update kernel to 5.10.198

hc

2024-01-03 2f7c68cb55ecb7331f2381deb497c27155f32faf

 kernel/drivers/gpu/drm/i915/i915_memcpy.c
..
..
@@ -25,16 +25,20 @@
25
25
 #include <linux/kernel.h>
26
26
 #include <asm/fpu/api.h>
27
27
 
28
-#include "i915_drv.h"
28
+#include "i915_memcpy.h"
29
+
30
+#if IS_ENABLED(CONFIG_DRM_I915_DEBUG)
31
+#define CI_BUG_ON(expr) BUG_ON(expr)
32
+#else
33
+#define CI_BUG_ON(expr) BUILD_BUG_ON_INVALID(expr)
34
+#endif
29
35
 
30
36
 static DEFINE_STATIC_KEY_FALSE(has_movntdqa);
31
37
 
32
-#ifdef CONFIG_AS_MOVNTDQA
33
38
 static void __memcpy_ntdqa(void *dst, const void *src, unsigned long len)
34
39
 {
35
40
 	kernel_fpu_begin();
36
41
 
37
-	len >>= 4;
38
42
 	while (len >= 4) {
39
43
 		asm("movntdqa   (%0), %%xmm0\n"
40
44
 		    "movntdqa 16(%0), %%xmm1\n"
..
..
@@ -59,7 +63,35 @@
59
63
 
60
64
 	kernel_fpu_end();
61
65
 }
62
-#endif
66
+
67
+static void __memcpy_ntdqu(void *dst, const void *src, unsigned long len)
68
+{
69
+	kernel_fpu_begin();
70
+
71
+	while (len >= 4) {
72
+		asm("movntdqa   (%0), %%xmm0\n"
73
+		    "movntdqa 16(%0), %%xmm1\n"
74
+		    "movntdqa 32(%0), %%xmm2\n"
75
+		    "movntdqa 48(%0), %%xmm3\n"
76
+		    "movups %%xmm0,   (%1)\n"
77
+		    "movups %%xmm1, 16(%1)\n"
78
+		    "movups %%xmm2, 32(%1)\n"
79
+		    "movups %%xmm3, 48(%1)\n"
80
+		    :: "r" (src), "r" (dst) : "memory");
81
+		src += 64;
82
+		dst += 64;
83
+		len -= 4;
84
+	}
85
+	while (len--) {
86
+		asm("movntdqa (%0), %%xmm0\n"
87
+		    "movups %%xmm0, (%1)\n"
88
+		    :: "r" (src), "r" (dst) : "memory");
89
+		src += 16;
90
+		dst += 16;
91
+	}
92
+
93
+	kernel_fpu_end();
94
+}
63
95
 
64
96
 /**
65
97
  * i915_memcpy_from_wc: perform an accelerated *aligned* read from WC
..
..
@@ -83,17 +115,47 @@
83
115
 	if (unlikely(((unsigned long)dst | (unsigned long)src | len) & 15))
84
116
 		return false;
85
117
 
86
-#ifdef CONFIG_AS_MOVNTDQA
87
118
 	if (static_branch_likely(&has_movntdqa)) {
88
119
 		if (likely(len))
89
-			__memcpy_ntdqa(dst, src, len);
120
+			__memcpy_ntdqa(dst, src, len >> 4);
90
121
 		return true;
91
122
 	}
92
-#endif
93
123
 
94
124
 	return false;
95
125
 }
96
126
 
127
+/**
128
+ * i915_unaligned_memcpy_from_wc: perform a mostly accelerated read from WC
129
+ * @dst: destination pointer
130
+ * @src: source pointer
131
+ * @len: how many bytes to copy
132
+ *
133
+ * Like i915_memcpy_from_wc(), the unaligned variant copies @len bytes from
134
+ * @src to @dst using * non-temporal instructions where available, but
135
+ * accepts that its arguments may not be aligned, but are valid for the
136
+ * potential 16-byte read past the end.
137
+ */
138
+void i915_unaligned_memcpy_from_wc(void *dst, void *src, unsigned long len)
139
+{
140
+	unsigned long addr;
141
+
142
+	CI_BUG_ON(!i915_has_memcpy_from_wc());
143
+
144
+	addr = (unsigned long)src;
145
+	if (!IS_ALIGNED(addr, 16)) {
146
+		unsigned long x = min(ALIGN(addr, 16) - addr, len);
147
+
148
+		memcpy(dst, src, x);
149
+
150
+		len -= x;
151
+		dst += x;
152
+		src += x;
153
+	}
154
+
155
+	if (likely(len))
156
+		__memcpy_ntdqu(dst, src, DIV_ROUND_UP(len, 16));
157
+}
158
+
97
159
 void i915_memcpy_init_early(struct drm_i915_private *dev_priv)
98
160
 {
99
161
 	/*