[RFC][PATCH 2/2 -mm] kexec based hibernation: kexec restore

[RFC][PATCH 2/2 -mm] kexec based hibernation: kexec restore

[Huang, Ying]

This patch adds writing support for /dev/oldmem. This is used to
restore the memory contents of hibernated system.

Signed-off-by: Huang Ying <ying.huang@intel.com>

---

 arch/i386/kernel/crash_dump.c |   27 +++++++++++++++++++++++++++
 drivers/char/mem.c            |   32 ++++++++++++++++++++++++++++++++
 include/linux/crash_dump.h    |    2 ++
 3 files changed, 61 insertions(+)

Index: linux-2.6.23-rc3/arch/i386/kernel/crash_dump.c
===================================================================
--- linux-2.6.23-rc3.orig/arch/i386/kernel/crash_dump.c	2007-08-25 21:56:52.000000000 +0800
+++ linux-2.6.23-rc3/arch/i386/kernel/crash_dump.c	2007-08-25 21:57:11.000000000 +0800
@@ -58,6 +58,33 @@
 	return csize;
 }
 
+ssize_t write_oldmem_page(unsigned long pfn, const char *buf,
+			  size_t csize, unsigned long offset, int userbuf)
+{
+	void  *vaddr;
+
+	if (!csize)
+		return 0;
+
+	if (!userbuf) {
+		vaddr = kmap_atomic_pfn(pfn, KM_PTE0);
+		memcpy(vaddr + offset, buf, csize);
+	} else {
+		if (!kdump_buf_page) {
+			printk(KERN_WARNING "Kdump: Kdump buffer page not"
+				" allocated\n");
+			return -EFAULT;
+		}
+		if (copy_from_user(kdump_buf_page, buf, csize))
+			return -EFAULT;
+		vaddr = kmap_atomic_pfn(pfn, KM_PTE0);
+		memcpy(vaddr + offset, kdump_buf_page, csize);
+	}
+	kunmap_atomic(vaddr, KM_PTE0);
+
+	return csize;
+}
+
 static int __init kdump_buf_page_init(void)
 {
 	int ret = 0;
Index: linux-2.6.23-rc3/include/linux/crash_dump.h
===================================================================
--- linux-2.6.23-rc3.orig/include/linux/crash_dump.h	2007-08-25 21:56:52.000000000 +0800
+++ linux-2.6.23-rc3/include/linux/crash_dump.h	2007-08-25 21:57:11.000000000 +0800
@@ -11,6 +11,8 @@
 extern unsigned long long elfcorehdr_addr;
 extern ssize_t copy_oldmem_page(unsigned long, char *, size_t,
 						unsigned long, int);
+extern ssize_t write_oldmem_page(unsigned long, const char *, size_t,
+				 unsigned long, int);
 extern const struct file_operations proc_vmcore_operations;
 extern struct proc_dir_entry *proc_vmcore;
 
Index: linux-2.6.23-rc3/drivers/char/mem.c
===================================================================
--- linux-2.6.23-rc3.orig/drivers/char/mem.c	2007-08-25 21:56:52.000000000 +0800
+++ linux-2.6.23-rc3/drivers/char/mem.c	2007-08-25 21:57:11.000000000 +0800
@@ -348,6 +348,37 @@
 	}
 	return read;
 }
+
+/*
+ * Write memory corresponding to the old kernel.
+ */
+static ssize_t write_oldmem(struct file *file, const char __user *buf,
+			    size_t count, loff_t *ppos)
+{
+	unsigned long pfn, offset;
+	size_t write = 0, csize;
+	int rc = 0;
+
+	while (count) {
+		pfn = *ppos / PAGE_SIZE;
+		if (pfn > saved_max_pfn)
+			return write;
+
+		offset = (unsigned long)(*ppos % PAGE_SIZE);
+		if (count > PAGE_SIZE - offset)
+			csize = PAGE_SIZE - offset;
+		else
+			csize = count;
+		rc = write_oldmem_page(pfn, buf, csize, offset, 1);
+		if (rc < 0)
+			return rc;
+		buf += csize;
+		*ppos += csize;
+		write += csize;
+		count -= csize;
+	}
+	return write;
+}
 #endif
 
 extern long vread(char *buf, char *addr, unsigned long count);
@@ -783,6 +814,7 @@
 #ifdef CONFIG_CRASH_DUMP
 static const struct file_operations oldmem_fops = {
 	.read	= read_oldmem,
+	.write	= write_oldmem,
 	.open	= open_oldmem,
 };
 #endif

Re: [RFC][PATCH 2/2 -mm] kexec based hibernation: kexec restore

[Pavel Machek]

Hi!

> This patch adds writing support for /dev/oldmem. This is used to
> restore the memory contents of hibernated system.
> 
> Signed-off-by: Huang Ying <ying.huang@intel.com>

> +ssize_t write_oldmem_page(unsigned long pfn, const char *buf,
> +			  size_t csize, unsigned long offset, int userbuf)

Hmm, int userbuf is only ever set to one... Does it make sense to have
write_oldmem_page in the separate file? The onl user is mem.c, perhaps
it should go there?

> +
> +/*
> + * Write memory corresponding to the old kernel.
> + */
> +static ssize_t write_oldmem(struct file *file, const char __user *buf,
> +			    size_t count, loff_t *ppos)
> +{
> +	unsigned long pfn, offset;
> +	size_t write = 0, csize;
> +	int rc = 0;
> +
> +	while (count) {
> +		pfn = *ppos / PAGE_SIZE;
> +		if (pfn > saved_max_pfn)
> +			return write;
> +
> +		offset = (unsigned long)(*ppos % PAGE_SIZE);
> +		if (count > PAGE_SIZE - offset)
> +			csize = PAGE_SIZE - offset;
> +		else
> +			csize = count;
> +		rc = write_oldmem_page(pfn, buf, csize, offset, 1);
> +		if (rc < 0)
> +			return rc;
> +		buf += csize;
> +		*ppos += csize;
> +		write += csize;
> +		count -= csize;
> +	}
> +	return write;
> +}

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

Re: [RFC][PATCH 2/2 -mm] kexec based hibernation: kexec restore

[Huang, Ying]

On Mon, 2007-08-27 at 23:31 +0200, Pavel Machek wrote:
> Hi!
> 
> > This patch adds writing support for /dev/oldmem. This is used to
> > restore the memory contents of hibernated system.
> > 
> > Signed-off-by: Huang Ying <ying.huang@intel.com>
> 
> > +ssize_t write_oldmem_page(unsigned long pfn, const char *buf,
> > +			  size_t csize, unsigned long offset, int userbuf)
> 
> Hmm, int userbuf is only ever set to one... Does it make sense to have
> write_oldmem_page in the separate file? The onl user is mem.c, perhaps
> it should go there?
> 

write_oldmem_page is kept to be consistent with copy_oldmem_page as much
as possible. The userbuf is used by copy_oldmem_page too, and
write_oldmem_page is in the same file as copy_oldmem_page. I think the
consistence between them is reasonable.

And the copy_oldmem_page/write_oldmem_page is considered to be
architecture dependent. Now, there are different implementations for
copy_oldmem_page on different architectures. So I think the
copy_oldmem_page/write_oldmem_page should be kept in separate file
instead of go mem.c.

Best Regards,
Huang Ying

Re: [RFC][PATCH 2/2 -mm] kexec based hibernation: kexec restore

[Pavel Machek]

> On Mon, 2007-08-27 at 23:31 +0200, Pavel Machek wrote:
> > Hi!
> > 
> > > This patch adds writing support for /dev/oldmem. This is used to
> > > restore the memory contents of hibernated system.
> > > 
> > > Signed-off-by: Huang Ying <ying.huang@intel.com>
> > 
> > > +ssize_t write_oldmem_page(unsigned long pfn, const char *buf,
> > > +			  size_t csize, unsigned long offset, int userbuf)
> > 
> > Hmm, int userbuf is only ever set to one... Does it make sense to have
> > write_oldmem_page in the separate file? The onl user is mem.c, perhaps
> > it should go there?
> > 
> 
> write_oldmem_page is kept to be consistent with copy_oldmem_page as much
> as possible. The userbuf is used by copy_oldmem_page too, and
> write_oldmem_page is in the same file as copy_oldmem_page. I think the
> consistence between them is reasonable.
> 
> And the copy_oldmem_page/write_oldmem_page is considered to be
> architecture dependent. Now, there are different implementations for
> copy_oldmem_page on different architectures. So I think the
> copy_oldmem_page/write_oldmem_page should be kept in separate file
> instead of go mem.c.

I thought it may have something to do with consistency. ACK on that
one. Perhaps it should be pushed to akpm?
									Pavel
-- 
(english) http://www.livejournal.com/~pavelmachek
(cesky, pictures) http://atrey.karlin.mff.cuni.cz/~pavel/picture/horses/blog.html