|
[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index] [Xen-changelog] [xen-unstable] [IA64] Make P2M semantics match x86
# HG changeset patch
# User awilliam@xxxxxxxxxxxx
# Date 1174922117 21600
# Node ID fa77ec03afd609b095ce5813a776123b176c8b67
# Parent be1017157768e8d6d5d5552cf0008d81d611b1bb
[IA64] Make P2M semantics match x86
So far the ia64 p2m has the semantic similar to
(the x86 p2m) + (page reference count).
But the differece from the x86 p2m have caused the breakage and work around.
This patch make the ia64 p2m semantic same as x86 to solve it.
- get_page() only for grant table page mapping and foreign domain page
mapping.
- put_page() only for those pages.
- guest_physmap_remove_page() doesn't touch PGC_allocated bit.
Signed-off-by: Isaku Yamahata <yamahata@xxxxxxxxxxxxx>
---
xen/arch/ia64/xen/mm.c | 175 ++++++++++++++++++----------------------------
xen/include/asm-ia64/mm.h | 2
2 files changed, 73 insertions(+), 104 deletions(-)
diff -r be1017157768 -r fa77ec03afd6 xen/arch/ia64/xen/mm.c
--- a/xen/arch/ia64/xen/mm.c Thu Mar 22 09:30:54 2007 -0600
+++ b/xen/arch/ia64/xen/mm.c Mon Mar 26 09:15:17 2007 -0600
@@ -208,46 +208,6 @@ alloc_dom_xen_and_dom_io(void)
BUG_ON(dom_io == NULL);
}
-// heavily depends on the struct page_info layout.
-// if (page_get_owner(page) == d &&
-// test_and_clear_bit(_PGC_allocated, &page->count_info)) {
-// put_page(page);
-// }
-static void
-try_to_clear_PGC_allocate(struct domain* d, struct page_info* page)
-{
- u32 _d, _nd;
- u64 x, nx, y;
-
- _d = pickle_domptr(d);
- y = *((u64*)&page->count_info);
- do {
- x = y;
- _nd = x >> 32;
- nx = x - 1;
- __clear_bit(_PGC_allocated, &nx);
-
- if (unlikely(!(x & PGC_allocated)) || unlikely(_nd != _d)) {
- struct domain* nd = unpickle_domptr(_nd);
- if (nd == NULL) {
- gdprintk(XENLOG_INFO, "gnttab_transfer: "
- "Bad page %p: ed=%p(%u) 0x%x, "
- "sd=%p 0x%x,"
- " caf=%016lx, taf=%" PRtype_info "\n",
- (void *) page_to_mfn(page),
- d, d->domain_id, _d,
- nd, _nd,
- x,
- page->u.inuse.type_info);
- }
- break;
- }
-
- BUG_ON((nx & PGC_count_mask) < 1);
- y = cmpxchg((u64*)&page->count_info, x, nx);
- } while (unlikely(y != x));
-}
-
static void
mm_teardown_pte(struct domain* d, volatile pte_t* pte, unsigned long offset)
{
@@ -267,22 +227,25 @@ mm_teardown_pte(struct domain* d, volati
if (!mfn_valid(mfn))
return;
page = mfn_to_page(mfn);
- // struct page_info corresponding to mfn may exist or not depending
- // on CONFIG_VIRTUAL_FRAME_TABLE.
- // This check is too easy.
- // The right way is to check whether this page is of io area or acpi pages
+ // page might be pte page for p2m exposing. check it.
if (page_get_owner(page) == NULL) {
BUG_ON(page->count_info != 0);
return;
}
+ // struct page_info corresponding to mfn may exist or not depending
+ // on CONFIG_VIRTUAL_FRAME_TABLE.
+ // The above check is too easy.
+ // The right way is to check whether this page is of io area or acpi pages
if (pte_pgc_allocated(old_pte)) {
BUG_ON(page_get_owner(page) != d);
BUG_ON(get_gpfn_from_mfn(mfn) == INVALID_M2P_ENTRY);
set_gpfn_from_mfn(mfn, INVALID_M2P_ENTRY);
- try_to_clear_PGC_allocate(d, page);
- }
- put_page(page);
+ if (test_and_clear_bit(_PGC_allocated, &page->count_info))
+ put_page(page);
+ } else {
+ put_page(page);
+ }
}
static void
@@ -770,7 +733,6 @@ __assign_new_domain_page(struct domain *
{
struct page_info *p;
unsigned long maddr;
- int ret;
BUG_ON(!pte_none(*pte));
@@ -791,8 +753,6 @@ __assign_new_domain_page(struct domain *
maddr);
}
- ret = get_page(p, d);
- BUG_ON(ret == 0);
set_gpfn_from_mfn(page_to_mfn(p), mpaddr >> PAGE_SHIFT);
// clear_page() and set_gpfn_from_mfn() become visible before set_pte_rel()
// because set_pte_rel() has release semantics
@@ -893,11 +853,9 @@ assign_domain_page(struct domain *d,
unsigned long mpaddr, unsigned long physaddr)
{
struct page_info* page = mfn_to_page(physaddr >> PAGE_SHIFT);
- int ret;
BUG_ON((physaddr & GPFN_IO_MASK) != GPFN_MEM);
- ret = get_page(page, d);
- BUG_ON(ret == 0);
+ BUG_ON(page->count_info != (PGC_allocated | 1));
set_gpfn_from_mfn(physaddr >> PAGE_SHIFT, mpaddr >> PAGE_SHIFT);
// because __assign_domain_page() uses set_pte_rel() which has
// release semantics, smp_mb() isn't needed.
@@ -1086,6 +1044,29 @@ assign_domain_mach_page(struct domain *d
}
static void
+adjust_page_count_info(struct page_info* page)
+{
+ struct domain* d = page_get_owner(page);
+ BUG_ON((page->count_info & PGC_count_mask) != 1);
+ if (d != NULL) {
+ int ret = get_page(page, d);
+ BUG_ON(ret == 0);
+ } else {
+ u64 x, nx, y;
+
+ y = *((u64*)&page->count_info);
+ do {
+ x = y;
+ nx = x + 1;
+
+ BUG_ON((x >> 32) != 0);
+ BUG_ON((nx & PGC_count_mask) != 2);
+ y = cmpxchg((u64*)&page->count_info, x, nx);
+ } while (unlikely(y != x));
+ }
+}
+
+static void
domain_put_page(struct domain* d, unsigned long mpaddr,
volatile pte_t* ptep, pte_t old_pte, int clear_PGC_allocate)
{
@@ -1100,8 +1081,19 @@ domain_put_page(struct domain* d, unsign
BUG();
}
- if (clear_PGC_allocate)
- try_to_clear_PGC_allocate(d, page);
+ if (likely(clear_PGC_allocate)) {
+ if (!test_and_clear_bit(_PGC_allocated, &page->count_info))
+ BUG();
+ /* put_page() is done by domain_page_flush_and_put() */
+ } else {
+ // In this case, page reference count mustn't touched.
+ // domain_page_flush_and_put() decrements it, we increment
+ // it in advence. This patch is slow path.
+ //
+ // guest_remove_page(): owner = d, count_info = 1
+ // memory_exchange(): owner = NULL, count_info = 1
+ adjust_page_count_info(page);
+ }
}
domain_page_flush_and_put(d, mpaddr, ptep, old_pte, page);
}
@@ -1148,7 +1140,7 @@ assign_domain_page_cmpxchg_rel(struct do
assign_domain_page_cmpxchg_rel(struct domain* d, unsigned long mpaddr,
struct page_info* old_page,
struct page_info* new_page,
- unsigned long flags)
+ unsigned long flags, int clear_PGC_allocate)
{
struct mm_struct *mm = &d->arch.mm;
volatile pte_t* pte;
@@ -1160,6 +1152,7 @@ assign_domain_page_cmpxchg_rel(struct do
pte_t new_pte;
pte_t ret_pte;
+ BUG_ON((flags & ASSIGN_pgc_allocated) == 0);
pte = lookup_alloc_domain_pte(d, mpaddr);
again:
@@ -1200,6 +1193,18 @@ assign_domain_page_cmpxchg_rel(struct do
BUG_ON(old_mfn == new_mfn);
set_gpfn_from_mfn(old_mfn, INVALID_M2P_ENTRY);
+ if (likely(clear_PGC_allocate)) {
+ if (!test_and_clear_bit(_PGC_allocated, &old_page->count_info))
+ BUG();
+ } else {
+ int ret;
+ // adjust for count_info for domain_page_flush_and_put()
+ // This is slow path.
+ BUG_ON(!test_bit(_PGC_allocated, &old_page->count_info));
+ BUG_ON(d == NULL);
+ ret = get_page(old_page, d);
+ BUG_ON(ret == 0);
+ }
domain_page_flush_and_put(d, mpaddr, pte, old_pte, old_page);
perfc_incrc(assign_domain_pge_cmpxchg_rel);
@@ -1207,7 +1212,8 @@ assign_domain_page_cmpxchg_rel(struct do
}
static void
-zap_domain_page_one(struct domain *d, unsigned long mpaddr, unsigned long mfn)
+zap_domain_page_one(struct domain *d, unsigned long mpaddr,
+ int clear_PGC_allocate, unsigned long mfn)
{
struct mm_struct *mm = &d->arch.mm;
volatile pte_t *pte;
@@ -1258,12 +1264,8 @@ zap_domain_page_one(struct domain *d, un
page = mfn_to_page(mfn);
BUG_ON((page->count_info & PGC_count_mask) == 0);
- // exchange_memory() calls
- // steal_page()
- // page owner is set to NULL
- // guest_physmap_remove_page()
- // zap_domain_page_one()
- domain_put_page(d, mpaddr, pte, old_pte, (page_get_owner(page) != NULL));
+ BUG_ON(clear_PGC_allocate && (page_get_owner(page) == NULL));
+ domain_put_page(d, mpaddr, pte, old_pte, clear_PGC_allocate);
perfc_incrc(zap_dcomain_page_one);
}
@@ -1276,7 +1278,7 @@ dom0vp_zap_physmap(struct domain *d, uns
return -ENOSYS;
}
- zap_domain_page_one(d, gpfn << PAGE_SHIFT, INVALID_MFN);
+ zap_domain_page_one(d, gpfn << PAGE_SHIFT, 1, INVALID_MFN);
perfc_incrc(dom0vp_zap_physmap);
return 0;
}
@@ -1563,6 +1565,7 @@ destroy_grant_host_mapping(unsigned long
/* try_to_clear_PGC_allocate(d, page) is not needed. */
BUG_ON(page_get_owner(page) == d &&
get_gpfn_from_mfn(mfn) == gpfn);
+ BUG_ON(pte_pgc_allocated(old_pte));
domain_page_flush_and_put(d, gpaddr, pte, old_pte, page);
perfc_incrc(destroy_grant_host_mapping);
@@ -1606,9 +1609,6 @@ steal_page(struct domain *d, struct page
// assign_domain_page_cmpxchg_rel() has release semantics
// so smp_mb() isn't needed.
- ret = get_page(new, d);
- BUG_ON(ret == 0);
-
gpfn = get_gpfn_from_mfn(page_to_mfn(page));
if (gpfn == INVALID_M2P_ENTRY) {
free_domheap_page(new);
@@ -1621,7 +1621,7 @@ steal_page(struct domain *d, struct page
ret = assign_domain_page_cmpxchg_rel(d, gpfn << PAGE_SHIFT, page, new,
ASSIGN_writable |
- ASSIGN_pgc_allocated);
+ ASSIGN_pgc_allocated, 0);
if (ret < 0) {
gdprintk(XENLOG_INFO, "assign_domain_page_cmpxchg_rel failed %d\n",
ret);
@@ -1648,18 +1648,8 @@ steal_page(struct domain *d, struct page
// page->u.inused._domain = 0;
_nd = x >> 32;
- if (
- // when !MEMF_no_refcount, page might be put_page()'d or
- // it will be put_page()'d later depending on queued.
- unlikely(!(memflags & MEMF_no_refcount) &&
- ((x & (PGC_count_mask | PGC_allocated)) !=
+ if (unlikely(((x & (PGC_count_mask | PGC_allocated)) !=
(1 | PGC_allocated))) ||
- // when MEMF_no_refcount, page isn't de-assigned from
- // this domain yet. So count_info = 2
- unlikely((memflags & MEMF_no_refcount) &&
- ((x & (PGC_count_mask | PGC_allocated)) !=
- (2 | PGC_allocated))) ||
-
unlikely(_nd != _d)) {
struct domain* nd = unpickle_domptr(_nd);
if (nd == NULL) {
@@ -1711,11 +1701,8 @@ guest_physmap_add_page(struct domain *d,
guest_physmap_add_page(struct domain *d, unsigned long gpfn,
unsigned long mfn)
{
- int ret;
-
BUG_ON(!mfn_valid(mfn));
- ret = get_page(mfn_to_page(mfn), d);
- BUG_ON(ret == 0);
+ BUG_ON(mfn_to_page(mfn)->count_info != (PGC_allocated | 1));
set_gpfn_from_mfn(mfn, gpfn);
smp_mb();
assign_domain_page_replace(d, gpfn << PAGE_SHIFT, mfn,
@@ -1731,7 +1718,7 @@ guest_physmap_remove_page(struct domain
unsigned long mfn)
{
BUG_ON(mfn == 0);//XXX
- zap_domain_page_one(d, gpfn << PAGE_SHIFT, mfn);
+ zap_domain_page_one(d, gpfn << PAGE_SHIFT, 0, mfn);
perfc_incrc(guest_physmap_remove_page);
}
@@ -2142,26 +2129,8 @@ arch_memory_op(int op, XEN_GUEST_HANDLE(
/* Unmap from old location, if any. */
gpfn = get_gpfn_from_mfn(mfn);
- if (gpfn != INVALID_M2P_ENTRY) {
- /*
- * guest_physmap_remove_page() (for IPF) descrements page
- * counter and unset PGC_allocated flag,
- * so pre-increment page counter and post-set flag inserted
- */
- /* pre-increment page counter */
- if (!get_page(mfn_to_page(mfn), d))
- goto out;
-
+ if (gpfn != INVALID_M2P_ENTRY)
guest_physmap_remove_page(d, gpfn, mfn);
-
- /* post-set PGC_allocated flag */
- if ((mfn_to_page(mfn)->count_info & PGC_count_mask) != 1) {
- /* no one but us is using this page */
- put_page(mfn_to_page(mfn));
- goto out;
- }
- set_bit(_PGC_allocated, &mfn_to_page(mfn)->count_info);
- }
/* Map at new location. */
guest_physmap_add_page(d, xatp.gpfn, mfn);
diff -r be1017157768 -r fa77ec03afd6 xen/include/asm-ia64/mm.h
--- a/xen/include/asm-ia64/mm.h Thu Mar 22 09:30:54 2007 -0600
+++ b/xen/include/asm-ia64/mm.h Mon Mar 26 09:15:17 2007 -0600
@@ -212,7 +212,7 @@ static inline int get_page_and_type(stru
static inline int page_is_removable(struct page_info *page)
{
- return ((page->count_info & PGC_count_mask) == 2);
+ return ((page->count_info & PGC_count_mask) == 1);
}
#define set_machinetophys(_mfn, _pfn) do { } while(0);
_______________________________________________
Xen-changelog mailing list
Xen-changelog@xxxxxxxxxxxxxxxxxxx
http://lists.xensource.com/xen-changelog
|
 |
Lists.xenproject.org is hosted with RackSpace, monitoring our |