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
|
// SPDX-License-Identifier: MIT
/*
* Copyright © 2023 Intel Corporation
*/
#include <drm/drm_managed.h>
#include "regs/xe_gt_regs.h"
#include "xe_assert.h"
#include "xe_gt.h"
#include "xe_gt_ccs_mode.h"
#include "xe_gt_sysfs.h"
#include "xe_mmio.h"
static void __xe_gt_apply_ccs_mode(struct xe_gt *gt, u32 num_engines)
{
u32 mode = CCS_MODE_CSLICE_0_3_MASK; /* disable all by default */
int num_slices = hweight32(CCS_MASK(gt));
struct xe_device *xe = gt_to_xe(gt);
int width, cslice = 0;
u32 config = 0;
xe_assert(xe, xe_gt_ccs_mode_enabled(gt));
xe_assert(xe, num_engines && num_engines <= num_slices);
xe_assert(xe, !(num_slices % num_engines));
/*
* Loop over all available slices and assign each a user engine.
* For example, if there are four compute slices available, the
* assignment of compute slices to compute engines would be,
*
* With 1 engine (ccs0):
* slice 0, 1, 2, 3: ccs0
*
* With 2 engines (ccs0, ccs1):
* slice 0, 2: ccs0
* slice 1, 3: ccs1
*
* With 4 engines (ccs0, ccs1, ccs2, ccs3):
* slice 0: ccs0
* slice 1: ccs1
* slice 2: ccs2
* slice 3: ccs3
*/
for (width = num_slices / num_engines; width; width--) {
struct xe_hw_engine *hwe;
enum xe_hw_engine_id id;
for_each_hw_engine(hwe, gt, id) {
if (hwe->class != XE_ENGINE_CLASS_COMPUTE)
continue;
if (hwe->logical_instance >= num_engines)
break;
config |= BIT(hwe->instance) << XE_HW_ENGINE_CCS0;
/* If a slice is fused off, leave disabled */
while ((CCS_MASK(gt) & BIT(cslice)) == 0)
cslice++;
mode &= ~CCS_MODE_CSLICE(cslice, CCS_MODE_CSLICE_MASK);
mode |= CCS_MODE_CSLICE(cslice, hwe->instance);
cslice++;
}
}
xe_mmio_write32(gt, CCS_MODE, mode);
xe_gt_info(gt, "CCS_MODE=%x config:%08x, num_engines:%d, num_slices:%d\n",
mode, config, num_engines, num_slices);
}
void xe_gt_apply_ccs_mode(struct xe_gt *gt)
{
if (!gt->ccs_mode)
return;
__xe_gt_apply_ccs_mode(gt, gt->ccs_mode);
}
static ssize_t
num_cslices_show(struct device *kdev,
struct device_attribute *attr, char *buf)
{
struct xe_gt *gt = kobj_to_gt(&kdev->kobj);
return sysfs_emit(buf, "%u\n", hweight32(CCS_MASK(gt)));
}
static DEVICE_ATTR_RO(num_cslices);
static ssize_t
ccs_mode_show(struct device *kdev,
struct device_attribute *attr, char *buf)
{
struct xe_gt *gt = kobj_to_gt(&kdev->kobj);
return sysfs_emit(buf, "%u\n", gt->ccs_mode);
}
static ssize_t
ccs_mode_store(struct device *kdev, struct device_attribute *attr,
const char *buff, size_t count)
{
struct xe_gt *gt = kobj_to_gt(&kdev->kobj);
struct xe_device *xe = gt_to_xe(gt);
u32 num_engines, num_slices;
int ret;
ret = kstrtou32(buff, 0, &num_engines);
if (ret)
return ret;
/*
* Ensure number of engines specified is valid and there is an
* exact multiple of engines for slices.
*/
num_slices = hweight32(CCS_MASK(gt));
if (!num_engines || num_engines > num_slices || num_slices % num_engines) {
xe_gt_dbg(gt, "Invalid compute config, %d engines %d slices\n",
num_engines, num_slices);
return -EINVAL;
}
/* CCS mode can only be updated when there are no drm clients */
spin_lock(&xe->clients.lock);
if (xe->clients.count) {
spin_unlock(&xe->clients.lock);
return -EBUSY;
}
if (gt->ccs_mode != num_engines) {
xe_gt_info(gt, "Setting compute mode to %d\n", num_engines);
gt->ccs_mode = num_engines;
xe_gt_reset_async(gt);
}
spin_unlock(&xe->clients.lock);
return count;
}
static DEVICE_ATTR_RW(ccs_mode);
static const struct attribute *gt_ccs_mode_attrs[] = {
&dev_attr_ccs_mode.attr,
&dev_attr_num_cslices.attr,
NULL,
};
static void xe_gt_ccs_mode_sysfs_fini(struct drm_device *drm, void *arg)
{
struct xe_gt *gt = arg;
sysfs_remove_files(gt->sysfs, gt_ccs_mode_attrs);
}
/**
* xe_gt_ccs_mode_sysfs_init - Initialize CCS mode sysfs interfaces
* @gt: GT structure
*
* Through a per-gt 'ccs_mode' sysfs interface, the user can enable a fixed
* number of compute hardware engines to which the available compute slices
* are to be allocated. This user configuration change triggers a gt reset
* and it is expected that there are no open drm clients while doing so.
* The number of available compute slices is exposed to user through a per-gt
* 'num_cslices' sysfs interface.
*/
void xe_gt_ccs_mode_sysfs_init(struct xe_gt *gt)
{
struct xe_device *xe = gt_to_xe(gt);
int err;
if (!xe_gt_ccs_mode_enabled(gt))
return;
err = sysfs_create_files(gt->sysfs, gt_ccs_mode_attrs);
if (err) {
drm_warn(&xe->drm, "Sysfs creation for ccs_mode failed err: %d\n", err);
return;
}
err = drmm_add_action_or_reset(&xe->drm, xe_gt_ccs_mode_sysfs_fini, gt);
if (err) {
sysfs_remove_files(gt->sysfs, gt_ccs_mode_attrs);
drm_warn(&xe->drm, "%s: drmm_add_action_or_reset failed, err: %d\n",
__func__, err);
}
}
|
