|
[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index] [Xen-changelog] Fix the balloon driver to do work in smaller batches, and
# HG changeset patch
# User kaf24@xxxxxxxxxxxxxxxxxxxx
# Node ID d5bd2c583cb01e1442827d5565b1cc19398b3f32
# Parent 0856c511a83ec51fc37aa22ab98257470e20b4fb
Fix the balloon driver to do work in smaller batches, and
not use a vmalloc'ed mfn list which may not be mapped into
all page tables.
Signed-off-by: Keir Fraser <keir@xxxxxxxxxxxxx>
diff -r 0856c511a83e -r d5bd2c583cb0
linux-2.6-xen-sparse/drivers/xen/balloon/balloon.c
--- a/linux-2.6-xen-sparse/drivers/xen/balloon/balloon.c Mon Sep 5
17:21:08 2005
+++ b/linux-2.6-xen-sparse/drivers/xen/balloon/balloon.c Mon Sep 5
17:51:55 2005
@@ -58,6 +58,12 @@
static struct proc_dir_entry *balloon_pde;
static DECLARE_MUTEX(balloon_mutex);
+
+/*
+ * Protects atomic reservation decrease/increase against concurrent increases.
+ * Also protects non-atomic updates of current_pages and driver_pages, and
+ * balloon lists.
+ */
spinlock_t balloon_lock = SPIN_LOCK_UNLOCKED;
/* We aim for 'current allocation' == 'target allocation'. */
@@ -157,6 +163,146 @@
return target;
}
+static int increase_reservation(unsigned long nr_pages)
+{
+ unsigned long *mfn_list, pfn, i, flags;
+ struct page *page;
+ long rc;
+ struct xen_memory_reservation reservation = {
+ .address_bits = 0,
+ .extent_order = 0,
+ .domid = DOMID_SELF
+ };
+
+ if (nr_pages > (PAGE_SIZE / sizeof(unsigned long)))
+ nr_pages = PAGE_SIZE / sizeof(unsigned long);
+
+ mfn_list = (unsigned long *)__get_free_page(GFP_KERNEL);
+ if (mfn_list == NULL)
+ return -ENOMEM;
+
+ balloon_lock(flags);
+
+ reservation.extent_start = mfn_list;
+ reservation.nr_extents = nr_pages;
+ rc = HYPERVISOR_memory_op(
+ XENMEM_increase_reservation, &reservation);
+ if (rc < nr_pages) {
+ /* We hit the Xen hard limit: reprobe. */
+ reservation.extent_start = mfn_list;
+ reservation.nr_extents = rc;
+ BUG_ON(HYPERVISOR_memory_op(
+ XENMEM_decrease_reservation,
+ &reservation) != rc);
+ hard_limit = current_pages + rc - driver_pages;
+ goto out;
+ }
+
+ for (i = 0; i < nr_pages; i++) {
+ page = balloon_retrieve();
+ BUG_ON(page == NULL);
+
+ pfn = page - mem_map;
+ BUG_ON(phys_to_machine_mapping[pfn] != INVALID_P2M_ENTRY);
+
+ /* Update P->M and M->P tables. */
+ phys_to_machine_mapping[pfn] = mfn_list[i];
+ xen_machphys_update(mfn_list[i], pfn);
+
+ /* Link back into the page tables if not highmem. */
+ if (pfn < max_low_pfn)
+ BUG_ON(HYPERVISOR_update_va_mapping(
+ (unsigned long)__va(pfn << PAGE_SHIFT),
+ pfn_pte_ma(mfn_list[i], PAGE_KERNEL),
+ 0));
+
+ /* Relinquish the page back to the allocator. */
+ ClearPageReserved(page);
+ set_page_count(page, 1);
+ __free_page(page);
+ }
+
+ current_pages += nr_pages;
+
+ out:
+ balloon_unlock(flags);
+
+ free_page((unsigned long)mfn_list);
+
+ return 0;
+}
+
+static int decrease_reservation(unsigned long nr_pages)
+{
+ unsigned long *mfn_list, pfn, i, flags;
+ struct page *page;
+ void *v;
+ int need_sleep = 0;
+ struct xen_memory_reservation reservation = {
+ .address_bits = 0,
+ .extent_order = 0,
+ .domid = DOMID_SELF
+ };
+
+ if (nr_pages > (PAGE_SIZE / sizeof(unsigned long)))
+ nr_pages = PAGE_SIZE / sizeof(unsigned long);
+
+ mfn_list = (unsigned long *)__get_free_page(GFP_KERNEL);
+ if (mfn_list == NULL)
+ return -ENOMEM;
+
+ for (i = 0; i < nr_pages; i++) {
+ if ((page = alloc_page(GFP_HIGHUSER)) == NULL) {
+ nr_pages = i;
+ need_sleep = 1;
+ break;
+ }
+
+ pfn = page - mem_map;
+ mfn_list[i] = phys_to_machine_mapping[pfn];
+
+ if (!PageHighMem(page)) {
+ v = phys_to_virt(pfn << PAGE_SHIFT);
+ scrub_pages(v, 1);
+ BUG_ON(HYPERVISOR_update_va_mapping(
+ (unsigned long)v, __pte_ma(0), 0));
+ }
+#ifdef CONFIG_XEN_SCRUB_PAGES
+ else {
+ v = kmap(page);
+ scrub_pages(v, 1);
+ kunmap(page);
+ }
+#endif
+ }
+
+ /* Ensure that ballooned highmem pages don't have kmaps. */
+ kmap_flush_unused();
+ flush_tlb_all();
+
+ balloon_lock(flags);
+
+ /* No more mappings: invalidate P2M and add to balloon. */
+ for (i = 0; i < nr_pages; i++) {
+ pfn = mfn_to_pfn(mfn_list[i]);
+ phys_to_machine_mapping[pfn] = INVALID_P2M_ENTRY;
+ balloon_append(pfn_to_page(pfn));
+ }
+
+ reservation.extent_start = mfn_list;
+ reservation.nr_extents = nr_pages;
+ BUG_ON(HYPERVISOR_memory_op(
+ XENMEM_decrease_reservation, &reservation) != nr_pages);
+
+ current_pages -= nr_pages;
+
+ balloon_unlock(flags);
+
+ free_page((unsigned long)mfn_list);
+
+ return need_sleep;
+}
+
/*
* We avoid multiple worker processes conflicting via the balloon mutex.
* We may of course race updates of the target counts (which are protected
@@ -165,123 +311,23 @@
*/
static void balloon_process(void *unused)
{
- unsigned long *mfn_list, pfn, i, flags;
- struct page *page;
- long credit, debt, rc;
- void *v;
- struct xen_memory_reservation reservation = {
- .address_bits = 0,
- .extent_order = 0,
- .domid = DOMID_SELF
- };
+ int need_sleep = 0;
+ long credit;
down(&balloon_mutex);
- retry:
- mfn_list = NULL;
-
- if ((credit = current_target() - current_pages) > 0) {
- mfn_list = vmalloc(credit * sizeof(*mfn_list));
- if (mfn_list == NULL)
- goto out;
-
- balloon_lock(flags);
- reservation.extent_start = mfn_list;
- reservation.nr_extents = credit;
- rc = HYPERVISOR_memory_op(
- XENMEM_increase_reservation, &reservation);
- balloon_unlock(flags);
- if (rc < credit) {
- /* We hit the Xen hard limit: reprobe. */
- reservation.extent_start = mfn_list;
- reservation.nr_extents = rc;
- BUG_ON(HYPERVISOR_memory_op(
- XENMEM_decrease_reservation,
- &reservation) != rc);
- hard_limit = current_pages + rc - driver_pages;
- vfree(mfn_list);
- goto retry;
- }
-
- for (i = 0; i < credit; i++) {
- page = balloon_retrieve();
- BUG_ON(page == NULL);
-
- pfn = page - mem_map;
- if (phys_to_machine_mapping[pfn] != INVALID_P2M_ENTRY)
- BUG();
-
- /* Update P->M and M->P tables. */
- phys_to_machine_mapping[pfn] = mfn_list[i];
- xen_machphys_update(mfn_list[i], pfn);
-
- /* Link back into the page tables if not highmem. */
- if (pfn < max_low_pfn)
- BUG_ON(HYPERVISOR_update_va_mapping(
- (unsigned long)__va(pfn << PAGE_SHIFT),
- pfn_pte_ma(mfn_list[i], PAGE_KERNEL),
- 0));
-
- /* Relinquish the page back to the allocator. */
- ClearPageReserved(page);
- set_page_count(page, 1);
- __free_page(page);
- }
-
- current_pages += credit;
- } else if (credit < 0) {
- debt = -credit;
-
- mfn_list = vmalloc(debt * sizeof(*mfn_list));
- if (mfn_list == NULL)
- goto out;
-
- for (i = 0; i < debt; i++) {
- if ((page = alloc_page(GFP_HIGHUSER)) == NULL) {
- debt = i;
- break;
- }
-
- pfn = page - mem_map;
- mfn_list[i] = phys_to_machine_mapping[pfn];
-
- if (!PageHighMem(page)) {
- v = phys_to_virt(pfn << PAGE_SHIFT);
- scrub_pages(v, 1);
- BUG_ON(HYPERVISOR_update_va_mapping(
- (unsigned long)v, __pte_ma(0), 0));
- }
-#ifdef CONFIG_XEN_SCRUB_PAGES
- else {
- v = kmap(page);
- scrub_pages(v, 1);
- kunmap(page);
- }
+ do {
+ credit = current_target() - current_pages;
+ if (credit > 0)
+ need_sleep = (increase_reservation(credit) != 0);
+ if (credit < 0)
+ need_sleep = (decrease_reservation(-credit) != 0);
+
+#ifndef CONFIG_PREEMPT
+ if (need_resched())
+ schedule();
#endif
- }
-
- /* Ensure that ballooned highmem pages don't have kmaps. */
- kmap_flush_unused();
- flush_tlb_all();
-
- /* No more mappings: invalidate P2M and add to balloon. */
- for (i = 0; i < debt; i++) {
- pfn = mfn_to_pfn(mfn_list[i]);
- phys_to_machine_mapping[pfn] = INVALID_P2M_ENTRY;
- balloon_append(pfn_to_page(pfn));
- }
-
- reservation.extent_start = mfn_list;
- reservation.nr_extents = debt;
- BUG_ON(HYPERVISOR_memory_op(
- XENMEM_decrease_reservation, &reservation) != debt);
-
- current_pages -= debt;
- }
-
- out:
- if (mfn_list != NULL)
- vfree(mfn_list);
+ } while ((credit != 0) && !need_sleep);
/* Schedule more work if there is some still to be done. */
if (current_target() != current_pages)
@@ -441,8 +487,9 @@
void balloon_update_driver_allowance(long delta)
{
unsigned long flags;
+
balloon_lock(flags);
- driver_pages += delta; /* non-atomic update */
+ driver_pages += delta;
balloon_unlock(flags);
}
@@ -475,9 +522,10 @@
scrub_pages(vstart, 1 << order);
- balloon_lock(flags);
BUG_ON(generic_page_range(
&init_mm, vstart, PAGE_SIZE << order, dealloc_pte_fn, NULL));
+
+ balloon_lock(flags);
current_pages -= 1UL << order;
balloon_unlock(flags);
_______________________________________________
Xen-changelog mailing list
Xen-changelog@xxxxxxxxxxxxxxxxxxx
http://lists.xensource.com/xen-changelog
|
 |
Lists.xenproject.org is hosted with RackSpace, monitoring our |