Message-ID: <20190815044155.88483-19-gaoxiang25@huawei.com>
Date:   Thu, 15 Aug 2019 12:41:49 +0800
From:   Gao Xiang <gaoxiang25@...wei.com>
To:     <linux-fsdevel@...r.kernel.org>, <devel@...verdev.osuosl.org>,
        "Alexander Viro" <viro@...iv.linux.org.uk>
CC:     LKML <linux-kernel@...r.kernel.org>,
        Greg Kroah-Hartman <gregkh@...uxfoundation.org>,
        Andrew Morton <akpm@...ux-foundation.org>,
        Stephen Rothwell <sfr@...b.auug.org.au>,
        Theodore Ts'o <tytso@....edu>, "Pavel Machek" <pavel@...x.de>,
        David Sterba <dsterba@...e.cz>,
        Amir Goldstein <amir73il@...il.com>,
        Christoph Hellwig <hch@...radead.org>,
        "Darrick J . Wong" <darrick.wong@...cle.com>,
        Dave Chinner <david@...morbit.com>,
        "Jaegeuk Kim" <jaegeuk@...nel.org>, Jan Kara <jack@...e.cz>,
        Richard Weinberger <richard@....at>,
        Linus Torvalds <torvalds@...ux-foundation.org>,
        <linux-erofs@...ts.ozlabs.org>, Chao Yu <yuchao0@...wei.com>,
        Miao Xie <miaoxie@...wei.com>,
        Li Guifu <bluce.liguifu@...wei.com>,
        Fang Wei <fangwei1@...wei.com>,
        Gao Xiang <gaoxiang25@...wei.com>
Subject: [PATCH v8 18/24] erofs: introduce pagevec for decompression subsystem

For each physical cluster, there is a straight-forward
way of allocating a fixed or variable-sized array to
record the corresponding file pages for its decompression
if we decide to decompress these pages asynchronously
(eg. read-ahead case), however it will take variable-sized
on-heap memory compared with traditional uncompressed
filesystems.

This patch introduces a pagevec solution to reuse some
allocated file page in the time-sharing approach to store
parts of the array itself in order to minimize the extra
memory overhead, thus only a small-sized constant array
used for booting the whole array itself up will be needed.

Signed-off-by: Gao Xiang <gaoxiang25@...wei.com>
---
 fs/erofs/zpvec.h | 159 +++++++++++++++++++++++++++++++++++++++++++++++
 1 file changed, 159 insertions(+)
 create mode 100644 fs/erofs/zpvec.h

diff --git a/fs/erofs/zpvec.h b/fs/erofs/zpvec.h
new file mode 100644
index 000000000000..bb7689e67836
--- /dev/null
+++ b/fs/erofs/zpvec.h
@@ -0,0 +1,159 @@
+/* SPDX-License-Identifier: GPL-2.0-only */
+/*
+ * linux/fs/erofs/zpvec.h
+ *
+ * Copyright (C) 2018 HUAWEI, Inc.
+ *             http://www.huawei.com/
+ * Created by Gao Xiang <gaoxiang25@...wei.com>
+ */
+#ifndef __EROFS_FS_ZPVEC_H
+#define __EROFS_FS_ZPVEC_H
+
+#include "tagptr.h"
+
+/* page type in pagevec for decompress subsystem */
+enum z_erofs_page_type {
+	/* including Z_EROFS_VLE_PAGE_TAIL_EXCLUSIVE */
+	Z_EROFS_PAGE_TYPE_EXCLUSIVE,
+
+	Z_EROFS_VLE_PAGE_TYPE_TAIL_SHARED,
+
+	Z_EROFS_VLE_PAGE_TYPE_HEAD,
+	Z_EROFS_VLE_PAGE_TYPE_MAX
+};
+
+extern void __compiletime_error("Z_EROFS_PAGE_TYPE_EXCLUSIVE != 0")
+	__bad_page_type_exclusive(void);
+
+/* pagevec tagged pointer */
+typedef tagptr2_t	erofs_vtptr_t;
+
+/* pagevec collector */
+struct z_erofs_pagevec_ctor {
+	struct page *curr, *next;
+	erofs_vtptr_t *pages;
+
+	unsigned int nr, index;
+};
+
+static inline void z_erofs_pagevec_ctor_exit(struct z_erofs_pagevec_ctor *ctor,
+					     bool atomic)
+{
+	if (!ctor->curr)
+		return;
+
+	if (atomic)
+		kunmap_atomic(ctor->pages);
+	else
+		kunmap(ctor->curr);
+}
+
+static inline struct page *
+z_erofs_pagevec_ctor_next_page(struct z_erofs_pagevec_ctor *ctor,
+			       unsigned int nr)
+{
+	unsigned int index;
+
+	/* keep away from occupied pages */
+	if (ctor->next)
+		return ctor->next;
+
+	for (index = 0; index < nr; ++index) {
+		const erofs_vtptr_t t = ctor->pages[index];
+		const unsigned int tags = tagptr_unfold_tags(t);
+
+		if (tags == Z_EROFS_PAGE_TYPE_EXCLUSIVE)
+			return tagptr_unfold_ptr(t);
+	}
+	DBG_BUGON(nr >= ctor->nr);
+	return NULL;
+}
+
+static inline void
+z_erofs_pagevec_ctor_pagedown(struct z_erofs_pagevec_ctor *ctor,
+			      bool atomic)
+{
+	struct page *next = z_erofs_pagevec_ctor_next_page(ctor, ctor->nr);
+
+	z_erofs_pagevec_ctor_exit(ctor, atomic);
+
+	ctor->curr = next;
+	ctor->next = NULL;
+	ctor->pages = atomic ?
+		kmap_atomic(ctor->curr) : kmap(ctor->curr);
+
+	ctor->nr = PAGE_SIZE / sizeof(struct page *);
+	ctor->index = 0;
+}
+
+static inline void z_erofs_pagevec_ctor_init(struct z_erofs_pagevec_ctor *ctor,
+					     unsigned int nr,
+					     erofs_vtptr_t *pages,
+					     unsigned int i)
+{
+	ctor->nr = nr;
+	ctor->curr = ctor->next = NULL;
+	ctor->pages = pages;
+
+	if (i >= nr) {
+		i -= nr;
+		z_erofs_pagevec_ctor_pagedown(ctor, false);
+		while (i > ctor->nr) {
+			i -= ctor->nr;
+			z_erofs_pagevec_ctor_pagedown(ctor, false);
+		}
+	}
+	ctor->next = z_erofs_pagevec_ctor_next_page(ctor, i);
+	ctor->index = i;
+}
+
+static inline bool z_erofs_pagevec_enqueue(struct z_erofs_pagevec_ctor *ctor,
+					   struct page *page,
+					   enum z_erofs_page_type type,
+					   bool *occupied)
+{
+	*occupied = false;
+	if (unlikely(!ctor->next && type))
+		if (ctor->index + 1 == ctor->nr)
+			return false;
+
+	if (unlikely(ctor->index >= ctor->nr))
+		z_erofs_pagevec_ctor_pagedown(ctor, false);
+
+	/* exclusive page type must be 0 */
+	if (Z_EROFS_PAGE_TYPE_EXCLUSIVE != (uintptr_t)NULL)
+		__bad_page_type_exclusive();
+
+	/* should remind that collector->next never equal to 1, 2 */
+	if (type == (uintptr_t)ctor->next) {
+		ctor->next = page;
+		*occupied = true;
+	}
+	ctor->pages[ctor->index++] = tagptr_fold(erofs_vtptr_t, page, type);
+	return true;
+}
+
+static inline struct page *
+z_erofs_pagevec_dequeue(struct z_erofs_pagevec_ctor *ctor,
+			enum z_erofs_page_type *type)
+{
+	erofs_vtptr_t t;
+
+	if (unlikely(ctor->index >= ctor->nr)) {
+		DBG_BUGON(!ctor->next);
+		z_erofs_pagevec_ctor_pagedown(ctor, true);
+	}
+
+	t = ctor->pages[ctor->index];
+
+	*type = tagptr_unfold_tags(t);
+
+	/* should remind that collector->next never equal to 1, 2 */
+	if (*type == (uintptr_t)ctor->next)
+		ctor->next = tagptr_unfold_ptr(t);
+
+	ctor->pages[ctor->index++] = tagptr_fold(erofs_vtptr_t, NULL, 0);
+	return tagptr_unfold_ptr(t);
+}
+#endif
+
-- 
2.17.1

Powered by blists - more mailing lists