drm/radeon: add support for ASPM on SI asics (v2)

Enables PCIE ASPM (Active State Power Management) on
SI asics.

v2: fix typo

Signed-off-by: Alex Deucher <alexander.deucher@amd.com>

2 files changed, 314 insertions, 0 deletions

diff --git a/drivers/gpu/drm/radeon/si.c b/drivers/gpu/drm/radeon/si.c
index b073b2caf54d..9fd0bc379f36 100644
--- a/drivers/gpu/drm/radeon/si.c
+++ b/drivers/gpu/drm/radeon/si.c
@@ -67,6 +67,7 @@ MODULE_FIRMWARE("radeon/HAINAN_mc.bin");
 MODULE_FIRMWARE("radeon/HAINAN_rlc.bin");
 
 static void si_pcie_gen3_enable(struct radeon_device *rdev);
+static void si_program_aspm(struct radeon_device *rdev);
 extern int r600_ih_ring_alloc(struct radeon_device *rdev);
 extern void r600_ih_ring_fini(struct radeon_device *rdev);
 extern void evergreen_fix_pci_max_read_req_size(struct radeon_device *rdev);
@@ -5319,6 +5320,8 @@ static int si_startup(struct radeon_device *rdev)
 
 	/* enable pcie gen2/3 link */
 	si_pcie_gen3_enable(rdev);
+	/* enable aspm */
+	si_program_aspm(rdev);
 
 	if (!rdev->me_fw || !rdev->pfp_fw || !rdev->ce_fw ||
 	    !rdev->rlc_fw || !rdev->mc_fw) {
@@ -5943,3 +5946,203 @@ static void si_pcie_gen3_enable(struct radeon_device *rdev)
 	}
 }
 
+static void si_program_aspm(struct radeon_device *rdev)
+{
+	u32 data, orig;
+	bool disable_l0s = false, disable_l1 = false, disable_plloff_in_l1 = false;
+	bool disable_clkreq = false;
+
+	if (!(rdev->flags & RADEON_IS_PCIE))
+		return;
+
+	orig = data = RREG32_PCIE_PORT(PCIE_LC_N_FTS_CNTL);
+	data &= ~LC_XMIT_N_FTS_MASK;
+	data |= LC_XMIT_N_FTS(0x24) | LC_XMIT_N_FTS_OVERRIDE_EN;
+	if (orig != data)
+		WREG32_PCIE_PORT(PCIE_LC_N_FTS_CNTL, data);
+
+	orig = data = RREG32_PCIE_PORT(PCIE_LC_CNTL3);
+	data |= LC_GO_TO_RECOVERY;
+	if (orig != data)
+		WREG32_PCIE_PORT(PCIE_LC_CNTL3, data);
+
+	orig = data = RREG32_PCIE(PCIE_P_CNTL);
+	data |= P_IGNORE_EDB_ERR;
+	if (orig != data)
+		WREG32_PCIE(PCIE_P_CNTL, data);
+
+	orig = data = RREG32_PCIE_PORT(PCIE_LC_CNTL);
+	data &= ~(LC_L0S_INACTIVITY_MASK | LC_L1_INACTIVITY_MASK);
+	data |= LC_PMI_TO_L1_DIS;
+	if (!disable_l0s)
+		data |= LC_L0S_INACTIVITY(7);
+
+	if (!disable_l1) {
+		data |= LC_L1_INACTIVITY(7);
+		data &= ~LC_PMI_TO_L1_DIS;
+		if (orig != data)
+			WREG32_PCIE_PORT(PCIE_LC_CNTL, data);
+
+		if (!disable_plloff_in_l1) {
+			bool clk_req_support;
+
+			orig = data = RREG32_PIF_PHY0(PB0_PIF_PWRDOWN_0);
+			data &= ~(PLL_POWER_STATE_IN_OFF_0_MASK | PLL_POWER_STATE_IN_TXS2_0_MASK);
+			data |= PLL_POWER_STATE_IN_OFF_0(7) | PLL_POWER_STATE_IN_TXS2_0(7);
+			if (orig != data)
+				WREG32_PIF_PHY0(PB0_PIF_PWRDOWN_0, data);
+
+			orig = data = RREG32_PIF_PHY0(PB0_PIF_PWRDOWN_1);
+			data &= ~(PLL_POWER_STATE_IN_OFF_1_MASK | PLL_POWER_STATE_IN_TXS2_1_MASK);
+			data |= PLL_POWER_STATE_IN_OFF_1(7) | PLL_POWER_STATE_IN_TXS2_1(7);
+			if (orig != data)
+				WREG32_PIF_PHY0(PB0_PIF_PWRDOWN_1, data);
+
+			orig = data = RREG32_PIF_PHY1(PB1_PIF_PWRDOWN_0);
+			data &= ~(PLL_POWER_STATE_IN_OFF_0_MASK | PLL_POWER_STATE_IN_TXS2_0_MASK);
+			data |= PLL_POWER_STATE_IN_OFF_0(7) | PLL_POWER_STATE_IN_TXS2_0(7);
+			if (orig != data)
+				WREG32_PIF_PHY1(PB1_PIF_PWRDOWN_0, data);
+
+			orig = data = RREG32_PIF_PHY1(PB1_PIF_PWRDOWN_1);
+			data &= ~(PLL_POWER_STATE_IN_OFF_1_MASK | PLL_POWER_STATE_IN_TXS2_1_MASK);
+			data |= PLL_POWER_STATE_IN_OFF_1(7) | PLL_POWER_STATE_IN_TXS2_1(7);
+			if (orig != data)
+				WREG32_PIF_PHY1(PB1_PIF_PWRDOWN_1, data);
+
+			if ((rdev->family != CHIP_OLAND) && (rdev->family != CHIP_HAINAN)) {
+				orig = data = RREG32_PIF_PHY0(PB0_PIF_PWRDOWN_0);
+				data &= ~PLL_RAMP_UP_TIME_0_MASK;
+				if (orig != data)
+					WREG32_PIF_PHY0(PB0_PIF_PWRDOWN_0, data);
+
+				orig = data = RREG32_PIF_PHY0(PB0_PIF_PWRDOWN_1);
+				data &= ~PLL_RAMP_UP_TIME_1_MASK;
+				if (orig != data)
+					WREG32_PIF_PHY0(PB0_PIF_PWRDOWN_1, data);
+
+				orig = data = RREG32_PIF_PHY0(PB0_PIF_PWRDOWN_2);
+				data &= ~PLL_RAMP_UP_TIME_2_MASK;
+				if (orig != data)
+					WREG32_PIF_PHY0(PB0_PIF_PWRDOWN_2, data);
+
+				orig = data = RREG32_PIF_PHY0(PB0_PIF_PWRDOWN_3);
+				data &= ~PLL_RAMP_UP_TIME_3_MASK;
+				if (orig != data)
+					WREG32_PIF_PHY0(PB0_PIF_PWRDOWN_3, data);
+
+				orig = data = RREG32_PIF_PHY1(PB1_PIF_PWRDOWN_0);
+				data &= ~PLL_RAMP_UP_TIME_0_MASK;
+				if (orig != data)
+					WREG32_PIF_PHY1(PB1_PIF_PWRDOWN_0, data);
+
+				orig = data = RREG32_PIF_PHY1(PB1_PIF_PWRDOWN_1);
+				data &= ~PLL_RAMP_UP_TIME_1_MASK;
+				if (orig != data)
+					WREG32_PIF_PHY1(PB1_PIF_PWRDOWN_1, data);
+
+				orig = data = RREG32_PIF_PHY1(PB1_PIF_PWRDOWN_2);
+				data &= ~PLL_RAMP_UP_TIME_2_MASK;
+				if (orig != data)
+					WREG32_PIF_PHY1(PB1_PIF_PWRDOWN_2, data);
+
+				orig = data = RREG32_PIF_PHY1(PB1_PIF_PWRDOWN_3);
+				data &= ~PLL_RAMP_UP_TIME_3_MASK;
+				if (orig != data)
+					WREG32_PIF_PHY1(PB1_PIF_PWRDOWN_3, data);
+			}
+			orig = data = RREG32_PCIE_PORT(PCIE_LC_LINK_WIDTH_CNTL);
+			data &= ~LC_DYN_LANES_PWR_STATE_MASK;
+			data |= LC_DYN_LANES_PWR_STATE(3);
+			if (orig != data)
+				WREG32_PCIE_PORT(PCIE_LC_LINK_WIDTH_CNTL, data);
+
+			orig = data = RREG32_PIF_PHY0(PB0_PIF_CNTL);
+			data &= ~LS2_EXIT_TIME_MASK;
+			if ((rdev->family == CHIP_OLAND) || (rdev->family == CHIP_HAINAN))
+				data |= LS2_EXIT_TIME(5);
+			if (orig != data)
+				WREG32_PIF_PHY0(PB0_PIF_CNTL, data);
+
+			orig = data = RREG32_PIF_PHY1(PB1_PIF_CNTL);
+			data &= ~LS2_EXIT_TIME_MASK;
+			if ((rdev->family == CHIP_OLAND) || (rdev->family == CHIP_HAINAN))
+				data |= LS2_EXIT_TIME(5);
+			if (orig != data)
+				WREG32_PIF_PHY1(PB1_PIF_CNTL, data);
+
+			if (!disable_clkreq) {
+				struct pci_dev *root = rdev->pdev->bus->self;
+				u32 lnkcap;
+
+				clk_req_support = false;
+				pcie_capability_read_dword(root, PCI_EXP_LNKCAP, &lnkcap);
+				if (lnkcap & PCI_EXP_LNKCAP_CLKPM)
+					clk_req_support = true;
+			} else {
+				clk_req_support = false;
+			}
+
+			if (clk_req_support) {
+				orig = data = RREG32_PCIE_PORT(PCIE_LC_CNTL2);
+				data |= LC_ALLOW_PDWN_IN_L1 | LC_ALLOW_PDWN_IN_L23;
+				if (orig != data)
+					WREG32_PCIE_PORT(PCIE_LC_CNTL2, data);
+
+				orig = data = RREG32(THM_CLK_CNTL);
+				data &= ~(CMON_CLK_SEL_MASK | TMON_CLK_SEL_MASK);
+				data |= CMON_CLK_SEL(1) | TMON_CLK_SEL(1);
+				if (orig != data)
+					WREG32(THM_CLK_CNTL, data);
+
+				orig = data = RREG32(MISC_CLK_CNTL);
+				data &= ~(DEEP_SLEEP_CLK_SEL_MASK | ZCLK_SEL_MASK);
+				data |= DEEP_SLEEP_CLK_SEL(1) | ZCLK_SEL(1);
+				if (orig != data)
+					WREG32(MISC_CLK_CNTL, data);
+
+				orig = data = RREG32(CG_CLKPIN_CNTL);
+				data &= ~BCLK_AS_XCLK;
+				if (orig != data)
+					WREG32(CG_CLKPIN_CNTL, data);
+
+				orig = data = RREG32(CG_CLKPIN_CNTL_2);
+				data &= ~FORCE_BIF_REFCLK_EN;
+				if (orig != data)
+					WREG32(CG_CLKPIN_CNTL_2, data);
+
+				orig = data = RREG32(MPLL_BYPASSCLK_SEL);
+				data &= ~MPLL_CLKOUT_SEL_MASK;
+				data |= MPLL_CLKOUT_SEL(4);
+				if (orig != data)
+					WREG32(MPLL_BYPASSCLK_SEL, data);
+
+				orig = data = RREG32(SPLL_CNTL_MODE);
+				data &= ~SPLL_REFCLK_SEL_MASK;
+				if (orig != data)
+					WREG32(SPLL_CNTL_MODE, data);
+			}
+		}
+	} else {
+		if (orig != data)
+			WREG32_PCIE_PORT(PCIE_LC_CNTL, data);
+	}
+
+	orig = data = RREG32_PCIE(PCIE_CNTL2);
+	data |= SLV_MEM_LS_EN | MST_MEM_LS_EN | REPLAY_MEM_LS_EN;
+	if (orig != data)
+		WREG32_PCIE(PCIE_CNTL2, data);
+
+	if (!disable_l0s) {
+		data = RREG32_PCIE_PORT(PCIE_LC_N_FTS_CNTL);
+		if((data & LC_N_FTS_MASK) == LC_N_FTS_MASK) {
+			data = RREG32_PCIE(PCIE_LC_STATUS1);
+			if ((data & LC_REVERSE_XMIT) && (data & LC_REVERSE_RCVR)) {
+				orig = data = RREG32_PCIE_PORT(PCIE_LC_CNTL);
+				data &= ~LC_L0S_INACTIVITY_MASK;
+				if (orig != data)
+					WREG32_PCIE_PORT(PCIE_LC_CNTL, data);
+			}
+		}
+	}
+}
diff --git a/drivers/gpu/drm/radeon/sid.h b/drivers/gpu/drm/radeon/sid.h
index 6d4bdbc797a4..5f29d81d8db2 100644
--- a/drivers/gpu/drm/radeon/sid.h
+++ b/drivers/gpu/drm/radeon/sid.h
@@ -88,11 +88,32 @@
 #define VGA_HDP_CONTROL  				0x328
 #define		VGA_MEMORY_DISABLE				(1 << 4)
 
+#define	SPLL_CNTL_MODE					0x618
+#	define SPLL_REFCLK_SEL(x)			((x) << 8)
+#	define SPLL_REFCLK_SEL_MASK			0xFF00
+
+#define	MPLL_BYPASSCLK_SEL				0x65c
+#	define MPLL_CLKOUT_SEL(x)			((x) << 8)
+#	define MPLL_CLKOUT_SEL_MASK			0xFF00
+
 #define CG_CLKPIN_CNTL                                    0x660
 #       define XTALIN_DIVIDE                              (1 << 1)
+#       define BCLK_AS_XCLK                               (1 << 2)
 #define CG_CLKPIN_CNTL_2                                  0x664
+#       define FORCE_BIF_REFCLK_EN                        (1 << 3)
 #       define MUX_TCLK_TO_XCLK                           (1 << 8)
 
+#define	THM_CLK_CNTL					0x66c
+#	define CMON_CLK_SEL(x)				((x) << 0)
+#	define CMON_CLK_SEL_MASK			0xFF
+#	define TMON_CLK_SEL(x)				((x) << 8)
+#	define TMON_CLK_SEL_MASK			0xFF00
+#define	MISC_CLK_CNTL					0x670
+#	define DEEP_SLEEP_CLK_SEL(x)			((x) << 0)
+#	define DEEP_SLEEP_CLK_SEL_MASK			0xFF
+#	define ZCLK_SEL(x)				((x) << 8)
+#	define ZCLK_SEL_MASK				0xFF00
+
 #define DMIF_ADDR_CONFIG  				0xBD4
 
 #define DMIF_ADDR_CALC  				0xC00
@@ -829,14 +850,88 @@
 #       define THREAD_TRACE_FLUSH                       (54 << 0)
 #       define THREAD_TRACE_FINISH                      (55 << 0)
 
+/* PIF PHY0 registers idx/data 0x8/0xc */
+#define PB0_PIF_CNTL                                      0x10
+#       define LS2_EXIT_TIME(x)                           ((x) << 17)
+#       define LS2_EXIT_TIME_MASK                         (0x7 << 17)
+#       define LS2_EXIT_TIME_SHIFT                        17
+#define PB0_PIF_PAIRING                                   0x11
+#       define MULTI_PIF                                  (1 << 25)
+#define PB0_PIF_PWRDOWN_0                                 0x12
+#       define PLL_POWER_STATE_IN_TXS2_0(x)               ((x) << 7)
+#       define PLL_POWER_STATE_IN_TXS2_0_MASK             (0x7 << 7)
+#       define PLL_POWER_STATE_IN_TXS2_0_SHIFT            7
+#       define PLL_POWER_STATE_IN_OFF_0(x)                ((x) << 10)
+#       define PLL_POWER_STATE_IN_OFF_0_MASK              (0x7 << 10)
+#       define PLL_POWER_STATE_IN_OFF_0_SHIFT             10
+#       define PLL_RAMP_UP_TIME_0(x)                      ((x) << 24)
+#       define PLL_RAMP_UP_TIME_0_MASK                    (0x7 << 24)
+#       define PLL_RAMP_UP_TIME_0_SHIFT                   24
+#define PB0_PIF_PWRDOWN_1                                 0x13
+#       define PLL_POWER_STATE_IN_TXS2_1(x)               ((x) << 7)
+#       define PLL_POWER_STATE_IN_TXS2_1_MASK             (0x7 << 7)
+#       define PLL_POWER_STATE_IN_TXS2_1_SHIFT            7
+#       define PLL_POWER_STATE_IN_OFF_1(x)                ((x) << 10)
+#       define PLL_POWER_STATE_IN_OFF_1_MASK              (0x7 << 10)
+#       define PLL_POWER_STATE_IN_OFF_1_SHIFT             10
+#       define PLL_RAMP_UP_TIME_1(x)                      ((x) << 24)
+#       define PLL_RAMP_UP_TIME_1_MASK                    (0x7 << 24)
+#       define PLL_RAMP_UP_TIME_1_SHIFT                   24
+
+#define PB0_PIF_PWRDOWN_2                                 0x17
+#       define PLL_POWER_STATE_IN_TXS2_2(x)               ((x) << 7)
+#       define PLL_POWER_STATE_IN_TXS2_2_MASK             (0x7 << 7)
+#       define PLL_POWER_STATE_IN_TXS2_2_SHIFT            7
+#       define PLL_POWER_STATE_IN_OFF_2(x)                ((x) << 10)
+#       define PLL_POWER_STATE_IN_OFF_2_MASK              (0x7 << 10)
+#       define PLL_POWER_STATE_IN_OFF_2_SHIFT             10
+#       define PLL_RAMP_UP_TIME_2(x)                      ((x) << 24)
+#       define PLL_RAMP_UP_TIME_2_MASK                    (0x7 << 24)
+#       define PLL_RAMP_UP_TIME_2_SHIFT                   24
+#define PB0_PIF_PWRDOWN_3                                 0x18
+#       define PLL_POWER_STATE_IN_TXS2_3(x)               ((x) << 7)
+#       define PLL_POWER_STATE_IN_TXS2_3_MASK             (0x7 << 7)
+#       define PLL_POWER_STATE_IN_TXS2_3_SHIFT            7
+#       define PLL_POWER_STATE_IN_OFF_3(x)                ((x) << 10)
+#       define PLL_POWER_STATE_IN_OFF_3_MASK              (0x7 << 10)
+#       define PLL_POWER_STATE_IN_OFF_3_SHIFT             10
+#       define PLL_RAMP_UP_TIME_3(x)                      ((x) << 24)
+#       define PLL_RAMP_UP_TIME_3_MASK                    (0x7 << 24)
+#       define PLL_RAMP_UP_TIME_3_SHIFT                   24
+/* PIF PHY1 registers idx/data 0x10/0x14 */
+#define PB1_PIF_CNTL                                      0x10
+#define PB1_PIF_PAIRING                                   0x11
+#define PB1_PIF_PWRDOWN_0                                 0x12
+#define PB1_PIF_PWRDOWN_1                                 0x13
+
+#define PB1_PIF_PWRDOWN_2                                 0x17
+#define PB1_PIF_PWRDOWN_3                                 0x18
 /* PCIE registers idx/data 0x30/0x34 */
+#define PCIE_CNTL2                                        0x1c /* PCIE */
+#       define SLV_MEM_LS_EN                              (1 << 16)
+#       define MST_MEM_LS_EN                              (1 << 18)
+#       define REPLAY_MEM_LS_EN                           (1 << 19)
 #define PCIE_LC_STATUS1                                   0x28 /* PCIE */
+#       define LC_REVERSE_RCVR                            (1 << 0)
+#       define LC_REVERSE_XMIT                            (1 << 1)
 #       define LC_OPERATING_LINK_WIDTH_MASK               (0x7 << 2)
 #       define LC_OPERATING_LINK_WIDTH_SHIFT              2
 #       define LC_DETECTED_LINK_WIDTH_MASK                (0x7 << 5)
 #       define LC_DETECTED_LINK_WIDTH_SHIFT               5
 
+#define PCIE_P_CNTL                                       0x40 /* PCIE */
+#       define P_IGNORE_EDB_ERR                           (1 << 6)
+
 /* PCIE PORT registers idx/data 0x38/0x3c */
+#define PCIE_LC_CNTL                                      0xa0
+#       define LC_L0S_INACTIVITY(x)                       ((x) << 8)
+#       define LC_L0S_INACTIVITY_MASK                     (0xf << 8)
+#       define LC_L0S_INACTIVITY_SHIFT                    8
+#       define LC_L1_INACTIVITY(x)                        ((x) << 12)
+#       define LC_L1_INACTIVITY_MASK                      (0xf << 12)
+#       define LC_L1_INACTIVITY_SHIFT                     12
+#       define LC_PMI_TO_L1_DIS                           (1 << 16)
+#       define LC_ASPM_TO_L1_DIS                          (1 << 24)
 #define PCIE_LC_LINK_WIDTH_CNTL                           0xa2 /* PCIE_P */
 #       define LC_LINK_WIDTH_SHIFT                        0
 #       define LC_LINK_WIDTH_MASK                         0x7
@@ -855,6 +950,15 @@
 #       define LC_SHORT_RECONFIG_EN                       (1 << 11)
 #       define LC_UPCONFIGURE_SUPPORT                     (1 << 12)
 #       define LC_UPCONFIGURE_DIS                         (1 << 13)
+#       define LC_DYN_LANES_PWR_STATE(x)                  ((x) << 21)
+#       define LC_DYN_LANES_PWR_STATE_MASK                (0x3 << 21)
+#       define LC_DYN_LANES_PWR_STATE_SHIFT               21
+#define PCIE_LC_N_FTS_CNTL                                0xa3 /* PCIE_P */
+#       define LC_XMIT_N_FTS(x)                           ((x) << 0)
+#       define LC_XMIT_N_FTS_MASK                         (0xff << 0)
+#       define LC_XMIT_N_FTS_SHIFT                        0
+#       define LC_XMIT_N_FTS_OVERRIDE_EN                  (1 << 8)
+#       define LC_N_FTS_MASK                              (0xff << 24)
 #define PCIE_LC_SPEED_CNTL                                0xa4 /* PCIE_P */
 #       define LC_GEN2_EN_STRAP                           (1 << 0)
 #       define LC_GEN3_EN_STRAP                           (1 << 1)
@@ -875,6 +979,13 @@
 #       define LC_OTHER_SIDE_SUPPORTS_GEN2                (1 << 19)
 #       define LC_OTHER_SIDE_EVER_SENT_GEN3               (1 << 20)
 #       define LC_OTHER_SIDE_SUPPORTS_GEN3                (1 << 21)
+
+#define PCIE_LC_CNTL2                                     0xb1
+#       define LC_ALLOW_PDWN_IN_L1                        (1 << 17)
+#       define LC_ALLOW_PDWN_IN_L23                       (1 << 18)
+
+#define PCIE_LC_CNTL3                                     0xb5 /* PCIE_P */
+#       define LC_GO_TO_RECOVERY                          (1 << 30)
 #define PCIE_LC_CNTL4                                     0xb6 /* PCIE_P */
 #       define LC_REDO_EQ                                 (1 << 5)
 #       define LC_SET_QUIESCE                             (1 << 13)