Deleted Added
sdiff udiff text old ( 341477 ) new ( 342033 )
full compact
netmap_virt.h (341477) netmap_virt.h (342033)
1/*
2 * Copyright (C) 2013-2016 Luigi Rizzo
3 * Copyright (C) 2013-2016 Giuseppe Lettieri
1/*
2 * Copyright (C) 2013-2016 Luigi Rizzo
3 * Copyright (C) 2013-2016 Giuseppe Lettieri
4 * Copyright (C) 2013-2016 Vincenzo Maffione
4 * Copyright (C) 2013-2018 Vincenzo Maffione
5 * Copyright (C) 2015 Stefano Garzarella
6 * All rights reserved.
7 *
8 * Redistribution and use in source and binary forms, with or without
9 * modification, are permitted provided that the following conditions
10 * are met:
11 * 1. Redistributions of source code must retain the above copyright
12 * notice, this list of conditions and the following disclaimer.

--- 8 unchanged lines hidden (view full) ---

21 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
22 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
23 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
24 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
25 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
26 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
27 * SUCH DAMAGE.
28 *
5 * Copyright (C) 2015 Stefano Garzarella
6 * All rights reserved.
7 *
8 * Redistribution and use in source and binary forms, with or without
9 * modification, are permitted provided that the following conditions
10 * are met:
11 * 1. Redistributions of source code must retain the above copyright
12 * notice, this list of conditions and the following disclaimer.

--- 8 unchanged lines hidden (view full) ---

21 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
22 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
23 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
24 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
25 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
26 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
27 * SUCH DAMAGE.
28 *
29 * $FreeBSD: stable/11/sys/net/netmap_virt.h 341477 2018-12-04 17:40:56Z vmaffione $
29 * $FreeBSD: stable/11/sys/net/netmap_virt.h 342033 2018-12-13 10:13:29Z vmaffione $
30 */
31
32#ifndef NETMAP_VIRT_H
33#define NETMAP_VIRT_H
34
35/*
30 */
31
32#ifndef NETMAP_VIRT_H
33#define NETMAP_VIRT_H
34
35/*
36 * ptnetmap_memdev: device used to expose memory into the guest VM
36 * Register offsets and other macros for the ptnetmap paravirtual devices:
37 * ptnetmap-memdev: device used to expose memory into the guest
38 * ptnet: paravirtualized NIC exposing a netmap port in the guest
37 *
38 * These macros are used in the hypervisor frontend (QEMU, bhyve) and in the
39 * guest device driver.
40 */
41
39 *
40 * These macros are used in the hypervisor frontend (QEMU, bhyve) and in the
41 * guest device driver.
42 */
43
42/* PCI identifiers and PCI BARs for the ptnetmap memdev
43 * and ptnetmap network interface. */
44/* PCI identifiers and PCI BARs for ptnetmap-memdev and ptnet. */
44#define PTNETMAP_MEMDEV_NAME "ptnetmap-memdev"
45#define PTNETMAP_PCI_VENDOR_ID 0x1b36 /* QEMU virtual devices */
46#define PTNETMAP_PCI_DEVICE_ID 0x000c /* memory device */
47#define PTNETMAP_PCI_NETIF_ID 0x000d /* ptnet network interface */
48#define PTNETMAP_IO_PCI_BAR 0
49#define PTNETMAP_MEM_PCI_BAR 1
50#define PTNETMAP_MSIX_PCI_BAR 2
51
45#define PTNETMAP_MEMDEV_NAME "ptnetmap-memdev"
46#define PTNETMAP_PCI_VENDOR_ID 0x1b36 /* QEMU virtual devices */
47#define PTNETMAP_PCI_DEVICE_ID 0x000c /* memory device */
48#define PTNETMAP_PCI_NETIF_ID 0x000d /* ptnet network interface */
49#define PTNETMAP_IO_PCI_BAR 0
50#define PTNETMAP_MEM_PCI_BAR 1
51#define PTNETMAP_MSIX_PCI_BAR 2
52
52/* Registers for the ptnetmap memdev */
53/* Device registers for ptnetmap-memdev */
53#define PTNET_MDEV_IO_MEMSIZE_LO 0 /* netmap memory size (low) */
54#define PTNET_MDEV_IO_MEMSIZE_HI 4 /* netmap_memory_size (high) */
55#define PTNET_MDEV_IO_MEMID 8 /* memory allocator ID in the host */
56#define PTNET_MDEV_IO_IF_POOL_OFS 64
57#define PTNET_MDEV_IO_IF_POOL_OBJNUM 68
58#define PTNET_MDEV_IO_IF_POOL_OBJSZ 72
59#define PTNET_MDEV_IO_RING_POOL_OFS 76
60#define PTNET_MDEV_IO_RING_POOL_OBJNUM 80
61#define PTNET_MDEV_IO_RING_POOL_OBJSZ 84
62#define PTNET_MDEV_IO_BUF_POOL_OFS 88
63#define PTNET_MDEV_IO_BUF_POOL_OBJNUM 92
64#define PTNET_MDEV_IO_BUF_POOL_OBJSZ 96
65#define PTNET_MDEV_IO_END 100
66
54#define PTNET_MDEV_IO_MEMSIZE_LO 0 /* netmap memory size (low) */
55#define PTNET_MDEV_IO_MEMSIZE_HI 4 /* netmap_memory_size (high) */
56#define PTNET_MDEV_IO_MEMID 8 /* memory allocator ID in the host */
57#define PTNET_MDEV_IO_IF_POOL_OFS 64
58#define PTNET_MDEV_IO_IF_POOL_OBJNUM 68
59#define PTNET_MDEV_IO_IF_POOL_OBJSZ 72
60#define PTNET_MDEV_IO_RING_POOL_OFS 76
61#define PTNET_MDEV_IO_RING_POOL_OBJNUM 80
62#define PTNET_MDEV_IO_RING_POOL_OBJSZ 84
63#define PTNET_MDEV_IO_BUF_POOL_OFS 88
64#define PTNET_MDEV_IO_BUF_POOL_OBJNUM 92
65#define PTNET_MDEV_IO_BUF_POOL_OBJSZ 96
66#define PTNET_MDEV_IO_END 100
67
67/*
68 * ptnetmap configuration
69 *
70 * The ptnet kthreads (running in host kernel-space) need to be configured
71 * in order to know how to intercept guest kicks (I/O register writes) and
72 * how to inject MSI-X interrupts to the guest. The configuration may vary
73 * depending on the hypervisor. Currently, we support QEMU/KVM on Linux and
74 * and bhyve on FreeBSD.
75 * The configuration is passed by the hypervisor to the host netmap module
76 * by means of an ioctl() with nr_cmd=NETMAP_PT_HOST_CREATE, and it is
77 * specified by the ptnetmap_cfg struct. This struct contains an header
78 * with general informations and an array of entries whose size depends
79 * on the hypervisor. The NETMAP_PT_HOST_CREATE command is issued every
80 * time the kthreads are started.
81 */
82struct ptnetmap_cfg {
83#define PTNETMAP_CFGTYPE_QEMU 0x1
84#define PTNETMAP_CFGTYPE_BHYVE 0x2
85 uint16_t cfgtype; /* how to interpret the cfg entries */
86 uint16_t entry_size; /* size of a config entry */
87 uint32_t num_rings; /* number of config entries */
88 void *csb_gh; /* CSB for guest --> host communication */
89 void *csb_hg; /* CSB for host --> guest communication */
90 /* Configuration entries are allocated right after the struct. */
91};
92
93/* Configuration of a ptnetmap ring for QEMU. */
94struct ptnetmap_cfgentry_qemu {
95 uint32_t ioeventfd; /* to intercept guest register access */
96 uint32_t irqfd; /* to inject guest interrupts */
97};
98
99/* Configuration of a ptnetmap ring for bhyve. */
100struct ptnetmap_cfgentry_bhyve {
101 uint64_t wchan; /* tsleep() parameter, to wake up kthread */
102 uint32_t ioctl_fd; /* ioctl fd */
103 /* ioctl parameters to send irq */
104 uint32_t ioctl_cmd;
105 /* vmm.ko MSIX parameters for IOCTL */
106 struct {
107 uint64_t msg_data;
108 uint64_t addr;
109 } ioctl_data;
110};
111
112/*
113 * Pass a pointer to a userspace buffer to be passed to kernelspace for write
114 * or read. Used by NETMAP_PT_HOST_CREATE.
115 * XXX deprecated
116 */
117static inline void
118nmreq_pointer_put(struct nmreq *nmr, void *userptr)
119{
120 uintptr_t *pp = (uintptr_t *)&nmr->nr_arg1;
121 *pp = (uintptr_t)userptr;
122}
123
124static inline void *
125nmreq_pointer_get(const struct nmreq *nmr)
126{
127 const uintptr_t *pp = (const uintptr_t *)&nmr->nr_arg1;
128 return (void *)*pp;
129}
130
131/* ptnetmap features */
132#define PTNETMAP_F_VNET_HDR 1
133
68/* ptnetmap features */
69#define PTNETMAP_F_VNET_HDR 1
70
134/* I/O registers for the ptnet device. */
71/* Device registers for the ptnet network device. */
135#define PTNET_IO_PTFEAT 0
136#define PTNET_IO_PTCTL 4
137#define PTNET_IO_MAC_LO 8
138#define PTNET_IO_MAC_HI 12
139#define PTNET_IO_CSBBAH 16 /* deprecated */
140#define PTNET_IO_CSBBAL 20 /* deprecated */
141#define PTNET_IO_NIFP_OFS 24
142#define PTNET_IO_NUM_TX_RINGS 28

--- 5 unchanged lines hidden (view full) ---

148#define PTNET_IO_CSB_GH_BAH 52
149#define PTNET_IO_CSB_GH_BAL 56
150#define PTNET_IO_CSB_HG_BAH 60
151#define PTNET_IO_CSB_HG_BAL 64
152#define PTNET_IO_END 68
153#define PTNET_IO_KICK_BASE 128
154#define PTNET_IO_MASK 0xff
155
72#define PTNET_IO_PTFEAT 0
73#define PTNET_IO_PTCTL 4
74#define PTNET_IO_MAC_LO 8
75#define PTNET_IO_MAC_HI 12
76#define PTNET_IO_CSBBAH 16 /* deprecated */
77#define PTNET_IO_CSBBAL 20 /* deprecated */
78#define PTNET_IO_NIFP_OFS 24
79#define PTNET_IO_NUM_TX_RINGS 28

--- 5 unchanged lines hidden (view full) ---

85#define PTNET_IO_CSB_GH_BAH 52
86#define PTNET_IO_CSB_GH_BAL 56
87#define PTNET_IO_CSB_HG_BAH 60
88#define PTNET_IO_CSB_HG_BAL 64
89#define PTNET_IO_END 68
90#define PTNET_IO_KICK_BASE 128
91#define PTNET_IO_MASK 0xff
92
156/* ptnetmap control commands (values for PTCTL register) */
93/* ptnet control commands (values for PTCTL register):
94 * - CREATE starts the host sync-kloop
95 * - DELETE stops the host sync-kloop
96 */
157#define PTNETMAP_PTCTL_CREATE 1
158#define PTNETMAP_PTCTL_DELETE 2
159
97#define PTNETMAP_PTCTL_CREATE 1
98#define PTNETMAP_PTCTL_DELETE 2
99
160/* ptnetmap synchronization variables shared between guest and host */
161struct ptnet_csb_gh {
162 uint32_t head; /* GW+ HR+ the head of the guest netmap_ring */
163 uint32_t cur; /* GW+ HR+ the cur of the guest netmap_ring */
164 uint32_t guest_need_kick; /* GW+ HR+ host-->guest notification enable */
165 uint32_t sync_flags; /* GW+ HR+ the flags of the guest [tx|rx]sync() */
166 char pad[48]; /* pad to a 64 bytes cacheline */
167};
168struct ptnet_csb_hg {
169 uint32_t hwcur; /* GR+ HW+ the hwcur of the host netmap_kring */
170 uint32_t hwtail; /* GR+ HW+ the hwtail of the host netmap_kring */
171 uint32_t host_need_kick; /* GR+ HW+ guest-->host notification enable */
172 char pad[4+48];
173};
174
175#ifdef WITH_PTNETMAP_GUEST
176
177/* ptnetmap_memdev routines used to talk with ptnetmap_memdev device driver */
178struct ptnetmap_memdev;
179int nm_os_pt_memdev_iomap(struct ptnetmap_memdev *, vm_paddr_t *, void **,
180 uint64_t *);
181void nm_os_pt_memdev_iounmap(struct ptnetmap_memdev *);
182uint32_t nm_os_pt_memdev_ioread(struct ptnetmap_memdev *, unsigned int);
183
184/* Guest driver: Write kring pointers (cur, head) to the CSB.
185 * This routine is coupled with ptnetmap_host_read_kring_csb(). */
186static inline void
187ptnetmap_guest_write_kring_csb(struct ptnet_csb_gh *ptr, uint32_t cur,
188 uint32_t head)
189{
190 /*
191 * We need to write cur and head to the CSB but we cannot do it atomically.
192 * There is no way we can prevent the host from reading the updated value
193 * of one of the two and the old value of the other. However, if we make
194 * sure that the host never reads a value of head more recent than the
195 * value of cur we are safe. We can allow the host to read a value of cur
196 * more recent than the value of head, since in the netmap ring cur can be
197 * ahead of head and cur cannot wrap around head because it must be behind
198 * tail. Inverting the order of writes below could instead result into the
199 * host to think head went ahead of cur, which would cause the sync
200 * prologue to fail.
201 *
202 * The following memory barrier scheme is used to make this happen:
203 *
204 * Guest Host
205 *
206 * STORE(cur) LOAD(head)
207 * mb() <-----------> mb()
208 * STORE(head) LOAD(cur)
209 */
210 ptr->cur = cur;
211 mb();
212 ptr->head = head;
213}
214
215/* Guest driver: Read kring pointers (hwcur, hwtail) from the CSB.
216 * This routine is coupled with ptnetmap_host_write_kring_csb(). */
217static inline void
218ptnetmap_guest_read_kring_csb(struct ptnet_csb_hg *pthg, struct netmap_kring *kring)
219{
220 /*
221 * We place a memory barrier to make sure that the update of hwtail never
222 * overtakes the update of hwcur.
223 * (see explanation in ptnetmap_host_write_kring_csb).
224 */
225 kring->nr_hwtail = pthg->hwtail;
226 mb();
227 kring->nr_hwcur = pthg->hwcur;
228}
229
230#endif /* WITH_PTNETMAP_GUEST */
231
232#ifdef WITH_PTNETMAP_HOST
233/*
234 * ptnetmap kernel thread routines
235 * */
236
237/* Functions to read and write CSB fields in the host */
238#if defined (linux)
239#define CSB_READ(csb, field, r) (get_user(r, &csb->field))
240#define CSB_WRITE(csb, field, v) (put_user(v, &csb->field))
241#else /* ! linux */
242#define CSB_READ(csb, field, r) (r = fuword32(&csb->field))
243#define CSB_WRITE(csb, field, v) (suword32(&csb->field, v))
244#endif /* ! linux */
245
246/* Host netmap: Write kring pointers (hwcur, hwtail) to the CSB.
247 * This routine is coupled with ptnetmap_guest_read_kring_csb(). */
248static inline void
249ptnetmap_host_write_kring_csb(struct ptnet_csb_hg __user *ptr, uint32_t hwcur,
250 uint32_t hwtail)
251{
252 /*
253 * The same scheme used in ptnetmap_guest_write_kring_csb() applies here.
254 * We allow the guest to read a value of hwcur more recent than the value
255 * of hwtail, since this would anyway result in a consistent view of the
256 * ring state (and hwcur can never wraparound hwtail, since hwcur must be
257 * behind head).
258 *
259 * The following memory barrier scheme is used to make this happen:
260 *
261 * Guest Host
262 *
263 * STORE(hwcur) LOAD(hwtail)
264 * mb() <-------------> mb()
265 * STORE(hwtail) LOAD(hwcur)
266 */
267 CSB_WRITE(ptr, hwcur, hwcur);
268 mb();
269 CSB_WRITE(ptr, hwtail, hwtail);
270}
271
272/* Host netmap: Read kring pointers (head, cur, sync_flags) from the CSB.
273 * This routine is coupled with ptnetmap_guest_write_kring_csb(). */
274static inline void
275ptnetmap_host_read_kring_csb(struct ptnet_csb_gh __user *ptr,
276 struct netmap_ring *shadow_ring,
277 uint32_t num_slots)
278{
279 /*
280 * We place a memory barrier to make sure that the update of head never
281 * overtakes the update of cur.
282 * (see explanation in ptnetmap_guest_write_kring_csb).
283 */
284 CSB_READ(ptr, head, shadow_ring->head);
285 mb();
286 CSB_READ(ptr, cur, shadow_ring->cur);
287 CSB_READ(ptr, sync_flags, shadow_ring->flags);
288}
289
290#endif /* WITH_PTNETMAP_HOST */
291
292#endif /* NETMAP_VIRT_H */
100#endif /* NETMAP_VIRT_H */