From: Andi Kleen <a...@linux.intel.com>
This implements the freeing back of large chunks in the ggc madvise path
Richard Guenther asked for. This way on systems with limited
address space malloc() and other allocators still have
a chance to get back at some of the memory ggc freed. The
fragmented pages are still just given back, but the address space
stays allocated.
I tried freeing only aligned 2MB areas to optimize for 2MB huge
pages, but the hit rate was quite low, so I switched to 1MB+
unaligned areas. The target size is a param now.
Passed bootstrap and testing on x86_64-linux
gcc/:
2011-10-18 Andi Kleen <a...@linux.intel.com>
* ggc-page (release_pages): First free large continuous
chunks in the madvise path.
* params.def (GGC_FREE_UNIT): Add.
* doc/invoke.texi (ggc-free-unit): Add.
---
gcc/doc/invoke.texi | 5 +++++
gcc/ggc-page.c | 48 ++++++++++++++++++++++++++++++++++++++++++++++++
gcc/params.def | 5 +++++
3 files changed, 58 insertions(+), 0 deletions(-)
diff --git a/gcc/doc/invoke.texi b/gcc/doc/invoke.texi
index 4f55dbc..e622552 100644
--- a/gcc/doc/invoke.texi
+++ b/gcc/doc/invoke.texi
@@ -8858,6 +8858,11 @@ very large effectively disables garbage collection.
Setting this
parameter and @option{ggc-min-expand} to zero causes a full collection
to occur at every opportunity.
+@item ggc-free-unit
+
+Continuous areas in OS pages to free back to OS immediately. Default is 256
+pages, which is 1MB with 4K pages.
+
@item max-reload-search-insns
The maximum number of instruction reload should look backward for equivalent
register. Increasing values mean more aggressive optimization, making the
diff --git a/gcc/ggc-page.c b/gcc/ggc-page.c
index ba88e3f..eb0eeef 100644
--- a/gcc/ggc-page.c
+++ b/gcc/ggc-page.c
@@ -972,6 +972,54 @@ release_pages (void)
page_entry *p, *start_p;
char *start;
size_t len;
+ size_t mapped_len;
+ page_entry *next, *prev, *newprev;
+ size_t free_unit = PARAM_VALUE (GGC_FREE_UNIT) * G.pagesize;
+
+ /* First free larger continuous areas to the OS.
+ This allows other allocators to grab these areas if needed.
+ This is only done on larger chunks to avoid fragmentation.
+ This does not always work because the free_pages list is only
+ sorted over a single GC cycle. */
+
+ p = G.free_pages;
+ prev = NULL;
+ while (p)
+ {
+ start = p->page;
+ start_p = p;
+ len = 0;
+ mapped_len = 0;
+ newprev = prev;
+ while (p && p->page == start + len)
+ {
+ len += p->bytes;
+ if (!p->discarded)
+ mapped_len += p->bytes;
+ newprev = p;
+ p = p->next;
+ }
+ if (len >= free_unit)
+ {
+ while (start_p != p)
+ {
+ next = start_p->next;
+ free (start_p);
+ start_p = next;
+ }
+ munmap (start, len);
+ if (prev)
+ prev->next = p;
+ else
+ G.free_pages = p;
+ G.bytes_mapped -= mapped_len;
+ continue;
+ }
+ prev = newprev;
+ }
+
+ /* Now give back the fragmented pages to the OS, but keep the address
+ space to reuse it next time. */
for (p = G.free_pages; p; )
{
diff --git a/gcc/params.def b/gcc/params.def
index 5e49c48..edbf0de 100644
--- a/gcc/params.def
+++ b/gcc/params.def
@@ -561,6 +561,11 @@ DEFPARAM(GGC_MIN_HEAPSIZE,
#undef GGC_MIN_EXPAND_DEFAULT
#undef GGC_MIN_HEAPSIZE_DEFAULT
+DEFPARAM(GGC_FREE_UNIT,
+ "ggc-free-unit",
+ "Continuous areas in OS pages to free back immediately",
+ 256, 0, 0)
+
DEFPARAM(PARAM_MAX_RELOAD_SEARCH_INSNS,
"max-reload-search-insns",
"The maximum number of instructions to search backward when looking
for equivalent reload",
--
1.7.5.4
