1/*
2 * Copyright (C) 2013 Luigi Rizzo. All rights reserved.
3 *
4 * Redistribution and use in source and binary forms, with or without
5 * modification, are permitted provided that the following conditions
6 * are met:
7 *   1. Redistributions of source code must retain the above copyright
8 *      notice, this list of conditions and the following disclaimer.
9 *   2. Redistributions in binary form must reproduce the above copyright
10 *      notice, this list of conditions and the following disclaimer in the
11 *    documentation and/or other materials provided with the distribution.
12 *
13 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
14 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
15 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
16 * ARE DISCLAIMED.  IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
17 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
18 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
19 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
20 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
21 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
22 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
23 * SUCH DAMAGE.
24 */
25
26#ifndef NET_PARAVIRT_H
27#define NET_PARAVIRT_H
28
29 /*
30  * $FreeBSD$
31  *
32 Support for virtio-like communication between host (H) and guest (G) NICs.
33
34 THIS IS EXPERIMENTAL CODE AND SUBJECT TO CHANGE.
35
36 The guest allocates the shared Communication Status Block (csb) and
37 write its physical address at CSBAL and CSBAH (data is little endian).
38 csb->csb_on enables the mode. If disabled, the device acts a regular one.
39
40 Notifications for tx and rx are exchanged without vm exits
41 if possible. In particular (only mentioning csb mode below),
42 the following actions are performed. In the description below,
43 "double check" means verifying again the condition that caused
44 the previous action, and reverting the action if the condition has
45 changed. The condition typically depends on a variable set by the
46 other party, and the double check is done to avoid races. E.g.
47
48	// start with A=0
49    again:
50	// do something
51	if ( cond(C) ) { // C is written by the other side
52	    A = 1;
53	    // barrier
54	    if ( !cond(C) ) {
55		A = 0;
56		goto again;
57	    }
58	}
59
60 TX: start from idle:
61    H starts with host_need_txkick=1 when the I/O thread bh is idle. Upon new
62    transmissions, G always updates guest_tdt.  If host_need_txkick == 1,
63    G also writes to the TDT, which acts as a kick to H (so pending
64    writes are always dispatched to H as soon as possible.)
65
66 TX: active state:
67    On the kick (TDT write) H sets host_need_txkick == 0 (if not
68    done already by G), and starts an I/O thread trying to consume
69    packets from TDH to guest_tdt, periodically refreshing host_tdh
70    and TDH.  When host_tdh == guest_tdt, H sets host_need_txkick=1,
71    and then does the "double check" for race avoidance.
72
73 TX: G runs out of buffers
74    XXX there are two mechanisms, one boolean (using guest_need_txkick)
75    and one with a threshold (using guest_txkick_at). They are mutually
76    exclusive.
77    BOOLEAN: when G has no space, it sets guest_need_txkick=1 and does
78        the double check. If H finds guest_need_txkick== 1 on a write
79        to TDH, it also generates an interrupt.
80    THRESHOLD: G sets guest_txkick_at to the TDH value for which it
81	wants to receive an interrupt. When H detects that TDH moves
82	across guest_txkick_at, it generates an interrupt.
83	This second mechanism reduces the number of interrupts and
84	TDT writes on the transmit side when the host is too slow.
85
86 RX: start from idle
87    G starts with guest_need_rxkick = 1 when the receive ring is empty.
88    As packets arrive, H updates host_rdh (and RDH) and also generates an
89    interrupt when guest_need_rxkick == 1 (so incoming packets are
90    always reported to G as soon as possible, apart from interrupt
91    moderation delays). It also tracks guest_rdt for new buffers.
92
93 RX: active state
94    As the interrupt arrives, G sets guest_need_rxkick = 0 and starts
95    draining packets from the receive ring, while updating guest_rdt
96    When G runs out of packets it sets guest_need_rxkick=1 and does the
97    double check.
98
99 RX: H runs out of buffers
100    XXX there are two mechanisms, one boolean (using host_need_rxkick)
101    and one with a threshold (using host_xxkick_at). They are mutually
102    exclusive.
103    BOOLEAN: when H has no space, it sets host_need_rxkick=1 and does the
104	double check. If G finds host_need_rxkick==1 on updating guest_rdt,
105        it also writes to RDT causing a kick to H.
106    THRESHOLD: H sets host_rxkick_at to the RDT value for which it wants
107	to receive a kick. When G detects that guest_rdt moves across
108	host_rxkick_at, it writes to RDT thus generates a kick.
109	This second mechanism reduces the number of kicks and
110        RDT writes on the receive side when the guest is too slow and
111	would free only a few buffers at a time.
112
113 */
114struct paravirt_csb {
115    /* XXX revise the layout to minimize cache bounces.
116     * Usage is described as follows:
117     * 	[GH][RW][+-0]	guest/host reads/writes frequently/rarely/almost never
118     */
119    /* these are (mostly) written by the guest */
120    uint32_t guest_tdt;            /* GW+ HR+ pkt to transmit */
121    uint32_t guest_need_txkick;    /* GW- HR+ G ran out of tx bufs, request kick */
122    uint32_t guest_need_rxkick;    /* GW- HR+ G ran out of rx pkts, request kick  */
123    uint32_t guest_csb_on;         /* GW- HR+ enable paravirtual mode */
124    uint32_t guest_rdt;            /* GW+ HR+ rx buffers available */
125    uint32_t guest_txkick_at;      /* GW- HR+ tx ring pos. where G expects an intr */
126    uint32_t guest_use_msix;        /* GW0 HR0 guest uses MSI-X interrupts. */
127    uint32_t pad[9];
128
129    /* these are (mostly) written by the host */
130    uint32_t host_tdh;             /* GR0 HW- shadow register, mostly unused */
131    uint32_t host_need_txkick;     /* GR+ HW- start the iothread */
132    uint32_t host_txcycles_lim;    /* GW- HR- how much to spin before  sleep.
133				    * set by the guest */
134    uint32_t host_txcycles;        /* GR0 HW- counter, but no need to be exported */
135    uint32_t host_rdh;             /* GR0 HW- shadow register, mostly unused */
136    uint32_t host_need_rxkick;     /* GR+ HW- flush rx queued packets */
137    uint32_t host_isr;             /* GR* HW* shadow copy of ISR */
138    uint32_t host_rxkick_at;       /* GR+ HW- rx ring pos where H expects a kick */
139    uint32_t vnet_ring_high;	/* Vnet ring physical address high. */
140    uint32_t vnet_ring_low;	/* Vnet ring physical address low. */
141};
142
143#define NET_PARAVIRT_CSB_SIZE   4096
144#define NET_PARAVIRT_NONE   (~((uint32_t)0))
145
146#ifdef	QEMU_PCI_H
147
148/*
149 * API functions only available within QEMU
150 */
151
152void paravirt_configure_csb(struct paravirt_csb** csb, uint32_t csbbal,
153			uint32_t csbbah, QEMUBH* tx_bh, AddressSpace *as);
154
155#endif /* QEMU_PCI_H */
156
157#endif /* NET_PARAVIRT_H */
158