drm/nouveau/fb/ramnva3: Link training for DDR3

V2: fix whitespace errors in memx.fuc
Signed-off-by: Roy Spliet <rspliet@eclipso.eu>
Signed-off-by: Ben Skeggs <bskeggs@redhat.com>

drivers/gpu/drm/nouveau/core/include/subdev/pwr.h

@@ -48,6 +48,8 @@ void nouveau_memx_wait(struct nouveau_memx *,
 		       u32 addr, u32 mask, u32 data, u32 nsec);
 void nouveau_memx_nsec(struct nouveau_memx *, u32 nsec);
 void nouveau_memx_wait_vblank(struct nouveau_memx *);
+void nouveau_memx_train(struct nouveau_memx *);
+int  nouveau_memx_train_result(struct nouveau_pwr *, u32 *, int);
 void nouveau_memx_block(struct nouveau_memx *);
 void nouveau_memx_unblock(struct nouveau_memx *);
 

drivers/gpu/drm/nouveau/core/subdev/fb/ramfuc.h

@@ -140,6 +140,20 @@ ramfuc_wait_vblank(struct ramfuc *ram)
 	nouveau_memx_wait_vblank(ram->memx);
 }
 
+static inline void
+ramfuc_train(struct ramfuc *ram)
+{
+	nouveau_memx_train(ram->memx);
+}
+
+static inline int
+ramfuc_train_result(struct nouveau_fb *pfb, u32 *result, u32 rsize)
+{
+	struct nouveau_pwr *ppwr = nouveau_pwr(pfb);
+
+	return nouveau_memx_train_result(ppwr, result, rsize);
+}
+
 static inline void
 ramfuc_block(struct ramfuc *ram)
 {
@@ -162,6 +176,8 @@ ramfuc_unblock(struct ramfuc *ram)
 #define ram_wait(s,r,m,d,n)  ramfuc_wait(&(s)->base, (r), (m), (d), (n))
 #define ram_nsec(s,n)        ramfuc_nsec(&(s)->base, (n))
 #define ram_wait_vblank(s)   ramfuc_wait_vblank(&(s)->base)
+#define ram_train(s)         ramfuc_train(&(s)->base)
+#define ram_train_result(s,r,l) ramfuc_train_result((s), (r), (l))
 #define ram_block(s)         ramfuc_block(&(s)->base)
 #define ram_unblock(s)       ramfuc_unblock(&(s)->base)
 

drivers/gpu/drm/nouveau/core/subdev/pwr/fuc/memx.fuc

@@ -50,6 +50,7 @@ handler(WR32  , 0x0000, 0x0002, #memx_func_wr32)
 handler(WAIT  , 0x0004, 0x0000, #memx_func_wait)
 handler(DELAY , 0x0001, 0x0000, #memx_func_delay)
 handler(VBLANK, 0x0001, 0x0000, #memx_func_wait_vblank)
+handler(TRAIN , 0x0000, 0x0000, #memx_func_train)
 memx_func_tail:
 
 .equ #memx_func_size #memx_func_next - #memx_func_head
@@ -63,6 +64,10 @@ memx_ts_end:
 memx_data_head:
 .skip 0x0800
 memx_data_tail:
+
+memx_train_head:
+.skip 0x0100
+memx_train_tail:
 #endif
 
 /******************************************************************************
@@ -257,6 +262,101 @@ memx_func_delay:
 	call(nsec)
 	ret
 
+// description
+//
+// $r15 - current (memx)
+// $r4  - packet length
+// $r3  - opcode desciption
+// $r0  - zero
+memx_func_train:
+#if NVKM_PPWR_CHIPSET == GT215
+// $r5 - outer loop counter
+// $r6 - inner loop counter
+// $r7 - entry counter (#memx_train_head + $r7)
+	movw $r5 0x3
+	movw $r7 0x0
+
+// Read random memory to wake up... things
+	imm32($r9, 0x700000)
+	nv_rd32($r8,$r9)
+	movw $r14 0x2710
+	call(nsec)
+
+	memx_func_train_loop_outer:
+		mulu $r8 $r5 0x101
+		sethi $r8 0x02000000
+		imm32($r9, 0x1111e0)
+		nv_wr32($r9, $r8)
+		push $r5
+
+		movw $r6 0x0
+		memx_func_train_loop_inner:
+			movw $r8 0x1111
+			mulu $r9 $r6 $r8
+			shl b32 $r8 $r9 0x10
+			or $r8 $r9
+			imm32($r9, 0x100720)
+			nv_wr32($r9, $r8)
+
+			imm32($r9, 0x100080)
+			nv_rd32($r8, $r9)
+			or $r8 $r8 0x20
+			nv_wr32($r9, $r8)
+
+			imm32($r9, 0x10053c)
+			imm32($r8, 0x80003002)
+			nv_wr32($r9, $r8)
+
+			imm32($r14, 0x100560)
+			imm32($r13, 0x80000000)
+			add b32 $r12 $r13 0
+			imm32($r11, 0x001e8480)
+			call(wait)
+
+			// $r5 - inner inner loop counter
+			// $r9 - result
+			movw $r5 0
+			imm32($r9, 0x8300ffff)
+			memx_func_train_loop_4x:
+				imm32($r10, 0x100080)
+				nv_rd32($r8, $r10)
+				imm32($r11, 0xffffffdf)
+				and $r8 $r11
+				nv_wr32($r10, $r8)
+
+				imm32($r10, 0x10053c)
+				imm32($r8, 0x80003002)
+				nv_wr32($r10, $r8)
+
+				imm32($r14, 0x100560)
+				imm32($r13, 0x80000000)
+				mov b32 $r12 $r13
+				imm32($r11, 0x00002710)
+				call(wait)
+
+				nv_rd32($r13, $r14)
+				and $r9 $r9 $r13
+
+				add b32 $r5 1
+				cmp b16 $r5 0x4
+				bra l #memx_func_train_loop_4x
+
+			add b32 $r10 $r7 #memx_train_head
+			st b32 D[$r10 + 0] $r9
+			add b32 $r6 1
+			add b32 $r7 4
+
+			cmp b16 $r6 0x10
+			bra l #memx_func_train_loop_inner
+
+		pop $r5
+		add b32 $r5 1
+		cmp b16 $r5 7
+		bra l #memx_func_train_loop_outer
+
+#endif
+	ret
+
 // description
 //
 // $r15 - current (memx)
@@ -307,8 +407,19 @@ memx_exec:
 // $r11 - data1
 // $r0  - zero
 memx_info:
+	cmp b16 $r12 0x1
+	bra e #memx_info_train
+
+	memx_info_data:
 	mov $r12 #memx_data_head
 	mov $r11 #memx_data_tail - #memx_data_head
+	bra #memx_info_send
+
+	memx_info_train:
+	mov $r12 #memx_train_head
+	mov $r11 #memx_train_tail - #memx_train_head
+
+	memx_info_send:
 	call(send)
 	ret
 

drivers/gpu/drm/nouveau/core/subdev/pwr/fuc/os.h

@@ -18,6 +18,10 @@
 #define MEMX_MSG_INFO 0
 #define MEMX_MSG_EXEC 1
 
+/* MEMX: info types */
+#define MEMX_INFO_DATA  0
+#define MEMX_INFO_TRAIN 1
+
 /* MEMX: script opcode definitions */
 #define MEMX_ENTER  1
 #define MEMX_LEAVE  2
@@ -25,6 +29,7 @@
 #define MEMX_WAIT   4
 #define MEMX_DELAY  5
 #define MEMX_VBLANK 6
+#define MEMX_TRAIN  7
 
 /* I2C_: message identifiers */
 #define I2C__MSG_RD08 0

drivers/gpu/drm/nouveau/core/subdev/pwr/memx.c

@@ -47,7 +47,8 @@ nouveau_memx_init(struct nouveau_pwr *ppwr, struct nouveau_memx **pmemx)
 	u32 reply[2];
 	int ret;
 
-	ret = ppwr->message(ppwr, reply, PROC_MEMX, MEMX_MSG_INFO, 0, 0);
+	ret = ppwr->message(ppwr, reply, PROC_MEMX, MEMX_MSG_INFO,
+					MEMX_INFO_DATA, 0);
 	if (ret)
 		return ret;
 
@@ -151,6 +152,38 @@ nouveau_memx_wait_vblank(struct nouveau_memx *memx)
 	memx_out(memx); /* fuc can't handle multiple */
 }
 
+void
+nouveau_memx_train(struct nouveau_memx *memx)
+{
+	nv_debug(memx->ppwr, "   MEM TRAIN\n");
+	memx_cmd(memx, MEMX_TRAIN, 0, NULL);
+}
+
+int
+nouveau_memx_train_result(struct nouveau_pwr *ppwr, u32 *res, int rsize)
+{
+	u32 reply[2], base, size, i;
+	int ret;
+
+	ret = ppwr->message(ppwr, reply, PROC_MEMX, MEMX_MSG_INFO,
+					MEMX_INFO_TRAIN, 0);
+	if (ret)
+		return ret;
+
+	base = reply[0];
+	size = reply[1] >> 2;
+	if (size > rsize)
+		return -ENOMEM;
+
+	/* read the packet */
+	nv_wr32(ppwr, 0x10a1c0, 0x02000000 | base);
+
+	for (i = 0; i < size; i++)
+		res[i] = nv_rd32(ppwr, 0x10a1c4);
+
+	return 0;
+}
+
 void
 nouveau_memx_block(struct nouveau_memx *memx)
 {