Received-SPF: pass (google.com: domain of linux-kernel+bounces-63405-linux.lists.archive=gmail.com@vger.kernel.org designates 147.75.80.249 as permitted sender) client-ip=147.75.80.249;
From: Gang Li <gang.li@linux.dev>
To: Andrew Morton <akpm@linux-foundation.org>,
	Steffen Klassert <steffen.klassert@secunet.com>,
	Daniel Jordan <daniel.m.jordan@oracle.com>,
	Jane Chu <jane.chu@oracle.com>,
	"Paul E . McKenney" <paulmck@kernel.org>,
	Muchun Song <muchun.song@linux.dev>,
	Gang Li <gang.li@linux.dev>,
	Randy Dunlap <rdunlap@infradead.org>,
	kernel test robot <lkp@intel.com>,
	Gang Li <ligang.bdlg@bytedance.com>
Cc: linux-kernel@vger.kernel.org,
	linux-mm@kvack.org
Subject: [PATCH v1 2/2] hugetlb: process multiple lists in gather_bootmem_prealloc_parallel
Date: Tue, 13 Feb 2024 19:13:47 +0800
Message-Id: <20240213111347.3189206-3-gang.li@linux.dev>
In-Reply-To: <20240213111347.3189206-1-gang.li@linux.dev>
References: <20240213111347.3189206-1-gang.li@linux.dev>
Precedence: bulk
MIME-Version: 1.0
Content-Transfer-Encoding: 8bit

gather_bootmem_prealloc_node currently only process one list in
huge_boot_pages array. So gather_bootmem_prealloc expects
padata_do_multithreaded to run num_node_state(N_MEMORY) instances of
gather_bootmem_prealloc_node to process all lists in huge_boot_pages.

This works well in current padata_do_multithreaded implementation.
It guarantees that size/min_chunk <= thread num <= max_threads.

```
/* Ensure at least one thread when size < min_chunk. */
nworks = max(job->size / max(job->min_chunk, job->align), 1ul);
nworks = min(nworks, job->max_threads);

ps.nworks      = padata_work_alloc_mt(nworks, &ps, &works);
```

However, the comment of padata_do_multithreaded API only promises a
maximum value for the number of threads and does not specify a
minimum value. Which may pass multiple nodes to
gather_bootmem_prealloc_node and only one node will be processed.

To avoid potential errors, introduce gather_bootmem_prealloc_parallel
to handle the case where the number of threads does not meet the
requirement of max_threads.

Fixes: 0306f03dcbd7 ("hugetlb: parallelize 1G hugetlb initialization")
Signed-off-by: Gang Li <ligang.bdlg@bytedance.com>
---
 mm/hugetlb.c | 15 +++++++++++----
 1 file changed, 11 insertions(+), 4 deletions(-)

diff --git a/mm/hugetlb.c b/mm/hugetlb.c
index 25069ca6ec248..2799a7ea098c1 100644
--- a/mm/hugetlb.c
+++ b/mm/hugetlb.c
@@ -3414,10 +3414,8 @@ static void __init prep_and_add_bootmem_folios(struct hstate *h,
  * Put bootmem huge pages into the standard lists after mem_map is up.
  * Note: This only applies to gigantic (order > MAX_PAGE_ORDER) pages.
  */
-static void __init gather_bootmem_prealloc_node(unsigned long start, unsigned long end, void *arg)
-
+static void __init gather_bootmem_prealloc_node(unsigned long nid)
 {
-	int nid = start;
 	LIST_HEAD(folio_list);
 	struct huge_bootmem_page *m;
 	struct hstate *h = NULL, *prev_h = NULL;
@@ -3455,10 +3453,19 @@ static void __init gather_bootmem_prealloc_node(unsigned long start, unsigned lo
 	prep_and_add_bootmem_folios(h, &folio_list);
 }
 
+static void __init gather_bootmem_prealloc_parallel(unsigned long start,
+						    unsigned long end, void *arg)
+{
+	int nid;
+
+	for (nid = start; nid < end; nid++)
+		gather_bootmem_prealloc_node(nid);
+}
+
 static void __init gather_bootmem_prealloc(void)
 {
 	struct padata_mt_job job = {
-		.thread_fn	= gather_bootmem_prealloc_node,
+		.thread_fn	= gather_bootmem_prealloc_parallel,
 		.fn_arg		= NULL,
 		.start		= 0,
 		.size		= num_node_state(N_MEMORY),
-- 
2.20.1