Skip to contentSkip to conversation

drm/amdgpu: correct tcp harvest setting

PATCH In review Evan Quan 3 participants 14 messages Jun 21, 2021 – Jun 23, 2021

Subsystems (2)
lore
Share & tools
lore

Evan Quan

author
#0 lore raw

PATCH v3 1/7 drm/amdgpu: correct tcp harvest setting

Add missing settings for SQC bits. And correct some confusing logics around active wgp bitmap calculation.

Change-Id If4992e175fd61d5609b00328cbe21f487517d039

---
V1->V2:
- restore correct tcp_harvest setting for NV10 and NV12
- move asic type guard upper layer for better readability

---
 drivers/gpu/drm/amd/amdgpu/gfx_v10_0.c | 103 ++++++++++++++-----------
 1 file changed, 57 insertions(+), 46 deletions(-)
drivers/gpu/drm/amd/amdgpu/gfx_v10_0.c +57 -46
index 15ae9e33b925..384b95fbad8b 100644
@@ -5090,47 +5090,50 @@ static void gfx_v10_0_tcp_harvest(struct amdgpu_device *adev)
 		4 + /* RMI */
 		1); /* SQG */
 
-	if (adev->asic_type == CHIP_NAVI10 ||
-	    adev->asic_type == CHIP_NAVI14 ||
-	    adev->asic_type == CHIP_NAVI12) {
-		mutex_lock(&adev->grbm_idx_mutex);
-		for (i = 0; i < adev->gfx.config.max_shader_engines; i++) {
-			for (j = 0; j < adev->gfx.config.max_sh_per_se; j++) {
-				gfx_v10_0_select_se_sh(adev, i, j, 0xffffffff);
-				wgp_active_bitmap = gfx_v10_0_get_wgp_active_bitmap_per_sh(adev);
-				/*
-				 * Set corresponding TCP bits for the inactive WGPs in
-				 * GCRD_SA_TARGETS_DISABLE
-				 */
-				gcrd_targets_disable_tcp = 0;
-				/* Set TCP & SQC bits in UTCL1_UTCL0_INVREQ_DISABLE */
-				utcl_invreq_disable = 0;
-
-				for (k = 0; k < max_wgp_per_sh; k++) {
-					if (!(wgp_active_bitmap & (1 << k))) {
-						gcrd_targets_disable_tcp |= 3 << (2 * k);
-						utcl_invreq_disable |= (3 << (2 * k)) |
-							(3 << (2 * (max_wgp_per_sh + k)));
-					}
+	mutex_lock(&adev->grbm_idx_mutex);
+	for (i = 0; i < adev->gfx.config.max_shader_engines; i++) {
+		for (j = 0; j < adev->gfx.config.max_sh_per_se; j++) {
+			gfx_v10_0_select_se_sh(adev, i, j, 0xffffffff);
+			wgp_active_bitmap = gfx_v10_0_get_wgp_active_bitmap_per_sh(adev);
+			/*
+			 * Set corresponding TCP bits for the inactive WGPs in
+			 * GCRD_SA_TARGETS_DISABLE
+			 */
+			gcrd_targets_disable_tcp = 0;
+			/* Set TCP & SQC bits in UTCL1_UTCL0_INVREQ_DISABLE */
+			utcl_invreq_disable = 0;
+
+			for (k = 0; k < max_wgp_per_sh; k++) {
+				if (!(wgp_active_bitmap & (1 << k))) {
+					gcrd_targets_disable_tcp |= 3 << (2 * k);
+					gcrd_targets_disable_tcp |= 1 << (k + (max_wgp_per_sh * 2));
+					utcl_invreq_disable |= (3 << (2 * k)) |
+						(3 << (2 * (max_wgp_per_sh + k)));
 				}
-
-				tmp = RREG32_SOC15(GC, 0, mmUTCL1_UTCL0_INVREQ_DISABLE);
-				/* only override TCP & SQC bits */
-				tmp &= 0xffffffff << (4 * max_wgp_per_sh);
-				tmp |= (utcl_invreq_disable & utcl_invreq_disable_mask);
-				WREG32_SOC15(GC, 0, mmUTCL1_UTCL0_INVREQ_DISABLE, tmp);
-
-				tmp = RREG32_SOC15(GC, 0, mmGCRD_SA_TARGETS_DISABLE);
-				/* only override TCP bits */
-				tmp &= 0xffffffff << (2 * max_wgp_per_sh);
-				tmp |= (gcrd_targets_disable_tcp & gcrd_targets_disable_mask);
-				WREG32_SOC15(GC, 0, mmGCRD_SA_TARGETS_DISABLE, tmp);
 			}
-		}
 
-		gfx_v10_0_select_se_sh(adev, 0xffffffff, 0xffffffff, 0xffffffff);
-		mutex_unlock(&adev->grbm_idx_mutex);
+			tmp = RREG32_SOC15(GC, 0, mmUTCL1_UTCL0_INVREQ_DISABLE);
+			/* only override TCP & SQC bits */
+			if (adev->asic_type == CHIP_NAVI14)
+				tmp &= 0xff000000;
+			else
+				tmp &=0xfff00000;
+			tmp |= (utcl_invreq_disable & utcl_invreq_disable_mask);
+			WREG32_SOC15(GC, 0, mmUTCL1_UTCL0_INVREQ_DISABLE, tmp);
+
+			tmp = RREG32_SOC15(GC, 0, mmGCRD_SA_TARGETS_DISABLE);
+			/* only override TCP & SQC bits */
+			if (adev->asic_type == CHIP_NAVI14)
+				tmp &= 0xfffc0000;
+			else
+				tmp &= 0xffff8000;
+			tmp |= (gcrd_targets_disable_tcp & gcrd_targets_disable_mask);
+			WREG32_SOC15(GC, 0, mmGCRD_SA_TARGETS_DISABLE, tmp);
+		}
 	}
+
+	gfx_v10_0_select_se_sh(adev, 0xffffffff, 0xffffffff, 0xffffffff);
+	mutex_unlock(&adev->grbm_idx_mutex);
 }
 
 static void gfx_v10_0_get_tcc_info(struct amdgpu_device *adev)
@@ -7408,7 +7411,10 @@ static int gfx_v10_0_hw_init(void *handle)
 	 * init golden registers and rlc resume may override some registers,
 	 * reconfig them here
 	 */
-	gfx_v10_0_tcp_harvest(adev);
+	if (adev->asic_type == CHIP_NAVI10 ||
+	    adev->asic_type == CHIP_NAVI14 ||
+	    adev->asic_type == CHIP_NAVI12)
+		gfx_v10_0_tcp_harvest(adev);
 
 	r = gfx_v10_0_cp_resume(adev);
 	if (r)
@@ -9328,17 +9334,22 @@ static void gfx_v10_0_set_user_wgp_inactive_bitmap_per_sh(struct amdgpu_device *
 
 static u32 gfx_v10_0_get_wgp_active_bitmap_per_sh(struct amdgpu_device *adev)
 {
-	u32 data, wgp_bitmask;
-	data = RREG32_SOC15(GC, 0, mmCC_GC_SHADER_ARRAY_CONFIG);
-	data |= RREG32_SOC15(GC, 0, mmGC_USER_SHADER_ARRAY_CONFIG);
+	u32 disabled_mask =
+		~amdgpu_gfx_create_bitmask(adev->gfx.config.max_cu_per_sh >> 1);
+	u32 efuse_setting = 0;
+	u32 vbios_setting = 0;
+
+	efuse_setting = RREG32_SOC15(GC, 0, mmCC_GC_SHADER_ARRAY_CONFIG);
+	efuse_setting &= CC_GC_SHADER_ARRAY_CONFIG__INACTIVE_WGPS_MASK;
+	efuse_setting >>= CC_GC_SHADER_ARRAY_CONFIG__INACTIVE_WGPS__SHIFT;
 
-	data &= CC_GC_SHADER_ARRAY_CONFIG__INACTIVE_WGPS_MASK;
-	data >>= CC_GC_SHADER_ARRAY_CONFIG__INACTIVE_WGPS__SHIFT;
+	vbios_setting = RREG32_SOC15(GC, 0, mmGC_USER_SHADER_ARRAY_CONFIG);
+	vbios_setting &= GC_USER_SHADER_ARRAY_CONFIG__INACTIVE_WGPS_MASK;
+	vbios_setting >>= GC_USER_SHADER_ARRAY_CONFIG__INACTIVE_WGPS__SHIFT;
 
-	wgp_bitmask =
-		amdgpu_gfx_create_bitmask(adev->gfx.config.max_cu_per_sh >> 1);
+	disabled_mask |= efuse_setting | vbios_setting;
 
-	return (~data) & wgp_bitmask;
+	return (~disabled_mask);
 }
 
 static u32 gfx_v10_0_get_cu_active_bitmap_per_sh(struct amdgpu_device *adev)
Signature · 6 lines
2.29.0

_______________________________________________
amd-gfx mailing list
[email protected]
https://lists.freedesktop.org/mailman/listinfo/amd-gfx
author
#1 lore raw

PATCH v3 2/7 drm/amdgpu: fix Navi1x tcp power gating hang when issuing lightweight invalidaiton

Fix TCP hang when a lightweight invalidation happens on Navi1x.

Change-Id I5000fefa9ec48a5e863372d298354bed1562b332

---
V1->V2:
- Alex: use ARRAY_SIZE instead of hard code
- limit the changes for Navi1x only

---
 drivers/gpu/drm/amd/amdgpu/gfx_v10_0.c | 95 ++++++++++++++++++++++++++
 1 file changed, 95 insertions(+)
drivers/gpu/drm/amd/amdgpu/gfx_v10_0.c +95 -0
index 384b95fbad8b..b170d5fb8993 100644
@@ -7971,6 +7971,97 @@ static void gfx_v10_0_update_fine_grain_clock_gating(struct amdgpu_device *adev,
 	}
 }
 
+static void gfx_v10_0_apply_medium_grain_clock_gating_workaround(struct amdgpu_device *adev)
+{
+	uint32_t reg_data = 0;
+	uint32_t reg_idx = 0;
+	uint32_t i;
+
+	const uint32_t tcp_ctrl_regs[] = {
+		mmCGTS_SA0_WGP00_CU0_TCP_CTRL_REG,
+		mmCGTS_SA0_WGP00_CU1_TCP_CTRL_REG,
+		mmCGTS_SA0_WGP01_CU0_TCP_CTRL_REG,
+		mmCGTS_SA0_WGP01_CU1_TCP_CTRL_REG,
+		mmCGTS_SA0_WGP02_CU0_TCP_CTRL_REG,
+		mmCGTS_SA0_WGP02_CU1_TCP_CTRL_REG,
+		mmCGTS_SA0_WGP10_CU0_TCP_CTRL_REG,
+		mmCGTS_SA0_WGP10_CU1_TCP_CTRL_REG,
+		mmCGTS_SA0_WGP11_CU0_TCP_CTRL_REG,
+		mmCGTS_SA0_WGP11_CU1_TCP_CTRL_REG,
+		mmCGTS_SA0_WGP12_CU0_TCP_CTRL_REG,
+		mmCGTS_SA0_WGP12_CU1_TCP_CTRL_REG,
+		mmCGTS_SA1_WGP00_CU0_TCP_CTRL_REG,
+		mmCGTS_SA1_WGP00_CU1_TCP_CTRL_REG,
+		mmCGTS_SA1_WGP01_CU0_TCP_CTRL_REG,
+		mmCGTS_SA1_WGP01_CU1_TCP_CTRL_REG,
+		mmCGTS_SA1_WGP02_CU0_TCP_CTRL_REG,
+		mmCGTS_SA1_WGP02_CU1_TCP_CTRL_REG,
+		mmCGTS_SA1_WGP10_CU0_TCP_CTRL_REG,
+		mmCGTS_SA1_WGP10_CU1_TCP_CTRL_REG,
+		mmCGTS_SA1_WGP11_CU0_TCP_CTRL_REG,
+		mmCGTS_SA1_WGP11_CU1_TCP_CTRL_REG,
+		mmCGTS_SA1_WGP12_CU0_TCP_CTRL_REG,
+		mmCGTS_SA1_WGP12_CU1_TCP_CTRL_REG
+	};
+
+	const uint32_t tcp_ctrl_regs_nv12[] = {
+		mmCGTS_SA0_WGP00_CU0_TCP_CTRL_REG,
+		mmCGTS_SA0_WGP00_CU1_TCP_CTRL_REG,
+		mmCGTS_SA0_WGP01_CU0_TCP_CTRL_REG,
+		mmCGTS_SA0_WGP01_CU1_TCP_CTRL_REG,
+		mmCGTS_SA0_WGP02_CU0_TCP_CTRL_REG,
+		mmCGTS_SA0_WGP02_CU1_TCP_CTRL_REG,
+		mmCGTS_SA0_WGP10_CU0_TCP_CTRL_REG,
+		mmCGTS_SA0_WGP10_CU1_TCP_CTRL_REG,
+		mmCGTS_SA0_WGP11_CU0_TCP_CTRL_REG,
+		mmCGTS_SA0_WGP11_CU1_TCP_CTRL_REG,
+		mmCGTS_SA1_WGP00_CU0_TCP_CTRL_REG,
+		mmCGTS_SA1_WGP00_CU1_TCP_CTRL_REG,
+		mmCGTS_SA1_WGP01_CU0_TCP_CTRL_REG,
+		mmCGTS_SA1_WGP01_CU1_TCP_CTRL_REG,
+		mmCGTS_SA1_WGP02_CU0_TCP_CTRL_REG,
+		mmCGTS_SA1_WGP02_CU1_TCP_CTRL_REG,
+		mmCGTS_SA1_WGP10_CU0_TCP_CTRL_REG,
+		mmCGTS_SA1_WGP10_CU1_TCP_CTRL_REG,
+		mmCGTS_SA1_WGP11_CU0_TCP_CTRL_REG,
+		mmCGTS_SA1_WGP11_CU1_TCP_CTRL_REG,
+	};
+
+	const uint32_t sm_ctlr_regs[] = {
+		mmCGTS_SA0_QUAD0_SM_CTRL_REG,
+		mmCGTS_SA0_QUAD1_SM_CTRL_REG,
+		mmCGTS_SA1_QUAD0_SM_CTRL_REG,
+		mmCGTS_SA1_QUAD1_SM_CTRL_REG
+	};
+
+	if (adev->asic_type == CHIP_NAVI12) {
+		for (i = 0; i < ARRAY_SIZE(tcp_ctrl_regs_nv12); i++) {
+			reg_idx = adev->reg_offset[GC_HWIP][0][mmCGTS_SA0_WGP00_CU0_TCP_CTRL_REG_BASE_IDX] +
+				  tcp_ctrl_regs_nv12[i];
+			reg_data = RREG32(reg_idx);
+			reg_data |= CGTS_SA0_WGP00_CU0_TCP_CTRL_REG__TCPI_LS_OVERRIDE_MASK;
+			WREG32(reg_idx, reg_data);
+		}
+	} else {
+		for (i = 0; i < ARRAY_SIZE(tcp_ctrl_regs); i++) {
+			reg_idx = adev->reg_offset[GC_HWIP][0][mmCGTS_SA0_WGP00_CU0_TCP_CTRL_REG_BASE_IDX] +
+				  tcp_ctrl_regs[i];
+			reg_data = RREG32(reg_idx);
+			reg_data |= CGTS_SA0_WGP00_CU0_TCP_CTRL_REG__TCPI_LS_OVERRIDE_MASK;
+			WREG32(reg_idx, reg_data);
+		}
+	}
+
+	for (i = 0; i < ARRAY_SIZE(sm_ctlr_regs); i++) {
+		reg_idx = adev->reg_offset[GC_HWIP][0][mmCGTS_SA0_QUAD0_SM_CTRL_REG_BASE_IDX] +
+			  sm_ctlr_regs[i];
+		reg_data = RREG32(reg_idx);
+		reg_data &= ~CGTS_SA0_QUAD0_SM_CTRL_REG__SM_MODE_MASK;
+		reg_data |= 2 << CGTS_SA0_QUAD0_SM_CTRL_REG__SM_MODE__SHIFT;
+		WREG32(reg_idx, reg_data);
+	}
+}
+
 static int gfx_v10_0_update_gfx_clock_gating(struct amdgpu_device *adev,
 					    bool enable)
 {
@@ -7987,6 +8078,10 @@ static int gfx_v10_0_update_gfx_clock_gating(struct amdgpu_device *adev,
 		gfx_v10_0_update_3d_clock_gating(adev, enable);
 		/* ===  CGCG + CGLS === */
 		gfx_v10_0_update_coarse_grain_clock_gating(adev, enable);
+
+		if ((adev->asic_type >= CHIP_NAVI10) &&
+		     (adev->asic_type <= CHIP_NAVI12))
+			gfx_v10_0_apply_medium_grain_clock_gating_workaround(adev);
 	} else {
 		/* CGCG/CGLS should be disabled before MGCG/MGLS
 		 * ===  CGCG + CGLS ===
Signature · 6 lines
2.29.0

_______________________________________________
amd-gfx mailing list
[email protected]
https://lists.freedesktop.org/mailman/listinfo/amd-gfx
author
#2 lore raw

PATCH v3 3/7 drm/amdgpu: fix NAK-G generation during PCI-e link width switch

In reply to Evan Quan

A lot of NAK-G being generated when link widht switching is happening. WA for this issue is to program the SPC to 4 symbols per clock during bootup when the native PCIE width is x4.

Change-Id I7a4d751e44bddc4bd1e97860cb4f53dfadc02a2c

---
V1->V2:
- move the code to nbio_v2_3.c as that's where the pcie related changes reside

---
 drivers/gpu/drm/amd/amdgpu/amdgpu_nbio.h |  1 +
 drivers/gpu/drm/amd/amdgpu/nbio_v2_3.c   | 28 ++++++++++++++++++++++++
 drivers/gpu/drm/amd/amdgpu/nv.c          |  3 +++
 3 files changed, 32 insertions(+)
drivers/gpu/drm/amd/amdgpu/amdgpu_nbio.h +1 -0
index 25ee53545837..43d074bb00a1 100644
@@ -93,6 +93,7 @@ struct amdgpu_nbio_funcs {
 	void (*enable_aspm)(struct amdgpu_device *adev,
 			    bool enable);
 	void (*program_aspm)(struct amdgpu_device *adev);
+	void (*apply_lc_spc_mode_wa)(struct amdgpu_device *adev);
 };
 
 struct amdgpu_nbio {
drivers/gpu/drm/amd/amdgpu/nbio_v2_3.c +28 -0
index 05ddec7ba7e2..315d57bb373d 100644
@@ -51,6 +51,8 @@
 #define mmBIF_MMSCH1_DOORBELL_RANGE		0x01d8
 #define mmBIF_MMSCH1_DOORBELL_RANGE_BASE_IDX	2
 
+#define smnPCIE_LC_LINK_WIDTH_CNTL		0x11140288
+
 static void nbio_v2_3_remap_hdp_registers(struct amdgpu_device *adev)
 {
 	WREG32_SOC15(NBIO, 0, mmREMAP_HDP_MEM_FLUSH_CNTL,
@@ -463,6 +465,31 @@ static void nbio_v2_3_program_aspm(struct amdgpu_device *adev)
 		WREG32_PCIE(smnPCIE_LC_CNTL3, data);
 }
 
+static void nbio_v2_3_apply_lc_spc_mode_wa(struct amdgpu_device *adev)
+{
+	uint32_t reg_data = 0;
+	uint32_t link_width = 0;
+
+	if (!((adev->asic_type >= CHIP_NAVI10) &&
+	     (adev->asic_type <= CHIP_NAVI12)))
+		return;
+
+	reg_data = RREG32_PCIE(smnPCIE_LC_LINK_WIDTH_CNTL);
+	link_width = (reg_data & PCIE_LC_LINK_WIDTH_CNTL__LC_LINK_WIDTH_RD_MASK)
+		>> PCIE_LC_LINK_WIDTH_CNTL__LC_LINK_WIDTH_RD__SHIFT;
+
+	/*
+	 * Program PCIE_LC_CNTL6.LC_SPC_MODE_8GT to 0x2 (4 symbols per clock data)
+	 * if link_width is 0x3 (x4)
+	 */
+	if (0x3 == link_width) {
+		reg_data = RREG32_PCIE(smnPCIE_LC_CNTL6);
+		reg_data &= ~PCIE_LC_CNTL6__LC_SPC_MODE_8GT_MASK;
+		reg_data |= (0x2 << PCIE_LC_CNTL6__LC_SPC_MODE_8GT__SHIFT);
+		WREG32_PCIE(smnPCIE_LC_CNTL6, reg_data);
+	}
+}
+
 const struct amdgpu_nbio_funcs nbio_v2_3_funcs = {
 	.get_hdp_flush_req_offset = nbio_v2_3_get_hdp_flush_req_offset,
 	.get_hdp_flush_done_offset = nbio_v2_3_get_hdp_flush_done_offset,
@@ -484,4 +511,5 @@ const struct amdgpu_nbio_funcs nbio_v2_3_funcs = {
 	.remap_hdp_registers = nbio_v2_3_remap_hdp_registers,
 	.enable_aspm = nbio_v2_3_enable_aspm,
 	.program_aspm =  nbio_v2_3_program_aspm,
+	.apply_lc_spc_mode_wa = nbio_v2_3_apply_lc_spc_mode_wa,
 };
drivers/gpu/drm/amd/amdgpu/nv.c +3 -0
index 455d0425787c..63c96ca8d2a2 100644
@@ -1411,6 +1411,9 @@ static int nv_common_hw_init(void *handle)
 {
 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
 
+	if (adev->nbio.funcs->apply_lc_spc_mode_wa)
+		adev->nbio.funcs->apply_lc_spc_mode_wa(adev);
+
 	/* enable pcie gen2/3 link */
 	nv_pcie_gen3_enable(adev);
 	/* enable aspm */
Signature · 6 lines
2.29.0

_______________________________________________
amd-gfx mailing list
[email protected]
https://lists.freedesktop.org/mailman/listinfo/amd-gfx
author
#3 lore raw

PATCH v3 4/7 drm/amdgpu: fix the hang caused by PCIe link width switch

In reply to Evan Quan

SMU had set all the necessary fields for a link width switch but the width switch wasn't occurring because the link was idle in the L1 state. Setting LC_L1_RECONFIG_EN=0x1 will allow width switches to also be initiated while in L1 instead of waiting until the link is back in L0.

Change-Id I6315681f6fb194036b20991512dd88fa65bc0d56

---
V1->V2:
- limit the change for Navi10 only
V2->V3:
- move the code to nbio_v2_3.c

---
 drivers/gpu/drm/amd/amdgpu/amdgpu_nbio.h |  1 +
 drivers/gpu/drm/amd/amdgpu/nbio_v2_3.c   | 13 +++++++++++++
 drivers/gpu/drm/amd/amdgpu/nv.c          |  3 +++
 3 files changed, 17 insertions(+)
drivers/gpu/drm/amd/amdgpu/amdgpu_nbio.h +1 -0
index 43d074bb00a1..45295dce5c3e 100644
@@ -94,6 +94,7 @@ struct amdgpu_nbio_funcs {
 			    bool enable);
 	void (*program_aspm)(struct amdgpu_device *adev);
 	void (*apply_lc_spc_mode_wa)(struct amdgpu_device *adev);
+	void (*apply_l1_link_width_reconfig_wa)(struct amdgpu_device *adev);
 };
 
 struct amdgpu_nbio {
drivers/gpu/drm/amd/amdgpu/nbio_v2_3.c +13 -0
index 315d57bb373d..754b11dea6f0 100644
@@ -490,6 +490,18 @@ static void nbio_v2_3_apply_lc_spc_mode_wa(struct amdgpu_device *adev)
 	}
 }
 
+static void nbio_v2_3_apply_l1_link_width_reconfig_wa(struct amdgpu_device *adev)
+{
+	uint32_t reg_data = 0;
+
+	if (adev->asic_type != CHIP_NAVI10)
+		return;
+
+	reg_data = RREG32_PCIE(smnPCIE_LC_LINK_WIDTH_CNTL);
+	reg_data |= PCIE_LC_LINK_WIDTH_CNTL__LC_L1_RECONFIG_EN_MASK;
+	WREG32_PCIE(smnPCIE_LC_LINK_WIDTH_CNTL, reg_data);
+}
+
 const struct amdgpu_nbio_funcs nbio_v2_3_funcs = {
 	.get_hdp_flush_req_offset = nbio_v2_3_get_hdp_flush_req_offset,
 	.get_hdp_flush_done_offset = nbio_v2_3_get_hdp_flush_done_offset,
@@ -512,4 +524,5 @@ const struct amdgpu_nbio_funcs nbio_v2_3_funcs = {
 	.enable_aspm = nbio_v2_3_enable_aspm,
 	.program_aspm =  nbio_v2_3_program_aspm,
 	.apply_lc_spc_mode_wa = nbio_v2_3_apply_lc_spc_mode_wa,
+	.apply_l1_link_width_reconfig_wa = nbio_v2_3_apply_l1_link_width_reconfig_wa,
 };
drivers/gpu/drm/amd/amdgpu/nv.c +3 -0
index 63c96ca8d2a2..5231b3402990 100644
@@ -1414,6 +1414,9 @@ static int nv_common_hw_init(void *handle)
 	if (adev->nbio.funcs->apply_lc_spc_mode_wa)
 		adev->nbio.funcs->apply_lc_spc_mode_wa(adev);
 
+	if (adev->nbio.funcs->apply_l1_link_width_reconfig_wa)
+		adev->nbio.funcs->apply_l1_link_width_reconfig_wa(adev);
+
 	/* enable pcie gen2/3 link */
 	nv_pcie_gen3_enable(adev);
 	/* enable aspm */
Signature · 6 lines
2.29.0

_______________________________________________
amd-gfx mailing list
[email protected]
https://lists.freedesktop.org/mailman/listinfo/amd-gfx
author
#4 lore raw

PATCH v3 5/7 drm/amdgpu: correct clock gating settings on feature unsupported

In reply to Evan Quan

Clock gating setting is still performed even when the corresponding CG feature is not supported. And the tricky part is disablement is actually performed no matter for enablement or disablement request. That seems not logically right.
Considering HW should already properly take care of the CG state, we will just skip the corresponding clock gating setting when the feature is not supported.

Change-Id Ic0995cf3de9f36b59316a90a28b7c95a08f4dccd
---
 drivers/gpu/drm/amd/amdgpu/athub_v2_0.c  | 12 +++--
 drivers/gpu/drm/amd/amdgpu/gfx_v10_0.c   | 69 ++++++++++++++++++------
 drivers/gpu/drm/amd/amdgpu/mmhub_v2_0.c  | 10 +++-
 drivers/gpu/drm/amd/amdgpu/nbio_v2_3.c   | 10 +++-
 drivers/gpu/drm/amd/amdgpu/smuio_v11_0.c |  5 +-
 5 files changed, 83 insertions(+), 23 deletions(-)
drivers/gpu/drm/amd/amdgpu/athub_v2_0.c +9 -3
index 5b90efd6f6d0..3ac505d954c4 100644
@@ -36,9 +36,12 @@ athub_v2_0_update_medium_grain_clock_gating(struct amdgpu_device *adev,
 {
 	uint32_t def, data;
 
+	if (!(adev->cg_flags & AMD_CG_SUPPORT_MC_MGCG))
+		return;
+
 	def = data = RREG32_SOC15(ATHUB, 0, mmATHUB_MISC_CNTL);
 
-	if (enable && (adev->cg_flags & AMD_CG_SUPPORT_MC_MGCG))
+	if (enable)
 		data |= ATHUB_MISC_CNTL__CG_ENABLE_MASK;
 	else
 		data &= ~ATHUB_MISC_CNTL__CG_ENABLE_MASK;
@@ -53,10 +56,13 @@ athub_v2_0_update_medium_grain_light_sleep(struct amdgpu_device *adev,
 {
 	uint32_t def, data;
 
+	if (!((adev->cg_flags & AMD_CG_SUPPORT_MC_LS) &&
+	       (adev->cg_flags & AMD_CG_SUPPORT_HDP_LS)))
+		return;
+
 	def = data = RREG32_SOC15(ATHUB, 0, mmATHUB_MISC_CNTL);
 
-	if (enable && (adev->cg_flags & AMD_CG_SUPPORT_MC_LS) &&
-	    (adev->cg_flags & AMD_CG_SUPPORT_HDP_LS))
+	if (enable)
 		data |= ATHUB_MISC_CNTL__CG_MEM_LS_ENABLE_MASK;
 	else
 		data &= ~ATHUB_MISC_CNTL__CG_MEM_LS_ENABLE_MASK;
drivers/gpu/drm/amd/amdgpu/gfx_v10_0.c +54 -15

133 lines not on this page · raw message

drivers/gpu/drm/amd/amdgpu/mmhub_v2_0.c +8 -2
index f7e93bbc4e15..7ded6b2f058e 100644
@@ -568,6 +568,9 @@ static void mmhub_v2_0_update_medium_grain_clock_gating(struct amdgpu_device *ad
 {
 	uint32_t def, data, def1, data1;
 
+	if (!(adev->cg_flags & AMD_CG_SUPPORT_MC_MGCG))
+		return;
+
 	switch (adev->asic_type) {
 	case CHIP_SIENNA_CICHLID:
 	case CHIP_NAVY_FLOUNDER:
@@ -582,7 +585,7 @@ static void mmhub_v2_0_update_medium_grain_clock_gating(struct amdgpu_device *ad
 		break;
 	}
 
-	if (enable && (adev->cg_flags & AMD_CG_SUPPORT_MC_MGCG)) {
+	if (enable) {
 		data |= MM_ATC_L2_MISC_CG__ENABLE_MASK;
 
 		data1 &= ~(DAGB0_CNTL_MISC2__DISABLE_WRREQ_CG_MASK |
@@ -627,6 +630,9 @@ static void mmhub_v2_0_update_medium_grain_light_sleep(struct amdgpu_device *ade
 {
 	uint32_t def, data;
 
+	if (!(adev->cg_flags & AMD_CG_SUPPORT_MC_LS))
+		return;
+
 	switch (adev->asic_type) {
 	case CHIP_SIENNA_CICHLID:
 	case CHIP_NAVY_FLOUNDER:
@@ -639,7 +645,7 @@ static void mmhub_v2_0_update_medium_grain_light_sleep(struct amdgpu_device *ade
 		break;
 	}
 
-	if (enable && (adev->cg_flags & AMD_CG_SUPPORT_MC_LS))
+	if (enable)
 		data |= MM_ATC_L2_MISC_CG__MEM_LS_ENABLE_MASK;
 	else
 		data &= ~MM_ATC_L2_MISC_CG__MEM_LS_ENABLE_MASK;
drivers/gpu/drm/amd/amdgpu/nbio_v2_3.c +8 -2
index 754b11dea6f0..7b79eeaa88aa 100644
@@ -220,8 +220,11 @@ static void nbio_v2_3_update_medium_grain_clock_gating(struct amdgpu_device *ade
 {
 	uint32_t def, data;
 
+	if (!(adev->cg_flags & AMD_CG_SUPPORT_BIF_MGCG))
+		return;
+
 	def = data = RREG32_PCIE(smnCPM_CONTROL);
-	if (enable && (adev->cg_flags & AMD_CG_SUPPORT_BIF_MGCG)) {
+	if (enable) {
 		data |= (CPM_CONTROL__LCLK_DYN_GATE_ENABLE_MASK |
 			 CPM_CONTROL__TXCLK_DYN_GATE_ENABLE_MASK |
 			 CPM_CONTROL__TXCLK_LCNT_GATE_ENABLE_MASK |
@@ -246,8 +249,11 @@ static void nbio_v2_3_update_medium_grain_light_sleep(struct amdgpu_device *adev
 {
 	uint32_t def, data;
 
+	if (!(adev->cg_flags & AMD_CG_SUPPORT_BIF_LS))
+		return;
+
 	def = data = RREG32_PCIE(smnPCIE_CNTL2);
-	if (enable && (adev->cg_flags & AMD_CG_SUPPORT_BIF_LS)) {
+	if (enable) {
 		data |= (PCIE_CNTL2__SLV_MEM_LS_EN_MASK |
 			 PCIE_CNTL2__MST_MEM_LS_EN_MASK |
 			 PCIE_CNTL2__REPLAY_MEM_LS_EN_MASK);
drivers/gpu/drm/amd/amdgpu/smuio_v11_0.c +4 -1
index e9c474c217ec..b6f1322f908c 100644
@@ -43,9 +43,12 @@ static void smuio_v11_0_update_rom_clock_gating(struct amdgpu_device *adev, bool
 	if (adev->flags & AMD_IS_APU)
 		return;
 
+	if (!(adev->cg_flags & AMD_CG_SUPPORT_ROM_MGCG))
+		return;
+
 	def = data = RREG32_SOC15(SMUIO, 0, mmCGTT_ROM_CLK_CTRL0);
 
-	if (enable && (adev->cg_flags & AMD_CG_SUPPORT_ROM_MGCG))
+	if (enable)
 		data &= ~(CGTT_ROM_CLK_CTRL0__SOFT_OVERRIDE0_MASK |
 			CGTT_ROM_CLK_CTRL0__SOFT_OVERRIDE1_MASK);
 	else
Signature · 6 lines
2.29.0

_______________________________________________
amd-gfx mailing list
[email protected]
https://lists.freedesktop.org/mailman/listinfo/amd-gfx
author
#5 lore raw

PATCH v3 6/7 drm/amdgpu: update GFX MGCG settings

In reply to Evan Quan

Update GFX MGCG related settings.

Change-Id I0b7b8e7c97859f99db5f52026abbb4d226c179df
---
 drivers/gpu/drm/amd/amdgpu/gfx_v10_0.c | 11 +++++++----
 1 file changed, 7 insertions(+), 4 deletions(-)
drivers/gpu/drm/amd/amdgpu/gfx_v10_0.c +7 -4
index 5d0062f820c1..b855a7c5bc4e 100644
@@ -7787,11 +7787,11 @@ static void gfx_v10_0_update_medium_grain_clock_gating(struct amdgpu_device *ade
 {
 	uint32_t data, def;
 
-	if (!(adev->cg_flags & AMD_CG_SUPPORT_GFX_MGCG))
+	if (!(adev->cg_flags & (AMD_CG_SUPPORT_GFX_MGCG | AMD_CG_SUPPORT_GFX_MGLS)))
 		return;
 
 	/* It is disabled by HW by default */
-	if (enable) {
+	if (enable && (adev->cg_flags & AMD_CG_SUPPORT_GFX_MGCG)) {
 		/* 0 - Disable some blocks' MGCG */
 		WREG32_SOC15(GC, 0, mmGRBM_GFX_INDEX, 0xe0000000);
 		WREG32_SOC15(GC, 0, mmCGTT_WD_CLK_CTRL, 0xff000000);
@@ -7804,6 +7804,7 @@ static void gfx_v10_0_update_medium_grain_clock_gating(struct amdgpu_device *ade
 			  RLC_CGTT_MGCG_OVERRIDE__RLC_CGTT_SCLK_OVERRIDE_MASK |
 			  RLC_CGTT_MGCG_OVERRIDE__GFXIP_MGCG_OVERRIDE_MASK |
 			  RLC_CGTT_MGCG_OVERRIDE__GFXIP_MGLS_OVERRIDE_MASK |
+			  RLC_CGTT_MGCG_OVERRIDE__GFXIP_FGCG_OVERRIDE_MASK |
 			  RLC_CGTT_MGCG_OVERRIDE__ENABLE_CGTS_LEGACY_MASK);
 
 		if (def != data)
@@ -7826,13 +7827,15 @@ static void gfx_v10_0_update_medium_grain_clock_gating(struct amdgpu_device *ade
 					WREG32_SOC15(GC, 0, mmCP_MEM_SLP_CNTL, data);
 			}
 		}
-	} else {
+	} else if (!enable || !(adev->cg_flags & AMD_CG_SUPPORT_GFX_MGCG)) {
 		/* 1 - MGCG_OVERRIDE */
 		def = data = RREG32_SOC15(GC, 0, mmRLC_CGTT_MGCG_OVERRIDE);
 		data |= (RLC_CGTT_MGCG_OVERRIDE__RLC_CGTT_SCLK_OVERRIDE_MASK |
 			 RLC_CGTT_MGCG_OVERRIDE__GRBM_CGTT_SCLK_OVERRIDE_MASK |
 			 RLC_CGTT_MGCG_OVERRIDE__GFXIP_MGCG_OVERRIDE_MASK |
-			 RLC_CGTT_MGCG_OVERRIDE__GFXIP_MGLS_OVERRIDE_MASK);
+			 RLC_CGTT_MGCG_OVERRIDE__GFXIP_MGLS_OVERRIDE_MASK |
+			 RLC_CGTT_MGCG_OVERRIDE__GFXIP_FGCG_OVERRIDE_MASK |
+			 RLC_CGTT_MGCG_OVERRIDE__ENABLE_CGTS_LEGACY_MASK);
 		if (def != data)
 			WREG32_SOC15(GC, 0, mmRLC_CGTT_MGCG_OVERRIDE, data);
 
Signature · 6 lines
2.29.0

_______________________________________________
amd-gfx mailing list
[email protected]
https://lists.freedesktop.org/mailman/listinfo/amd-gfx
author
#6 lore raw

PATCH v3 7/7 drm/amdgpu: update HDP LS settings

In reply to Evan Quan

Avoid unnecessary register programming on feature disablement.

Change-Id Ia8ad4fb28cb23f80ddcf1399eace284e4d33bd90
---
 drivers/gpu/drm/amd/amdgpu/hdp_v5_0.c | 85 +++++++++++++++------------
 1 file changed, 48 insertions(+), 37 deletions(-)
drivers/gpu/drm/amd/amdgpu/hdp_v5_0.c +48 -37
index 7a15e669b68d..5793977953cc 100644
@@ -90,45 +90,56 @@ static void hdp_v5_0_update_mem_power_gating(struct amdgpu_device *adev,
 					 RC_MEM_POWER_SD_EN, 0);
 	WREG32_SOC15(HDP, 0, mmHDP_MEM_POWER_CTRL, hdp_mem_pwr_cntl);
 
-	/* only one clock gating mode (LS/DS/SD) can be enabled */
-	if (adev->cg_flags & AMD_CG_SUPPORT_HDP_LS) {
-		hdp_mem_pwr_cntl = REG_SET_FIELD(hdp_mem_pwr_cntl,
-						 HDP_MEM_POWER_CTRL,
-						 IPH_MEM_POWER_LS_EN, enable);
-		hdp_mem_pwr_cntl = REG_SET_FIELD(hdp_mem_pwr_cntl,
-						 HDP_MEM_POWER_CTRL,
-						 RC_MEM_POWER_LS_EN, enable);
-	} else if (adev->cg_flags & AMD_CG_SUPPORT_HDP_DS) {
-		hdp_mem_pwr_cntl = REG_SET_FIELD(hdp_mem_pwr_cntl,
-						 HDP_MEM_POWER_CTRL,
-						 IPH_MEM_POWER_DS_EN, enable);
-		hdp_mem_pwr_cntl = REG_SET_FIELD(hdp_mem_pwr_cntl,
-						 HDP_MEM_POWER_CTRL,
-						 RC_MEM_POWER_DS_EN, enable);
-	} else if (adev->cg_flags & AMD_CG_SUPPORT_HDP_SD) {
-		hdp_mem_pwr_cntl = REG_SET_FIELD(hdp_mem_pwr_cntl,
-						 HDP_MEM_POWER_CTRL,
-						 IPH_MEM_POWER_SD_EN, enable);
-		/* RC should not use shut down mode, fallback to ds */
-		hdp_mem_pwr_cntl = REG_SET_FIELD(hdp_mem_pwr_cntl,
-						 HDP_MEM_POWER_CTRL,
-						 RC_MEM_POWER_DS_EN, enable);
-	}
-
-	/* confirmed that IPH_MEM_POWER_CTRL_EN and RC_MEM_POWER_CTRL_EN have to
-	 * be set for SRAM LS/DS/SD */
-	if (adev->cg_flags & (AMD_CG_SUPPORT_HDP_LS | AMD_CG_SUPPORT_HDP_DS |
-			      AMD_CG_SUPPORT_HDP_SD)) {
-		hdp_mem_pwr_cntl = REG_SET_FIELD(hdp_mem_pwr_cntl, HDP_MEM_POWER_CTRL,
-						 IPH_MEM_POWER_CTRL_EN, 1);
-		hdp_mem_pwr_cntl = REG_SET_FIELD(hdp_mem_pwr_cntl, HDP_MEM_POWER_CTRL,
-						 RC_MEM_POWER_CTRL_EN, 1);
+	/* Already disabled above. The actions below are for "enabled" only */
+	if (enable) {
+		/* only one clock gating mode (LS/DS/SD) can be enabled */
+		if (adev->cg_flags & AMD_CG_SUPPORT_HDP_LS) {
+			hdp_mem_pwr_cntl = REG_SET_FIELD(hdp_mem_pwr_cntl,
+							 HDP_MEM_POWER_CTRL,
+							 IPH_MEM_POWER_LS_EN, 1);
+			hdp_mem_pwr_cntl = REG_SET_FIELD(hdp_mem_pwr_cntl,
+							 HDP_MEM_POWER_CTRL,
+							 RC_MEM_POWER_LS_EN, 1);
+		} else if (adev->cg_flags & AMD_CG_SUPPORT_HDP_DS) {
+			hdp_mem_pwr_cntl = REG_SET_FIELD(hdp_mem_pwr_cntl,
+							 HDP_MEM_POWER_CTRL,
+							 IPH_MEM_POWER_DS_EN, 1);
+			hdp_mem_pwr_cntl = REG_SET_FIELD(hdp_mem_pwr_cntl,
+							 HDP_MEM_POWER_CTRL,
+							 RC_MEM_POWER_DS_EN, 1);
+		} else if (adev->cg_flags & AMD_CG_SUPPORT_HDP_SD) {
+			hdp_mem_pwr_cntl = REG_SET_FIELD(hdp_mem_pwr_cntl,
+							 HDP_MEM_POWER_CTRL,
+							 IPH_MEM_POWER_SD_EN, 1);
+			/* RC should not use shut down mode, fallback to ds  or ls if allowed */
+			if (adev->cg_flags & AMD_CG_SUPPORT_HDP_DS)
+				hdp_mem_pwr_cntl = REG_SET_FIELD(hdp_mem_pwr_cntl,
+								 HDP_MEM_POWER_CTRL,
+								 RC_MEM_POWER_DS_EN, 1);
+			else if (adev->cg_flags & AMD_CG_SUPPORT_HDP_LS)
+				hdp_mem_pwr_cntl = REG_SET_FIELD(hdp_mem_pwr_cntl,
+								 HDP_MEM_POWER_CTRL,
+								 RC_MEM_POWER_LS_EN, 1);
+		}
+
+		/* confirmed that IPH_MEM_POWER_CTRL_EN and RC_MEM_POWER_CTRL_EN have to
+		 * be set for SRAM LS/DS/SD */
+		if (adev->cg_flags & (AMD_CG_SUPPORT_HDP_LS | AMD_CG_SUPPORT_HDP_DS |
+				      AMD_CG_SUPPORT_HDP_SD)) {
+			hdp_mem_pwr_cntl = REG_SET_FIELD(hdp_mem_pwr_cntl, HDP_MEM_POWER_CTRL,
+							 IPH_MEM_POWER_CTRL_EN, 1);
+			hdp_mem_pwr_cntl = REG_SET_FIELD(hdp_mem_pwr_cntl, HDP_MEM_POWER_CTRL,
+							 RC_MEM_POWER_CTRL_EN, 1);
+			WREG32_SOC15(HDP, 0, mmHDP_MEM_POWER_CTRL, hdp_mem_pwr_cntl);
+		}
 	}
 
-	WREG32_SOC15(HDP, 0, mmHDP_MEM_POWER_CTRL, hdp_mem_pwr_cntl);
-
-	/* restore IPH & RC clock override after clock/power mode changing */
-	WREG32_SOC15(HDP, 0, mmHDP_CLK_CNTL, hdp_clk_cntl1);
+	/* disable IPH & RC clock override after clock/power mode changing */
+	hdp_clk_cntl = REG_SET_FIELD(hdp_clk_cntl, HDP_CLK_CNTL,
+				     IPH_MEM_CLK_SOFT_OVERRIDE, 0);
+	hdp_clk_cntl = REG_SET_FIELD(hdp_clk_cntl, HDP_CLK_CNTL,
+				     RC_MEM_CLK_SOFT_OVERRIDE, 0);
+	WREG32_SOC15(HDP, 0, mmHDP_CLK_CNTL, hdp_clk_cntl);
 }
 
 static void hdp_v5_0_update_medium_grain_clock_gating(struct amdgpu_device *adev,
Signature · 6 lines
2.29.0

_______________________________________________
amd-gfx mailing list
[email protected]
https://lists.freedesktop.org/mailman/listinfo/amd-gfx
Reviewed-by on v3 1/7
#7 lore raw
In reply to Evan Quan

One minor comment below, apart from that the series looks good.

Quoted 76 lines · Evan Quan

Add missing settings for SQC bits. And correct some confusing logics around active wgp bitmap calculation.

Change-Id If4992e175fd61d5609b00328cbe21f487517d039

---
V1->V2:
- restore correct tcp_harvest setting for NV10 and NV12
- move asic type guard upper layer for better readability

---
  drivers/gpu/drm/amd/amdgpu/gfx_v10_0.c | 103 ++++++++++++++-----------
  1 file changed, 57 insertions(+), 46 deletions(-)
drivers/gpu/drm/amd/amdgpu/gfx_v10_0.c +19 -34
index 15ae9e33b925..384b95fbad8b 100644
@@ -5090,47 +5090,50 @@ static void gfx_v10_0_tcp_harvest(struct amdgpu_device *adev)
  		4 + /* RMI */
  		1); /* SQG */
 
-	if (adev->asic_type == CHIP_NAVI10 ||
-	    adev->asic_type == CHIP_NAVI14 ||
-	    adev->asic_type == CHIP_NAVI12) {
-		mutex_lock(&adev->grbm_idx_mutex);
-		for (i = 0; i < adev->gfx.config.max_shader_engines; i++) {
-			for (j = 0; j < adev->gfx.config.max_sh_per_se; j++) {
-				gfx_v10_0_select_se_sh(adev, i, j, 0xffffffff);
-				wgp_active_bitmap = gfx_v10_0_get_wgp_active_bitmap_per_sh(adev);
-				/*
-				 * Set corresponding TCP bits for the inactive WGPs in
-				 * GCRD_SA_TARGETS_DISABLE
-				 */
-				gcrd_targets_disable_tcp = 0;
-				/* Set TCP & SQC bits in UTCL1_UTCL0_INVREQ_DISABLE */
-				utcl_invreq_disable = 0;
-
-				for (k = 0; k < max_wgp_per_sh; k++) {
-					if (!(wgp_active_bitmap & (1 << k))) {
-						gcrd_targets_disable_tcp |= 3 << (2 * k);
-						utcl_invreq_disable |= (3 << (2 * k)) |
-							(3 << (2 * (max_wgp_per_sh + k)));
-					}
+	mutex_lock(&adev->grbm_idx_mutex);
+	for (i = 0; i < adev->gfx.config.max_shader_engines; i++) {
+		for (j = 0; j < adev->gfx.config.max_sh_per_se; j++) {
+			gfx_v10_0_select_se_sh(adev, i, j, 0xffffffff);
+			wgp_active_bitmap = gfx_v10_0_get_wgp_active_bitmap_per_sh(adev);
+			/*
+			 * Set corresponding TCP bits for the inactive WGPs in
+			 * GCRD_SA_TARGETS_DISABLE
+			 */
+			gcrd_targets_disable_tcp = 0;
+			/* Set TCP & SQC bits in UTCL1_UTCL0_INVREQ_DISABLE */
+			utcl_invreq_disable = 0;
+
+			for (k = 0; k < max_wgp_per_sh; k++) {
+				if (!(wgp_active_bitmap & (1 << k))) {
+					gcrd_targets_disable_tcp |= 3 << (2 * k);
+					gcrd_targets_disable_tcp |= 1 << (k + (max_wgp_per_sh * 2));
+					utcl_invreq_disable |= (3 << (2 * k)) |
+						(3 << (2 * (max_wgp_per_sh + k)));
  				}
-
-				tmp = RREG32_SOC15(GC, 0, mmUTCL1_UTCL0_INVREQ_DISABLE);
-				/* only override TCP & SQC bits */
-				tmp &= 0xffffffff << (4 * max_wgp_per_sh);
-				tmp |= (utcl_invreq_disable & utcl_invreq_disable_mask);
-				WREG32_SOC15(GC, 0, mmUTCL1_UTCL0_INVREQ_DISABLE, tmp);
-
-				tmp = RREG32_SOC15(GC, 0, mmGCRD_SA_TARGETS_DISABLE);
-				/* only override TCP bits */
-				tmp &= 0xffffffff << (2 * max_wgp_per_sh);
-				tmp |= (gcrd_targets_disable_tcp & gcrd_targets_disable_mask);
-				WREG32_SOC15(GC, 0, mmGCRD_SA_TARGETS_DISABLE, tmp);
  			}

drivers/gpu/drm/amd/amdgpu/gfx_v10_0.c static void gfx_v10_0_tcp_harvest(struct amdgpu_device *adev) · line 5120

Quoted 10 lines
drivers/gpu/drm/amd/amdgpu/gfx_v10_0.c +6 -3
-		}
 
-		gfx_v10_0_select_se_sh(adev, 0xffffffff, 0xffffffff, 0xffffffff);
-		mutex_unlock(&adev->grbm_idx_mutex);
+			tmp = RREG32_SOC15(GC, 0, mmUTCL1_UTCL0_INVREQ_DISABLE);
+			/* only override TCP & SQC bits */
+			if (adev->asic_type == CHIP_NAVI14)
+				tmp &= 0xff000000;
+			else
+				tmp &=0xfff00000;

For the disable field mask calculation (which is the value that is applied finally), there is no ASIC check. The above code may utilize the same method as in the original code without ASIC check.

tmp &= 0xffffffff << (4 * max_wgp_per_sh);

Same for below case also - 3*max_wgp_per_sh.

Thanks,
Lijo

Quoted 53 lines
drivers/gpu/drm/amd/amdgpu/gfx_v10_0.c +30 -6
+			tmp |= (utcl_invreq_disable & utcl_invreq_disable_mask);
+			WREG32_SOC15(GC, 0, mmUTCL1_UTCL0_INVREQ_DISABLE, tmp);
+
+			tmp = RREG32_SOC15(GC, 0, mmGCRD_SA_TARGETS_DISABLE);
+			/* only override TCP & SQC bits */
+			if (adev->asic_type == CHIP_NAVI14)
+				tmp &= 0xfffc0000;
+			else
+				tmp &= 0xffff8000;
+			tmp |= (gcrd_targets_disable_tcp & gcrd_targets_disable_mask);
+			WREG32_SOC15(GC, 0, mmGCRD_SA_TARGETS_DISABLE, tmp);
+		}
  	}
+
+	gfx_v10_0_select_se_sh(adev, 0xffffffff, 0xffffffff, 0xffffffff);
+	mutex_unlock(&adev->grbm_idx_mutex);
  }
 
  static void gfx_v10_0_get_tcc_info(struct amdgpu_device *adev)
@@ -7408,7 +7411,10 @@ static int gfx_v10_0_hw_init(void *handle)
  	 * init golden registers and rlc resume may override some registers,
  	 * reconfig them here
  	 */
-	gfx_v10_0_tcp_harvest(adev);
+	if (adev->asic_type == CHIP_NAVI10 ||
+	    adev->asic_type == CHIP_NAVI14 ||
+	    adev->asic_type == CHIP_NAVI12)
+		gfx_v10_0_tcp_harvest(adev);
 
  	r = gfx_v10_0_cp_resume(adev);
  	if (r)
@@ -9328,17 +9334,22 @@ static void gfx_v10_0_set_user_wgp_inactive_bitmap_per_sh(struct amdgpu_device *
 
  static u32 gfx_v10_0_get_wgp_active_bitmap_per_sh(struct amdgpu_device *adev)
  {
-	u32 data, wgp_bitmask;
-	data = RREG32_SOC15(GC, 0, mmCC_GC_SHADER_ARRAY_CONFIG);
-	data |= RREG32_SOC15(GC, 0, mmGC_USER_SHADER_ARRAY_CONFIG);
+	u32 disabled_mask =
+		~amdgpu_gfx_create_bitmask(adev->gfx.config.max_cu_per_sh >> 1);
+	u32 efuse_setting = 0;
+	u32 vbios_setting = 0;
+
+	efuse_setting = RREG32_SOC15(GC, 0, mmCC_GC_SHADER_ARRAY_CONFIG);
+	efuse_setting &= CC_GC_SHADER_ARRAY_CONFIG__INACTIVE_WGPS_MASK;
+	efuse_setting >>= CC_GC_SHADER_ARRAY_CONFIG__INACTIVE_WGPS__SHIFT;
 
-	data &= CC_GC_SHADER_ARRAY_CONFIG__INACTIVE_WGPS_MASK;
-	data >>= CC_GC_SHADER_ARRAY_CONFIG__INACTIVE_WGPS__SHIFT;
+	vbios_setting = RREG32_SOC15(GC, 0, mmGC_USER_SHADER_ARRAY_CONFIG);
+	vbios_setting &= GC_USER_SHADER_ARRAY_CONFIG__INACTIVE_WGPS_MASK;
+	vbios_setting >>= GC_USER_SHADER_ARRAY_CONFIG__INACTIVE_WGPS__SHIFT;

drivers/gpu/drm/amd/amdgpu/gfx_v10_0.c static void gfx_v10_0_set_user_wgp_inactive_bitmap_per_sh(struct amdgpu_device * · line 9355

Quoted 10 lines
drivers/gpu/drm/amd/amdgpu/gfx_v10_0.c +2 -3
 
-	wgp_bitmask =
-		amdgpu_gfx_create_bitmask(adev->gfx.config.max_cu_per_sh >> 1);
+	disabled_mask |= efuse_setting | vbios_setting;
 
-	return (~data) & wgp_bitmask;
+	return (~disabled_mask);
  }
 
  static u32 gfx_v10_0_get_cu_active_bitmap_per_sh(struct amdgpu_device *adev)

_______________________________________________
amd-gfx mailing list
[email protected]
https://lists.freedesktop.org/mailman/listinfo/amd-gfx

author
#8 lore raw

[AMD Official Use Only]

Thanks Lijo.
However, I'm not quite sure whether " 0xffffffff << (4 * max_wgp_per_sh);" is a valid expression since it kind of triggers some overflow. Can that work for non-x86 platform or even work reliably for x86 platform?

BR
Evan

Quoted 206 lines

-----Original Message-----
From: Lazar, Lijo <[email protected]>
Sent: Monday, June 21, 2021 8:08 PM
To: Quan, Evan <[email protected]>; [email protected]
Cc: Deucher, Alexander <[email protected]>
Subject: Re: [PATCH V3 1/7] drm/amdgpu: correct tcp harvest setting

One minor comment below, apart from that the series looks good.

Quoted 11 lines · Evan Quan

Add missing settings for SQC bits. And correct some confusing logics around active wgp bitmap calculation.

Change-Id If4992e175fd61d5609b00328cbe21f487517d039

---
V1->V2:
- restore correct tcp_harvest setting for NV10 and NV12
- move asic type guard upper layer for better readability

---
  drivers/gpu/drm/amd/amdgpu/gfx_v10_0.c | 103 ++++++++++++++------

-----

drivers/gpu/drm/amd/amdgpu/gfx_v10_0.c static void gfx_v10_0_tcp_harvest(struct

Quoted 8 lines
  1 file changed, 57 insertions(+), 46 deletions(-)
drivers/gpu/drm/amd/amdgpu/gfx_v10_0.c +0 -0
index 15ae9e33b925..384b95fbad8b 100644
@@ -5090,47 +5090,50 @@ static void gfx_v10_0_tcp_harvest(struct

amdgpu_device *adev)

drivers/gpu/drm/amd/amdgpu/gfx_v10_0.c static void gfx_v10_0_tcp_harvest(struct amdgpu_device *adev)

Quoted 11 lines
drivers/gpu/drm/amd/amdgpu/gfx_v10_0.c +0 -8
  		4 + /* RMI */
  		1); /* SQG */
 
-	if (adev->asic_type == CHIP_NAVI10 ||
-	    adev->asic_type == CHIP_NAVI14 ||
-	    adev->asic_type == CHIP_NAVI12) {
-		mutex_lock(&adev->grbm_idx_mutex);
-		for (i = 0; i < adev->gfx.config.max_shader_engines; i++) {
-			for (j = 0; j < adev->gfx.config.max_sh_per_se; j++) {
-				gfx_v10_0_select_se_sh(adev, i, j, 0xffffffff);
-				wgp_active_bitmap =

gfx_v10_0_get_wgp_active_bitmap_per_sh(adev);

drivers/gpu/drm/amd/amdgpu/gfx_v10_0.c static void gfx_v10_0_tcp_harvest(struct · line 5102

Quoted 2 lines
drivers/gpu/drm/amd/amdgpu/gfx_v10_0.c +0 -2
-				/*
-				 * Set corresponding TCP bits for the inactive

WGPs in

drivers/gpu/drm/amd/amdgpu/gfx_v10_0.c static void gfx_v10_0_tcp_harvest(struct amdgpu_device *adev)

Quoted 4 lines
drivers/gpu/drm/amd/amdgpu/gfx_v10_0.c +0 -4
-				 * GCRD_SA_TARGETS_DISABLE
-				 */
-				gcrd_targets_disable_tcp = 0;
-				/* Set TCP & SQC bits in

UTCL1_UTCL0_INVREQ_DISABLE */

drivers/gpu/drm/amd/amdgpu/gfx_v10_0.c static void gfx_v10_0_tcp_harvest(struct amdgpu_device *adev)

Quoted 5 lines
drivers/gpu/drm/amd/amdgpu/gfx_v10_0.c +0 -5
-				utcl_invreq_disable = 0;
-
-				for (k = 0; k < max_wgp_per_sh; k++) {
-					if (!(wgp_active_bitmap & (1 << k))) {
-						gcrd_targets_disable_tcp |=

3 << (2 * k);

drivers/gpu/drm/amd/amdgpu/gfx_v10_0.c static void gfx_v10_0_tcp_harvest(struct · line 5112

Quoted 1 line
drivers/gpu/drm/amd/amdgpu/gfx_v10_0.c +0 -1
-						utcl_invreq_disable |= (3 <<

(2 * k)) |

drivers/gpu/drm/amd/amdgpu/gfx_v10_0.c static void gfx_v10_0_tcp_harvest(struct · line 5113

Quoted 1 line
drivers/gpu/drm/amd/amdgpu/gfx_v10_0.c +0 -1
-							(3 << (2 *

(max_wgp_per_sh + k)));

drivers/gpu/drm/amd/amdgpu/gfx_v10_0.c static void gfx_v10_0_tcp_harvest(struct amdgpu_device *adev)

Quoted 6 lines
drivers/gpu/drm/amd/amdgpu/gfx_v10_0.c +5 -1
-					}
+	mutex_lock(&adev->grbm_idx_mutex);
+	for (i = 0; i < adev->gfx.config.max_shader_engines; i++) {
+		for (j = 0; j < adev->gfx.config.max_sh_per_se; j++) {
+			gfx_v10_0_select_se_sh(adev, i, j, 0xffffffff);
+			wgp_active_bitmap =

gfx_v10_0_get_wgp_active_bitmap_per_sh(adev);

drivers/gpu/drm/amd/amdgpu/gfx_v10_0.c static void gfx_v10_0_tcp_harvest(struct amdgpu_device *adev)

Quoted 6 lines
drivers/gpu/drm/amd/amdgpu/gfx_v10_0.c +6 -0
+			/*
+			 * Set corresponding TCP bits for the inactive WGPs in
+			 * GCRD_SA_TARGETS_DISABLE
+			 */
+			gcrd_targets_disable_tcp = 0;
+			/* Set TCP & SQC bits in

UTCL1_UTCL0_INVREQ_DISABLE */

drivers/gpu/drm/amd/amdgpu/gfx_v10_0.c static void gfx_v10_0_tcp_harvest(struct amdgpu_device *adev)

Quoted 5 lines
drivers/gpu/drm/amd/amdgpu/gfx_v10_0.c +5 -0
+			utcl_invreq_disable = 0;
+
+			for (k = 0; k < max_wgp_per_sh; k++) {
+				if (!(wgp_active_bitmap & (1 << k))) {
+					gcrd_targets_disable_tcp |= 3 << (2 *

k);

drivers/gpu/drm/amd/amdgpu/gfx_v10_0.c static void gfx_v10_0_tcp_harvest(struct · line 5109

Quoted 1 line
drivers/gpu/drm/amd/amdgpu/gfx_v10_0.c +1 -0
+					gcrd_targets_disable_tcp |= 1 << (k +

(max_wgp_per_sh * 2));

drivers/gpu/drm/amd/amdgpu/gfx_v10_0.c static void gfx_v10_0_tcp_harvest(struct · line 5111

Quoted 2 lines
drivers/gpu/drm/amd/amdgpu/gfx_v10_0.c +2 -0
+					utcl_invreq_disable |= (3 << (2 * k)) |
+						(3 << (2 * (max_wgp_per_sh

+ k)));

drivers/gpu/drm/amd/amdgpu/gfx_v10_0.c static void gfx_v10_0_tcp_harvest(struct amdgpu_device *adev)

Quoted 3 lines
drivers/gpu/drm/amd/amdgpu/gfx_v10_0.c +0 -2
  				}
-
-				tmp = RREG32_SOC15(GC, 0,

mmUTCL1_UTCL0_INVREQ_DISABLE);

drivers/gpu/drm/amd/amdgpu/gfx_v10_0.c static void gfx_v10_0_tcp_harvest(struct amdgpu_device *adev)

Quoted 3 lines
drivers/gpu/drm/amd/amdgpu/gfx_v10_0.c +0 -3
-				/* only override TCP & SQC bits */
-				tmp &= 0xffffffff << (4 * max_wgp_per_sh);
-				tmp |= (utcl_invreq_disable &

utcl_invreq_disable_mask);

drivers/gpu/drm/amd/amdgpu/gfx_v10_0.c static void gfx_v10_0_tcp_harvest(struct · line 5121

Quoted 1 line
drivers/gpu/drm/amd/amdgpu/gfx_v10_0.c +0 -1
-				WREG32_SOC15(GC, 0,

mmUTCL1_UTCL0_INVREQ_DISABLE, tmp);

drivers/gpu/drm/amd/amdgpu/gfx_v10_0.c static void gfx_v10_0_tcp_harvest(struct · line 5123

Quoted 2 lines
drivers/gpu/drm/amd/amdgpu/gfx_v10_0.c +0 -2
-
-				tmp = RREG32_SOC15(GC, 0,

mmGCRD_SA_TARGETS_DISABLE);

drivers/gpu/drm/amd/amdgpu/gfx_v10_0.c static void gfx_v10_0_tcp_harvest(struct amdgpu_device *adev)

Quoted 3 lines
drivers/gpu/drm/amd/amdgpu/gfx_v10_0.c +0 -3
-				/* only override TCP bits */
-				tmp &= 0xffffffff << (2 * max_wgp_per_sh);
-				tmp |= (gcrd_targets_disable_tcp &

gcrd_targets_disable_mask);

drivers/gpu/drm/amd/amdgpu/gfx_v10_0.c static void gfx_v10_0_tcp_harvest(struct · line 5127

Quoted 1 line
drivers/gpu/drm/amd/amdgpu/gfx_v10_0.c +0 -1
-				WREG32_SOC15(GC, 0,

mmGCRD_SA_TARGETS_DISABLE, tmp);

drivers/gpu/drm/amd/amdgpu/gfx_v10_0.c static void gfx_v10_0_tcp_harvest(struct amdgpu_device *adev)

Quoted 6 lines
drivers/gpu/drm/amd/amdgpu/gfx_v10_0.c +1 -3
  			}
-		}
 
-		gfx_v10_0_select_se_sh(adev, 0xffffffff, 0xffffffff, 0xffffffff);
-		mutex_unlock(&adev->grbm_idx_mutex);
+			tmp = RREG32_SOC15(GC, 0,

mmUTCL1_UTCL0_INVREQ_DISABLE);

drivers/gpu/drm/amd/amdgpu/gfx_v10_0.c static void gfx_v10_0_tcp_harvest(struct amdgpu_device *adev) · line 5120

Quoted 5 lines
drivers/gpu/drm/amd/amdgpu/gfx_v10_0.c +5 -0
+			/* only override TCP & SQC bits */
+			if (adev->asic_type == CHIP_NAVI14)
+				tmp &= 0xff000000;
+			else
+				tmp &=0xfff00000;

For the disable field mask calculation (which is the value that is applied finally), there is no ASIC check. The above code may utilize the same method as in the original code without ASIC check.

tmp &= 0xffffffff << (4 * max_wgp_per_sh);

Same for below case also - 3*max_wgp_per_sh.

Thanks,
Lijo

drivers/gpu/drm/amd/amdgpu/gfx_v10_0.c static void gfx_v10_0_tcp_harvest(struct · line 5121

Quoted 1 line
drivers/gpu/drm/amd/amdgpu/gfx_v10_0.c +1 -0
+			tmp |= (utcl_invreq_disable &

utcl_invreq_disable_mask);

drivers/gpu/drm/amd/amdgpu/gfx_v10_0.c static void gfx_v10_0_tcp_harvest(struct · line 5122

Quoted 1 line
drivers/gpu/drm/amd/amdgpu/gfx_v10_0.c +1 -0
+			WREG32_SOC15(GC, 0,

mmUTCL1_UTCL0_INVREQ_DISABLE, tmp);

drivers/gpu/drm/amd/amdgpu/gfx_v10_0.c static void gfx_v10_0_tcp_harvest(struct · line 5124

Quoted 2 lines
drivers/gpu/drm/amd/amdgpu/gfx_v10_0.c +2 -0
+
+			tmp = RREG32_SOC15(GC, 0,

mmGCRD_SA_TARGETS_DISABLE);

drivers/gpu/drm/amd/amdgpu/gfx_v10_0.c static void gfx_v10_0_tcp_harvest(struct amdgpu_device *adev)

Quoted 6 lines
drivers/gpu/drm/amd/amdgpu/gfx_v10_0.c +6 -0
+			/* only override TCP & SQC bits */
+			if (adev->asic_type == CHIP_NAVI14)
+				tmp &= 0xfffc0000;
+			else
+				tmp &= 0xffff8000;
+			tmp |= (gcrd_targets_disable_tcp &

gcrd_targets_disable_mask);

drivers/gpu/drm/amd/amdgpu/gfx_v10_0.c static void gfx_v10_0_tcp_harvest(struct · line 5131

Quoted 1 line
drivers/gpu/drm/amd/amdgpu/gfx_v10_0.c +1 -0
+			WREG32_SOC15(GC, 0,

mmGCRD_SA_TARGETS_DISABLE, tmp);

Quoted 21 lines
drivers/gpu/drm/amd/amdgpu/gfx_v10_0.c +8 -2
+		}
  	}
+
+	gfx_v10_0_select_se_sh(adev, 0xffffffff, 0xffffffff, 0xffffffff);
+	mutex_unlock(&adev->grbm_idx_mutex);
  }
 
  static void gfx_v10_0_get_tcc_info(struct amdgpu_device *adev) @@
-7408,7 +7411,10 @@ static int gfx_v10_0_hw_init(void *handle)
  	 * init golden registers and rlc resume may override some registers,
  	 * reconfig them here
  	 */
-	gfx_v10_0_tcp_harvest(adev);
+	if (adev->asic_type == CHIP_NAVI10 ||
+	    adev->asic_type == CHIP_NAVI14 ||
+	    adev->asic_type == CHIP_NAVI12)
+		gfx_v10_0_tcp_harvest(adev);
 
  	r = gfx_v10_0_cp_resume(adev);
  	if (r)
@@ -9328,17 +9334,22 @@ static void
Quoted 1 line

gfx_v10_0_set_user_wgp_inactive_bitmap_per_sh(struct amdgpu_device

*

Quoted 2 lines

static u32 gfx_v10_0_get_wgp_active_bitmap_per_sh(struct

amdgpu_device *adev)

drivers/gpu/drm/amd/amdgpu/gfx_v10_0.c static void gfx_v10_0_set_user_wgp_inactive_bitmap_per_sh(struct amdgpu_device *

Quoted 4 lines
  {
drivers/gpu/drm/amd/amdgpu/gfx_v10_0.c +0 -3
-	u32 data, wgp_bitmask;
-	data = RREG32_SOC15(GC, 0, mmCC_GC_SHADER_ARRAY_CONFIG);
-	data |= RREG32_SOC15(GC, 0,

mmGC_USER_SHADER_ARRAY_CONFIG);

drivers/gpu/drm/amd/amdgpu/gfx_v10_0.c static void gfx_v10_0_set_user_wgp_inactive_bitmap_per_sh(struct amdgpu_device *

Quoted 7 lines
drivers/gpu/drm/amd/amdgpu/gfx_v10_0.c +2 -0
+	u32 disabled_mask =
+		~amdgpu_gfx_create_bitmask(adev-

gfx.config.max_cu_per_sh >> 1);

drivers/gpu/drm/amd/amdgpu/gfx_v10_0.c +4 -0
+	u32 efuse_setting = 0;
+	u32 vbios_setting = 0;
+
+	efuse_setting = RREG32_SOC15(GC, 0,

mmCC_GC_SHADER_ARRAY_CONFIG);

drivers/gpu/drm/amd/amdgpu/gfx_v10_0.c

Quoted 1 line
drivers/gpu/drm/amd/amdgpu/gfx_v10_0.c +1 -0
+	efuse_setting &=

CC_GC_SHADER_ARRAY_CONFIG__INACTIVE_WGPS_MASK;

drivers/gpu/drm/amd/amdgpu/gfx_v10_0.c

Quoted 1 line
drivers/gpu/drm/amd/amdgpu/gfx_v10_0.c +1 -0
+	efuse_setting >>=

CC_GC_SHADER_ARRAY_CONFIG__INACTIVE_WGPS__SHIFT;

drivers/gpu/drm/amd/amdgpu/gfx_v10_0.c

Quoted 3 lines
drivers/gpu/drm/amd/amdgpu/gfx_v10_0.c +0 -2
-	data &= CC_GC_SHADER_ARRAY_CONFIG__INACTIVE_WGPS_MASK;
-	data >>=

CC_GC_SHADER_ARRAY_CONFIG__INACTIVE_WGPS__SHIFT;

drivers/gpu/drm/amd/amdgpu/gfx_v10_0.c

Quoted 1 line
drivers/gpu/drm/amd/amdgpu/gfx_v10_0.c +1 -0
+	vbios_setting = RREG32_SOC15(GC, 0,

mmGC_USER_SHADER_ARRAY_CONFIG);

drivers/gpu/drm/amd/amdgpu/gfx_v10_0.c

Quoted 1 line
drivers/gpu/drm/amd/amdgpu/gfx_v10_0.c +1 -0
+	vbios_setting &=

GC_USER_SHADER_ARRAY_CONFIG__INACTIVE_WGPS_MASK;

drivers/gpu/drm/amd/amdgpu/gfx_v10_0.c

Quoted 1 line
drivers/gpu/drm/amd/amdgpu/gfx_v10_0.c +1 -0
+	vbios_setting >>=

GC_USER_SHADER_ARRAY_CONFIG__INACTIVE_WGPS__SHIFT;

Quoted 6 lines
drivers/gpu/drm/amd/amdgpu/gfx_v10_0.c +0 -2
-	wgp_bitmask =
-		amdgpu_gfx_create_bitmask(adev-

gfx.config.max_cu_per_sh >> 1);

drivers/gpu/drm/amd/amdgpu/gfx_v10_0.c +1 -0
+	disabled_mask |= efuse_setting | vbios_setting;
Quoted 7 lines
Quoted 7 lines
drivers/gpu/drm/amd/amdgpu/gfx_v10_0.c +1 -1
-	return (~data) & wgp_bitmask;
+	return (~disabled_mask);
  }
 
  static u32 gfx_v10_0_get_cu_active_bitmap_per_sh(struct

amdgpu_device *adev)

_______________________________________________
amd-gfx mailing list
[email protected]
https://lists.freedesktop.org/mailman/listinfo/amd-gfx

#9 lore raw

[Public]

AFAIK, that expression is legal (some code analyzer may warn on value of 4*max_wgp_per_sh); similar kind is used in rotate shift operations.

Thanks,
Lijo

-----Original Message-----
From: Quan, Evan <[email protected]>
Sent: Tuesday, June 22, 2021 7:56 AM
To: Lazar, Lijo <[email protected]>; [email protected]
Cc: Deucher, Alexander <[email protected]>
Subject: RE: [PATCH V3 1/7] drm/amdgpu: correct tcp harvest setting

[AMD Official Use Only]

Thanks Lijo.
However, I'm not quite sure whether " 0xffffffff << (4 * max_wgp_per_sh);" is a valid expression since it kind of triggers some overflow. Can that work for non-x86 platform or even work reliably for x86 platform?

BR
Evan

Quoted 206 lines

-----Original Message-----
From: Lazar, Lijo <[email protected]>
Sent: Monday, June 21, 2021 8:08 PM
To: Quan, Evan <[email protected]>; [email protected]
Cc: Deucher, Alexander <[email protected]>
Subject: Re: [PATCH V3 1/7] drm/amdgpu: correct tcp harvest setting

One minor comment below, apart from that the series looks good.

Quoted 11 lines · Evan Quan

Add missing settings for SQC bits. And correct some confusing logics around active wgp bitmap calculation.

Change-Id If4992e175fd61d5609b00328cbe21f487517d039

---
V1->V2:
- restore correct tcp_harvest setting for NV10 and NV12
- move asic type guard upper layer for better readability

---
  drivers/gpu/drm/amd/amdgpu/gfx_v10_0.c | 103 ++++++++++++++------

-----

drivers/gpu/drm/amd/amdgpu/gfx_v10_0.c static void gfx_v10_0_tcp_harvest(struct

Quoted 8 lines
  1 file changed, 57 insertions(+), 46 deletions(-)
drivers/gpu/drm/amd/amdgpu/gfx_v10_0.c +0 -0
index 15ae9e33b925..384b95fbad8b 100644
@@ -5090,47 +5090,50 @@ static void gfx_v10_0_tcp_harvest(struct

amdgpu_device *adev)

drivers/gpu/drm/amd/amdgpu/gfx_v10_0.c static void gfx_v10_0_tcp_harvest(struct amdgpu_device *adev)

Quoted 11 lines
drivers/gpu/drm/amd/amdgpu/gfx_v10_0.c +0 -8
  		4 + /* RMI */
  		1); /* SQG */
 
-	if (adev->asic_type == CHIP_NAVI10 ||
-	    adev->asic_type == CHIP_NAVI14 ||
-	    adev->asic_type == CHIP_NAVI12) {
-		mutex_lock(&adev->grbm_idx_mutex);
-		for (i = 0; i < adev->gfx.config.max_shader_engines; i++) {
-			for (j = 0; j < adev->gfx.config.max_sh_per_se; j++) {
-				gfx_v10_0_select_se_sh(adev, i, j, 0xffffffff);
-				wgp_active_bitmap =

gfx_v10_0_get_wgp_active_bitmap_per_sh(adev);

drivers/gpu/drm/amd/amdgpu/gfx_v10_0.c static void gfx_v10_0_tcp_harvest(struct · line 5102

Quoted 2 lines
drivers/gpu/drm/amd/amdgpu/gfx_v10_0.c +0 -2
-				/*
-				 * Set corresponding TCP bits for the inactive

WGPs in

drivers/gpu/drm/amd/amdgpu/gfx_v10_0.c static void gfx_v10_0_tcp_harvest(struct amdgpu_device *adev)

Quoted 4 lines
drivers/gpu/drm/amd/amdgpu/gfx_v10_0.c +0 -4
-				 * GCRD_SA_TARGETS_DISABLE
-				 */
-				gcrd_targets_disable_tcp = 0;
-				/* Set TCP & SQC bits in

UTCL1_UTCL0_INVREQ_DISABLE */

drivers/gpu/drm/amd/amdgpu/gfx_v10_0.c static void gfx_v10_0_tcp_harvest(struct amdgpu_device *adev)

Quoted 5 lines
drivers/gpu/drm/amd/amdgpu/gfx_v10_0.c +0 -5
-				utcl_invreq_disable = 0;
-
-				for (k = 0; k < max_wgp_per_sh; k++) {
-					if (!(wgp_active_bitmap & (1 << k))) {
-						gcrd_targets_disable_tcp |=

3 << (2 * k);

drivers/gpu/drm/amd/amdgpu/gfx_v10_0.c static void gfx_v10_0_tcp_harvest(struct · line 5112

Quoted 1 line
drivers/gpu/drm/amd/amdgpu/gfx_v10_0.c +0 -1
-						utcl_invreq_disable |= (3 <<

(2 * k)) |

drivers/gpu/drm/amd/amdgpu/gfx_v10_0.c static void gfx_v10_0_tcp_harvest(struct · line 5113

Quoted 1 line
drivers/gpu/drm/amd/amdgpu/gfx_v10_0.c +0 -1
-							(3 << (2 *

(max_wgp_per_sh + k)));

drivers/gpu/drm/amd/amdgpu/gfx_v10_0.c static void gfx_v10_0_tcp_harvest(struct amdgpu_device *adev)

Quoted 6 lines
drivers/gpu/drm/amd/amdgpu/gfx_v10_0.c +5 -1
-					}
+	mutex_lock(&adev->grbm_idx_mutex);
+	for (i = 0; i < adev->gfx.config.max_shader_engines; i++) {
+		for (j = 0; j < adev->gfx.config.max_sh_per_se; j++) {
+			gfx_v10_0_select_se_sh(adev, i, j, 0xffffffff);
+			wgp_active_bitmap =

gfx_v10_0_get_wgp_active_bitmap_per_sh(adev);

drivers/gpu/drm/amd/amdgpu/gfx_v10_0.c static void gfx_v10_0_tcp_harvest(struct amdgpu_device *adev)

Quoted 6 lines
drivers/gpu/drm/amd/amdgpu/gfx_v10_0.c +6 -0
+			/*
+			 * Set corresponding TCP bits for the inactive WGPs in
+			 * GCRD_SA_TARGETS_DISABLE
+			 */
+			gcrd_targets_disable_tcp = 0;
+			/* Set TCP & SQC bits in

UTCL1_UTCL0_INVREQ_DISABLE */

drivers/gpu/drm/amd/amdgpu/gfx_v10_0.c static void gfx_v10_0_tcp_harvest(struct amdgpu_device *adev)

Quoted 5 lines
drivers/gpu/drm/amd/amdgpu/gfx_v10_0.c +5 -0
+			utcl_invreq_disable = 0;
+
+			for (k = 0; k < max_wgp_per_sh; k++) {
+				if (!(wgp_active_bitmap & (1 << k))) {
+					gcrd_targets_disable_tcp |= 3 << (2 *

k);

drivers/gpu/drm/amd/amdgpu/gfx_v10_0.c static void gfx_v10_0_tcp_harvest(struct · line 5109

Quoted 1 line
drivers/gpu/drm/amd/amdgpu/gfx_v10_0.c +1 -0
+					gcrd_targets_disable_tcp |= 1 << (k +

(max_wgp_per_sh * 2));

drivers/gpu/drm/amd/amdgpu/gfx_v10_0.c static void gfx_v10_0_tcp_harvest(struct · line 5111

Quoted 2 lines
drivers/gpu/drm/amd/amdgpu/gfx_v10_0.c +2 -0
+					utcl_invreq_disable |= (3 << (2 * k)) |
+						(3 << (2 * (max_wgp_per_sh

+ k)));

drivers/gpu/drm/amd/amdgpu/gfx_v10_0.c static void gfx_v10_0_tcp_harvest(struct amdgpu_device *adev)

Quoted 3 lines
drivers/gpu/drm/amd/amdgpu/gfx_v10_0.c +0 -2
  				}
-
-				tmp = RREG32_SOC15(GC, 0,

mmUTCL1_UTCL0_INVREQ_DISABLE);

drivers/gpu/drm/amd/amdgpu/gfx_v10_0.c static void gfx_v10_0_tcp_harvest(struct amdgpu_device *adev)

Quoted 3 lines
drivers/gpu/drm/amd/amdgpu/gfx_v10_0.c +0 -3
-				/* only override TCP & SQC bits */
-				tmp &= 0xffffffff << (4 * max_wgp_per_sh);
-				tmp |= (utcl_invreq_disable &

utcl_invreq_disable_mask);

drivers/gpu/drm/amd/amdgpu/gfx_v10_0.c static void gfx_v10_0_tcp_harvest(struct · line 5121

Quoted 1 line
drivers/gpu/drm/amd/amdgpu/gfx_v10_0.c +0 -1
-				WREG32_SOC15(GC, 0,

mmUTCL1_UTCL0_INVREQ_DISABLE, tmp);

drivers/gpu/drm/amd/amdgpu/gfx_v10_0.c static void gfx_v10_0_tcp_harvest(struct · line 5123

Quoted 2 lines
drivers/gpu/drm/amd/amdgpu/gfx_v10_0.c +0 -2
-
-				tmp = RREG32_SOC15(GC, 0,

mmGCRD_SA_TARGETS_DISABLE);

drivers/gpu/drm/amd/amdgpu/gfx_v10_0.c static void gfx_v10_0_tcp_harvest(struct amdgpu_device *adev)

Quoted 3 lines
drivers/gpu/drm/amd/amdgpu/gfx_v10_0.c +0 -3
-				/* only override TCP bits */
-				tmp &= 0xffffffff << (2 * max_wgp_per_sh);
-				tmp |= (gcrd_targets_disable_tcp &

gcrd_targets_disable_mask);

drivers/gpu/drm/amd/amdgpu/gfx_v10_0.c static void gfx_v10_0_tcp_harvest(struct · line 5127

Quoted 1 line
drivers/gpu/drm/amd/amdgpu/gfx_v10_0.c +0 -1
-				WREG32_SOC15(GC, 0,

mmGCRD_SA_TARGETS_DISABLE, tmp);

drivers/gpu/drm/amd/amdgpu/gfx_v10_0.c static void gfx_v10_0_tcp_harvest(struct amdgpu_device *adev)

Quoted 6 lines
drivers/gpu/drm/amd/amdgpu/gfx_v10_0.c +1 -3
  			}
-		}
 
-		gfx_v10_0_select_se_sh(adev, 0xffffffff, 0xffffffff, 0xffffffff);
-		mutex_unlock(&adev->grbm_idx_mutex);
+			tmp = RREG32_SOC15(GC, 0,

mmUTCL1_UTCL0_INVREQ_DISABLE);

drivers/gpu/drm/amd/amdgpu/gfx_v10_0.c static void gfx_v10_0_tcp_harvest(struct amdgpu_device *adev) · line 5120

Quoted 5 lines
drivers/gpu/drm/amd/amdgpu/gfx_v10_0.c +5 -0
+			/* only override TCP & SQC bits */
+			if (adev->asic_type == CHIP_NAVI14)
+				tmp &= 0xff000000;
+			else
+				tmp &=0xfff00000;

For the disable field mask calculation (which is the value that is applied finally), there is no ASIC check. The above code may utilize the same method as in the original code without ASIC check.

tmp &= 0xffffffff << (4 * max_wgp_per_sh);

Same for below case also - 3*max_wgp_per_sh.

Thanks,
Lijo

drivers/gpu/drm/amd/amdgpu/gfx_v10_0.c static void gfx_v10_0_tcp_harvest(struct · line 5121

Quoted 1 line
drivers/gpu/drm/amd/amdgpu/gfx_v10_0.c +1 -0
+			tmp |= (utcl_invreq_disable &

utcl_invreq_disable_mask);

drivers/gpu/drm/amd/amdgpu/gfx_v10_0.c static void gfx_v10_0_tcp_harvest(struct · line 5122

Quoted 1 line
drivers/gpu/drm/amd/amdgpu/gfx_v10_0.c +1 -0
+			WREG32_SOC15(GC, 0,

mmUTCL1_UTCL0_INVREQ_DISABLE, tmp);

drivers/gpu/drm/amd/amdgpu/gfx_v10_0.c static void gfx_v10_0_tcp_harvest(struct · line 5124

Quoted 2 lines
drivers/gpu/drm/amd/amdgpu/gfx_v10_0.c +2 -0
+
+			tmp = RREG32_SOC15(GC, 0,

mmGCRD_SA_TARGETS_DISABLE);

drivers/gpu/drm/amd/amdgpu/gfx_v10_0.c static void gfx_v10_0_tcp_harvest(struct amdgpu_device *adev)

Quoted 6 lines
drivers/gpu/drm/amd/amdgpu/gfx_v10_0.c +6 -0
+			/* only override TCP & SQC bits */
+			if (adev->asic_type == CHIP_NAVI14)
+				tmp &= 0xfffc0000;
+			else
+				tmp &= 0xffff8000;
+			tmp |= (gcrd_targets_disable_tcp &

gcrd_targets_disable_mask);

drivers/gpu/drm/amd/amdgpu/gfx_v10_0.c static void gfx_v10_0_tcp_harvest(struct · line 5131

Quoted 1 line
drivers/gpu/drm/amd/amdgpu/gfx_v10_0.c +1 -0
+			WREG32_SOC15(GC, 0,

mmGCRD_SA_TARGETS_DISABLE, tmp);

Quoted 21 lines
drivers/gpu/drm/amd/amdgpu/gfx_v10_0.c +8 -2
+		}
  	}
+
+	gfx_v10_0_select_se_sh(adev, 0xffffffff, 0xffffffff, 0xffffffff);
+	mutex_unlock(&adev->grbm_idx_mutex);
  }
 
  static void gfx_v10_0_get_tcc_info(struct amdgpu_device *adev) @@
-7408,7 +7411,10 @@ static int gfx_v10_0_hw_init(void *handle)
  	 * init golden registers and rlc resume may override some registers,
  	 * reconfig them here
  	 */
-	gfx_v10_0_tcp_harvest(adev);
+	if (adev->asic_type == CHIP_NAVI10 ||
+	    adev->asic_type == CHIP_NAVI14 ||
+	    adev->asic_type == CHIP_NAVI12)
+		gfx_v10_0_tcp_harvest(adev);
 
  	r = gfx_v10_0_cp_resume(adev);
  	if (r)
@@ -9328,17 +9334,22 @@ static void
Quoted 1 line

gfx_v10_0_set_user_wgp_inactive_bitmap_per_sh(struct amdgpu_device

*

Quoted 2 lines

static u32 gfx_v10_0_get_wgp_active_bitmap_per_sh(struct

amdgpu_device *adev)

drivers/gpu/drm/amd/amdgpu/gfx_v10_0.c static void gfx_v10_0_set_user_wgp_inactive_bitmap_per_sh(struct amdgpu_device *

Quoted 4 lines
  {
drivers/gpu/drm/amd/amdgpu/gfx_v10_0.c +0 -3
-	u32 data, wgp_bitmask;
-	data = RREG32_SOC15(GC, 0, mmCC_GC_SHADER_ARRAY_CONFIG);
-	data |= RREG32_SOC15(GC, 0,

mmGC_USER_SHADER_ARRAY_CONFIG);

drivers/gpu/drm/amd/amdgpu/gfx_v10_0.c static void gfx_v10_0_set_user_wgp_inactive_bitmap_per_sh(struct amdgpu_device *

Quoted 7 lines
drivers/gpu/drm/amd/amdgpu/gfx_v10_0.c +2 -0
+	u32 disabled_mask =
+		~amdgpu_gfx_create_bitmask(adev-

gfx.config.max_cu_per_sh >> 1);

drivers/gpu/drm/amd/amdgpu/gfx_v10_0.c +4 -0
+	u32 efuse_setting = 0;
+	u32 vbios_setting = 0;
+
+	efuse_setting = RREG32_SOC15(GC, 0,

mmCC_GC_SHADER_ARRAY_CONFIG);

drivers/gpu/drm/amd/amdgpu/gfx_v10_0.c

Quoted 1 line
drivers/gpu/drm/amd/amdgpu/gfx_v10_0.c +1 -0
+	efuse_setting &=

CC_GC_SHADER_ARRAY_CONFIG__INACTIVE_WGPS_MASK;

drivers/gpu/drm/amd/amdgpu/gfx_v10_0.c

Quoted 1 line
drivers/gpu/drm/amd/amdgpu/gfx_v10_0.c +1 -0
+	efuse_setting >>=

CC_GC_SHADER_ARRAY_CONFIG__INACTIVE_WGPS__SHIFT;

drivers/gpu/drm/amd/amdgpu/gfx_v10_0.c

Quoted 3 lines
drivers/gpu/drm/amd/amdgpu/gfx_v10_0.c +0 -2
-	data &= CC_GC_SHADER_ARRAY_CONFIG__INACTIVE_WGPS_MASK;
-	data >>=

CC_GC_SHADER_ARRAY_CONFIG__INACTIVE_WGPS__SHIFT;

drivers/gpu/drm/amd/amdgpu/gfx_v10_0.c

Quoted 1 line
drivers/gpu/drm/amd/amdgpu/gfx_v10_0.c +1 -0
+	vbios_setting = RREG32_SOC15(GC, 0,

mmGC_USER_SHADER_ARRAY_CONFIG);

drivers/gpu/drm/amd/amdgpu/gfx_v10_0.c

Quoted 1 line
drivers/gpu/drm/amd/amdgpu/gfx_v10_0.c +1 -0
+	vbios_setting &=

GC_USER_SHADER_ARRAY_CONFIG__INACTIVE_WGPS_MASK;

drivers/gpu/drm/amd/amdgpu/gfx_v10_0.c

Quoted 1 line
drivers/gpu/drm/amd/amdgpu/gfx_v10_0.c +1 -0
+	vbios_setting >>=

GC_USER_SHADER_ARRAY_CONFIG__INACTIVE_WGPS__SHIFT;

Quoted 6 lines
drivers/gpu/drm/amd/amdgpu/gfx_v10_0.c +0 -2
-	wgp_bitmask =
-		amdgpu_gfx_create_bitmask(adev-

gfx.config.max_cu_per_sh >> 1);

drivers/gpu/drm/amd/amdgpu/gfx_v10_0.c +1 -0
+	disabled_mask |= efuse_setting | vbios_setting;
Quoted 7 lines
Quoted 7 lines
drivers/gpu/drm/amd/amdgpu/gfx_v10_0.c +1 -1
-	return (~data) & wgp_bitmask;
+	return (~disabled_mask);
  }
 
  static u32 gfx_v10_0_get_cu_active_bitmap_per_sh(struct

amdgpu_device *adev)

_______________________________________________
amd-gfx mailing list
[email protected]
https://lists.freedesktop.org/mailman/listinfo/amd-gfx

#10 lore raw
Quoted 3 lines · Lijo

[Public]

AFAIK, that expression is legal (some code analyzer may warn on value of 4*max_wgp_per_sh); similar kind is used in rotate shift operations.

The default type for constants in C is int, so 0xffffffff is a 32-bit signed integer.

The C99 specification lists this under J.2 Undefined behavior:

— An expression having signed promoted type is left-shifted and either the value of the
expression is negative or the result of shifting would be not be representable in the promoted type (6.5.7).

So it would be safer to make it unsigned: 0xffffffffu (or just ~0u).

Quoted 12 lines

-----Original Message-----
From: Quan, Evan <[email protected]>
Sent: Tuesday, June 22, 2021 7:56 AM
To: Lazar, Lijo <[email protected]>; [email protected]
Cc: Deucher, Alexander <[email protected]>
Subject: RE: [PATCH V3 1/7] drm/amdgpu: correct tcp harvest setting

[AMD Official Use Only]

Thanks Lijo.
However, I'm not quite sure whether " 0xffffffff << (4 * max_wgp_per_sh);" is a valid expression since it kind of triggers some overflow. Can that work for non-x86 platform or even work reliably for x86 platform?

Signature · 6 lines
Earthling Michel Dänzer               |               https://redhat.com
Libre software enthusiast             |             Mesa and X developer
_______________________________________________
amd-gfx mailing list
[email protected]
https://lists.freedesktop.org/mailman/listinfo/amd-gfx
#11 lore raw
Quoted 6 lines · Michel Dänzer
Quoted 3 lines · Lijo

[Public]

AFAIK, that expression is legal (some code analyzer may warn on value of 4*max_wgp_per_sh); similar kind is used in rotate shift operations.

The default type for constants in C is int, so 0xffffffff is a 32-bit signed integer.

Probably not as per section 6.4.4.

"The type of an integer constant is the first of the corresponding list in which its value can be represented."

It is a hexadecimal constant and the first to fit this value is unsigned int. Regardless, adding u suffix will avoid any ambiguity.

Thanks,
Lijo

Quoted 25 lines

The C99 specification lists this under J.2 Undefined behavior:

— An expression having signed promoted type is left-shifted and either the value of the
expression is negative or the result of shifting would be not be representable in the promoted type (6.5.7).

So it would be safer to make it unsigned: 0xffffffffu (or just ~0u).

Quoted 12 lines

-----Original Message-----
From: Quan, Evan <[email protected]>
Sent: Tuesday, June 22, 2021 7:56 AM
To: Lazar, Lijo <[email protected]>; [email protected]
Cc: Deucher, Alexander <[email protected]>
Subject: RE: [PATCH V3 1/7] drm/amdgpu: correct tcp harvest setting

[AMD Official Use Only]

Thanks Lijo.
However, I'm not quite sure whether " 0xffffffff << (4 * max_wgp_per_sh);" is a valid expression since it kind of triggers some overflow. Can that work for non-x86 platform or even work reliably for x86 platform?

_______________________________________________
amd-gfx mailing list
[email protected]
https://lists.freedesktop.org/mailman/listinfo/amd-gfx

author
#12 lore raw

[AMD Official Use Only]

Thanks for the information. I will use 0xffffffffu then.

BR
Evan

Quoted 51 lines

-----Original Message-----
From: Lazar, Lijo <[email protected]>
Sent: Tuesday, June 22, 2021 6:40 PM
To: Michel Dänzer <[email protected]>; Quan, Evan <[email protected]>
Cc: Deucher, Alexander <[email protected]>; amd- [email protected]
Subject: Re: [PATCH V3 1/7] drm/amdgpu: correct tcp harvest setting

Quoted 4 lines · Michel Dänzer
Quoted 3 lines · Lijo

[Public]

AFAIK, that expression is legal (some code analyzer may warn on value of

4*max_wgp_per_sh); similar kind is used in rotate shift operations.

Quoted 2 lines

The default type for constants in C is int, so 0xffffffff is a 32-bit signed

integer.

Probably not as per section 6.4.4.

"The type of an integer constant is the first of the corresponding list in which its value can be represented."

It is a hexadecimal constant and the first to fit this value is unsigned int. Regardless, adding u suffix will avoid any ambiguity.

Thanks,
Lijo

Quoted 4 lines

The C99 specification lists this under J.2 Undefined behavior:

— An expression having signed promoted type is left-shifted and either

the value of the

Quoted 1 line

expression is negative or the result of shifting would be not be

representable in the

Quoted 12 lines

promoted type (6.5.7).

So it would be safer to make it unsigned: 0xffffffffu (or just ~0u).

Quoted 7 lines

-----Original Message-----
From: Quan, Evan <[email protected]>
Sent: Tuesday, June 22, 2021 7:56 AM
To: Lazar, Lijo <[email protected]>; [email protected]
Cc: Deucher, Alexander <[email protected]>
Subject: RE: [PATCH V3 1/7] drm/amdgpu: correct tcp harvest setting

Quoted 10 lines
Quoted 4 lines
Quoted 4 lines

[AMD Official Use Only]

Thanks Lijo.
However, I'm not quite sure whether " 0xffffffff << (4 *

max_wgp_per_sh);" is a valid expression since it kind of triggers some overflow.

Quoted 4 lines
Quoted 1 line

Can that work for non-x86 platform or even work reliably for x86 platform?

_______________________________________________
amd-gfx mailing list
[email protected]
https://lists.freedesktop.org/mailman/listinfo/amd-gfx

#13 lore raw
In reply to Lazar, Lijo
Quoted 6 lines · Lijo
Quoted 5 lines · Michel Dänzer
Quoted 3 lines · Lijo

[Public]

AFAIK, that expression is legal (some code analyzer may warn on value of 4*max_wgp_per_sh); similar kind is used in rotate shift operations.

Quoted 7 lines
Quoted 1 line

The default type for constants in C is int, so 0xffffffff is a 32-bit signed integer.

Probably not as per section 6.4.4.

"The type of an integer constant is the first of the corresponding list in which its value can be represented."

It is a hexadecimal constant and the first to fit this value is unsigned int.

Ah, indeed, thanks.

Signature · 6 lines
Earthling Michel Dänzer               |               https://redhat.com
Libre software enthusiast             |             Mesa and X developer
_______________________________________________
amd-gfx mailing list
[email protected]
https://lists.freedesktop.org/mailman/listinfo/amd-gfx