1*4882a593Smuzhiyun #ifndef _UAPI_LINUX_VIRTIO_RING_H
2*4882a593Smuzhiyun #define _UAPI_LINUX_VIRTIO_RING_H
3*4882a593Smuzhiyun /* An interface for efficient virtio implementation, currently for use by KVM,
4*4882a593Smuzhiyun * but hopefully others soon. Do NOT change this since it will
5*4882a593Smuzhiyun * break existing servers and clients.
6*4882a593Smuzhiyun *
7*4882a593Smuzhiyun * This header is BSD licensed so anyone can use the definitions to implement
8*4882a593Smuzhiyun * compatible drivers/servers.
9*4882a593Smuzhiyun *
10*4882a593Smuzhiyun * Redistribution and use in source and binary forms, with or without
11*4882a593Smuzhiyun * modification, are permitted provided that the following conditions
12*4882a593Smuzhiyun * are met:
13*4882a593Smuzhiyun * 1. Redistributions of source code must retain the above copyright
14*4882a593Smuzhiyun * notice, this list of conditions and the following disclaimer.
15*4882a593Smuzhiyun * 2. Redistributions in binary form must reproduce the above copyright
16*4882a593Smuzhiyun * notice, this list of conditions and the following disclaimer in the
17*4882a593Smuzhiyun * documentation and/or other materials provided with the distribution.
18*4882a593Smuzhiyun * 3. Neither the name of IBM nor the names of its contributors
19*4882a593Smuzhiyun * may be used to endorse or promote products derived from this software
20*4882a593Smuzhiyun * without specific prior written permission.
21*4882a593Smuzhiyun * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS ``AS IS'' AND
22*4882a593Smuzhiyun * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
23*4882a593Smuzhiyun * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
24*4882a593Smuzhiyun * ARE DISCLAIMED. IN NO EVENT SHALL IBM OR CONTRIBUTORS BE LIABLE
25*4882a593Smuzhiyun * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
26*4882a593Smuzhiyun * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
27*4882a593Smuzhiyun * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
28*4882a593Smuzhiyun * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
29*4882a593Smuzhiyun * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
30*4882a593Smuzhiyun * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
31*4882a593Smuzhiyun * SUCH DAMAGE.
32*4882a593Smuzhiyun *
33*4882a593Smuzhiyun * Copyright Rusty Russell IBM Corporation 2007. */
34*4882a593Smuzhiyun #ifndef __KERNEL__
35*4882a593Smuzhiyun #include <stdint.h>
36*4882a593Smuzhiyun #endif
37*4882a593Smuzhiyun #include <linux/types.h>
38*4882a593Smuzhiyun #include <linux/virtio_types.h>
39*4882a593Smuzhiyun
40*4882a593Smuzhiyun /* This marks a buffer as continuing via the next field. */
41*4882a593Smuzhiyun #define VRING_DESC_F_NEXT 1
42*4882a593Smuzhiyun /* This marks a buffer as write-only (otherwise read-only). */
43*4882a593Smuzhiyun #define VRING_DESC_F_WRITE 2
44*4882a593Smuzhiyun /* This means the buffer contains a list of buffer descriptors. */
45*4882a593Smuzhiyun #define VRING_DESC_F_INDIRECT 4
46*4882a593Smuzhiyun
47*4882a593Smuzhiyun /*
48*4882a593Smuzhiyun * Mark a descriptor as available or used in packed ring.
49*4882a593Smuzhiyun * Notice: they are defined as shifts instead of shifted values.
50*4882a593Smuzhiyun */
51*4882a593Smuzhiyun #define VRING_PACKED_DESC_F_AVAIL 7
52*4882a593Smuzhiyun #define VRING_PACKED_DESC_F_USED 15
53*4882a593Smuzhiyun
54*4882a593Smuzhiyun /* The Host uses this in used->flags to advise the Guest: don't kick me when
55*4882a593Smuzhiyun * you add a buffer. It's unreliable, so it's simply an optimization. Guest
56*4882a593Smuzhiyun * will still kick if it's out of buffers. */
57*4882a593Smuzhiyun #define VRING_USED_F_NO_NOTIFY 1
58*4882a593Smuzhiyun /* The Guest uses this in avail->flags to advise the Host: don't interrupt me
59*4882a593Smuzhiyun * when you consume a buffer. It's unreliable, so it's simply an
60*4882a593Smuzhiyun * optimization. */
61*4882a593Smuzhiyun #define VRING_AVAIL_F_NO_INTERRUPT 1
62*4882a593Smuzhiyun
63*4882a593Smuzhiyun /* Enable events in packed ring. */
64*4882a593Smuzhiyun #define VRING_PACKED_EVENT_FLAG_ENABLE 0x0
65*4882a593Smuzhiyun /* Disable events in packed ring. */
66*4882a593Smuzhiyun #define VRING_PACKED_EVENT_FLAG_DISABLE 0x1
67*4882a593Smuzhiyun /*
68*4882a593Smuzhiyun * Enable events for a specific descriptor in packed ring.
69*4882a593Smuzhiyun * (as specified by Descriptor Ring Change Event Offset/Wrap Counter).
70*4882a593Smuzhiyun * Only valid if VIRTIO_RING_F_EVENT_IDX has been negotiated.
71*4882a593Smuzhiyun */
72*4882a593Smuzhiyun #define VRING_PACKED_EVENT_FLAG_DESC 0x2
73*4882a593Smuzhiyun
74*4882a593Smuzhiyun /*
75*4882a593Smuzhiyun * Wrap counter bit shift in event suppression structure
76*4882a593Smuzhiyun * of packed ring.
77*4882a593Smuzhiyun */
78*4882a593Smuzhiyun #define VRING_PACKED_EVENT_F_WRAP_CTR 15
79*4882a593Smuzhiyun
80*4882a593Smuzhiyun /* We support indirect buffer descriptors */
81*4882a593Smuzhiyun #define VIRTIO_RING_F_INDIRECT_DESC 28
82*4882a593Smuzhiyun
83*4882a593Smuzhiyun /* The Guest publishes the used index for which it expects an interrupt
84*4882a593Smuzhiyun * at the end of the avail ring. Host should ignore the avail->flags field. */
85*4882a593Smuzhiyun /* The Host publishes the avail index for which it expects a kick
86*4882a593Smuzhiyun * at the end of the used ring. Guest should ignore the used->flags field. */
87*4882a593Smuzhiyun #define VIRTIO_RING_F_EVENT_IDX 29
88*4882a593Smuzhiyun
89*4882a593Smuzhiyun /* Alignment requirements for vring elements.
90*4882a593Smuzhiyun * When using pre-virtio 1.0 layout, these fall out naturally.
91*4882a593Smuzhiyun */
92*4882a593Smuzhiyun #define VRING_AVAIL_ALIGN_SIZE 2
93*4882a593Smuzhiyun #define VRING_USED_ALIGN_SIZE 4
94*4882a593Smuzhiyun #define VRING_DESC_ALIGN_SIZE 16
95*4882a593Smuzhiyun
96*4882a593Smuzhiyun /* Virtio ring descriptors: 16 bytes. These can chain together via "next". */
97*4882a593Smuzhiyun struct vring_desc {
98*4882a593Smuzhiyun /* Address (guest-physical). */
99*4882a593Smuzhiyun __virtio64 addr;
100*4882a593Smuzhiyun /* Length. */
101*4882a593Smuzhiyun __virtio32 len;
102*4882a593Smuzhiyun /* The flags as indicated above. */
103*4882a593Smuzhiyun __virtio16 flags;
104*4882a593Smuzhiyun /* We chain unused descriptors via this, too */
105*4882a593Smuzhiyun __virtio16 next;
106*4882a593Smuzhiyun };
107*4882a593Smuzhiyun
108*4882a593Smuzhiyun struct vring_avail {
109*4882a593Smuzhiyun __virtio16 flags;
110*4882a593Smuzhiyun __virtio16 idx;
111*4882a593Smuzhiyun __virtio16 ring[];
112*4882a593Smuzhiyun };
113*4882a593Smuzhiyun
114*4882a593Smuzhiyun /* u32 is used here for ids for padding reasons. */
115*4882a593Smuzhiyun struct vring_used_elem {
116*4882a593Smuzhiyun /* Index of start of used descriptor chain. */
117*4882a593Smuzhiyun __virtio32 id;
118*4882a593Smuzhiyun /* Total length of the descriptor chain which was used (written to) */
119*4882a593Smuzhiyun __virtio32 len;
120*4882a593Smuzhiyun };
121*4882a593Smuzhiyun
122*4882a593Smuzhiyun typedef struct vring_used_elem __attribute__((aligned(VRING_USED_ALIGN_SIZE)))
123*4882a593Smuzhiyun vring_used_elem_t;
124*4882a593Smuzhiyun
125*4882a593Smuzhiyun struct vring_used {
126*4882a593Smuzhiyun __virtio16 flags;
127*4882a593Smuzhiyun __virtio16 idx;
128*4882a593Smuzhiyun vring_used_elem_t ring[];
129*4882a593Smuzhiyun };
130*4882a593Smuzhiyun
131*4882a593Smuzhiyun /*
132*4882a593Smuzhiyun * The ring element addresses are passed between components with different
133*4882a593Smuzhiyun * alignments assumptions. Thus, we might need to decrease the compiler-selected
134*4882a593Smuzhiyun * alignment, and so must use a typedef to make sure the aligned attribute
135*4882a593Smuzhiyun * actually takes hold:
136*4882a593Smuzhiyun *
137*4882a593Smuzhiyun * https://gcc.gnu.org/onlinedocs//gcc/Common-Type-Attributes.html#Common-Type-Attributes
138*4882a593Smuzhiyun *
139*4882a593Smuzhiyun * When used on a struct, or struct member, the aligned attribute can only
140*4882a593Smuzhiyun * increase the alignment; in order to decrease it, the packed attribute must
141*4882a593Smuzhiyun * be specified as well. When used as part of a typedef, the aligned attribute
142*4882a593Smuzhiyun * can both increase and decrease alignment, and specifying the packed
143*4882a593Smuzhiyun * attribute generates a warning.
144*4882a593Smuzhiyun */
145*4882a593Smuzhiyun typedef struct vring_desc __attribute__((aligned(VRING_DESC_ALIGN_SIZE)))
146*4882a593Smuzhiyun vring_desc_t;
147*4882a593Smuzhiyun typedef struct vring_avail __attribute__((aligned(VRING_AVAIL_ALIGN_SIZE)))
148*4882a593Smuzhiyun vring_avail_t;
149*4882a593Smuzhiyun typedef struct vring_used __attribute__((aligned(VRING_USED_ALIGN_SIZE)))
150*4882a593Smuzhiyun vring_used_t;
151*4882a593Smuzhiyun
152*4882a593Smuzhiyun struct vring {
153*4882a593Smuzhiyun unsigned int num;
154*4882a593Smuzhiyun
155*4882a593Smuzhiyun vring_desc_t *desc;
156*4882a593Smuzhiyun
157*4882a593Smuzhiyun vring_avail_t *avail;
158*4882a593Smuzhiyun
159*4882a593Smuzhiyun vring_used_t *used;
160*4882a593Smuzhiyun };
161*4882a593Smuzhiyun
162*4882a593Smuzhiyun #ifndef VIRTIO_RING_NO_LEGACY
163*4882a593Smuzhiyun
164*4882a593Smuzhiyun /* The standard layout for the ring is a continuous chunk of memory which looks
165*4882a593Smuzhiyun * like this. We assume num is a power of 2.
166*4882a593Smuzhiyun *
167*4882a593Smuzhiyun * struct vring
168*4882a593Smuzhiyun * {
169*4882a593Smuzhiyun * // The actual descriptors (16 bytes each)
170*4882a593Smuzhiyun * struct vring_desc desc[num];
171*4882a593Smuzhiyun *
172*4882a593Smuzhiyun * // A ring of available descriptor heads with free-running index.
173*4882a593Smuzhiyun * __virtio16 avail_flags;
174*4882a593Smuzhiyun * __virtio16 avail_idx;
175*4882a593Smuzhiyun * __virtio16 available[num];
176*4882a593Smuzhiyun * __virtio16 used_event_idx;
177*4882a593Smuzhiyun *
178*4882a593Smuzhiyun * // Padding to the next align boundary.
179*4882a593Smuzhiyun * char pad[];
180*4882a593Smuzhiyun *
181*4882a593Smuzhiyun * // A ring of used descriptor heads with free-running index.
182*4882a593Smuzhiyun * __virtio16 used_flags;
183*4882a593Smuzhiyun * __virtio16 used_idx;
184*4882a593Smuzhiyun * struct vring_used_elem used[num];
185*4882a593Smuzhiyun * __virtio16 avail_event_idx;
186*4882a593Smuzhiyun * };
187*4882a593Smuzhiyun */
188*4882a593Smuzhiyun /* We publish the used event index at the end of the available ring, and vice
189*4882a593Smuzhiyun * versa. They are at the end for backwards compatibility. */
190*4882a593Smuzhiyun #define vring_used_event(vr) ((vr)->avail->ring[(vr)->num])
191*4882a593Smuzhiyun #define vring_avail_event(vr) (*(__virtio16 *)&(vr)->used->ring[(vr)->num])
192*4882a593Smuzhiyun
vring_init(struct vring * vr,unsigned int num,void * p,unsigned long align)193*4882a593Smuzhiyun static inline void vring_init(struct vring *vr, unsigned int num, void *p,
194*4882a593Smuzhiyun unsigned long align)
195*4882a593Smuzhiyun {
196*4882a593Smuzhiyun vr->num = num;
197*4882a593Smuzhiyun vr->desc = p;
198*4882a593Smuzhiyun vr->avail = (struct vring_avail *)((char *)p + num * sizeof(struct vring_desc));
199*4882a593Smuzhiyun vr->used = (void *)(((uintptr_t)&vr->avail->ring[num] + sizeof(__virtio16)
200*4882a593Smuzhiyun + align-1) & ~(align - 1));
201*4882a593Smuzhiyun }
202*4882a593Smuzhiyun
vring_size(unsigned int num,unsigned long align)203*4882a593Smuzhiyun static inline unsigned vring_size(unsigned int num, unsigned long align)
204*4882a593Smuzhiyun {
205*4882a593Smuzhiyun return ((sizeof(struct vring_desc) * num + sizeof(__virtio16) * (3 + num)
206*4882a593Smuzhiyun + align - 1) & ~(align - 1))
207*4882a593Smuzhiyun + sizeof(__virtio16) * 3 + sizeof(struct vring_used_elem) * num;
208*4882a593Smuzhiyun }
209*4882a593Smuzhiyun
210*4882a593Smuzhiyun #endif /* VIRTIO_RING_NO_LEGACY */
211*4882a593Smuzhiyun
212*4882a593Smuzhiyun /* The following is used with USED_EVENT_IDX and AVAIL_EVENT_IDX */
213*4882a593Smuzhiyun /* Assuming a given event_idx value from the other side, if
214*4882a593Smuzhiyun * we have just incremented index from old to new_idx,
215*4882a593Smuzhiyun * should we trigger an event? */
vring_need_event(__u16 event_idx,__u16 new_idx,__u16 old)216*4882a593Smuzhiyun static inline int vring_need_event(__u16 event_idx, __u16 new_idx, __u16 old)
217*4882a593Smuzhiyun {
218*4882a593Smuzhiyun /* Note: Xen has similar logic for notification hold-off
219*4882a593Smuzhiyun * in include/xen/interface/io/ring.h with req_event and req_prod
220*4882a593Smuzhiyun * corresponding to event_idx + 1 and new_idx respectively.
221*4882a593Smuzhiyun * Note also that req_event and req_prod in Xen start at 1,
222*4882a593Smuzhiyun * event indexes in virtio start at 0. */
223*4882a593Smuzhiyun return (__u16)(new_idx - event_idx - 1) < (__u16)(new_idx - old);
224*4882a593Smuzhiyun }
225*4882a593Smuzhiyun
226*4882a593Smuzhiyun struct vring_packed_desc_event {
227*4882a593Smuzhiyun /* Descriptor Ring Change Event Offset/Wrap Counter. */
228*4882a593Smuzhiyun __le16 off_wrap;
229*4882a593Smuzhiyun /* Descriptor Ring Change Event Flags. */
230*4882a593Smuzhiyun __le16 flags;
231*4882a593Smuzhiyun };
232*4882a593Smuzhiyun
233*4882a593Smuzhiyun struct vring_packed_desc {
234*4882a593Smuzhiyun /* Buffer Address. */
235*4882a593Smuzhiyun __le64 addr;
236*4882a593Smuzhiyun /* Buffer Length. */
237*4882a593Smuzhiyun __le32 len;
238*4882a593Smuzhiyun /* Buffer ID. */
239*4882a593Smuzhiyun __le16 id;
240*4882a593Smuzhiyun /* The flags depending on descriptor type. */
241*4882a593Smuzhiyun __le16 flags;
242*4882a593Smuzhiyun };
243*4882a593Smuzhiyun
244*4882a593Smuzhiyun #endif /* _UAPI_LINUX_VIRTIO_RING_H */
245