Merge tag 'for_linus' of git://git.kernel.org/pub/scm/linux/kernel/git/mst/vhost

Pull virtio/vhost fixes from Michael Tsirkin: - Fixes in the iommu and balloon devices. - Disable the meta-data optimization for now - I hope we can get it fixed shortly, but there's no point in making users suffer crashes while we are working on that. * tag 'for_linus' of git://git.kernel.org/pub/scm/linux/kernel/git/mst/vhost: vhost: disable metadata prefetch optimization iommu/virtio: Update to most recent specification balloon: fix up comments mm/balloon_compaction: avoid duplicate page removal

Linus Torvalds committed 7 years ago

Commit 2a11c762 parents 45aee68 73f628e4 files · +89 −54Browse files

Unified · Split

4 files

drivers/iommu/virtio-iommu.c +29 −11
drivers/vhost/vhost.h +1 −1
include/uapi/linux/virtio_iommu.h +18 −14
mm/balloon_compaction.c +41 −28

drivers/iommu/virtio-iommu.c+29−11

@@ -2,7 +2,7 @@ · 1 unmodified line
2	2	/*
3	3	* Virtio driver for the paravirtualized IOMMU
4	4	*
5		− * Copyright (C) 2018 Arm Limited
	5	+ * Copyright (C) 2019 Arm Limited
6	6	*/
7	7
8	8	#define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
@@ -47,7 +47,10 @@ struct viommu_dev { · 38 unmodified lines
47	47	/* Device configuration */
48	48	struct iommu_domain_geometry geometry;
49	49	u64 pgsize_bitmap;
50		− u8 domain_bits;
	50	+ u32 first_domain;
	51	+ u32 last_domain;
	52	+ /* Supported MAP flags */
	53	+ u32 map_flags;
51	54	u32 probe_size;
52	55	};
53	56
@@ -62,6 +65,7 @@ struct viommu_domain { · 8 unmodified lines
62	65	struct viommu_dev *viommu;
63	66	struct mutex mutex; /* protects viommu pointer */
64	67	unsigned int id;
	68	+ u32 map_flags;
65	69
66	70	spinlock_t mappings_lock;
67	71	struct rb_root_cached mappings;
@@ -113,6 +117,8 @@ static int viommu_get_req_errno(void *buf, size_t len) · 45 unmodified lines
113	117	return -ENOENT;
114	118	case VIRTIO_IOMMU_S_FAULT:
115	119	return -EFAULT;
	120	+ case VIRTIO_IOMMU_S_NOMEM:
	121	+ return -ENOMEM;
116	122	case VIRTIO_IOMMU_S_IOERR:
117	123	case VIRTIO_IOMMU_S_DEVERR:
118	124	default:
@@ -607,15 +613,15 @@ static int viommu_domain_finalise(struct viommu_dev *viommu, · 488 unmodified lines
607	613	{
608	614	int ret;
609	615	struct viommu_domain *vdomain = to_viommu_domain(domain);
610		− unsigned int max_domain = viommu->domain_bits > 31 ? ~0 :
611		− (1U << viommu->domain_bits) - 1;
612	616
613	617	vdomain->viommu = viommu;
	618	+ vdomain->map_flags = viommu->map_flags;
614	619
615	620	domain->pgsize_bitmap = viommu->pgsize_bitmap;
616	621	domain->geometry = viommu->geometry;
617	622
618		− ret = ida_alloc_max(&viommu->domain_ids, max_domain, GFP_KERNEL);
	623	+ ret = ida_alloc_range(&viommu->domain_ids, viommu->first_domain,
	624	+ viommu->last_domain, GFP_KERNEL);
619	625	if (ret >= 0)
620	626	vdomain->id = (unsigned int)ret;
621	627
@@ -710,7 +716,7 @@ static int viommu_map(struct iommu_domain *domain, unsigned long iova, · 88 unmodified lines
710	716	phys_addr_t paddr, size_t size, int prot)
711	717	{
712	718	int ret;
713		− int flags;
	719	+ u32 flags;
714	720	struct virtio_iommu_req_map map;
715	721	struct viommu_domain *vdomain = to_viommu_domain(domain);
716	722
@@ -718,6 +724,9 @@ static int viommu_map(struct iommu_domain *domain, unsigned long iova, · 1 unmodified line
718	724	(prot & IOMMU_WRITE ? VIRTIO_IOMMU_MAP_F_WRITE : 0) \|
719	725	(prot & IOMMU_MMIO ? VIRTIO_IOMMU_MAP_F_MMIO : 0);
720	726
	727	+ if (flags & ~vdomain->map_flags)
	728	+ return -EINVAL;
	729	+
721	730	ret = viommu_add_mapping(vdomain, iova, paddr, size, flags);
722	731	if (ret)
723	732	return ret;
@@ -1027,7 +1036,8 @@ static int viommu_probe(struct virtio_device *vdev) · 303 unmodified lines
1027	1036	goto err_free_vqs;
1028	1037	}
1029	1038
1030		− viommu->domain_bits = 32;
	1039	+ viommu->map_flags = VIRTIO_IOMMU_MAP_F_READ \| VIRTIO_IOMMU_MAP_F_WRITE;
	1040	+ viommu->last_domain = ~0U;
1031	1041
1032	1042	/* Optional features */
1033	1043	virtio_cread_feature(vdev, VIRTIO_IOMMU_F_INPUT_RANGE,
@@ -1038,9 +1048,13 @@ static int viommu_probe(struct virtio_device *vdev) · 4 unmodified lines
1038	1048	struct virtio_iommu_config, input_range.end,
1039	1049	&input_end);
1040	1050
1041		− virtio_cread_feature(vdev, VIRTIO_IOMMU_F_DOMAIN_BITS,
1042		− struct virtio_iommu_config, domain_bits,
1043		− &viommu->domain_bits);
	1051	+ virtio_cread_feature(vdev, VIRTIO_IOMMU_F_DOMAIN_RANGE,
	1052	+ struct virtio_iommu_config, domain_range.start,
	1053	+ &viommu->first_domain);
	1054	+
	1055	+ virtio_cread_feature(vdev, VIRTIO_IOMMU_F_DOMAIN_RANGE,
	1056	+ struct virtio_iommu_config, domain_range.end,
	1057	+ &viommu->last_domain);
1044	1058
1045	1059	virtio_cread_feature(vdev, VIRTIO_IOMMU_F_PROBE,
1046	1060	struct virtio_iommu_config, probe_size,
@@ -1052,6 +1066,9 @@ static int viommu_probe(struct virtio_device *vdev) · 5 unmodified lines
1052	1066	.force_aperture = true,
1053	1067	};
1054	1068
	1069	+ if (virtio_has_feature(vdev, VIRTIO_IOMMU_F_MMIO))
	1070	+ viommu->map_flags \|= VIRTIO_IOMMU_MAP_F_MMIO;
	1071	+
1055	1072	viommu_ops.pgsize_bitmap = viommu->pgsize_bitmap;
1056	1073
1057	1074	virtio_device_ready(vdev);
@@ -1130,9 +1147,10 @@ static void viommu_config_changed(struct virtio_device *vdev) · 72 unmodified lines
1130	1147
1131	1148	static unsigned int features[] = {
1132	1149	VIRTIO_IOMMU_F_MAP_UNMAP,
1133		− VIRTIO_IOMMU_F_DOMAIN_BITS,
1134	1150	VIRTIO_IOMMU_F_INPUT_RANGE,
	1151	+ VIRTIO_IOMMU_F_DOMAIN_RANGE,
1135	1152	VIRTIO_IOMMU_F_PROBE,
	1153	+ VIRTIO_IOMMU_F_MMIO,
1136	1154	};
1137	1155
1138	1156	static struct virtio_device_id id_table[] = {

drivers/vhost/vhost.h+1−1

@@ -96,7 +96,7 @@ struct vhost_uaddr { · 95 unmodified lines
96	96	};
97	97
98	98	#if defined(CONFIG_MMU_NOTIFIER) && ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE == 0
99		−#define VHOST_ARCH_CAN_ACCEL_UACCESS 1
	99	+#define VHOST_ARCH_CAN_ACCEL_UACCESS 0
100	100	#else
101	101	#define VHOST_ARCH_CAN_ACCEL_UACCESS 0
102	102	#endif

include/uapi/linux/virtio_iommu.h+18−14

@@ -1,8 +1,8 @@
1	1	/* SPDX-License-Identifier: BSD-3-Clause */
2	2	/*
3		− * Virtio-iommu definition v0.9
	3	+ * Virtio-iommu definition v0.12
4	4	*
5		− * Copyright (C) 2018 Arm Ltd.
	5	+ * Copyright (C) 2019 Arm Ltd.
6	6	*/
7	7	#ifndef _UAPI_LINUX_VIRTIO_IOMMU_H
8	8	#define _UAPI_LINUX_VIRTIO_IOMMU_H
@@ -11,26 +11,31 @@ · 2 unmodified lines
11	11
12	12	/* Feature bits */
13	13	#define VIRTIO_IOMMU_F_INPUT_RANGE 0
14		−#define VIRTIO_IOMMU_F_DOMAIN_BITS 1
	14	+#define VIRTIO_IOMMU_F_DOMAIN_RANGE 1
15	15	#define VIRTIO_IOMMU_F_MAP_UNMAP 2
16	16	#define VIRTIO_IOMMU_F_BYPASS 3
17	17	#define VIRTIO_IOMMU_F_PROBE 4
	18	+#define VIRTIO_IOMMU_F_MMIO 5
18	19
19		−struct virtio_iommu_range {
20		− __u64 start;
21		− __u64 end;
	20	+struct virtio_iommu_range_64 {
	21	+ __le64 start;
	22	+ __le64 end;
	23	+};
	24	+
	25	+struct virtio_iommu_range_32 {
	26	+ __le32 start;
	27	+ __le32 end;
22	28	};
23	29
24	30	struct virtio_iommu_config {
25	31	/* Supported page sizes */
26		− __u64 page_size_mask;
	32	+ __le64 page_size_mask;
27	33	/* Supported IOVA range */
28		− struct virtio_iommu_range input_range;
	34	+ struct virtio_iommu_range_64 input_range;
29	35	/* Max domain ID size */
30		− __u8 domain_bits;
31		− __u8 padding[3];
	36	+ struct virtio_iommu_range_32 domain_range;
32	37	/* Probe buffer size */
33		− __u32 probe_size;
	38	+ __le32 probe_size;
34	39	};
35	40
36	41	/* Request types */
@@ -49,6 +54,7 @@ struct virtio_iommu_config { · 12 unmodified lines
49	54	#define VIRTIO_IOMMU_S_RANGE 0x05
50	55	#define VIRTIO_IOMMU_S_NOENT 0x06
51	56	#define VIRTIO_IOMMU_S_FAULT 0x07
	57	+#define VIRTIO_IOMMU_S_NOMEM 0x08
52	58
53	59	struct virtio_iommu_req_head {
54	60	__u8 type;
@@ -78,12 +84,10 @@ struct virtio_iommu_req_detach { · 23 unmodified lines
78	84
79	85	#define VIRTIO_IOMMU_MAP_F_READ (1 << 0)
80	86	#define VIRTIO_IOMMU_MAP_F_WRITE (1 << 1)
81		−#define VIRTIO_IOMMU_MAP_F_EXEC (1 << 2)
82		−#define VIRTIO_IOMMU_MAP_F_MMIO (1 << 3)
	87	+#define VIRTIO_IOMMU_MAP_F_MMIO (1 << 2)
83	88
84	89	#define VIRTIO_IOMMU_MAP_F_MASK (VIRTIO_IOMMU_MAP_F_READ \| \
85	90	VIRTIO_IOMMU_MAP_F_WRITE \| \
86		− VIRTIO_IOMMU_MAP_F_EXEC \| \
87	91	VIRTIO_IOMMU_MAP_F_MMIO)
88	92
89	93	struct virtio_iommu_req_map {

mm/balloon_compaction.c+41−28

@@ -21,7 +21,6 @@ static void balloon_page_enqueue_one(struct balloon_dev_info *b_dev_info, · 20 unmodified lines
21	21	* memory corruption is possible and we should stop execution.
22	22	*/
23	23	BUG_ON(!trylock_page(page));
24		− list_del(&page->lru);
25	24	balloon_page_insert(b_dev_info, page);
26	25	unlock_page(page);
27	26	__count_vm_event(BALLOON_INFLATE);
@@ -33,8 +32,8 @@ static void balloon_page_enqueue_one(struct balloon_dev_info *b_dev_info, · 5 unmodified lines
33	32	* @b_dev_info: balloon device descriptor where we will insert a new page to
34	33	* @pages: pages to enqueue - allocated using balloon_page_alloc.
35	34	*
36		− * Driver must call it to properly enqueue a balloon pages before definitively
37		− * removing it from the guest system.
	35	+ * Driver must call this function to properly enqueue balloon pages before
	36	+ * definitively removing them from the guest system.
38	37	*
39	38	* Return: number of pages that were enqueued.
40	39	*/
@@ -47,6 +46,7 @@ size_t balloon_page_list_enqueue(struct balloon_dev_info *b_dev_info, · 6 unmodified lines
47	46
48	47	spin_lock_irqsave(&b_dev_info->pages_lock, flags);
49	48	list_for_each_entry_safe(page, tmp, pages, lru) {
	49	+ list_del(&page->lru);
50	50	balloon_page_enqueue_one(b_dev_info, page);
51	51	n_pages++;
52	52	}
@@ -63,12 +63,13 @@ EXPORT_SYMBOL_GPL(balloon_page_list_enqueue); · 10 unmodified lines
63	63	* @n_req_pages: number of requested pages.
64	64	*
65	65	* Driver must call this function to properly de-allocate a previous enlisted
66		− * balloon pages before definetively releasing it back to the guest system.
	66	+ * balloon pages before definitively releasing it back to the guest system.
67	67	* This function tries to remove @n_req_pages from the ballooned pages and
68	68	* return them to the caller in the @pages list.
69	69	*
70		− * Note that this function may fail to dequeue some pages temporarily empty due
71		− * to compaction isolated pages.
	70	+ * Note that this function may fail to dequeue some pages even if the balloon
	71	+ * isn't empty - since the page list can be temporarily empty due to compaction
	72	+ * of isolated pages.
72	73	*
73	74	* Return: number of pages that were added to the @pages list.
74	75	*/
@@ -112,12 +113,13 @@ EXPORT_SYMBOL_GPL(balloon_page_list_dequeue); · 37 unmodified lines
112	113
113	114	/*
114	115	* balloon_page_alloc - allocates a new page for insertion into the balloon
115		− * page list.
	116	+ * page list.
	117	+ *
	118	+ * Driver must call this function to properly allocate a new balloon page.
	119	+ * Driver must call balloon_page_enqueue before definitively removing the page
	120	+ * from the guest system.
116	121	*
117		− * Driver must call it to properly allocate a new enlisted balloon page.
118		− * Driver must call balloon_page_enqueue before definitively removing it from
119		− * the guest system. This function returns the page address for the recently
120		− * allocated page or NULL in the case we fail to allocate a new page this turn.
	122	+ * Return: struct page for the allocated page or NULL on allocation failure.
121	123	*/
122	124	struct page *balloon_page_alloc(void)
123	125	{
@@ -128,15 +130,17 @@ struct page *balloon_page_alloc(void) · 4 unmodified lines
128	130	EXPORT_SYMBOL_GPL(balloon_page_alloc);
129	131
130	132	/*
131		− * balloon_page_enqueue - allocates a new page and inserts it into the balloon
132		− * page list.
133		− * @b_dev_info: balloon device descriptor where we will insert a new page to
	133	+ * balloon_page_enqueue - inserts a new page into the balloon page list.
	134	+ *
	135	+ * @b_dev_info: balloon device descriptor where we will insert a new page
134	136	* @page: new page to enqueue - allocated using balloon_page_alloc.
135	137	*
136		− * Driver must call it to properly enqueue a new allocated balloon page
137		− * before definitively removing it from the guest system.
138		− * This function returns the page address for the recently enqueued page or
139		− * NULL in the case we fail to allocate a new page this turn.
	138	+ * Drivers must call this function to properly enqueue a new allocated balloon
	139	+ * page before definitively removing the page from the guest system.
	140	+ *
	141	+ * Drivers must not call balloon_page_enqueue on pages that have been pushed to
	142	+ * a list with balloon_page_push before removing them with balloon_page_pop. To
	143	+ * enqueue a list of pages, use balloon_page_list_enqueue instead.
140	144	*/
141	145	void balloon_page_enqueue(struct balloon_dev_info *b_dev_info,
142	146	struct page *page)
@@ -151,14 +155,23 @@ EXPORT_SYMBOL_GPL(balloon_page_enqueue); · 8 unmodified lines
151	155
152	156	/*
153	157	* balloon_page_dequeue - removes a page from balloon's page list and returns
154		− * the its address to allow the driver release the page.
	158	+ * its address to allow the driver to release the page.
155	159	* @b_dev_info: balloon device decriptor where we will grab a page from.
156	160	*
157		− * Driver must call it to properly de-allocate a previous enlisted balloon page
158		− * before definetively releasing it back to the guest system.
159		− * This function returns the page address for the recently dequeued page or
160		− * NULL in the case we find balloon's page list temporarily empty due to
161		− * compaction isolated pages.
	161	+ * Driver must call this function to properly dequeue a previously enqueued page
	162	+ * before definitively releasing it back to the guest system.
	163	+ *
	164	+ * Caller must perform its own accounting to ensure that this
	165	+ * function is called only if some pages are actually enqueued.
	166	+ *
	167	+ * Note that this function may fail to dequeue some pages even if there are
	168	+ * some enqueued pages - since the page list can be temporarily empty due to
	169	+ * the compaction of isolated pages.
	170	+ *
	171	+ * TODO: remove the caller accounting requirements, and allow caller to wait
	172	+ * until all pages can be dequeued.
	173	+ *
	174	+ * Return: struct page for the dequeued page, or NULL if no page was dequeued.
162	175	*/
163	176	struct page balloon_page_dequeue(struct balloon_dev_info b_dev_info)
164	177	{
@@ -171,9 +184,9 @@ struct page balloon_page_dequeue(struct balloon_dev_info b_dev_info) · 6 unmodified lines
171	184	if (n_pages != 1) {
172	185	/*
173	186	* If we are unable to dequeue a balloon page because the page
174		− * list is empty and there is no isolated pages, then something
	187	+ * list is empty and there are no isolated pages, then something
175	188	* went out of track and some balloon pages are lost.
176		− * BUG() here, otherwise the balloon driver may get stuck into
	189	+ * BUG() here, otherwise the balloon driver may get stuck in
177	190	* an infinite loop while attempting to release all its pages.
178	191	*/
179	192	spin_lock_irqsave(&b_dev_info->pages_lock, flags);
@@ -224,8 +237,8 @@ int balloon_page_migrate(struct address_space *mapping, · 44 unmodified lines
224	237
225	238	/*
226	239	* We can not easily support the no copy case here so ignore it as it
227		− * is unlikely to be use with ballon pages. See include/linux/hmm.h for
228		− * user of the MIGRATE_SYNC_NO_COPY mode.
	240	+ * is unlikely to be used with balloon pages. See include/linux/hmm.h
	241	+ * for a user of the MIGRATE_SYNC_NO_COPY mode.
229	242	*/
230	243	if (mode == MIGRATE_SYNC_NO_COPY)
231	244	return -EINVAL;