[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]
Message-ID: <20220615233420.3783-1-edwardwu@realtek.com>
Date: Thu, 16 Jun 2022 07:34:20 +0800
From: Edward Wu <edwardwu@...ltek.com>
To: Andrew Morton <akpm@...ux-foundation.org>, <linux-mm@...ck.org>,
<linux-kernel@...r.kernel.org>
CC: <surenb@...gle.com>, <minchan@...gle.com>, <edwardwu@...ltek.com>
Subject: [PATCH v3] mm: cma: sync everything after EBUSY
Since file-backed memory on CMA area could take long-term pinning.
By Minchan Kim's debug commit 151e084af494 ("mm: page_alloc:
dump migrate-failed pages only at -EBUSY")
We know the pinned page comes from buffer_head, ext4 journal, FS metadata.
Sync everything after EBUSY that can unpin most file-system pages.
And raise the success rate at next time try.
Link: https://lkml.kernel.org/r/20220615021504.23358-1-edwardwu@realtek.com
Signed-off-by: Edward Wu <edwardwu@...ltek.com>
---
V3:
- Use non-inline prototype for work function
v2:
- Fix compile warning
mm/cma.c | 25 +++++++++++++++++++++++++
1 file changed, 25 insertions(+)
diff --git a/mm/cma.c b/mm/cma.c
index eaa4b5c920a2..ddb9f0fbec27 100644
--- a/mm/cma.c
+++ b/mm/cma.c
@@ -31,6 +31,7 @@
#include <linux/highmem.h>
#include <linux/io.h>
#include <linux/kmemleak.h>
+#include <linux/syscalls.h>
#include <trace/events/cma.h>
#include "cma.h"
@@ -410,6 +411,24 @@ static void cma_debug_show_areas(struct cma *cma)
static inline void cma_debug_show_areas(struct cma *cma) { }
#endif
+static void cma_sync_work(struct work_struct *work)
+{
+ ksys_sync();
+ kfree(work);
+ pr_debug("%s(): EBUSY Sync complete\n", __func__);
+}
+
+static void cma_ebusy_sync_pinned_pages(void)
+{
+ struct work_struct *work;
+
+ work = kmalloc(sizeof(*work), GFP_ATOMIC);
+ if (work) {
+ INIT_WORK(work, cma_sync_work);
+ schedule_work(work);
+ }
+}
+
/**
* cma_alloc() - allocate pages from contiguous area
* @cma: Contiguous memory region for which the allocation is performed.
@@ -430,6 +449,7 @@ struct page *cma_alloc(struct cma *cma, unsigned long count,
unsigned long i;
struct page *page = NULL;
int ret = -ENOMEM;
+ bool sys_synchronized = false;
if (!cma || !cma->count || !cma->bitmap)
goto out;
@@ -480,6 +500,11 @@ struct page *cma_alloc(struct cma *cma, unsigned long count,
if (ret != -EBUSY)
break;
+ if (!sys_synchronized) {
+ sys_synchronized = true;
+ cma_ebusy_sync_pinned_pages();
+ }
+
pr_debug("%s(): memory range at %p is busy, retrying\n",
__func__, pfn_to_page(pfn));
--
2.17.1
Powered by blists - more mailing lists