kexec/i386: kexec page table code clean up - add arch_kimage

This patch add an architecture specific struct arch_kimage into struct kimage. Three pointers to page table pages used by kexec are added to struct arch_kimage. The page tables pages are dynamically allocated in machine_kexec_prepare instead of statically from BSS segment. This will save up to 20k memory when kexec image is not loaded.

Signed-off-by Huang Ying <[email protected]>

---
 arch/x86/kernel/machine_kexec_32.c |   68 +++++++++++++++++++++++++------------
 include/asm-x86/kexec_32.h         |   12 ++++++
 include/linux/kexec.h              |    4 ++
 3 files changed, 63 insertions(+), 21 deletions(-)

arch/x86/kernel/machine_kexec_32.c +1 -9

@@ -11,6 +11,7 @@
 #include <linux/delay.h>
 #include <linux/init.h>
 #include <linux/numa.h>
+#include <linux/gfp.h>
 #include <asm/pgtable.h>
 #include <asm/pgalloc.h>
 #include <asm/tlbflush.h>

@@ -21,15 +22,6 @@
 #include <asm/desc.h>
 #include <asm/system.h>
 
-#define PAGE_ALIGNED __attribute__ ((__aligned__(PAGE_SIZE)))
-static u32 kexec_pgd[1024] PAGE_ALIGNED;
-#ifdef CONFIG_X86_PAE
-static u32 kexec_pmd0[1024] PAGE_ALIGNED;
-static u32 kexec_pmd1[1024] PAGE_ALIGNED;
-#endif
-static u32 kexec_pte0[1024] PAGE_ALIGNED;
-static u32 kexec_pte1[1024] PAGE_ALIGNED;
-
 static void set_idt(void *newidt, __u16 limit)
 {
 	struct Xgt_desc_struct curidt;

arch/x86/kernel/machine_kexec_32.c +10 -1

+	free_page((unsigned long)image->arch_kimage.pgd);
+#ifdef CONFIG_X86_PAE
+	free_page((unsigned long)image->arch_kimage.pmd0);
+	free_page((unsigned long)image->arch_kimage.pmd1);
+#endif
+	free_page((unsigned long)image->arch_kimage.pte0);
+	free_page((unsigned long)image->arch_kimage.pte1);
+}
+
 /*
  * A architecture hook called to validate the
  * proposed image and prepare the control pages

@@ -83,10 +97,21 @@ static void load_segments(void)
  * reboot code buffer to allow us to avoid allocations
  * later.
  *
- * Currently nothing.
+ * - Allocate page tables
  */
 int machine_kexec_prepare(struct kimage *image)
 {

Review

Quoted 3 lines

(quoted diff) +2 -0

+static void alloc_page_tables(struct kimage *image)
+{

This is too generic a name. How about something like arch_alloc_kexec_page_tables()

Quoted 5 lines

arch/x86/kernel/machine_kexec_32.c +5 -0

+	image->arch_kimage.pgd = (pgd_t *)get_zeroed_page(GFP_KERNEL);
+#ifdef CONFIG_X86_PAE
+	image->arch_kimage.pmd0 = (pmd_t *)get_zeroed_page(GFP_KERNEL);
+	image->arch_kimage.pmd1 = (pmd_t *)get_zeroed_page(GFP_KERNEL);
+#endif

arch/x86/kernel/machine_kexec_32.c static void load_segments(void) · line 79

Quoted 6 lines

arch/x86/kernel/machine_kexec_32.c +6 -0

+	image->arch_kimage.pte0 = (pte_t *)get_zeroed_page(GFP_KERNEL);
+	image->arch_kimage.pte1 = (pte_t *)get_zeroed_page(GFP_KERNEL);
+}
+
+static void free_page_tables(struct kimage *image)
+{

How about arch_free_kexec_page_tables()

Quoted 26 lines

arch/x86/kernel/machine_kexec_32.c +15 -1

+	free_page((unsigned long)image->arch_kimage.pgd);
+#ifdef CONFIG_X86_PAE
+	free_page((unsigned long)image->arch_kimage.pmd0);
+	free_page((unsigned long)image->arch_kimage.pmd1);
+#endif
+	free_page((unsigned long)image->arch_kimage.pte0);
+	free_page((unsigned long)image->arch_kimage.pte1);
+}
+
 /*
  * A architecture hook called to validate the
  * proposed image and prepare the control pages

@@ -83,10 +97,21 @@ static void load_segments(void)
  * reboot code buffer to allow us to avoid allocations
  * later.
  *
- * Currently nothing.
+ * - Allocate page tables
  */
 int machine_kexec_prepare(struct kimage *image)
 {
+	alloc_page_tables(image);
+	if (!image->arch_kimage.pgd ||
+#ifdef CONFIG_X86_PAE
+	    !image->arch_kimage.pmd0 ||
+	    !image->arch_kimage.pmd1 ||

arch/x86/kernel/machine_kexec_32.c static void load_segments(void) · line 113

Quoted 5 lines

arch/x86/kernel/machine_kexec_32.c +5 -0

+#endif
+	    !image->arch_kimage.pte0 ||
+	    !image->arch_kimage.pte1) {
+		free_page_tables(image);
+		return -ENOMEM;

I think this error handling can be done in alloc_page_tables() itself and following will look neater.

Quoted 8 lines · Ying

This patch add an architecture specific struct arch_kimage into struct kimage. Three pointers to page table pages used by kexec are added to struct arch_kimage. The page tables pages are dynamically allocated in machine_kexec_prepare instead of statically from BSS segment. This will save up to 20k memory when kexec image is not loaded.

Signed-off-by Huang Ying <[email protected]>

There were some discussions in the past regarding static page table allocation vs dynamic. Can't recall it now though. Eric will know it better.

Anyway, this allocation is taking place at load time and not kexec execute time so as such should not be an issue for kdump.

Have you tested Xen too? Does it run into issues with this change?

Quoted 33 lines

---
 arch/x86/kernel/machine_kexec_32.c |   68 +++++++++++++++++++++++++------------
 include/asm-x86/kexec_32.h         |   12 ++++++
 include/linux/kexec.h              |    4 ++
 3 files changed, 63 insertions(+), 21 deletions(-)

arch/x86/kernel/machine_kexec_32.c +1 -9

@@ -11,6 +11,7 @@
 #include <linux/delay.h>
 #include <linux/init.h>
 #include <linux/numa.h>
+#include <linux/gfp.h>
 #include <asm/pgtable.h>
 #include <asm/pgalloc.h>
 #include <asm/tlbflush.h>

@@ -21,15 +22,6 @@
 #include <asm/desc.h>
 #include <asm/system.h>
 
-#define PAGE_ALIGNED __attribute__ ((__aligned__(PAGE_SIZE)))
-static u32 kexec_pgd[1024] PAGE_ALIGNED;
-#ifdef CONFIG_X86_PAE
-static u32 kexec_pmd0[1024] PAGE_ALIGNED;
-static u32 kexec_pmd1[1024] PAGE_ALIGNED;
-#endif
-static u32 kexec_pte0[1024] PAGE_ALIGNED;
-static u32 kexec_pte1[1024] PAGE_ALIGNED;
-
 static void set_idt(void *newidt, __u16 limit)
 {
 	struct Xgt_desc_struct curidt;

arch/x86/kernel/machine_kexec_32.c static void load_segments(void) · line 68

Quoted 5 lines

arch/x86/kernel/machine_kexec_32.c +2 -0

@@ -72,6 +64,28 @@ static void load_segments(void)
 #undef __STR
 }
 
+static void alloc_page_tables(struct kimage *image)
+{

This is too generic a name. How about something like arch_alloc_kexec_page_tables()

arch/x86/kernel/machine_kexec_32.c static void load_segments(void) · line 79

Quoted 11 lines

arch/x86/kernel/machine_kexec_32.c +11 -0

+	image->arch_kimage.pgd = (pgd_t *)get_zeroed_page(GFP_KERNEL);
+#ifdef CONFIG_X86_PAE
+	image->arch_kimage.pmd0 = (pmd_t *)get_zeroed_page(GFP_KERNEL);
+	image->arch_kimage.pmd1 = (pmd_t *)get_zeroed_page(GFP_KERNEL);
+#endif
+	image->arch_kimage.pte0 = (pte_t *)get_zeroed_page(GFP_KERNEL);
+	image->arch_kimage.pte1 = (pte_t *)get_zeroed_page(GFP_KERNEL);
+}
+
+static void free_page_tables(struct kimage *image)
+{

How about arch_free_kexec_page_tables()

Quoted 21 lines

arch/x86/kernel/machine_kexec_32.c +10 -1

+	free_page((unsigned long)image->arch_kimage.pgd);
+#ifdef CONFIG_X86_PAE
+	free_page((unsigned long)image->arch_kimage.pmd0);
+	free_page((unsigned long)image->arch_kimage.pmd1);
+#endif
+	free_page((unsigned long)image->arch_kimage.pte0);
+	free_page((unsigned long)image->arch_kimage.pte1);
+}
+
 /*
  * A architecture hook called to validate the
  * proposed image and prepare the control pages

@@ -83,10 +97,21 @@ static void load_segments(void)
  * reboot code buffer to allow us to avoid allocations
  * later.
  *
- * Currently nothing.
+ * - Allocate page tables
  */
 int machine_kexec_prepare(struct kimage *image)
 {

arch/x86/kernel/machine_kexec_32.c static void load_segments(void) · line 113

Quoted 10 lines

arch/x86/kernel/machine_kexec_32.c +10 -0

+	alloc_page_tables(image);
+	if (!image->arch_kimage.pgd ||
+#ifdef CONFIG_X86_PAE
+	    !image->arch_kimage.pmd0 ||
+	    !image->arch_kimage.pmd1 ||
+#endif
+	    !image->arch_kimage.pte0 ||
+	    !image->arch_kimage.pte1) {
+		free_page_tables(image);
+		return -ENOMEM;

I think this error handling can be done in alloc_page_tables() itself and following will look neater.

if (!alloc_page_tables)
return -ENOMEM.

Thanks
Vivek
--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in the body of a message to [email protected]
More majordomo info at http://vger.kernel.org/majordomo-info.html
Please read the FAQ at http://www.tux.org/lkml/

Quoted 1 line

Have you tested Xen too? Does it run into issues with this change?

Xen will need to be updated for this change but it shouldn't be a big deal. Basically right now it calls __ma() (which is kind of like __pa()) on kexec_pgd and friends. Updating the arguments to __ma() to use the new arch_kimage pointers should be easy enough.

N.B: As this is Dom0 code (not DomU) it is maintained in the xen repository and has been merged into Linux.

Quoted 6 lines · Simon Horman

Quoted 1 line

Have you tested Xen too? Does it run into issues with this change?

Xen will need to be updated for this change but it shouldn't be a big deal. Basically right now it calls __ma() (which is kind of like __pa()) on kexec_pgd and friends. Updating the arguments to __ma() to use the new arch_kimage pointers should be easy enough.

I think you mean the Xen specific parts of the kexec machinery in the kernel.c would need updating rather than the hypervisor, right?

I don't think these changes will cause anything more than minor problems for whoever ports the Xen kexec stuff upstream.

Quoted 2 lines

N.B: As this is Dom0 code (not DomU) it is maintained in the xen repository and has been merged into Linux.

I don't think we would bother backporting these changes to there, would we?