add wifi6 8852be driver

hc

2024-12-19 9370bb92b2d16684ee45cf24e879c93c509162da

 kernel/arch/x86/lib/delay.c
..
..
@@ -27,9 +27,20 @@
27
27
 # include <asm/smp.h>
28
28
 #endif
29
29
 
30
+static void delay_loop(u64 __loops);
31
+
32
+/*
33
+ * Calibration and selection of the delay mechanism happens only once
34
+ * during boot.
35
+ */
36
+static void (*delay_fn)(u64) __ro_after_init = delay_loop;
37
+static void (*delay_halt_fn)(u64 start, u64 cycles) __ro_after_init;
38
+
30
39
 /* simple loop based delay: */
31
-static void delay_loop(unsigned long loops)
40
+static void delay_loop(u64 __loops)
32
41
 {
42
+	unsigned long loops = (unsigned long)__loops;
43
+
33
44
 	asm volatile(
34
45
 		"	test %0,%0	\n"
35
46
 		"	jz 3f		\n"
..
..
@@ -43,15 +54,15 @@
43
54
 		"	jnz 2b		\n"
44
55
 		"3:	dec %0		\n"
45
56
 
46
-		: /* we don't need output */
47
-		:"a" (loops)
57
+		: "+a" (loops)
58
+		:
48
59
 	);
49
60
 }
50
61
 
51
62
 /* TSC based delay: */
52
-static void delay_tsc(unsigned long __loops)
63
+static void delay_tsc(u64 cycles)
53
64
 {
54
-	u64 bclock, now, loops = __loops;
65
+	u64 bclock, now;
55
66
 	int cpu;
56
67
 
57
68
 	preempt_disable();
..
..
@@ -59,7 +70,7 @@
59
70
 	bclock = rdtsc_ordered();
60
71
 	for (;;) {
61
72
 		now = rdtsc_ordered();
62
-		if ((now - bclock) >= loops)
73
+		if ((now - bclock) >= cycles)
63
74
 			break;
64
75
 
65
76
 		/* Allow RT tasks to run */
..
..
@@ -77,7 +88,7 @@
77
88
 		 * counter for this CPU.
78
89
 		 */
79
90
 		if (unlikely(cpu != smp_processor_id())) {
80
-			loops -= (now - bclock);
91
+			cycles -= (now - bclock);
81
92
 			cpu = smp_processor_id();
82
93
 			bclock = rdtsc_ordered();
83
94
 		}
..
..
@@ -86,65 +97,96 @@
86
97
 }
87
98
 
88
99
 /*
89
- * On some AMD platforms, MWAITX has a configurable 32-bit timer, that
90
- * counts with TSC frequency. The input value is the loop of the
91
- * counter, it will exit when the timer expires.
100
+ * On Intel the TPAUSE instruction waits until any of:
101
+ * 1) the TSC counter exceeds the value provided in EDX:EAX
102
+ * 2) global timeout in IA32_UMWAIT_CONTROL is exceeded
103
+ * 3) an external interrupt occurs
92
104
  */
93
-static void delay_mwaitx(unsigned long __loops)
105
+static void delay_halt_tpause(u64 start, u64 cycles)
94
106
 {
95
-	u64 start, end, delay, loops = __loops;
107
+	u64 until = start + cycles;
108
+	u32 eax, edx;
109
+
110
+	eax = lower_32_bits(until);
111
+	edx = upper_32_bits(until);
112
+
113
+	/*
114
+	 * Hard code the deeper (C0.2) sleep state because exit latency is
115
+	 * small compared to the "microseconds" that usleep() will delay.
116
+	 */
117
+	__tpause(TPAUSE_C02_STATE, edx, eax);
118
+}
119
+
120
+/*
121
+ * On some AMD platforms, MWAITX has a configurable 32-bit timer, that
122
+ * counts with TSC frequency. The input value is the number of TSC cycles
123
+ * to wait. MWAITX will also exit when the timer expires.
124
+ */
125
+static void delay_halt_mwaitx(u64 unused, u64 cycles)
126
+{
127
+	u64 delay;
128
+
129
+	delay = min_t(u64, MWAITX_MAX_WAIT_CYCLES, cycles);
130
+	/*
131
+	 * Use cpu_tss_rw as a cacheline-aligned, seldomly accessed per-cpu
132
+	 * variable as the monitor target.
133
+	 */
134
+	 __monitorx(raw_cpu_ptr(&cpu_tss_rw), 0, 0);
135
+
136
+	/*
137
+	 * AMD, like Intel, supports the EAX hint and EAX=0xf means, do not
138
+	 * enter any deep C-state and we use it here in delay() to minimize
139
+	 * wakeup latency.
140
+	 */
141
+	__mwaitx(MWAITX_DISABLE_CSTATES, delay, MWAITX_ECX_TIMER_ENABLE);
142
+}
143
+
144
+/*
145
+ * Call a vendor specific function to delay for a given amount of time. Because
146
+ * these functions may return earlier than requested, check for actual elapsed
147
+ * time and call again until done.
148
+ */
149
+static void delay_halt(u64 __cycles)
150
+{
151
+	u64 start, end, cycles = __cycles;
96
152
 
97
153
 	/*
98
154
 	 * Timer value of 0 causes MWAITX to wait indefinitely, unless there
99
155
 	 * is a store on the memory monitored by MONITORX.
100
156
 	 */
101
-	if (loops == 0)
157
+	if (!cycles)
102
158
 		return;
103
159
 
104
160
 	start = rdtsc_ordered();
105
161
 
106
162
 	for (;;) {
107
-		delay = min_t(u64, MWAITX_MAX_LOOPS, loops);
108
-
109
-		/*
110
-		 * Use cpu_tss_rw as a cacheline-aligned, seldomly
111
-		 * accessed per-cpu variable as the monitor target.
112
-		 */
113
-		__monitorx(raw_cpu_ptr(&cpu_tss_rw), 0, 0);
114
-
115
-		/*
116
-		 * AMD, like Intel's MWAIT version, supports the EAX hint and
117
-		 * EAX=0xf0 means, do not enter any deep C-state and we use it
118
-		 * here in delay() to minimize wakeup latency.
119
-		 */
120
-		__mwaitx(MWAITX_DISABLE_CSTATES, delay, MWAITX_ECX_TIMER_ENABLE);
121
-
163
+		delay_halt_fn(start, cycles);
122
164
 		end = rdtsc_ordered();
123
165
 
124
-		if (loops <= end - start)
166
+		if (cycles <= end - start)
125
167
 			break;
126
168
 
127
-		loops -= end - start;
128
-
169
+		cycles -= end - start;
129
170
 		start = end;
130
171
 	}
131
172
 }
132
173
 
133
-/*
134
- * Since we calibrate only once at boot, this
135
- * function should be set once at boot and not changed
136
- */
137
-static void (*delay_fn)(unsigned long) = delay_loop;
138
-
139
-void use_tsc_delay(void)
174
+void __init use_tsc_delay(void)
140
175
 {
141
176
 	if (delay_fn == delay_loop)
142
177
 		delay_fn = delay_tsc;
143
178
 }
144
179
 
180
+void __init use_tpause_delay(void)
181
+{
182
+	delay_halt_fn = delay_halt_tpause;
183
+	delay_fn = delay_halt;
184
+}
185
+
145
186
 void use_mwaitx_delay(void)
146
187
 {
147
-	delay_fn = delay_mwaitx;
188
+	delay_halt_fn = delay_halt_mwaitx;
189
+	delay_fn = delay_halt;
148
190
 }
149
191
 
150
192
 int read_current_timer(unsigned long *timer_val)
..
..
@@ -162,7 +204,7 @@
162
204
 }
163
205
 EXPORT_SYMBOL(__delay);
164
206
 
165
-void __const_udelay(unsigned long xloops)
207
+noinline void __const_udelay(unsigned long xloops)
166
208
 {
167
209
 	unsigned long lpj = this_cpu_read(cpu_info.loops_per_jiffy) ? : loops_per_jiffy;
168
210
 	int d0;