PATCH][1/8] 2.6.15 mlock: make_pages_wired/unwired

PATCH Stone Wang 3 participants 8 messages Mar 20, 2006 – Mar 22, 2006

Subsystems (2)

Memory management MM core

lore

Share & tools

lore

Stone Wang

author

2006-03-20 13:36 UTC #0 lore raw

1. Add make_pages_unwired routine.
2. Replace make_pages_present with make_pages_wired, support rollback.
3. Pass 1 more param ("wire") to get_user_pages.

Signed-off-by Shaoping Wang <[email protected]>

 include/linux/mm.h |    8 ++++--
 mm/memory.c        |   65 +++++++++++++++++++++++++++++++++++++++++++++--------
 2 files changed, 62 insertions(+), 11 deletions(-)

include/linux/mm.h +5 -1

diff -urN  linux-2.6.15.orig/include/linux/mm.h linux-2.6.15/include/linux/mm.h
@@ -59,6 +59,9 @@
 	unsigned long vm_start;		/* Our start address within vm_mm. */
 	unsigned long vm_end;		/* The first byte after our end address
 					   within vm_mm. */
+	int vm_wire_change;			/* VM_LOCKED bit of vm_flags was just changed.
+								 * For rollback support of sys_mlock series system calls.
+								 */
 
 	/* linked list of VM areas per task, sorted by address */
 	struct vm_area_struct *vm_next;

@@ -699,12 +706,13 @@
 	return __handle_mm_fault(mm, vma, address, write_access) & (~VM_FAULT_WRITE);
 }
 
-extern int make_pages_present(unsigned long addr, unsigned long end);
+extern int make_pages_wired(unsigned long addr, unsigned long end);
+void make_pages_unwired(struct mm_struct *mm, unsigned long addr,

unsigned long end);
extern int access_process_vm(struct task_struct *tsk, unsigned long
addr, void *buf, int len, int write);
void install_arg_page(struct vm_area_struct *, struct page *, unsigned long);

int get_user_pages(struct task_struct *tsk, struct mm_struct *mm,
unsigned long start,
- int len, int write, int force, struct page **pages, struct
vm_area_struct **vmas);
+ int len, int write, int force, int wire, struct page **pages,
struct vm_area_struct **vmas);
void print_bad_pte(struct vm_area_struct *, pte_t, unsigned long);

int __set_page_dirty_buffers(struct page *page);

mm/memory.c +34 -3

diff -urN  linux-2.6.15.orig/mm/memory.c linux-2.6.15/mm/memory.c
@@ -950,8 +950,30 @@
 	return page;
 }
 
+void make_pages_unwired(struct mm_struct *mm,
+					unsigned long start,unsigned long end)
+{
+	struct vm_area_struct *vma;
+	struct page *page;
+	unsigned int foll_flags;
+
+	foll_flags =0;
+
+	vma=find_vma(mm,start);
+	if(!vma)
+		BUG();
+	if(is_vm_hugetlb_page(vma))
+		return;
+	
+	for(; start<end ; start+=PAGE_SIZE) {
+		page=follow_page(vma,start,foll_flags);
+		if(page)
+			unwire_page(page);
+	}
+}
+
 int get_user_pages(struct task_struct *tsk, struct mm_struct *mm,
-		unsigned long start, int len, int write, int force,
+		unsigned long start, int len, int write,int force, int wire,
 		struct page **pages, struct vm_area_struct **vmas)
 {
 	int i;

@@ -973,6 +995,7 @@
 		if (!vma && in_gate_area(tsk, start)) {
 			unsigned long pg = start & PAGE_MASK;
 			struct vm_area_struct *gate_vma = get_gate_vma(tsk);
+			struct page *page;	
 			pgd_t *pgd;
 			pud_t *pud;
 			pmd_t *pmd;

@@ -994,6 +1017,7 @@
 				pte_unmap(pte);
 				return i ? : -EFAULT;
 			}
+			page = vm_normal_page(gate_vma, start, *pte);
 			if (pages) {
 				struct page *page = vm_normal_page(gate_vma, start, *pte);
 				pages[i] = page;

@@ -1003,9 +1027,12 @@
 			pte_unmap(pte);
 			if (vmas)
 				vmas[i] = gate_vma;
+			if(wire)
+				wire_page(page);
 			i++;
 			start += PAGE_SIZE;
 			len--;
+
 			continue;
 		}

@@ -1013,6 +1040,7 @@
 				|| !(vm_flags & vma->vm_flags))
 			return i ? : -EFAULT;
 
+		/* We dont account wired HugeTLB pages */
 		if (is_vm_hugetlb_page(vma)) {
 			i = follow_hugetlb_page(mm, vma, pages, vmas,
 						&start, &len, i);

@@ -1067,17 +1095,20 @@
 			}
 			if (vmas)
 				vmas[i] = vma;
+			if(wire)
+				wire_page(page);
 			i++;
 			start += PAGE_SIZE;
 			len--;
 		} while (len && start < vma->vm_end);
 	} while (len);
+
 	return i;
 }
 EXPORT_SYMBOL(get_user_pages);
 
-static int zeromap_pte_range(struct mm_struct *mm, pmd_t *pmd,
-			unsigned long addr, unsigned long end, pgprot_t prot)
+static int zeromap_pte_range(struct mm_struct *mm, struct vm_area_struct *vma,
+			pmd_t *pmd, unsigned long addr, unsigned long end, pgprot_t prot)
 {
 	pte_t *pte;
 	spinlock_t *ptl;

@@ -2306,10 +2338,13 @@
}
#endif /* __PAGETABLE_PMD_FOLDED */

-int make_pages_present(unsigned long addr, unsigned long end) +int make_pages_wired(unsigned long addr, unsigned long end)
{
int ret, len, write;
+ struct page *page;
struct vm_area_struct * vma;
+ struct mm_struct *mm=current->mm;
+ int wire_change;

vma = find_vma(current->mm, addr);
if (!vma)
@@ -2320,13 +2355,26 @@
if (end > vma->vm_end)
BUG();
len = (end+PAGE_SIZE-1)/PAGE_SIZE-addr/PAGE_SIZE;
- ret = get_user_pages(current, current->mm, addr,
- len, write, 0, NULL, NULL);
- if (ret < 0)
- return ret;
- return ret == len ? 0 : -1;
+ wire_change = vma->vm_wire_change;
+ vma->vm_wire_change = 1;
+ ret = get_user_pages(current, mm, addr,
+ len, write, 1, 1, NULL, NULL); /* write,set_wire */
+ vma->vm_wire_change = wire_change;
+ if(ret < len) {
+ for(; addr< end ; addr += PAGE_SIZE) {
+ page=follow_page(vma,addr,0);
+ if(page)
+ unwire_page(page);
+ else
+ BUG();
+ }
+ return -1;
+ }
+ else
+ return 0;
}

+
/*
* Map a vmalloc()-space virtual address to the physical page. */

--
To unsubscribe, send a message with 'unsubscribe linux-mm' in the body to [email protected]. For more info on Linux MM,
see: http://www.linux-mm.org/ .
Don't email: <a href=mailto:"[email protected]"> [email protected] </a>

Arjan van de Ven

2006-03-20 13:42 UTC #1 lore raw

Quoted 3 lines · Stone Wang

1. Add make_pages_unwired routine.
2. Replace make_pages_present with make_pages_wired, support rollback.
3. Pass 1 more param ("wire") to get_user_pages.

hmm again "wire" is a meaningless name
also.. get_user_pages ALWAYS pins the page ... so might as well make that automatic (with an unpin when the pinning is released)

Nick Piggin

2006-03-21 12:21 UTC #2 lore raw

In reply to Stone Wang

Quoted 1 line · Stone Wang

1. Add make_pages_unwired routine.

Unfortunately you forgot wire_page and unwire_page, so this patch will not even compile.

Quoted 1 line

2. Replace make_pages_present with make_pages_wired, support rollback.

What does support rollback mean?

Quoted 2 lines

3. Pass 1 more param ("wire") to get_user_pages.

As others have pointed out, wire may be a BSD / other unix thing, but it does not feature in Linux memory management terminology. If you want to introduce it, you need to do a better job of specifying it.

Quoted 15 lines

Signed-off-by Shaoping Wang <[email protected]>

mm/memory.c +12 -0

+void make_pages_unwired(struct mm_struct *mm,
+					unsigned long start,unsigned long end)
+{
+	struct vm_area_struct *vma;
+	struct page *page;
+	unsigned int foll_flags;
+
+	foll_flags =0;
+
+	vma=find_vma(mm,start);
+	if(!vma)
+		BUG();

mm/memory.c line 974

Quoted 10 lines

mm/memory.c +10 -0

+	if(is_vm_hugetlb_page(vma))
+		return;
+	
+	for(; start<end ; start+=PAGE_SIZE) {
+		page=follow_page(vma,start,foll_flags);
+		if(page)
+			unwire_page(page);
+	}
+}
+

What happens when start goes past vma->vm_end?

Quoted 15 lines

int get_user_pages(struct task_struct *tsk, struct mm_struct *mm,

mm/memory.c +2 -1

-		unsigned long start, int len, int write, int force,
+		unsigned long start, int len, int write,int force, int wire,
 		struct page **pages, struct vm_area_struct **vmas)
 {
 	int i;

@@ -973,6 +995,7 @@
 		if (!vma && in_gate_area(tsk, start)) {
 			unsigned long pg = start & PAGE_MASK;
 			struct vm_area_struct *gate_vma = get_gate_vma(tsk);
+			struct page *page;	
 			pgd_t *pgd;
 			pud_t *pud;
 			pmd_t *pmd;

mm/memory.c line 1020

Quoted 4 lines

mm/memory.c +1 -0

@@ -994,6 +1017,7 @@
 				pte_unmap(pte);
 				return i ? : -EFAULT;
 			}
+			page = vm_normal_page(gate_vma, start, *pte);

You wire gate_vma pages? But it doesn't look like you can unwire them with make_pages_unwired.

mm/memory.c line 1022

Quoted 2 lines

mm/memory.c +0 -0

 			if (pages) {
 				struct page *page = vm_normal_page(gate_vma, start, *pte);

This can go now?

Quoted 15 lines

mm/memory.c +3 -0

 				pages[i] = page;

@@ -1003,9 +1027,12 @@
 			pte_unmap(pte);
 			if (vmas)
 				vmas[i] = gate_vma;
+			if(wire)
+				wire_page(page);
 			i++;
 			start += PAGE_SIZE;
 			len--;
+
 			continue;
 		}

mm/memory.c line 1043

Quoted 4 lines

mm/memory.c +1 -0

@@ -1013,6 +1040,7 @@
 				|| !(vm_flags & vma->vm_flags))
 			return i ? : -EFAULT;
 
+		/* We dont account wired HugeTLB pages */

You don't account wired HugeTLB pages? If you can wire them you should be able to unwire them as well shouldn't you?

Signature · 9 lines

SUSE Labs, Novell Inc.

Send instant messages to your online friends http://au.messenger.yahoo.com 

--
To unsubscribe, send a message with 'unsubscribe linux-mm' in
the body to [email protected].  For more info on Linux MM,
see: http://www.linux-mm.org/ .
Don't email: <a href=mailto:"[email protected]"> [email protected] </a>

Stone Wang

author

2006-03-21 16:03 UTC #3 lore raw

We dont account HugeTLB pages for:

1. HugeTLB pages themselves are not reclaimable.

2. If we count HugeTLB pages in "Wired",then we would have no mind how many of the "Wired" are HugeTLB pages, and how many are
normal-size pages.
Thus, hard to get a clear map of physical memory use,for example:
how many pages are reclaimable?
If we must count HugeTLB pages,more fields should be added to
"/proc/meminfo",
for exmaple: "Wired HugeTLB:", "Wired Normal:".

Shaoping Wang

2006/3/21, Nick Piggin <[email protected]>:

Quoted 96 lines

Quoted 1 line · Stone Wang

1. Add make_pages_unwired routine.

Unfortunately you forgot wire_page and unwire_page, so this patch will not even compile.

Quoted 1 line

2. Replace make_pages_present with make_pages_wired, support rollback.

What does support rollback mean?

Quoted 2 lines

3. Pass 1 more param ("wire") to get_user_pages.

Quoted 15 lines

Signed-off-by Shaoping Wang <[email protected]>

mm/memory.c +12 -0

+void make_pages_unwired(struct mm_struct *mm,
+                                     unsigned long start,unsigned long end)
+{
+     struct vm_area_struct *vma;
+     struct page *page;
+     unsigned int foll_flags;
+
+     foll_flags =0;
+
+     vma=find_vma(mm,start);
+     if(!vma)
+             BUG();

mm/memory.c line 974

Quoted 10 lines

mm/memory.c +10 -0

+     if(is_vm_hugetlb_page(vma))
+             return;
+
+     for(; start<end ; start+=PAGE_SIZE) {
+             page=follow_page(vma,start,foll_flags);
+             if(page)
+                     unwire_page(page);
+     }
+}
+

What happens when start goes past vma->vm_end?

Quoted 15 lines

int get_user_pages(struct task_struct *tsk, struct mm_struct *mm,

mm/memory.c +2 -1

-             unsigned long start, int len, int write, int force,
+             unsigned long start, int len, int write,int force, int wire,
              struct page **pages, struct vm_area_struct **vmas)
 {
      int i;

@@ -973,6 +995,7 @@
              if (!vma && in_gate_area(tsk, start)) {
                      unsigned long pg = start & PAGE_MASK;
                      struct vm_area_struct *gate_vma = get_gate_vma(tsk);
+                     struct page *page;
                      pgd_t *pgd;
                      pud_t *pud;
                      pmd_t *pmd;

mm/memory.c line 1020

Quoted 4 lines

mm/memory.c +1 -0

@@ -994,6 +1017,7 @@
                              pte_unmap(pte);
                              return i ? : -EFAULT;
                      }
+                     page = vm_normal_page(gate_vma, start, *pte);

You wire gate_vma pages? But it doesn't look like you can unwire them with make_pages_unwired.

mm/memory.c line 1022

Quoted 2 lines

mm/memory.c +0 -0

                      if (pages) {
                              struct page *page = vm_normal_page(gate_vma, start, *pte);

This can go now?

Quoted 15 lines

mm/memory.c +3 -0

                              pages[i] = page;

@@ -1003,9 +1027,12 @@
                      pte_unmap(pte);
                      if (vmas)
                              vmas[i] = gate_vma;
+                     if(wire)
+                             wire_page(page);
                      i++;
                      start += PAGE_SIZE;
                      len--;
+
                      continue;
              }

mm/memory.c line 1043

Quoted 4 lines

mm/memory.c +1 -0

@@ -1013,6 +1040,7 @@
                              || !(vm_flags & vma->vm_flags))
                      return i ? : -EFAULT;
 
+             /* We dont account wired HugeTLB pages */

You don't account wired HugeTLB pages? If you can wire them you should be able to unwire them as well shouldn't you?

Quoted 6 lines

--
SUSE Labs, Novell Inc.
Send instant messages to your online friends http://au.messenger.yahoo.com

Nick Piggin

2006-03-22 00:28 UTC #4 lore raw

Quoted 4 lines · Stone Wang

We dont account HugeTLB pages for:
1. HugeTLB pages themselves are not reclaimable.

Quoted 9 lines

2. If we count HugeTLB pages in "Wired",then we would have no mind how many of the "Wired" are HugeTLB pages, and how many are
normal-size pages.
Thus, hard to get a clear map of physical memory use,for example:
how many pages are reclaimable?
If we must count HugeTLB pages,more fields should be added to
"/proc/meminfo",
for exmaple: "Wired HugeTLB:", "Wired Normal:".

Then why do you wire them at all? Your unwire function does not appear to be able to unwire them.

Signature · 8 lines

SUSE Labs, Novell Inc.
Send instant messages to your online friends http://au.messenger.yahoo.com 

--
To unsubscribe, send a message with 'unsubscribe linux-mm' in
the body to [email protected].  For more info on Linux MM,
see: http://www.linux-mm.org/ .
Don't email: <a href=mailto:"[email protected]"> [email protected] </a>

Stone Wang

author

2006-03-22 08:59 UTC #5 lore raw

2006/3/21, Nick Piggin <[email protected]>:

Quoted 15 lines

Quoted 4 lines · Stone Wang
We dont account HugeTLB pages for:
1. HugeTLB pages themselves are not reclaimable.
Quoted 9 lines
2. If we count HugeTLB pages in "Wired",then we would have no mind how many of the "Wired" are HugeTLB pages, and how many are
normal-size pages.
Thus, hard to get a clear map of physical memory use,for example:
how many pages are reclaimable?
If we must count HugeTLB pages,more fields should be added to
"/proc/meminfo",
for exmaple: "Wired HugeTLB:", "Wired Normal:".

Quoted 2 lines

Then why do you wire them at all? Your unwire function does not appear to be able to unwire them.

We didnt wire them.

Check get_user_pages():

        /* We dont account wired HugeTLB pages */
        if (is_vm_hugetlb_page(vma)) {
            i = follow_hugetlb_page(mm, vma, pages, vmas,
                        &start, &len, i);
            continue;
        }

Shaoping Wang

Quoted 6 lines

--
SUSE Labs, Novell Inc.
Send instant messages to your online friends http://au.messenger.yahoo.com

Nick Piggin

2006-03-22 10:13 UTC #6 lore raw

Quoted 5 lines · Stone Wang

2006/3/21, Nick Piggin <[email protected]>:
We didnt wire them.

But your comment said they were wired.

Signature · 8 lines

SUSE Labs, Novell Inc.
Send instant messages to your online friends http://au.messenger.yahoo.com 

--
To unsubscribe, send a message with 'unsubscribe linux-mm' in
the body to [email protected].  For more info on Linux MM,
see: http://www.linux-mm.org/ .
Don't email: <a href=mailto:"[email protected]"> [email protected] </a>

Stone Wang

author

2006-03-22 11:59 UTC #7 lore raw

Right, it made confusions.

I will correct it.

2006/3/22, Nick Piggin <[email protected]>:

Quoted 4 lines

Quoted 3 lines · Stone Wang
2006/3/21, Nick Piggin <[email protected]>:

Quoted 10 lines

Quoted 2 lines
We didnt wire them.
But your comment said they were wired.
--
SUSE Labs, Novell Inc.
Send instant messages to your online friends http://au.messenger.yahoo.com