Linux kernel mirror (for testing)
git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git
kernel
os
linux
accel/amdxdna: Fix race condition when checking rpm_on
When autosuspend is triggered, driver rpm_on flag is set to indicate that
a suspend/resume is already in progress. However, when a userspace
application submits a command during this narrow window,
amdxdna_pm_resume_get() may incorrectly skip the resume operation because
the rpm_on flag is still set. This results in commands being submitted
while the device has not actually resumed, causing unexpected behavior.
The set_dpm() is called by suspend/resume, it relied on rpm_on flag to
avoid calling into rpm suspend/resume recursivly. So to fix this, remove
the use of the rpm_on flag entirely. Instead, introduce aie2_pm_set_dpm()
which explicitly resumes the device before invoking set_dpm(). With this
change, set_dpm() is called directly inside the suspend or resume execution
path. Otherwise, aie2_pm_set_dpm() is called.
Fixes: 063db451832b ("accel/amdxdna: Enhance runtime power management")
Reviewed-by: Mario Limonciello (AMD) <superm1@kernel.org>
Reviewed-by: Maciej Falkowski <maciej.falkowski@linux.intel.com>
Signed-off-by: Lizhi Hou <lizhi.hou@amd.com>
Link: https://patch.msgid.link/20251208165356.1549237-1-lizhi.hou@amd.com
+24
-47
-1
drivers/accel/amdxdna/aie2_message.c
-1
drivers/accel/amdxdna/aie2_message.c
+1
-1
drivers/accel/amdxdna/aie2_pci.c
+1
-1
drivers/accel/amdxdna/aie2_pci.c
+1
drivers/accel/amdxdna/aie2_pci.h
+1
drivers/accel/amdxdna/aie2_pci.h
···
286
286
/* aie2_pm.c */
287
287
int aie2_pm_init(struct amdxdna_dev_hdl *ndev);
288
288
int aie2_pm_set_mode(struct amdxdna_dev_hdl *ndev, enum amdxdna_power_mode_type target);
289
+
int aie2_pm_set_dpm(struct amdxdna_dev_hdl *ndev, u32 dpm_level);
289
290
290
291
/* aie2_psp.c */
291
292
struct psp_device *aie2m_psp_create(struct drm_device *ddev, struct psp_config *conf);
+16
-1
drivers/accel/amdxdna/aie2_pm.c
+16
-1
drivers/accel/amdxdna/aie2_pm.c
···
10
10
11
11
#include "aie2_pci.h"
12
12
#include "amdxdna_pci_drv.h"
13
+
#include "amdxdna_pm.h"
13
14
14
15
#define AIE2_CLK_GATING_ENABLE 1
15
16
#define AIE2_CLK_GATING_DISABLE 0
···
25
24
26
25
ndev->clk_gating = val;
27
26
return 0;
27
+
}
28
+
29
+
int aie2_pm_set_dpm(struct amdxdna_dev_hdl *ndev, u32 dpm_level)
30
+
{
31
+
int ret;
32
+
33
+
ret = amdxdna_pm_resume_get(ndev->xdna);
34
+
if (ret)
35
+
return ret;
36
+
37
+
ret = ndev->priv->hw_ops.set_dpm(ndev, dpm_level);
38
+
amdxdna_pm_suspend_put(ndev->xdna);
39
+
40
+
return ret;
28
41
}
29
42
30
43
int aie2_pm_init(struct amdxdna_dev_hdl *ndev)
···
109
94
return -EOPNOTSUPP;
110
95
}
111
96
112
-
ret = ndev->priv->hw_ops.set_dpm(ndev, dpm_level);
97
+
ret = aie2_pm_set_dpm(ndev, dpm_level);
113
98
if (ret)
114
99
return ret;
115
100
+4
-23
drivers/accel/amdxdna/aie2_smu.c
+4
-23
drivers/accel/amdxdna/aie2_smu.c
···
11
11
12
12
#include "aie2_pci.h"
13
13
#include "amdxdna_pci_drv.h"
14
-
#include "amdxdna_pm.h"
15
14
16
15
#define SMU_RESULT_OK 1
17
16
···
66
67
u32 freq;
67
68
int ret;
68
69
69
-
ret = amdxdna_pm_resume_get(ndev->xdna);
70
-
if (ret)
71
-
return ret;
72
-
73
70
ret = aie2_smu_exec(ndev, AIE2_SMU_SET_MPNPUCLK_FREQ,
74
71
ndev->priv->dpm_clk_tbl[dpm_level].npuclk, &freq);
75
72
if (ret) {
76
73
XDNA_ERR(ndev->xdna, "Set npu clock to %d failed, ret %d\n",
77
74
ndev->priv->dpm_clk_tbl[dpm_level].npuclk, ret);
78
-
goto suspend_put;
75
+
return ret;
79
76
}
80
77
ndev->npuclk_freq = freq;
81
78
···
80
85
if (ret) {
81
86
XDNA_ERR(ndev->xdna, "Set h clock to %d failed, ret %d\n",
82
87
ndev->priv->dpm_clk_tbl[dpm_level].hclk, ret);
83
-
goto suspend_put;
88
+
return ret;
84
89
}
85
90
86
-
amdxdna_pm_suspend_put(ndev->xdna);
87
91
ndev->hclk_freq = freq;
88
92
ndev->dpm_level = dpm_level;
89
93
ndev->max_tops = 2 * ndev->total_col;
···
92
98
ndev->npuclk_freq, ndev->hclk_freq);
93
99
94
100
return 0;
95
-
96
-
suspend_put:
97
-
amdxdna_pm_suspend_put(ndev->xdna);
98
-
return ret;
99
101
}
100
102
101
103
int npu4_set_dpm(struct amdxdna_dev_hdl *ndev, u32 dpm_level)
102
104
{
103
105
int ret;
104
106
105
-
ret = amdxdna_pm_resume_get(ndev->xdna);
106
-
if (ret)
107
-
return ret;
108
-
109
107
ret = aie2_smu_exec(ndev, AIE2_SMU_SET_HARD_DPMLEVEL, dpm_level, NULL);
110
108
if (ret) {
111
109
XDNA_ERR(ndev->xdna, "Set hard dpm level %d failed, ret %d ",
112
110
dpm_level, ret);
113
-
goto suspend_put;
111
+
return ret;
114
112
}
115
113
116
114
ret = aie2_smu_exec(ndev, AIE2_SMU_SET_SOFT_DPMLEVEL, dpm_level, NULL);
117
115
if (ret) {
118
116
XDNA_ERR(ndev->xdna, "Set soft dpm level %d failed, ret %d",
119
117
dpm_level, ret);
120
-
goto suspend_put;
118
+
return ret;
121
119
}
122
120
123
-
amdxdna_pm_suspend_put(ndev->xdna);
124
121
ndev->npuclk_freq = ndev->priv->dpm_clk_tbl[dpm_level].npuclk;
125
122
ndev->hclk_freq = ndev->priv->dpm_clk_tbl[dpm_level].hclk;
126
123
ndev->dpm_level = dpm_level;
···
122
137
ndev->npuclk_freq, ndev->hclk_freq);
123
138
124
139
return 0;
125
-
126
-
suspend_put:
127
-
amdxdna_pm_suspend_put(ndev->xdna);
128
-
return ret;
129
140
}
130
141
131
142
int aie2_smu_init(struct amdxdna_dev_hdl *ndev)
-1
drivers/accel/amdxdna/amdxdna_pci_drv.h
-1
drivers/accel/amdxdna/amdxdna_pci_drv.h
+2
-20
drivers/accel/amdxdna/amdxdna_pm.c
+2
-20
drivers/accel/amdxdna/amdxdna_pm.c
···
15
15
{
16
16
struct amdxdna_dev *xdna = to_xdna_dev(dev_get_drvdata(dev));
17
17
int ret = -EOPNOTSUPP;
18
-
bool rpm;
19
18
20
-
if (xdna->dev_info->ops->suspend) {
21
-
rpm = xdna->rpm_on;
22
-
xdna->rpm_on = false;
19
+
if (xdna->dev_info->ops->suspend)
23
20
ret = xdna->dev_info->ops->suspend(xdna);
24
-
xdna->rpm_on = rpm;
25
-
}
26
21
27
22
XDNA_DBG(xdna, "Suspend done ret %d", ret);
28
23
return ret;
···
27
32
{
28
33
struct amdxdna_dev *xdna = to_xdna_dev(dev_get_drvdata(dev));
29
34
int ret = -EOPNOTSUPP;
30
-
bool rpm;
31
35
32
-
if (xdna->dev_info->ops->resume) {
33
-
rpm = xdna->rpm_on;
34
-
xdna->rpm_on = false;
36
+
if (xdna->dev_info->ops->resume)
35
37
ret = xdna->dev_info->ops->resume(xdna);
36
-
xdna->rpm_on = rpm;
37
-
}
38
38
39
39
XDNA_DBG(xdna, "Resume done ret %d", ret);
40
40
return ret;
···
39
49
{
40
50
struct device *dev = xdna->ddev.dev;
41
51
int ret;
42
-
43
-
if (!xdna->rpm_on)
44
-
return 0;
45
52
46
53
ret = pm_runtime_resume_and_get(dev);
47
54
if (ret) {
···
53
66
{
54
67
struct device *dev = xdna->ddev.dev;
55
68
56
-
if (!xdna->rpm_on)
57
-
return;
58
-
59
69
pm_runtime_put_autosuspend(dev);
60
70
}
61
71
···
65
81
pm_runtime_use_autosuspend(dev);
66
82
pm_runtime_allow(dev);
67
83
pm_runtime_put_autosuspend(dev);
68
-
xdna->rpm_on = true;
69
84
}
70
85
71
86
void amdxdna_pm_fini(struct amdxdna_dev *xdna)
72
87
{
73
88
struct device *dev = xdna->ddev.dev;
74
89
75
-
xdna->rpm_on = false;
76
90
pm_runtime_get_noresume(dev);
77
91
pm_runtime_forbid(dev);
78
92
}