Skip to contentSkip to conversation

xen: switch to post-init routines in xen mmu.c earlier

PATCH In review Juergen Gross 4 participants 7 messages Dec 10, 2014 – Dec 11, 2014

Subsystems (2)
lore
Share & tools
lore

Juergen Gross

author maintainer
#0 lore raw

PATCH 1/1 xen: switch to post-init routines in xen mmu.c earlier

With the virtual mapped linear p2m list the post-init mmu operations must be used for setting up the p2m mappings, as in case of CONFIG_FLATMEM the init routines may trigger BUGs.

---
 arch/x86/xen/mmu.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)
arch/x86/xen/mmu.c +1 -1
index 6ab6150..a1a429a 100644
@@ -1225,6 +1225,7 @@ static void __init xen_pagetable_p2m_setup(void)
 static void __init xen_pagetable_init(void)
 {
 	paging_init();
+	xen_post_allocator_init();
 
 	xen_pagetable_p2m_setup();
 
@@ -1236,7 +1237,6 @@ static void __init xen_pagetable_init(void)
 		xen_remap_memory();
 
 	xen_setup_shared_info();
-	xen_post_allocator_init();
 }
 static void xen_write_cr2(unsigned long cr2)
 {
Signature · 1 line
2.1.2
#1 lore raw
Quoted 3 lines · Juergen Gross

With the virtual mapped linear p2m list the post-init mmu operations must be used for setting up the p2m mappings, as in case of CONFIG_FLATMEM the init routines may trigger BUGs.

Um, could you explain a bit more of why the CONFIG_FLATMEM is such unique? What about the other CONFIG_*MEM cases?

Quoted 30 lines
---
 arch/x86/xen/mmu.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)
arch/x86/xen/mmu.c +1 -1
index 6ab6150..a1a429a 100644
@@ -1225,6 +1225,7 @@ static void __init xen_pagetable_p2m_setup(void)
 static void __init xen_pagetable_init(void)
 {
 	paging_init();
+	xen_post_allocator_init();
 
 	xen_pagetable_p2m_setup();
 
@@ -1236,7 +1237,6 @@ static void __init xen_pagetable_init(void)
 		xen_remap_memory();
 
 	xen_setup_shared_info();
-	xen_post_allocator_init();
 }
 static void xen_write_cr2(unsigned long cr2)
 {
Signature · 1 line
2.1.2
author maintainer
#2 lore raw
Quoted 7 lines · Konrad Rzeszutek Wilk
Quoted 3 lines · Juergen Gross

With the virtual mapped linear p2m list the post-init mmu operations must be used for setting up the p2m mappings, as in case of CONFIG_FLATMEM the init routines may trigger BUGs.

Um, could you explain a bit more of why the CONFIG_FLATMEM is such unique? What about the other CONFIG_*MEM cases?

CONFIG_FLATMEM is just the configuration hitting a BUG() in xen_alloc_pte_init() being used after finishing paging_init().

Juergen

Quoted 37 lines
Quoted 29 lines
---
  arch/x86/xen/mmu.c | 2 +-
  1 file changed, 1 insertion(+), 1 deletion(-)
arch/x86/xen/mmu.c +1 -2
index 6ab6150..a1a429a 100644
@@ -1225,6 +1225,7 @@ static void __init xen_pagetable_p2m_setup(void)
  static void __init xen_pagetable_init(void)
  {
  	paging_init();
+	xen_post_allocator_init();
 
  	xen_pagetable_p2m_setup();
 
@@ -1236,7 +1237,6 @@ static void __init xen_pagetable_init(void)
  		xen_remap_memory();
 
  	xen_setup_shared_info();
-	xen_post_allocator_init();
  }
  static void xen_write_cr2(unsigned long cr2)
  {
--
Quoted 2 lines

2.1.2

--
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/

#3 lore raw
In reply to Juergen Gross
Quoted 15 lines · Juergen Gross

With the virtual mapped linear p2m list the post-init mmu operations must be used for setting up the p2m mappings, as in case of CONFIG_FLATMEM the init routines may trigger BUGs.

---
 arch/x86/xen/mmu.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)
arch/x86/xen/mmu.c +0 -0
index 6ab6150..a1a429a 100644

arch/x86/xen/mmu.c static void __init xen_pagetable_p2m_setup(void) · line 1231

Quoted 7 lines
arch/x86/xen/mmu.c +1 -0
@@ -1225,6 +1225,7 @@ static void __init xen_pagetable_p2m_setup(void)
 static void __init xen_pagetable_init(void)
 {
 	paging_init();
+	xen_post_allocator_init();
 
 	xen_pagetable_p2m_setup();
 

This feels very chicken-and-egg to me: To setup the P2M we need to use the MMU ops that use the P2M...

Please explain very clearly why this is all safe.

Quoted 9 lines
arch/x86/xen/mmu.c +0 -1
@@ -1236,7 +1237,6 @@ static void __init xen_pagetable_init(void)
 		xen_remap_memory();
 
 	xen_setup_shared_info();
-	xen_post_allocator_init();
 }
 static void xen_write_cr2(unsigned long cr2)
 {
author maintainer
#4 lore raw
Quoted 21 lines · David Vrabel
Quoted 20 lines · Juergen Gross

With the virtual mapped linear p2m list the post-init mmu operations must be used for setting up the p2m mappings, as in case of CONFIG_FLATMEM the init routines may trigger BUGs.

---
  arch/x86/xen/mmu.c | 2 +-
  1 file changed, 1 insertion(+), 1 deletion(-)
arch/x86/xen/mmu.c +1 -0
index 6ab6150..a1a429a 100644
@@ -1225,6 +1225,7 @@ static void __init xen_pagetable_p2m_setup(void)
  static void __init xen_pagetable_init(void)
  {
  	paging_init();
+	xen_post_allocator_init();
 
Quoted 7 lines

arch/x86/xen/mmu.c static void __init xen_pagetable_p2m_setup(void) · line 1231

Quoted 2 lines
arch/x86/xen/mmu.c +0 -0
  	xen_pagetable_p2m_setup();
 

This feels very chicken-and-egg to me: To setup the P2M we need to use the MMU ops that use the P2M...

Please explain very clearly why this is all safe.

Okay. paging_init() sets up all infrastructure needed to switch to the post-init mmu ops done by xen_post_allocator_init(). With the virtual mapped linear p2m list we need some mmu ops during setup of this list, so we have to switch to the correct mmu ops as soon as possible.

The p2m list is usable from the beginning, just expansion requires to have established the new linear mapping. So the call of xen_remap_memory() had to be introduced, but this is not due to the mmu ops requiring this.

Summing it up: calling xen_post_allocator_init() not directly after paging_init() was conceptually wrong in the beginning, it just didn't matter up to now as no functions used between the two calls needed some critical mmu ops (e.g. alloc_pte). This has changed now, so I corrected it.

Juergen

Quoted 17 lines

arch/x86/xen/mmu.c static void __init xen_pagetable_init(void) · line 1242

Quoted 10 lines
arch/x86/xen/mmu.c +0 -1
@@ -1236,7 +1237,6 @@ static void __init xen_pagetable_init(void)
  		xen_remap_memory();
 
  	xen_setup_shared_info();
-	xen_post_allocator_init();
  }
  static void xen_write_cr2(unsigned long cr2)
  {

--
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/

#5 lore raw
Quoted 40 lines · Juergen Gross
Quoted 21 lines · David Vrabel
Quoted 20 lines · Juergen Gross

With the virtual mapped linear p2m list the post-init mmu operations must be used for setting up the p2m mappings, as in case of CONFIG_FLATMEM the init routines may trigger BUGs.

---
  arch/x86/xen/mmu.c | 2 +-
  1 file changed, 1 insertion(+), 1 deletion(-)
arch/x86/xen/mmu.c +1 -0
index 6ab6150..a1a429a 100644
@@ -1225,6 +1225,7 @@ static void __init xen_pagetable_p2m_setup(void)
  static void __init xen_pagetable_init(void)
  {
      paging_init();
+    xen_post_allocator_init();
 
Quoted 7 lines

arch/x86/xen/mmu.c static void __init xen_pagetable_p2m_setup(void) · line 1231

Quoted 2 lines
arch/x86/xen/mmu.c +0 -0
      xen_pagetable_p2m_setup();
 

This feels very chicken-and-egg to me: To setup the P2M we need to use the MMU ops that use the P2M...

Please explain very clearly why this is all safe.

Okay. paging_init() sets up all infrastructure needed to switch to the post-init mmu ops done by xen_post_allocator_init(). With the virtual mapped linear p2m list we need some mmu ops during setup of this list, so we have to switch to the correct mmu ops as soon as possible.

The p2m list is usable from the beginning, just expansion requires to have established the new linear mapping. So the call of xen_remap_memory() had to be introduced, but this is not due to the mmu ops requiring this.

Quoted 5 lines

Summing it up: calling xen_post_allocator_init() not directly after paging_init() was conceptually wrong in the beginning, it just didn't matter up to now as no functions used between the two calls needed some critical mmu ops (e.g. alloc_pte). This has changed now, so I corrected it.

I've added this to the commit message and applied to devel/for-linus-3.19. If the tests pass I will consider sending a further pull request for 3.19 including the linear p2m changes early next week.

Thanks.

David

#6 lore raw
In reply to Juergen Gross
Quoted 40 lines · Juergen Gross
Quoted 21 lines · David Vrabel
Quoted 20 lines · Juergen Gross

With the virtual mapped linear p2m list the post-init mmu operations must be used for setting up the p2m mappings, as in case of CONFIG_FLATMEM the init routines may trigger BUGs.

---
  arch/x86/xen/mmu.c | 2 +-
  1 file changed, 1 insertion(+), 1 deletion(-)
arch/x86/xen/mmu.c +1 -0
index 6ab6150..a1a429a 100644
@@ -1225,6 +1225,7 @@ static void __init xen_pagetable_p2m_setup(void)
  static void __init xen_pagetable_init(void)
  {
      paging_init();
+    xen_post_allocator_init();
 
Quoted 7 lines

arch/x86/xen/mmu.c static void __init xen_pagetable_p2m_setup(void) · line 1231

Quoted 2 lines
arch/x86/xen/mmu.c +0 -0
      xen_pagetable_p2m_setup();
 

This feels very chicken-and-egg to me: To setup the P2M we need to use the MMU ops that use the P2M...

Please explain very clearly why this is all safe.

Okay. paging_init() sets up all infrastructure needed to switch to the post-init mmu ops done by xen_post_allocator_init(). With the virtual mapped linear p2m list we need some mmu ops during setup of this list, so we have to switch to the correct mmu ops as soon as possible.

The p2m list is usable from the beginning, just expansion requires to have established the new linear mapping. So the call of xen_remap_memory() had to be introduced, but this is not due to the mmu ops requiring this.

Quoted 5 lines

Summing it up: calling xen_post_allocator_init() not directly after paging_init() was conceptually wrong in the beginning, it just didn't matter up to now as no functions used between the two calls needed some critical mmu ops (e.g. alloc_pte). This has changed now, so I corrected it.

I've added this to the commit message and applied to devel/for-linus-3.19. If the tests pass I will consider sending a further pull request for 3.19 including the linear p2m changes early next week.

Thanks.

David