1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
|
// SPDX-License-Identifier: GPL-2.0
/*
* Xen hypercall batching.
*
* Xen allows multiple hypercalls to be issued at once, using the
* multicall interface. This allows the cost of trapping into the
* hypervisor to be amortized over several calls.
*
* This file implements a simple interface for multicalls. There's a
* per-cpu buffer of outstanding multicalls. When you want to queue a
* multicall for issuing, you can allocate a multicall slot for the
* call and its arguments, along with storage for space which is
* pointed to by the arguments (for passing pointers to structures,
* etc). When the multicall is actually issued, all the space for the
* commands and allocated memory is freed for reuse.
*
* Multicalls are flushed whenever any of the buffers get full, or
* when explicitly requested. There's no way to get per-multicall
* return results back. It will BUG if any of the multicalls fail.
*
* Jeremy Fitzhardinge <jeremy@xensource.com>, XenSource Inc, 2007
*/
#include <linux/percpu.h>
#include <linux/hardirq.h>
#include <linux/debugfs.h>
#include <asm/xen/hypercall.h>
#include "multicalls.h"
#include "debugfs.h"
#define MC_BATCH 32
#define MC_DEBUG 0
#define MC_ARGS (MC_BATCH * 16)
struct mc_buffer {
unsigned mcidx, argidx, cbidx;
struct multicall_entry entries[MC_BATCH];
#if MC_DEBUG
struct multicall_entry debug[MC_BATCH];
void *caller[MC_BATCH];
#endif
unsigned char args[MC_ARGS];
struct callback {
void (*fn)(void *);
void *data;
} callbacks[MC_BATCH];
};
static DEFINE_PER_CPU(struct mc_buffer, mc_buffer);
DEFINE_PER_CPU(unsigned long, xen_mc_irq_flags);
void xen_mc_flush(void)
{
struct mc_buffer *b = this_cpu_ptr(&mc_buffer);
struct multicall_entry *mc;
int ret = 0;
unsigned long flags;
int i;
BUG_ON(preemptible());
/* Disable interrupts in case someone comes in and queues
something in the middle */
local_irq_save(flags);
trace_xen_mc_flush(b->mcidx, b->argidx, b->cbidx);
#if MC_DEBUG
memcpy(b->debug, b->entries,
b->mcidx * sizeof(struct multicall_entry));
#endif
switch (b->mcidx) {
case 0:
/* no-op */
BUG_ON(b->argidx != 0);
break;
case 1:
/* Singleton multicall - bypass multicall machinery
and just do the call directly. */
mc = &b->entries[0];
mc->result = xen_single_call(mc->op, mc->args[0], mc->args[1],
mc->args[2], mc->args[3],
mc->args[4]);
ret = mc->result < 0;
break;
default:
if (HYPERVISOR_multicall(b->entries, b->mcidx) != 0)
BUG();
for (i = 0; i < b->mcidx; i++)
if (b->entries[i].result < 0)
ret++;
}
if (WARN_ON(ret)) {
pr_err("%d of %d multicall(s) failed: cpu %d\n",
ret, b->mcidx, smp_processor_id());
for (i = 0; i < b->mcidx; i++) {
if (b->entries[i].result < 0) {
#if MC_DEBUG
pr_err(" call %2d: op=%lu arg=[%lx] result=%ld\t%pS\n",
i + 1,
b->debug[i].op,
b->debug[i].args[0],
b->entries[i].result,
b->caller[i]);
#else
pr_err(" call %2d: op=%lu arg=[%lx] result=%ld\n",
i + 1,
b->entries[i].op,
b->entries[i].args[0],
b->entries[i].result);
#endif
}
}
}
b->mcidx = 0;
b->argidx = 0;
for (i = 0; i < b->cbidx; i++) {
struct callback *cb = &b->callbacks[i];
(*cb->fn)(cb->data);
}
b->cbidx = 0;
local_irq_restore(flags);
}
struct multicall_space __xen_mc_entry(size_t args)
{
struct mc_buffer *b = this_cpu_ptr(&mc_buffer);
struct multicall_space ret;
unsigned argidx = roundup(b->argidx, sizeof(u64));
trace_xen_mc_entry_alloc(args);
BUG_ON(preemptible());
BUG_ON(b->argidx >= MC_ARGS);
if (unlikely(b->mcidx == MC_BATCH ||
(argidx + args) >= MC_ARGS)) {
trace_xen_mc_flush_reason((b->mcidx == MC_BATCH) ?
XEN_MC_FL_BATCH : XEN_MC_FL_ARGS);
xen_mc_flush();
argidx = roundup(b->argidx, sizeof(u64));
}
ret.mc = &b->entries[b->mcidx];
#if MC_DEBUG
b->caller[b->mcidx] = __builtin_return_address(0);
#endif
b->mcidx++;
ret.args = &b->args[argidx];
b->argidx = argidx + args;
BUG_ON(b->argidx >= MC_ARGS);
return ret;
}
struct multicall_space xen_mc_extend_args(unsigned long op, size_t size)
{
struct mc_buffer *b = this_cpu_ptr(&mc_buffer);
struct multicall_space ret = { NULL, NULL };
BUG_ON(preemptible());
BUG_ON(b->argidx >= MC_ARGS);
if (unlikely(b->mcidx == 0 ||
b->entries[b->mcidx - 1].op != op)) {
trace_xen_mc_extend_args(op, size, XEN_MC_XE_BAD_OP);
goto out;
}
if (unlikely((b->argidx + size) >= MC_ARGS)) {
trace_xen_mc_extend_args(op, size, XEN_MC_XE_NO_SPACE);
goto out;
}
ret.mc = &b->entries[b->mcidx - 1];
ret.args = &b->args[b->argidx];
b->argidx += size;
BUG_ON(b->argidx >= MC_ARGS);
trace_xen_mc_extend_args(op, size, XEN_MC_XE_OK);
out:
return ret;
}
void xen_mc_callback(void (*fn)(void *), void *data)
{
struct mc_buffer *b = this_cpu_ptr(&mc_buffer);
struct callback *cb;
if (b->cbidx == MC_BATCH) {
trace_xen_mc_flush_reason(XEN_MC_FL_CALLBACK);
xen_mc_flush();
}
trace_xen_mc_callback(fn, data);
cb = &b->callbacks[b->cbidx++];
cb->fn = fn;
cb->data = data;
}
|
