drm/radeon/kms: add support for MC ucode loading on SI

Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
Signed-off-by: Dave Airlie <airlied@redhat.com>

2 files changed, 211 insertions, 0 deletions

diff --git a/drivers/gpu/drm/radeon/si.c b/drivers/gpu/drm/radeon/si.c
index 0938c21e792d..a8789de8e6db 100644
--- a/drivers/gpu/drm/radeon/si.c
+++ b/drivers/gpu/drm/radeon/si.c
@@ -77,6 +77,201 @@ int si_get_temp(struct radeon_device *rdev)
 	return actual_temp;
 }
 
+#define TAHITI_IO_MC_REGS_SIZE 36
+
+static const u32 tahiti_io_mc_regs[TAHITI_IO_MC_REGS_SIZE][2] = {
+	{0x0000006f, 0x03044000},
+	{0x00000070, 0x0480c018},
+	{0x00000071, 0x00000040},
+	{0x00000072, 0x01000000},
+	{0x00000074, 0x000000ff},
+	{0x00000075, 0x00143400},
+	{0x00000076, 0x08ec0800},
+	{0x00000077, 0x040000cc},
+	{0x00000079, 0x00000000},
+	{0x0000007a, 0x21000409},
+	{0x0000007c, 0x00000000},
+	{0x0000007d, 0xe8000000},
+	{0x0000007e, 0x044408a8},
+	{0x0000007f, 0x00000003},
+	{0x00000080, 0x00000000},
+	{0x00000081, 0x01000000},
+	{0x00000082, 0x02000000},
+	{0x00000083, 0x00000000},
+	{0x00000084, 0xe3f3e4f4},
+	{0x00000085, 0x00052024},
+	{0x00000087, 0x00000000},
+	{0x00000088, 0x66036603},
+	{0x00000089, 0x01000000},
+	{0x0000008b, 0x1c0a0000},
+	{0x0000008c, 0xff010000},
+	{0x0000008e, 0xffffefff},
+	{0x0000008f, 0xfff3efff},
+	{0x00000090, 0xfff3efbf},
+	{0x00000094, 0x00101101},
+	{0x00000095, 0x00000fff},
+	{0x00000096, 0x00116fff},
+	{0x00000097, 0x60010000},
+	{0x00000098, 0x10010000},
+	{0x00000099, 0x00006000},
+	{0x0000009a, 0x00001000},
+	{0x0000009f, 0x00a77400}
+};
+
+static const u32 pitcairn_io_mc_regs[TAHITI_IO_MC_REGS_SIZE][2] = {
+	{0x0000006f, 0x03044000},
+	{0x00000070, 0x0480c018},
+	{0x00000071, 0x00000040},
+	{0x00000072, 0x01000000},
+	{0x00000074, 0x000000ff},
+	{0x00000075, 0x00143400},
+	{0x00000076, 0x08ec0800},
+	{0x00000077, 0x040000cc},
+	{0x00000079, 0x00000000},
+	{0x0000007a, 0x21000409},
+	{0x0000007c, 0x00000000},
+	{0x0000007d, 0xe8000000},
+	{0x0000007e, 0x044408a8},
+	{0x0000007f, 0x00000003},
+	{0x00000080, 0x00000000},
+	{0x00000081, 0x01000000},
+	{0x00000082, 0x02000000},
+	{0x00000083, 0x00000000},
+	{0x00000084, 0xe3f3e4f4},
+	{0x00000085, 0x00052024},
+	{0x00000087, 0x00000000},
+	{0x00000088, 0x66036603},
+	{0x00000089, 0x01000000},
+	{0x0000008b, 0x1c0a0000},
+	{0x0000008c, 0xff010000},
+	{0x0000008e, 0xffffefff},
+	{0x0000008f, 0xfff3efff},
+	{0x00000090, 0xfff3efbf},
+	{0x00000094, 0x00101101},
+	{0x00000095, 0x00000fff},
+	{0x00000096, 0x00116fff},
+	{0x00000097, 0x60010000},
+	{0x00000098, 0x10010000},
+	{0x00000099, 0x00006000},
+	{0x0000009a, 0x00001000},
+	{0x0000009f, 0x00a47400}
+};
+
+static const u32 verde_io_mc_regs[TAHITI_IO_MC_REGS_SIZE][2] = {
+	{0x0000006f, 0x03044000},
+	{0x00000070, 0x0480c018},
+	{0x00000071, 0x00000040},
+	{0x00000072, 0x01000000},
+	{0x00000074, 0x000000ff},
+	{0x00000075, 0x00143400},
+	{0x00000076, 0x08ec0800},
+	{0x00000077, 0x040000cc},
+	{0x00000079, 0x00000000},
+	{0x0000007a, 0x21000409},
+	{0x0000007c, 0x00000000},
+	{0x0000007d, 0xe8000000},
+	{0x0000007e, 0x044408a8},
+	{0x0000007f, 0x00000003},
+	{0x00000080, 0x00000000},
+	{0x00000081, 0x01000000},
+	{0x00000082, 0x02000000},
+	{0x00000083, 0x00000000},
+	{0x00000084, 0xe3f3e4f4},
+	{0x00000085, 0x00052024},
+	{0x00000087, 0x00000000},
+	{0x00000088, 0x66036603},
+	{0x00000089, 0x01000000},
+	{0x0000008b, 0x1c0a0000},
+	{0x0000008c, 0xff010000},
+	{0x0000008e, 0xffffefff},
+	{0x0000008f, 0xfff3efff},
+	{0x00000090, 0xfff3efbf},
+	{0x00000094, 0x00101101},
+	{0x00000095, 0x00000fff},
+	{0x00000096, 0x00116fff},
+	{0x00000097, 0x60010000},
+	{0x00000098, 0x10010000},
+	{0x00000099, 0x00006000},
+	{0x0000009a, 0x00001000},
+	{0x0000009f, 0x00a37400}
+};
+
+/* ucode loading */
+static int si_mc_load_microcode(struct radeon_device *rdev)
+{
+	const __be32 *fw_data;
+	u32 running, blackout = 0;
+	u32 *io_mc_regs;
+	int i, ucode_size, regs_size;
+
+	if (!rdev->mc_fw)
+		return -EINVAL;
+
+	switch (rdev->family) {
+	case CHIP_TAHITI:
+		io_mc_regs = (u32 *)&tahiti_io_mc_regs;
+		ucode_size = SI_MC_UCODE_SIZE;
+		regs_size = TAHITI_IO_MC_REGS_SIZE;
+		break;
+	case CHIP_PITCAIRN:
+		io_mc_regs = (u32 *)&pitcairn_io_mc_regs;
+		ucode_size = SI_MC_UCODE_SIZE;
+		regs_size = TAHITI_IO_MC_REGS_SIZE;
+		break;
+	case CHIP_VERDE:
+	default:
+		io_mc_regs = (u32 *)&verde_io_mc_regs;
+		ucode_size = SI_MC_UCODE_SIZE;
+		regs_size = TAHITI_IO_MC_REGS_SIZE;
+		break;
+	}
+
+	running = RREG32(MC_SEQ_SUP_CNTL) & RUN_MASK;
+
+	if (running == 0) {
+		if (running) {
+			blackout = RREG32(MC_SHARED_BLACKOUT_CNTL);
+			WREG32(MC_SHARED_BLACKOUT_CNTL, blackout | 1);
+		}
+
+		/* reset the engine and set to writable */
+		WREG32(MC_SEQ_SUP_CNTL, 0x00000008);
+		WREG32(MC_SEQ_SUP_CNTL, 0x00000010);
+
+		/* load mc io regs */
+		for (i = 0; i < regs_size; i++) {
+			WREG32(MC_SEQ_IO_DEBUG_INDEX, io_mc_regs[(i << 1)]);
+			WREG32(MC_SEQ_IO_DEBUG_DATA, io_mc_regs[(i << 1) + 1]);
+		}
+		/* load the MC ucode */
+		fw_data = (const __be32 *)rdev->mc_fw->data;
+		for (i = 0; i < ucode_size; i++)
+			WREG32(MC_SEQ_SUP_PGM, be32_to_cpup(fw_data++));
+
+		/* put the engine back into the active state */
+		WREG32(MC_SEQ_SUP_CNTL, 0x00000008);
+		WREG32(MC_SEQ_SUP_CNTL, 0x00000004);
+		WREG32(MC_SEQ_SUP_CNTL, 0x00000001);
+
+		/* wait for training to complete */
+		for (i = 0; i < rdev->usec_timeout; i++) {
+			if (RREG32(MC_SEQ_TRAIN_WAKEUP_CNTL) & TRAIN_DONE_D0)
+				break;
+			udelay(1);
+		}
+		for (i = 0; i < rdev->usec_timeout; i++) {
+			if (RREG32(MC_SEQ_TRAIN_WAKEUP_CNTL) & TRAIN_DONE_D1)
+				break;
+			udelay(1);
+		}
+
+		if (running)
+			WREG32(MC_SHARED_BLACKOUT_CNTL, blackout);
+	}
+
+	return 0;
+}
+
 static int si_init_microcode(struct radeon_device *rdev)
 {
 	struct platform_device *pdev;
diff --git a/drivers/gpu/drm/radeon/sid.h b/drivers/gpu/drm/radeon/sid.h
index ed1b1e5f5dd5..3cc25dbf8efd 100644
--- a/drivers/gpu/drm/radeon/sid.h
+++ b/drivers/gpu/drm/radeon/sid.h
@@ -142,6 +142,8 @@
 #define		SYSTEM_APERTURE_UNMAPPED_ACCESS_PASS_THRU	(0 << 5)
 #define		ENABLE_ADVANCED_DRIVER_MODEL			(1 << 6)
 
+#define MC_SHARED_BLACKOUT_CNTL           		0x20ac
+
 #define	MC_ARB_RAMCFG					0x2760
 #define		NOOFBANK_SHIFT					0
 #define		NOOFBANK_MASK					0x00000003
@@ -157,6 +159,20 @@
 #define		NOOFGROUPS_SHIFT				12
 #define		NOOFGROUPS_MASK					0x00001000
 
+#define	MC_SEQ_TRAIN_WAKEUP_CNTL			0x2808
+#define		TRAIN_DONE_D0      			(1 << 30)
+#define		TRAIN_DONE_D1      			(1 << 31)
+
+#define MC_SEQ_SUP_CNTL           			0x28c8
+#define		RUN_MASK      				(1 << 0)
+#define MC_SEQ_SUP_PGM           			0x28cc
+
+#define MC_IO_PAD_CNTL_D0           			0x29d0
+#define		MEM_FALL_OUT_CMD      			(1 << 8)
+
+#define MC_SEQ_IO_DEBUG_INDEX           		0x2a44
+#define MC_SEQ_IO_DEBUG_DATA           			0x2a48
+
 #define	HDP_HOST_PATH_CNTL				0x2C00
 #define	HDP_NONSURFACE_BASE				0x2C04
 #define	HDP_NONSURFACE_INFO				0x2C08