Page MenuHomeFreeBSD
Files F137040714

D8752.id22819.diff
No OneTemporary
Actions

D8752.id22819.diff
View Options

Index: vm/vm_page.c
===================================================================
--- vm/vm_page.c
+++ vm/vm_page.c
@@ -1409,9 +1409,7 @@
*
* Note: we *always* dirty the page. It is necessary both for the
* fact that we moved it, and because we may be invalidating
- * swap. If the page is on the cache, we have to deactivate it
- * or vm_page_dirty() will panic. Dirty pages are not allowed
- * on the cache.
+ * swap.
*
* The objects must be locked.
*/
@@ -2042,8 +2040,8 @@
} else if (level >= 0) {
/*
* The page is reserved but not yet allocated. In
- * other words, it is still cached or free. Extend
- * the current run by one page.
+ * other words, it is still free. Extend the current
+ * run by one page.
*/
run_ext = 1;
#endif
@@ -2050,10 +2048,10 @@
} else if ((order = m->order) < VM_NFREEORDER) {
/*
* The page is enqueued in the physical memory
- * allocator's cache/free page queues. Moreover, it
- * is the first page in a power-of-two-sized run of
- * contiguous cache/free pages. Add these pages to
- * the end of the current run, and jump ahead.
+ * allocator's free page queues. Moreover, it is the
+ * first page in a power-of-two-sized run of
+ * contiguous free pages. Add these pages to the end
+ * of the current run, and jump ahead.
*/
run_ext = 1 << order;
m_inc = 1 << order;
@@ -2061,16 +2059,15 @@
/*
* Skip the page for one of the following reasons: (1)
* It is enqueued in the physical memory allocator's
- * cache/free page queues. However, it is not the
- * first page in a run of contiguous cache/free pages.
- * (This case rarely occurs because the scan is
- * performed in ascending order.) (2) It is not
- * reserved, and it is transitioning from free to
- * allocated. (Conversely, the transition from
- * allocated to free for managed pages is blocked by
- * the page lock.) (3) It is allocated but not
- * contained by an object and not wired, e.g.,
- * allocated by Xen's balloon driver.
+ * free page queues. However, it is not the first
+ * page in a run of contiguous free pages. (This case
+ * rarely occurs because the scan is performed in
+ * ascending order.) (2) It is not reserved, and it is
+ * transitioning from free to allocated. (Conversely,
+ * the transition from allocated to free for managed
+ * pages is blocked by the page lock.) (3) It is
+ * allocated but not contained by an object and not
+ * wired, e.g., allocated by Xen's balloon driver.
*/
run_ext = 0;
}
@@ -2282,11 +2279,11 @@
if (order < VM_NFREEORDER) {
/*
* The page is enqueued in the physical memory
- * allocator's cache/free page queues.
- * Moreover, it is the first page in a power-
- * of-two-sized run of contiguous cache/free
- * pages. Jump ahead to the last page within
- * that run, and continue from there.
+ * allocator's free page queues. Moreover, it
+ * is the first page in a power-of-two-sized
+ * run of contiguous free pages. Jump ahead
+ * to the last page within that run, and
+ * continue from there.
*/
m += (1 << order) - 1;
}
@@ -2334,9 +2331,9 @@
* conditions by relocating the virtual pages using that physical memory.
* Returns true if reclamation is successful and false otherwise. Since
* relocation requires the allocation of physical pages, reclamation may
- * fail due to a shortage of cache/free pages. When reclamation fails,
- * callers are expected to perform VM_WAIT before retrying a failed
- * allocation operation, e.g., vm_page_alloc_contig().
+ * fail due to a shortage of free pages. When reclamation fails, callers
+ * are expected to perform VM_WAIT before retrying a failed allocation
+ * operation, e.g., vm_page_alloc_contig().
*
* The caller must always specify an allocation class through "req".
*
@@ -2371,8 +2368,8 @@
req_class = VM_ALLOC_SYSTEM;
/*
- * Return if the number of cached and free pages cannot satisfy the
- * requested allocation.
+ * Return if the number of free pages cannot satisfy the requested
+ * allocation.
*/
count = vm_cnt.v_free_count;
if (count < npages + vm_cnt.v_free_reserved || (count < npages +
@@ -2642,9 +2639,8 @@
/*
* vm_page_free_wakeup:
*
- * Helper routine for vm_page_free_toq() and vm_page_cache(). This
- * routine is called when a page has been added to the cache or free
- * queues.
+ * Helper routine for vm_page_free_toq(). This routine is called
+ * when a page is added to the free queues.
*
* The page queues must be locked.
*/
@@ -2733,7 +2729,7 @@
/*
* Insert the page into the physical memory allocator's
- * cache/free page queues.
+ * free page queues.
*/
mtx_lock(&vm_page_queue_free_mtx);
vm_phys_freecnt_adj(m, 1);
@@ -2833,21 +2829,10 @@
/*
* Move the specified page to the inactive queue.
*
- * Many pages placed on the inactive queue should actually go
- * into the cache, but it is difficult to figure out which. What
- * we do instead, if the inactive target is well met, is to put
- * clean pages at the head of the inactive queue instead of the tail.
- * This will cause them to be moved to the cache more quickly and
- * if not actively re-referenced, reclaimed more quickly. If we just
- * stick these pages at the end of the inactive queue, heavy filesystem
- * meta-data accesses can cause an unnecessary paging load on memory bound
- * processes. This optimization causes one-time-use metadata to be
- * reused more quickly.
- *
- * Normally noreuse is FALSE, resulting in LRU operation. noreuse is set
- * to TRUE if we want this page to be 'as if it were placed in the cache',
- * except without unmapping it from the process address space. In
- * practice this is implemented by inserting the page at the head of the
+ * Normally, "noreuse" is FALSE, resulting in LRU ordering of the inactive
+ * queue. However, setting "noreuse" to TRUE will accelerate the specified
+ * page's reclamation, but it will not unmap the page from any address space.
+ * This is implemented by inserting the page near the head of the inactive
* queue, using a marker page to guide FIFO insertion ordering.
*
* The page must be locked.
@@ -2974,16 +2959,9 @@
if (advice == MADV_FREE)
/*
* Mark the page clean. This will allow the page to be freed
- * up by the system. However, such pages are often reused
- * quickly by malloc() so we do not do anything that would
- * cause a page fault if we can help it.
- *
- * Specifically, we do not try to actually free the page now
- * nor do we try to put it in the cache (which would cause a
- * page fault on reuse).
- *
- * But we do make the page as freeable as we can without
- * actually taking the step of unmapping it.
+ * without first paging it out. MADV_FREE pages are often
+ * quickly reused by malloc(3), so we do not do anything that
+ * would result in a page fault on a later access.
*/
vm_page_undirty(m);
else if (advice != MADV_DONTNEED)

File Metadata

Mime Type
text/plain
Expires
Sat, Nov 22, 1:01 AM (11 h, 14 m)
Storage Engine
blob
Storage Format
Raw Data
Storage Handle
25852824
Default Alt Text
D8752.id22819.diff (6 KB)

Event Timeline