Loading...
1// SPDX-License-Identifier: GPL-2.0 OR BSD-3-Clause
2/*
3 * Copyright(c) 2015-2017 Intel Corporation.
4 */
5
6#include <linux/mm.h>
7#include <linux/sched/signal.h>
8#include <linux/device.h>
9#include <linux/module.h>
10
11#include "hfi.h"
12
13static unsigned long cache_size = 256;
14module_param(cache_size, ulong, S_IRUGO | S_IWUSR);
15MODULE_PARM_DESC(cache_size, "Send and receive side cache size limit (in MB)");
16
17/*
18 * Determine whether the caller can pin pages.
19 *
20 * This function should be used in the implementation of buffer caches.
21 * The cache implementation should call this function prior to attempting
22 * to pin buffer pages in order to determine whether they should do so.
23 * The function computes cache limits based on the configured ulimit and
24 * cache size. Use of this function is especially important for caches
25 * which are not limited in any other way (e.g. by HW resources) and, thus,
26 * could keeping caching buffers.
27 *
28 */
29bool hfi1_can_pin_pages(struct hfi1_devdata *dd, struct mm_struct *mm,
30 u32 nlocked, u32 npages)
31{
32 unsigned long ulimit_pages;
33 unsigned long cache_limit_pages;
34 unsigned int usr_ctxts;
35
36 /*
37 * Perform RLIMIT_MEMLOCK based checks unless CAP_IPC_LOCK is present.
38 */
39 if (!capable(CAP_IPC_LOCK)) {
40 ulimit_pages =
41 DIV_ROUND_DOWN_ULL(rlimit(RLIMIT_MEMLOCK), PAGE_SIZE);
42
43 /*
44 * Pinning these pages would exceed this process's locked memory
45 * limit.
46 */
47 if (atomic64_read(&mm->pinned_vm) + npages > ulimit_pages)
48 return false;
49
50 /*
51 * Only allow 1/4 of the user's RLIMIT_MEMLOCK to be used for HFI
52 * caches. This fraction is then equally distributed among all
53 * existing user contexts. Note that if RLIMIT_MEMLOCK is
54 * 'unlimited' (-1), the value of this limit will be > 2^42 pages
55 * (2^64 / 2^12 / 2^8 / 2^2).
56 *
57 * The effectiveness of this check may be reduced if I/O occurs on
58 * some user contexts before all user contexts are created. This
59 * check assumes that this process is the only one using this
60 * context (e.g., the corresponding fd was not passed to another
61 * process for concurrent access) as there is no per-context,
62 * per-process tracking of pinned pages. It also assumes that each
63 * user context has only one cache to limit.
64 */
65 usr_ctxts = dd->num_rcv_contexts - dd->first_dyn_alloc_ctxt;
66 if (nlocked + npages > (ulimit_pages / usr_ctxts / 4))
67 return false;
68 }
69
70 /*
71 * Pinning these pages would exceed the size limit for this cache.
72 */
73 cache_limit_pages = cache_size * (1024 * 1024) / PAGE_SIZE;
74 if (nlocked + npages > cache_limit_pages)
75 return false;
76
77 return true;
78}
79
80int hfi1_acquire_user_pages(struct mm_struct *mm, unsigned long vaddr, size_t npages,
81 bool writable, struct page **pages)
82{
83 int ret;
84 unsigned int gup_flags = FOLL_LONGTERM | (writable ? FOLL_WRITE : 0);
85
86 ret = pin_user_pages_fast(vaddr, npages, gup_flags, pages);
87 if (ret < 0)
88 return ret;
89
90 atomic64_add(ret, &mm->pinned_vm);
91
92 return ret;
93}
94
95void hfi1_release_user_pages(struct mm_struct *mm, struct page **p,
96 size_t npages, bool dirty)
97{
98 unpin_user_pages_dirty_lock(p, npages, dirty);
99
100 if (mm) { /* during close after signal, mm can be NULL */
101 atomic64_sub(npages, &mm->pinned_vm);
102 }
103}
1/*
2 * Copyright(c) 2015-2017 Intel Corporation.
3 *
4 * This file is provided under a dual BSD/GPLv2 license. When using or
5 * redistributing this file, you may do so under either license.
6 *
7 * GPL LICENSE SUMMARY
8 *
9 * This program is free software; you can redistribute it and/or modify
10 * it under the terms of version 2 of the GNU General Public License as
11 * published by the Free Software Foundation.
12 *
13 * This program is distributed in the hope that it will be useful, but
14 * WITHOUT ANY WARRANTY; without even the implied warranty of
15 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
16 * General Public License for more details.
17 *
18 * BSD LICENSE
19 *
20 * Redistribution and use in source and binary forms, with or without
21 * modification, are permitted provided that the following conditions
22 * are met:
23 *
24 * - Redistributions of source code must retain the above copyright
25 * notice, this list of conditions and the following disclaimer.
26 * - Redistributions in binary form must reproduce the above copyright
27 * notice, this list of conditions and the following disclaimer in
28 * the documentation and/or other materials provided with the
29 * distribution.
30 * - Neither the name of Intel Corporation nor the names of its
31 * contributors may be used to endorse or promote products derived
32 * from this software without specific prior written permission.
33 *
34 * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
35 * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
36 * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
37 * A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
38 * OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
39 * SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
40 * LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
41 * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
42 * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
43 * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
44 * OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
45 *
46 */
47
48#include <linux/mm.h>
49#include <linux/sched/signal.h>
50#include <linux/device.h>
51#include <linux/module.h>
52
53#include "hfi.h"
54
55static unsigned long cache_size = 256;
56module_param(cache_size, ulong, S_IRUGO | S_IWUSR);
57MODULE_PARM_DESC(cache_size, "Send and receive side cache size limit (in MB)");
58
59/*
60 * Determine whether the caller can pin pages.
61 *
62 * This function should be used in the implementation of buffer caches.
63 * The cache implementation should call this function prior to attempting
64 * to pin buffer pages in order to determine whether they should do so.
65 * The function computes cache limits based on the configured ulimit and
66 * cache size. Use of this function is especially important for caches
67 * which are not limited in any other way (e.g. by HW resources) and, thus,
68 * could keeping caching buffers.
69 *
70 */
71bool hfi1_can_pin_pages(struct hfi1_devdata *dd, struct mm_struct *mm,
72 u32 nlocked, u32 npages)
73{
74 unsigned long ulimit = rlimit(RLIMIT_MEMLOCK), pinned, cache_limit,
75 size = (cache_size * (1UL << 20)); /* convert to bytes */
76 unsigned int usr_ctxts =
77 dd->num_rcv_contexts - dd->first_dyn_alloc_ctxt;
78 bool can_lock = capable(CAP_IPC_LOCK);
79
80 /*
81 * Calculate per-cache size. The calculation below uses only a quarter
82 * of the available per-context limit. This leaves space for other
83 * pinning. Should we worry about shared ctxts?
84 */
85 cache_limit = (ulimit / usr_ctxts) / 4;
86
87 /* If ulimit isn't set to "unlimited" and is smaller than cache_size. */
88 if (ulimit != (-1UL) && size > cache_limit)
89 size = cache_limit;
90
91 /* Convert to number of pages */
92 size = DIV_ROUND_UP(size, PAGE_SIZE);
93
94 pinned = atomic64_read(&mm->pinned_vm);
95
96 /* First, check the absolute limit against all pinned pages. */
97 if (pinned + npages >= ulimit && !can_lock)
98 return false;
99
100 return ((nlocked + npages) <= size) || can_lock;
101}
102
103int hfi1_acquire_user_pages(struct mm_struct *mm, unsigned long vaddr, size_t npages,
104 bool writable, struct page **pages)
105{
106 int ret;
107 unsigned int gup_flags = FOLL_LONGTERM | (writable ? FOLL_WRITE : 0);
108
109 ret = get_user_pages_fast(vaddr, npages, gup_flags, pages);
110 if (ret < 0)
111 return ret;
112
113 atomic64_add(ret, &mm->pinned_vm);
114
115 return ret;
116}
117
118void hfi1_release_user_pages(struct mm_struct *mm, struct page **p,
119 size_t npages, bool dirty)
120{
121 put_user_pages_dirty_lock(p, npages, dirty);
122
123 if (mm) { /* during close after signal, mm can be NULL */
124 atomic64_sub(npages, &mm->pinned_vm);
125 }
126}