_copy_pages + nr_meta_pages + 1;
 	info->size = info->pages;
 	info->size <<= PAGE_SHIFT;
+	return init_header_complete(info);
 }
 
 /**
@@ -1303,7 +1315,11 @@ int snapshot_read_next(struct snapshot_h
 			return -ENOMEM;
 	}
 	if (!handle->offset) {
-		init_header((struct swsusp_info *)buffer);
+		int error;
+
+		error = init_header((struct swsusp_info *)buffer);
+		if (error)
+			return error;
 		handle->buffer = buffer;
 		memory_bm_position_reset(&orig_bm);
 		memory_bm_position_reset(&copy_bm);
@@ -1394,22 +1410,36 @@ duplicate_memory_bitmap(struct memory_bi
 	}
 }
 
-static inline int check_header(struct swsusp_info *info)
+#ifdef ARCH_HAS_HIBERNATION_HEADER
+static char *check_image_kernel(struct swsusp_info *info)
+{
+	return arch_hibernation_header_restore(info) ?
+			"architecture specific data" : NULL;
+}
+#else /* !ARCH_HAS_HIBERNATION_HEADER */
+static char *check_image_kernel(struct swsusp_info *info)
 {
-	char *reason = NULL;
-
 	if (info->version_code != LINUX_VERSION_CODE)
-		reason = "kernel version";
-	if (info->num_physpages != num_physpages)
-		reason = "memory size";
+		return "kernel version";
 	if (strcmp(info->uts.sysname,init_utsname()->sysname))
-		reason = "system type";
+		return "system type";
 	if (strcmp(info->uts.release,init_utsname()->release))
-		reason = "kernel release";
+		return "kernel release";
 	if (strcmp(info->uts.version,init_utsname()->version))
-		reason = "version";
+		return "version";
 	if (strcmp(info->uts.machine,init_utsname()->machine))
-		reason = "machine";
+		return "machine";
+	return NULL;
+}
+#endif /* !ARCH_HAS_HIBERNATION_HEADER */
+
+static int check_header(struct swsusp_info *info)
+{
+	char *reason;
+
+	reason = check_image_kernel(info);
+	if (!reason && info->num_physpages != num_physpages)
+		reason = "memory size";
 	if (reason) {
 		printk(KERN_ERR "swsusp: Resume mismatch: %s\n", reason);
 		return -EPERM;

Re: [RFC][PATCH -mm 1/4] Hibernation: Arbitrary boot kernel support - generic code

[Pavel Machek]

On Wed 2007-08-22 10:17:12, Rafael J. Wysocki wrote:
> From: Rafael J. Wysocki <rjw@sisk.pl>
> 
> Add the bits needed for supporting arbitrary boot kernels to the common
> hibernation code.
> 
> To support arbitrary boot kernels, make it possible to replace the 'struct
> new_utsname' and the kernel version in the hibernation image header by some
> architecture specific data that will be used to verify if the image is valid
> and to restore the image.
> 
> Signed-off-by: Rafael J. Wysocki <rjw@sisk.pl>

ACK.
									Pavel

-- 
(english) http://www.livejournal.com/~pavelmachek
(cesky, pictures) http://atrey.karlin.mff.cuni.cz/~pavel/picture/horses/blog.html

[RFC][PATCH -mm 2/4] Hibernation: Arbitrary boot kernel support on x86_64

[Rafael J. Wysocki]

From: Rafael J. Wysocki <rjw@sisk.pl>

Make it possible to restore a hibernation image on x86_64 with the help of a
kernel different from the one in the image.

The idea is to split the core restoration code into two separate parts and to
place each of them in a different page.  The first part belongs to the boot
kernel and is executed as the last step of the image kernel's memory restoration
procedure.  It restores all of the image kernel's memory that has not been
restored yet except for the one page containing the very code that is being
executed at that time.  The final operation performed by it is a jump to the
second part of the core restoration code that belongs to the image kernel and
has just been restored.  This code restores the last remaining page of the image
kernel's memory containing the first, already executed, part of the core
restoration code (temporary page tables created by the boot kernel are used at
this stage).  It also makes the CPU switch to the image kernel's page tables and
restores the state of general purpose registers (including the stack pointer)
from before the hibernation.

The main issue with this idea is that in order to jump to the second part of the
restoration code the boot kernel needs to know its address.  However, this
address may be passed to it in the image header.  Namely, the part of the image
header previously used for checking if the version of the image kernel is
correct can be replaced with some architecture specific data that will allow
the boot kernel to jump to the right address within the image kernel.  These
data should also be used for checking if the image kernel is compatible with
the boot kernel (as far as the memory restroration procedure is concerned).
It can be done, for example, with the help of a "magic" value that has to be
equal in both kernels, so that they can be regarded as compatible.

Signed-off-by: Rafael J. Wysocki <rjw@sisk.pl>
---
 arch/x86_64/kernel/suspend.c     |   43 ++++++++++++++++++++++++++
 arch/x86_64/kernel/suspend_asm.S |   63 +++++++++++++++++++++++++++++++++------
 include/asm-x86_64/suspend.h     |    6 +++
 3 files changed, 103 insertions(+), 9 deletions(-)

Index: linux-2.6.23-rc3/arch/x86_64/kernel/suspend_asm.S
===================================================================
--- linux-2.6.23-rc3.orig/arch/x86_64/kernel/suspend_asm.S	2007-08-21 20:36:49.000000000 +0200
+++ linux-2.6.23-rc3/arch/x86_64/kernel/suspend_asm.S	2007-08-21 21:16:01.000000000 +0200
@@ -2,8 +2,8 @@
  *
  * Distribute under GPLv2.
  *
- * swsusp_arch_resume may not use any stack, nor any variable that is
- * not "NoSave" during copying pages:
+ * swsusp_arch_resume must not use any stack or any nonlocal variables while
+ * copying pages:
  *
  * Its rewriting one kernel image with another. What is stack in "old"
  * image could very well be data page in "new" image, and overwriting
@@ -36,10 +36,20 @@ ENTRY(swsusp_arch_suspend)
 	pushfq
 	popq	pt_regs_eflags(%rax)
 
+	/* save the address of restore_registers */
+	movq	$restore_registers, %rax
+	movq	%rax, restore_jump_address(%rip)
+
 	call swsusp_save
 	ret
 
 ENTRY(restore_image)
+	/* compute the address of the page we are at and store it in R9 */
+	movq	$(restore_image - __START_KERNEL_map), %rax
+	movq	$__PAGE_OFFSET, %r9
+	addq	%rax, %r9
+	andq	$PAGE_MASK, %r9
+
 	/* switch to temporary page tables */
 	movq	$__PAGE_OFFSET, %rdx
 	movq	temp_level4_pgt(%rip), %rax
@@ -54,6 +64,11 @@ ENTRY(restore_image)
 	movq	%rcx, %cr3;
 	movq	%rax, %cr4;  # turn PGE back on
 
+	/* prepare to jump to the image kernel */
+	movq	restore_jump_address(%rip), %rax
+
+	/* copy image data to their original locations */
+	xorq	%r10, %r10
 	movq	restore_pblist(%rip), %rdx
 loop:
 	testq	%rdx, %rdx
@@ -62,16 +77,46 @@ loop:
 	/* get addresses from the pbe and copy the page */
 	movq	pbe_address(%rdx), %rsi
 	movq	pbe_orig_address(%rdx), %rdi
-	movq	$512, %rcx
+	/* skip the page we are at (address stored in R9) */
+	cmpq	%rdi, %r9
+	jne	1f
+	/* save the address of the data to be copied to the skipped page */
+	movq	%rsi, %r10
+	jmp	2f
+1:	movq	$(PAGE_SIZE >> 3), %rcx
 	rep
 	movsq
 
 	/* progress to the next pbe */
-	movq	pbe_next(%rdx), %rdx
+2:	movq	pbe_next(%rdx), %rdx
 	jmp	loop
 done:
+	/* jump to the restore_registers address from the image header */
+	jmpq	*%rax
+	/*
+	 * NOTE: This assumes that the boot kernel's text mapping covers the
+	 * image kernel's page containing restore_registers and the address of
+	 * this page is the same as in the image kernel's text mapping (it
+	 * should always be true, because the text mapping is linear, starting
+	 * from 0, and is supposed to cover the entire kernel text for every
+	 * kernel).
+	 */
+
+.balign PAGE_SIZE
+	/* code below belongs to the image kernel */
+ENTRY(restore_registers)
+	/* check if we have one more image page to copy */
+	testq	%r10, %r10
+	jz	1f
+	/* copy the skipped page */
+	movq	%r10, %rsi
+	movq	%r9, %rdi
+	movq	$(PAGE_SIZE >> 3), %rcx
+	rep
+	movsq
+
 	/* go back to the original page tables */
-	movq    $(init_level4_pgt - __START_KERNEL_map), %rax
+1:	movq    $(init_level4_pgt - __START_KERNEL_map), %rax
 	addq    phys_base(%rip), %rax
 	movq    %rax, %cr3
 
@@ -84,10 +129,7 @@ done:
 	movq	%rcx, %cr3
 	movq	%rax, %cr4;  # turn PGE back on
 
-	movl	$24, %eax
-	movl	%eax, %ds
-
-	/* We don't restore %rax, it must be 0 anyway */
+	/* restore GPRs (we don't restore %rax, it must be 0 anyway) */
 	movq	$saved_context, %rax
 	movq	pt_regs_rsp(%rax), %rsp
 	movq	pt_regs_rbp(%rax), %rbp
@@ -109,4 +151,7 @@ done:
 
 	xorq	%rax, %rax
 
+	/* tell the hibernation core that we've just restored the memory */
+	movq	%rax, in_suspend(%rip)
+
 	ret
Index: linux-2.6.23-rc3/arch/x86_64/kernel/suspend.c
===================================================================
--- linux-2.6.23-rc3.orig/arch/x86_64/kernel/suspend.c	2007-08-21 20:36:49.000000000 +0200
+++ linux-2.6.23-rc3/arch/x86_64/kernel/suspend.c	2007-08-21 21:07:45.000000000 +0200
@@ -144,6 +144,12 @@ void fix_processor_context(void)
 /* Defined in arch/x86_64/kernel/suspend_asm.S */
 extern int restore_image(void);
 
+/*
+ * Address to jump to in the last phase of restore in order to get to the image
+ * kernel's text (this value is passed in the image header).
+ */
+unsigned long restore_jump_address;
+
 pgd_t *temp_level4_pgt;
 
 static int res_phys_pud_init(pud_t *pud, unsigned long address, unsigned long end)
@@ -230,4 +236,41 @@ int pfn_is_nosave(unsigned long pfn)
 	unsigned long nosave_end_pfn = PAGE_ALIGN(__pa_symbol(&__nosave_end)) >> PAGE_SHIFT;
 	return (pfn >= nosave_begin_pfn) && (pfn < nosave_end_pfn);
 }
+
+struct restore_data_record {
+	unsigned long jump_address;
+	unsigned long control;
+};
+
+#define RESTORE_MAGIC	0x0123456789ABCDEFUL
+
+/**
+ *	arch_hibernation_header_save - populate the architecture specific part
+ *		of a hibernation image header
+ *	@addr: address to save the data at
+ */
+int arch_hibernation_header_save(void *addr, unsigned int max_size)
+{
+	struct restore_data_record *rdr = addr;
+
+	if (max_size < sizeof(struct restore_data_record))
+		return -EOVERFLOW;
+	rdr->jump_address = restore_jump_address;
+	rdr->control = (restore_jump_address ^ RESTORE_MAGIC);
+	return 0;
+}
+
+/**
+ *	arch_hibernation_header_restore - read the architecture specific data
+ *		from the hibernation image header
+ *	@addr: address to read the data from
+ */
+int arch_hibernation_header_restore(void *addr)
+{
+	struct restore_data_record *rdr = addr;
+
+	restore_jump_address = rdr->jump_address;
+	return (rdr->control == (restore_jump_address ^ RESTORE_MAGIC)) ?
+			0 : -EINVAL;
+}
 #endif /* CONFIG_HIBERNATION */
Index: linux-2.6.23-rc3/include/asm-x86_64/suspend.h
===================================================================
--- linux-2.6.23-rc3.orig/include/asm-x86_64/suspend.h	2007-08-21 20:36:49.000000000 +0200
+++ linux-2.6.23-rc3/include/asm-x86_64/suspend.h	2007-08-21 20:37:47.000000000 +0200
@@ -43,4 +43,10 @@ extern void fix_processor_context(void);
 /* routines for saving/restoring kernel state */
 extern int acpi_save_state_mem(void);
 
+#define ARCH_HAS_HIBERNATION_HEADER
+
+/* arch/x86_64/kernel/suspend.c */
+extern int arch_hibernation_header_save(void *addr, unsigned int max_size);
+extern int arch_hibernation_header_restore(void *addr);
+
 #endif /* __ASM_X86_64_SUSPEND_H */

Re: [RFC][PATCH -mm 2/4] Hibernation: Arbitrary boot kernel support on x86_64

[Pavel Machek]

On Wed 2007-08-22 10:19:03, Rafael J. Wysocki wrote:
> From: Rafael J. Wysocki <rjw@sisk.pl>
> 
> Make it possible to restore a hibernation image on x86_64 with the help of a
> kernel different from the one in the image.
> 
> The idea is to split the core restoration code into two separate parts and to
> place each of them in a different page.  The first part belongs to the boot
> kernel and is executed as the last step of the image kernel's memory restoration
> procedure.  It restores all of the image kernel's memory that has not been
> restored yet except for the one page containing the very code that is being
> executed at that time.  The final operation performed by it is a jump to the
> second part of the core restoration code that belongs to the image kernel and
> has just been restored.  This code restores the last remaining page of the image
> kernel's memory containing the first, already executed, part of the core
> restoration code (temporary page tables created by the boot kernel are used at
> this stage).  It also makes the CPU switch to the image kernel's page tables and
> restores the state of general purpose registers (including the stack pointer)
> from before the hibernation.
> 
> The main issue with this idea is that in order to jump to the second part of the
> restoration code the boot kernel needs to know its address.  However, this
> address may be passed to it in the image header.  Namely, the part of the image
> header previously used for checking if the version of the image kernel is
> correct can be replaced with some architecture specific data that will allow
> the boot kernel to jump to the right address within the image kernel.  These
> data should also be used for checking if the image kernel is compatible with
> the boot kernel (as far as the memory restroration procedure is concerned).
> It can be done, for example, with the help of a "magic" value that has to be
> equal in both kernels, so that they can be regarded as compatible.
> 
> Signed-off-by: Rafael J. Wysocki <rjw@sisk.pl>

ACK.
-- 
(english) http://www.livejournal.com/~pavelmachek
(cesky, pictures) http://atrey.karlin.mff.cuni.cz/~pavel/picture/horses/blog.html

[RFC][PATCH -mm 3/4] Hibernation: Pass CR3 in the image header on x86_64

[Rafael J. Wysocki]

From: Rafael J. Wysocki <rjw@sisk.pl>

Since we already pass the address of restore_registers() in the image header,
we can also pass the value of the CR3 register from before the hibernation in
the same way.  This will allow us to avoid using init_level4_pgt page tables
during the restore.

Signed-off-by: Rafael J. Wysocki <rjw@sisk.pl>
---
 arch/x86_64/kernel/suspend.c     |   26 +++++++++++++++++++++-----
 arch/x86_64/kernel/suspend_asm.S |    8 +++++---
 2 files changed, 26 insertions(+), 8 deletions(-)

Index: linux-2.6.23-rc3/arch/x86_64/kernel/suspend.c
===================================================================
--- linux-2.6.23-rc3.orig/arch/x86_64/kernel/suspend.c
+++ linux-2.6.23-rc3/arch/x86_64/kernel/suspend.c
@@ -9,6 +9,7 @@
 
 #include <linux/smp.h>
 #include <linux/suspend.h>
+#include <linux/crc32.h>
 #include <asm/proto.h>
 #include <asm/page.h>
 #include <asm/pgtable.h>
@@ -150,6 +151,12 @@ extern int restore_image(void);
  */
 unsigned long restore_jump_address;
 
+/*
+ * Value of the cr3 register from before the hibernation (this value is passed
+ * in the image header).
+ */
+unsigned long restore_cr3;
+
 pgd_t *temp_level4_pgt;
 
 static int res_phys_pud_init(pud_t *pud, unsigned long address, unsigned long end)
@@ -239,7 +246,9 @@ int pfn_is_nosave(unsigned long pfn)
 
 struct restore_data_record {
 	unsigned long jump_address;
-	unsigned long control;
+	unsigned long cr3;
+	unsigned long magic;
+	u32 crc;
 };
 
 #define RESTORE_MAGIC	0x0123456789ABCDEFUL
@@ -253,10 +262,13 @@ int arch_hibernation_header_save(void *a
 {
 	struct restore_data_record *rdr = addr;
 
-	if (max_size < sizeof(struct restore_data_record))
+	if (max_size < sizeof(*rdr))
 		return -EOVERFLOW;
 	rdr->jump_address = restore_jump_address;
-	rdr->control = (restore_jump_address ^ RESTORE_MAGIC);
+	rdr->cr3 = restore_cr3;
+	rdr->magic = RESTORE_MAGIC;
+	rdr->crc = 0;
+	rdr->crc = crc32_le(0, addr, sizeof(*rdr));
 	return 0;
 }
 
@@ -268,9 +280,13 @@ int arch_hibernation_header_save(void *a
 int arch_hibernation_header_restore(void *addr)
 {
 	struct restore_data_record *rdr = addr;
+	u32 crc;
 
 	restore_jump_address = rdr->jump_address;
-	return (rdr->control == (restore_jump_address ^ RESTORE_MAGIC)) ?
-			0 : -EINVAL;
+	restore_cr3 = rdr->cr3;
+	crc = rdr->crc;
+	rdr->crc = 0;
+	return (crc == crc32_le(0, addr, sizeof(*rdr))
+		&& rdr->magic == RESTORE_MAGIC) ? 0 : -EINVAL;
 }
 #endif /* CONFIG_HIBERNATION */
Index: linux-2.6.23-rc3/arch/x86_64/kernel/suspend_asm.S
===================================================================
--- linux-2.6.23-rc3.orig/arch/x86_64/kernel/suspend_asm.S
+++ linux-2.6.23-rc3/arch/x86_64/kernel/suspend_asm.S
@@ -39,6 +39,9 @@ ENTRY(swsusp_arch_suspend)
 	/* save the address of restore_registers */
 	movq	$restore_registers, %rax
 	movq	%rax, restore_jump_address(%rip)
+	/* save cr3 */
+	movq	%cr3, %rax
+	movq	%rax, restore_cr3(%rip)
 
 	call swsusp_save
 	ret
@@ -66,6 +69,7 @@ ENTRY(restore_image)
 
 	/* prepare to jump to the image kernel */
 	movq	restore_jump_address(%rip), %rax
+	movq	restore_cr3(%rip), %rbx
 
 	/* copy image data to their original locations */
 	xorq	%r10, %r10
@@ -116,9 +120,7 @@ ENTRY(restore_registers)
 	movsq
 
 	/* go back to the original page tables */
-1:	movq    $(init_level4_pgt - __START_KERNEL_map), %rax
-	addq    phys_base(%rip), %rax
-	movq    %rax, %cr3
+1:	movq    %rbx, %cr3
 
 	/* Flush TLB, including "global" things (vmalloc) */
 	movq	mmu_cr4_features(%rip), %rax

Re: [RFC][PATCH -mm 3/4] Hibernation: Pass CR3 in the image header on x86_64

[Pavel Machek]

Hi!

> From: Rafael J. Wysocki <rjw@sisk.pl>
> 
> Since we already pass the address of restore_registers() in the image header,
> we can also pass the value of the CR3 register from before the hibernation in
> the same way.  This will allow us to avoid using init_level4_pgt page tables
> during the restore.
> 
> Signed-off-by: Rafael J. Wysocki <rjw@sisk.pl>

> @@ -253,10 +262,13 @@ int arch_hibernation_header_save(void *a
>  {
>  	struct restore_data_record *rdr = addr;
>  
> -	if (max_size < sizeof(struct restore_data_record))
> +	if (max_size < sizeof(*rdr))
>  		return -EOVERFLOW;
>  	rdr->jump_address = restore_jump_address;
> -	rdr->control = (restore_jump_address ^ RESTORE_MAGIC);
> +	rdr->cr3 = restore_cr3;
> +	rdr->magic = RESTORE_MAGIC;
> +	rdr->crc = 0;
> +	rdr->crc = crc32_le(0, addr, sizeof(*rdr));
>  	return 0;
>  }

No, I do not think I like that. I believe both -> control and -> crc
is just useless paranoia. Bitflip in this area is not going to be any
worse than bitflip anywhere else, we should not pretend this is
somehow "more important".

-> control should really be "protocol version"... probably should
contain some field that is easy to increment.
								Pavel
-- 
(english) http://www.livejournal.com/~pavelmachek
(cesky, pictures) http://atrey.karlin.mff.cuni.cz/~pavel/picture/horses/blog.html

Re: [RFC][PATCH -mm 3/4] Hibernation: Pass CR3 in the image header on x86_64

[Rafael J. Wysocki]

On Wednesday, 22 August 2007 10:28, Pavel Machek wrote:
> Hi!
> 
> > From: Rafael J. Wysocki <rjw@sisk.pl>
> > 
> > Since we already pass the address of restore_registers() in the image header,
> > we can also pass the value of the CR3 register from before the hibernation in
> > the same way.  This will allow us to avoid using init_level4_pgt page tables
> > during the restore.
> > 
> > Signed-off-by: Rafael J. Wysocki <rjw@sisk.pl>
> 
> > @@ -253,10 +262,13 @@ int arch_hibernation_header_save(void *a
> >  {
> >  	struct restore_data_record *rdr = addr;
> >  
> > -	if (max_size < sizeof(struct restore_data_record))
> > +	if (max_size < sizeof(*rdr))
> >  		return -EOVERFLOW;
> >  	rdr->jump_address = restore_jump_address;
> > -	rdr->control = (restore_jump_address ^ RESTORE_MAGIC);
> > +	rdr->cr3 = restore_cr3;
> > +	rdr->magic = RESTORE_MAGIC;
> > +	rdr->crc = 0;
> > +	rdr->crc = crc32_le(0, addr, sizeof(*rdr));
> >  	return 0;
> >  }
> 
> No, I do not think I like that. I believe both -> control and -> crc
> is just useless paranoia. Bitflip in this area is not going to be any
> worse than bitflip anywhere else, we should not pretend this is
> somehow "more important".
> 
> -> control should really be "protocol version"... probably should
> contain some field that is easy to increment.

OK

Perhaps I'll just remove the crc field.  What do you think?

Rafael

Re: [linux-pm] Re: [RFC][PATCH -mm 3/4] Hibernation: Pass CR3 in the image header on x86_64

[Rafael J. Wysocki]

On Wednesday, 22 August 2007 18:17, Rafael J. Wysocki wrote:
> On Wednesday, 22 August 2007 10:28, Pavel Machek wrote:
> > Hi!
> > 
> > > From: Rafael J. Wysocki <rjw@sisk.pl>
> > > 
> > > Since we already pass the address of restore_registers() in the image header,
> > > we can also pass the value of the CR3 register from before the hibernation in
> > > the same way.  This will allow us to avoid using init_level4_pgt page tables
> > > during the restore.
> > > 
> > > Signed-off-by: Rafael J. Wysocki <rjw@sisk.pl>
> > 
> > > @@ -253,10 +262,13 @@ int arch_hibernation_header_save(void *a
> > >  {
> > >  	struct restore_data_record *rdr = addr;
> > >  
> > > -	if (max_size < sizeof(struct restore_data_record))
> > > +	if (max_size < sizeof(*rdr))
> > >  		return -EOVERFLOW;
> > >  	rdr->jump_address = restore_jump_address;
> > > -	rdr->control = (restore_jump_address ^ RESTORE_MAGIC);
> > > +	rdr->cr3 = restore_cr3;
> > > +	rdr->magic = RESTORE_MAGIC;
> > > +	rdr->crc = 0;
> > > +	rdr->crc = crc32_le(0, addr, sizeof(*rdr));
> > >  	return 0;
> > >  }
> > 
> > No, I do not think I like that. I believe both -> control and -> crc
> > is just useless paranoia. Bitflip in this area is not going to be any
> > worse than bitflip anywhere else, we should not pretend this is
> > somehow "more important".
> > 
> > -> control should really be "protocol version"... probably should
> > contain some field that is easy to increment.
> 
> OK
> 
> Perhaps I'll just remove the crc field.

Like in the patch below.

---
From: Rafael J. Wysocki <rjw@sisk.pl>

Since we already pass the address of restore_registers() in the image header,
we can also pass the value of the CR3 register from before the hibernation in
the same way.  This will allow us to avoid using init_level4_pgt page tables
during the restore.

Signed-off-by: Rafael J. Wysocki <rjw@sisk.pl>
---
 arch/x86_64/kernel/suspend.c     |   17 +++++++++++++----
 arch/x86_64/kernel/suspend_asm.S |    8 +++++---
 2 files changed, 18 insertions(+), 7 deletions(-)

Index: linux-2.6.23-rc3/arch/x86_64/kernel/suspend.c
===================================================================
--- linux-2.6.23-rc3.orig/arch/x86_64/kernel/suspend.c
+++ linux-2.6.23-rc3/arch/x86_64/kernel/suspend.c
@@ -9,6 +9,7 @@
 
 #include <linux/smp.h>
 #include <linux/suspend.h>
+#include <linux/crc32.h>
 #include <asm/proto.h>
 #include <asm/page.h>
 #include <asm/pgtable.h>
@@ -150,6 +151,12 @@ extern int restore_image(void);
  */
 unsigned long restore_jump_address;
 
+/*
+ * Value of the cr3 register from before the hibernation (this value is passed
+ * in the image header).
+ */
+unsigned long restore_cr3;
+
 pgd_t *temp_level4_pgt;
 
 static int res_phys_pud_init(pud_t *pud, unsigned long address, unsigned long end)
@@ -239,7 +246,8 @@ int pfn_is_nosave(unsigned long pfn)
 
 struct restore_data_record {
 	unsigned long jump_address;
-	unsigned long control;
+	unsigned long cr3;
+	unsigned long magic;
 };
 
 #define RESTORE_MAGIC	0x0123456789ABCDEFUL
@@ -256,7 +264,8 @@ int arch_hibernation_header_save(void *a
 	if (max_size < sizeof(struct restore_data_record))
 		return -EOVERFLOW;
 	rdr->jump_address = restore_jump_address;
-	rdr->control = (restore_jump_address ^ RESTORE_MAGIC);
+	rdr->cr3 = restore_cr3;
+	rdr->magic = RESTORE_MAGIC;
 	return 0;
 }
 
@@ -270,7 +279,7 @@ int arch_hibernation_header_restore(void
 	struct restore_data_record *rdr = addr;
 
 	restore_jump_address = rdr->jump_address;
-	return (rdr->control == (restore_jump_address ^ RESTORE_MAGIC)) ?
-			0 : -EINVAL;
+	restore_cr3 = rdr->cr3;
+	return (rdr->magic == RESTORE_MAGIC) ? 0 : -EINVAL;
 }
 #endif /* CONFIG_HIBERNATION */
Index: linux-2.6.23-rc3/arch/x86_64/kernel/suspend_asm.S
===================================================================
--- linux-2.6.23-rc3.orig/arch/x86_64/kernel/suspend_asm.S
+++ linux-2.6.23-rc3/arch/x86_64/kernel/suspend_asm.S
@@ -39,6 +39,9 @@ ENTRY(swsusp_arch_suspend)
 	/* save the address of restore_registers */
 	movq	$restore_registers, %rax
 	movq	%rax, restore_jump_address(%rip)
+	/* save cr3 */
+	movq	%cr3, %rax
+	movq	%rax, restore_cr3(%rip)
 
 	call swsusp_save
 	ret
@@ -66,6 +69,7 @@ ENTRY(restore_image)
 
 	/* prepare to jump to the image kernel */
 	movq	restore_jump_address(%rip), %rax
+	movq	restore_cr3(%rip), %rbx
 
 	/* copy image data to their original locations */
 	xorq	%r10, %r10
@@ -116,9 +120,7 @@ ENTRY(restore_registers)
 	movsq
 
 	/* go back to the original page tables */
-1:	movq    $(init_level4_pgt - __START_KERNEL_map), %rax
-	addq    phys_base(%rip), %rax
-	movq    %rax, %cr3
+1:	movq    %rbx, %cr3
 
 	/* Flush TLB, including "global" things (vmalloc) */
 	movq	mmu_cr4_features(%rip), %rax

Re: [linux-pm] Re: [RFC][PATCH -mm 3/4] Hibernation: Pass CR3 in the image header on x86_64

[Pavel Machek]

Hi!

> > > No, I do not think I like that. I believe both -> control and -> crc
> > > is just useless paranoia. Bitflip in this area is not going to be any
> > > worse than bitflip anywhere else, we should not pretend this is
> > > somehow "more important".
> > > 
> > > -> control should really be "protocol version"... probably should
> > > contain some field that is easy to increment.
> > 
> > OK
> > 
> > Perhaps I'll just remove the crc field.
> 
> Like in the patch below.

Yep, thanks.

> ---
> From: Rafael J. Wysocki <rjw@sisk.pl>
> 
> Since we already pass the address of restore_registers() in the image header,
> we can also pass the value of the CR3 register from before the hibernation in
> the same way.  This will allow us to avoid using init_level4_pgt page tables
> during the restore.

ACK.

> @@ -9,6 +9,7 @@
>  
>  #include <linux/smp.h>
>  #include <linux/suspend.h>
> +#include <linux/crc32.h>
>  #include <asm/proto.h>
>  #include <asm/page.h>
>  #include <asm/pgtable.h>

This one is no longer neccessary.


									Pavel
-- 
(english) http://www.livejournal.com/~pavelmachek
(cesky, pictures) http://atrey.karlin.mff.cuni.cz/~pavel/picture/horses/blog.html

Re: [linux-pm] Re: [RFC][PATCH -mm 3/4] Hibernation: Pass CR3 in the image header on x86_64

[Rafael J. Wysocki]

On Friday, 24 August 2007 11:02, Pavel Machek wrote:
> Hi!
> 
> > > > No, I do not think I like that. I believe both -> control and -> crc
> > > > is just useless paranoia. Bitflip in this area is not going to be any
> > > > worse than bitflip anywhere else, we should not pretend this is
> > > > somehow "more important".
> > > > 
> > > > -> control should really be "protocol version"... probably should
> > > > contain some field that is easy to increment.
> > > 
> > > OK
> > > 
> > > Perhaps I'll just remove the crc field.
> > 
> > Like in the patch below.
> 
> Yep, thanks.
> 
> > ---
> > From: Rafael J. Wysocki <rjw@sisk.pl>
> > 
> > Since we already pass the address of restore_registers() in the image header,
> > we can also pass the value of the CR3 register from before the hibernation in
> > the same way.  This will allow us to avoid using init_level4_pgt page tables
> > during the restore.
> 
> ACK.
> 
> > @@ -9,6 +9,7 @@
> >  
> >  #include <linux/smp.h>
> >  #include <linux/suspend.h>
> > +#include <linux/crc32.h>
> >  #include <asm/proto.h>
> >  #include <asm/page.h>
> >  #include <asm/pgtable.h>
> 
> This one is no longer neccessary.

I forgot about it.

I've already sent the 1/4 and 2/4 patches to Andrew, so I'll keep the remaining
two ones in the queue for a couple of days.

Thanks for the ACKs!

Greetings,
Rafael

[RFC][PATCH -mm 4/4] Hibernation: Use temporary page tables for kernel text mapping on x86_64

[Rafael J. Wysocki]

From: Rafael J. Wysocki <rjw@sisk.pl>

Use temporary page tables for the kernel text mapping during hibernation restore
on x86_64.

Without the patch, the original boot kernel's page tables that represent the
kernel text mapping are used while the core of the image kernel is being
restored.  However, in principle, if the boot kernel is not identical to the
image kernel, the location of these page tables in the image kernel need not be
the same, so we should create a safe copy of the kernel text mapping prior to
restoring the core of the image kernel.

Signed-off-by: Rafael J. Wysocki <rjw@sisk.pl>
---
 arch/x86_64/kernel/suspend.c |   41 ++++++++++++++++++++++++++++++++++-------
 1 file changed, 34 insertions(+), 7 deletions(-)

Index: linux-2.6.23-rc3/arch/x86_64/kernel/suspend.c
===================================================================
--- linux-2.6.23-rc3.orig/arch/x86_64/kernel/suspend.c
+++ linux-2.6.23-rc3/arch/x86_64/kernel/suspend.c
@@ -182,7 +182,7 @@ static int res_phys_pud_init(pud_t *pud,
 
 			if (paddr >= end)
 				break;
-			pe = _PAGE_NX | _PAGE_PSE | _KERNPG_TABLE | paddr;
+			pe = __PAGE_KERNEL_LARGE | paddr;
 			pe &= __supported_pte_mask;
 			set_pmd(pmd, __pmd(pe));
 		}
@@ -190,25 +190,42 @@ static int res_phys_pud_init(pud_t *pud,
 	return 0;
 }
 
+static int res_kernel_text_pud_init(pud_t *pud, unsigned long start)
+{
+	pmd_t *pmd;
+	unsigned long paddr;
+
+	pmd = (pmd_t *)get_safe_page(GFP_ATOMIC);
+	if (!pmd)
+		return -ENOMEM;
+	set_pud(pud + pud_index(start), __pud(__pa(pmd) | _KERNPG_TABLE));
+	for (paddr = 0; paddr < KERNEL_TEXT_SIZE; pmd++, paddr += PMD_SIZE) {
+		unsigned long pe;
+
+		pe = __PAGE_KERNEL_LARGE_EXEC | _PAGE_GLOBAL | paddr;
+		pe &= __supported_pte_mask;
+		set_pmd(pmd, __pmd(pe));
+	}
+
+	return 0;
+}
+
 static int set_up_temporary_mappings(void)
 {
 	unsigned long start, end, next;
+	pud_t *pud;
 	int error;
 
 	temp_level4_pgt = (pgd_t *)get_safe_page(GFP_ATOMIC);
 	if (!temp_level4_pgt)
 		return -ENOMEM;
 
-	/* It is safe to reuse the original kernel mapping */
-	set_pgd(temp_level4_pgt + pgd_index(__START_KERNEL_map),
-		init_level4_pgt[pgd_index(__START_KERNEL_map)]);
-
 	/* Set up the direct mapping from scratch */
 	start = (unsigned long)pfn_to_kaddr(0);
 	end = (unsigned long)pfn_to_kaddr(end_pfn);
 
 	for (; start < end; start = next) {
-		pud_t *pud = (pud_t *)get_safe_page(GFP_ATOMIC);
+		pud = (pud_t *)get_safe_page(GFP_ATOMIC);
 		if (!pud)
 			return -ENOMEM;
 		next = start + PGDIR_SIZE;
@@ -219,7 +236,17 @@ static int set_up_temporary_mappings(voi
 		set_pgd(temp_level4_pgt + pgd_index(start),
 			mk_kernel_pgd(__pa(pud)));
 	}
-	return 0;
+
+	/* Set up the kernel text mapping from scratch */
+	pud = (pud_t *)get_safe_page(GFP_ATOMIC);
+	if (!pud)
+		return -ENOMEM;
+	error = res_kernel_text_pud_init(pud, __START_KERNEL_map);
+	if (!error)
+		set_pgd(temp_level4_pgt + pgd_index(__START_KERNEL_map),
+			__pgd(__pa(pud) | _PAGE_TABLE));
+
+	return error;
 }
 
 int swsusp_arch_resume(void)

Re: [RFC][PATCH -mm 4/4] Hibernation: Use temporary page tables for kernel text mapping on x86_64

[Pavel Machek]

Hi!

> Use temporary page tables for the kernel text mapping during hibernation restore
> on x86_64.
> 
> Without the patch, the original boot kernel's page tables that represent the
> kernel text mapping are used while the core of the image kernel is being
> restored.  However, in principle, if the boot kernel is not identical to the
> image kernel, the location of these page tables in the image kernel need not be
> the same, so we should create a safe copy of the kernel text mapping prior to
> restoring the core of the image kernel.


> @@ -190,25 +190,42 @@ static int res_phys_pud_init(pud_t *pud,
>  	return 0;
>  }
>  
> +static int res_kernel_text_pud_init(pud_t *pud, unsigned long start)
> +{
> +	pmd_t *pmd;
> +	unsigned long paddr;
> +
> +	pmd = (pmd_t *)get_safe_page(GFP_ATOMIC);
> +	if (!pmd)
> +		return -ENOMEM;
> +	set_pud(pud + pud_index(start), __pud(__pa(pmd) | _KERNPG_TABLE));
> +	for (paddr = 0; paddr < KERNEL_TEXT_SIZE; pmd++, paddr += PMD_SIZE) {
> +		unsigned long pe;
> +
> +		pe = __PAGE_KERNEL_LARGE_EXEC | _PAGE_GLOBAL | paddr;
> +		pe &= __supported_pte_mask;
> +		set_pmd(pmd, __pmd(pe));
> +	}
> +
> +	return 0;
> +}
> +
>  static int set_up_temporary_mappings(void)
>  {
>  	unsigned long start, end, next;
> +	pud_t *pud;
>  	int error;
>  
>  	temp_level4_pgt = (pgd_t *)get_safe_page(GFP_ATOMIC);
>  	if (!temp_level4_pgt)
>  		return -ENOMEM;
>  
> -	/* It is safe to reuse the original kernel mapping */
> -	set_pgd(temp_level4_pgt + pgd_index(__START_KERNEL_map),
> -		init_level4_pgt[pgd_index(__START_KERNEL_map)]);
> -
>  	/* Set up the direct mapping from scratch */
>  	start = (unsigned long)pfn_to_kaddr(0);
>  	end = (unsigned long)pfn_to_kaddr(end_pfn);
>  
>  	for (; start < end; start = next) {
> -		pud_t *pud = (pud_t *)get_safe_page(GFP_ATOMIC);
> +		pud = (pud_t *)get_safe_page(GFP_ATOMIC);
>  		if (!pud)
>  			return -ENOMEM;

I believe we leak temp_level4_pgt here.

> +	/* Set up the kernel text mapping from scratch */
> +	pud = (pud_t *)get_safe_page(GFP_ATOMIC);
> +	if (!pud)
> +		return -ENOMEM;

...and some more memory here :-(. 
								Pavel
-- 
(english) http://www.livejournal.com/~pavelmachek
(cesky, pictures) http://atrey.karlin.mff.cuni.cz/~pavel/picture/horses/blog.html

Re: [RFC][PATCH -mm 4/4] Hibernation: Use temporary page tables for kernel text mapping on x86_64

[Rafael J. Wysocki]

On Wednesday, 22 August 2007 10:29, Pavel Machek wrote:
> Hi!
> 
> > Use temporary page tables for the kernel text mapping during hibernation restore
> > on x86_64.
> > 
> > Without the patch, the original boot kernel's page tables that represent the
> > kernel text mapping are used while the core of the image kernel is being
> > restored.  However, in principle, if the boot kernel is not identical to the
> > image kernel, the location of these page tables in the image kernel need not be
> > the same, so we should create a safe copy of the kernel text mapping prior to
> > restoring the core of the image kernel.
> 
> 
> > @@ -190,25 +190,42 @@ static int res_phys_pud_init(pud_t *pud,
> >  	return 0;
> >  }
> >  
> > +static int res_kernel_text_pud_init(pud_t *pud, unsigned long start)
> > +{
> > +	pmd_t *pmd;
> > +	unsigned long paddr;
> > +
> > +	pmd = (pmd_t *)get_safe_page(GFP_ATOMIC);
> > +	if (!pmd)
> > +		return -ENOMEM;
> > +	set_pud(pud + pud_index(start), __pud(__pa(pmd) | _KERNPG_TABLE));
> > +	for (paddr = 0; paddr < KERNEL_TEXT_SIZE; pmd++, paddr += PMD_SIZE) {
> > +		unsigned long pe;
> > +
> > +		pe = __PAGE_KERNEL_LARGE_EXEC | _PAGE_GLOBAL | paddr;
> > +		pe &= __supported_pte_mask;
> > +		set_pmd(pmd, __pmd(pe));
> > +	}
> > +
> > +	return 0;
> > +}
> > +
> >  static int set_up_temporary_mappings(void)
> >  {
> >  	unsigned long start, end, next;
> > +	pud_t *pud;
> >  	int error;
> >  
> >  	temp_level4_pgt = (pgd_t *)get_safe_page(GFP_ATOMIC);
> >  	if (!temp_level4_pgt)
> >  		return -ENOMEM;
> >  
> > -	/* It is safe to reuse the original kernel mapping */
> > -	set_pgd(temp_level4_pgt + pgd_index(__START_KERNEL_map),
> > -		init_level4_pgt[pgd_index(__START_KERNEL_map)]);
> > -
> >  	/* Set up the direct mapping from scratch */
> >  	start = (unsigned long)pfn_to_kaddr(0);
> >  	end = (unsigned long)pfn_to_kaddr(end_pfn);
> >  
> >  	for (; start < end; start = next) {
> > -		pud_t *pud = (pud_t *)get_safe_page(GFP_ATOMIC);
> > +		pud = (pud_t *)get_safe_page(GFP_ATOMIC);
> >  		if (!pud)
> >  			return -ENOMEM;
> 
> I believe we leak temp_level4_pgt here.
> 
> > +	/* Set up the kernel text mapping from scratch */
> > +	pud = (pud_t *)get_safe_page(GFP_ATOMIC);
> > +	if (!pud)
> > +		return -ENOMEM;
> 
> ...and some more memory here :-(. 

No, these pages will be freed by swsusp_free(), thanks to the get_safe_page().

I'm not _that_ careless. ;-)

Greetings,
Rafael

Re: [RFC][PATCH -mm 4/4] Hibernation: Use temporary page tables for kernel text mapping on x86_64

[Pavel Machek]

On Wed 2007-08-22 18:16:05, Rafael J. Wysocki wrote:
> On Wednesday, 22 August 2007 10:29, Pavel Machek wrote:
> > Hi!
> > 
> > > Use temporary page tables for the kernel text mapping during hibernation restore
> > > on x86_64.
> > > 
> > > Without the patch, the original boot kernel's page tables that represent the
> > > kernel text mapping are used while the core of the image kernel is being
> > > restored.  However, in principle, if the boot kernel is not identical to the
> > > image kernel, the location of these page tables in the image kernel need not be
> > > the same, so we should create a safe copy of the kernel text mapping prior to
> > > restoring the core of the image kernel.
> > 
> > 
> > > @@ -190,25 +190,42 @@ static int res_phys_pud_init(pud_t *pud,
> > >  	return 0;
> > >  }
> > >  
> > > +static int res_kernel_text_pud_init(pud_t *pud, unsigned long start)
> > > +{
> > > +	pmd_t *pmd;
> > > +	unsigned long paddr;
> > > +
> > > +	pmd = (pmd_t *)get_safe_page(GFP_ATOMIC);
> > > +	if (!pmd)
> > > +		return -ENOMEM;
> > > +	set_pud(pud + pud_index(start), __pud(__pa(pmd) | _KERNPG_TABLE));
> > > +	for (paddr = 0; paddr < KERNEL_TEXT_SIZE; pmd++, paddr += PMD_SIZE) {
> > > +		unsigned long pe;
> > > +
> > > +		pe = __PAGE_KERNEL_LARGE_EXEC | _PAGE_GLOBAL | paddr;
> > > +		pe &= __supported_pte_mask;
> > > +		set_pmd(pmd, __pmd(pe));
> > > +	}
> > > +
> > > +	return 0;
> > > +}
> > > +
> > >  static int set_up_temporary_mappings(void)
> > >  {
> > >  	unsigned long start, end, next;
> > > +	pud_t *pud;
> > >  	int error;
> > >  
> > >  	temp_level4_pgt = (pgd_t *)get_safe_page(GFP_ATOMIC);
> > >  	if (!temp_level4_pgt)
> > >  		return -ENOMEM;
> > >  
> > > -	/* It is safe to reuse the original kernel mapping */
> > > -	set_pgd(temp_level4_pgt + pgd_index(__START_KERNEL_map),
> > > -		init_level4_pgt[pgd_index(__START_KERNEL_map)]);
> > > -
> > >  	/* Set up the direct mapping from scratch */
> > >  	start = (unsigned long)pfn_to_kaddr(0);
> > >  	end = (unsigned long)pfn_to_kaddr(end_pfn);
> > >  
> > >  	for (; start < end; start = next) {
> > > -		pud_t *pud = (pud_t *)get_safe_page(GFP_ATOMIC);
> > > +		pud = (pud_t *)get_safe_page(GFP_ATOMIC);
> > >  		if (!pud)
> > >  			return -ENOMEM;
> > 
> > I believe we leak temp_level4_pgt here.
> > 
> > > +	/* Set up the kernel text mapping from scratch */
> > > +	pud = (pud_t *)get_safe_page(GFP_ATOMIC);
> > > +	if (!pud)
> > > +		return -ENOMEM;
> > 
> > ...and some more memory here :-(. 
> 
> No, these pages will be freed by swsusp_free(), thanks to the get_safe_page().
> 
> I'm not _that_ careless. ;-)

Ok, ACK.


-- 
(english) http://www.livejournal.com/~pavelmachek
(cesky, pictures) http://atrey.karlin.mff.cuni.cz/~pavel/picture/horses/blog.html