Adjust the dma-ranges for MTK IOMMU
Subsystems (7)
Yong Wu
maintainerv2 cover Adjust the dma-ranges for MTK IOMMU
After commit f1ad5338a4d5 ("of: Fix "dma-ranges" handling for bus controllers"), the dma-ranges is not allowed for dts leaf node. but we still would like to separate the different masters into different iova regions. Thus we adjust the internal flow, separate the 16GB iova range by the master HW larbid/portid and add the dma-ranges property in the parent "soc" node. This also could avoid the users forget/abuse the iova regions.
The commit f1ad5338a4d5 did affect the mt8195 venc, But it is not a fatal issue, it could also work well at 0-4GB iova. thus I don't add "Fixes:" tag.
In this series, I add functions for mt8192/mt8195/mt8186, mt8188 will be in its special patchset. and the previous mt8173/mt8183...support 0-4GB only, no need this function.
Change note:
v2:
a) Base on next-20230206 since mt8195 jpeg node was applied which affect this patchset.
b) Reword the commit message [1/10][2/10] to explain effect.
v1: https://lore.kernel.org/linux-mediatek/[email protected]/
Base on v6.2-rc3.
Yong Wu (10):
dt-bindings: media: mediatek,vcodec: Remove dma-ranges property
dt-bindings: media: mediatek,jpeg: Remove dma-ranges property iommu/mediatek: Get regionid from larb/port id iommu/mediatek: mt8195: Add larb_region_msk iommu/mediatek: mt8186: add larb_region_msk iommu/mediatek: mt8192: add larb_region_msk iommu/mediatek: Add a gap for the iova regions
arm64: dts: mt8195: Add dma-ranges for the parent "soc" node
arm64: dts: mt8195: Remove the unnecessary dma-ranges
arm64: dts: mt8186: Add dma-ranges for the parent "soc" node
.../media/mediatek,mt8195-jpegdec.yaml | 7 -- .../media/mediatek,mt8195-jpegenc.yaml | 7 -- .../media/mediatek,vcodec-decoder.yaml | 5 - .../media/mediatek,vcodec-encoder.yaml | 5 - .../media/mediatek,vcodec-subdev-decoder.yaml | 7 -- .../bindings/media/mediatek-jpeg-encoder.yaml | 5 - arch/arm64/boot/dts/mediatek/mt8186.dtsi | 1 + arch/arm64/boot/dts/mediatek/mt8195.dtsi | 4 +- drivers/iommu/mtk_iommu.c | 95 ++++++++++++++----- 9 files changed, 72 insertions(+), 64 deletions(-)
Signature · 1 line
2.18.0
Yong Wu
maintainerPATCH v2 1/10 dt-bindings: media: mediatek,vcodec: Remove dma-ranges property
After commit f1ad5338a4d5 ("of: Fix "dma-ranges" handling for bus controllers"), the dma-ranges of the leaf node doesn't work. Remove it for vcodec here.
1) For mediatek,vcodec-decoder.yaml and mediatek,vcodec-encoder.yaml, this property is in the leaf node, it is invalid as the above comment.
Currently there is only mt8195 VENC node has this property in upstream. Indeed, VENC is affected, but it is not a fatal issue. Originally it expects its iova range locate at 4GB-8GB. However after that commit, its expectation doesn't come true, it will fall back to 0-4GB iova and also could work well.
2) For mediatek,vcodec-subdev-decoder.yaml, It already uses parent-child node, this property works. Instead, MediaTek iommu will control the masters' iova ranges by the master's larb/port id internally, then the dma-ranges property is unnecessary for the master's node.
--- .../devicetree/bindings/media/mediatek,vcodec-decoder.yaml | 5 ----- .../devicetree/bindings/media/mediatek,vcodec-encoder.yaml | 5 ----- .../bindings/media/mediatek,vcodec-subdev-decoder.yaml | 7 ------- 3 files changed, 17 deletions(-)
Documentation/devicetree/bindings/media/mediatek,vcodec-decoder.yaml +0 -5
@@ -56,11 +56,6 @@ properties: List of the hardware port in respective IOMMU block for current Socs. Refer to bindings/iommu/mediatek,iommu.yaml. - dma-ranges: - maxItems: 1 - description: | - Describes the physical address space of IOMMU maps to memory. - mediatek,vpu: $ref: /schemas/types.yaml#/definitions/phandle description:
Documentation/devicetree/bindings/media/mediatek,vcodec-encoder.yaml +0 -5
@@ -49,11 +49,6 @@ properties: List of the hardware port in respective IOMMU block for current Socs. Refer to bindings/iommu/mediatek,iommu.yaml. - dma-ranges: - maxItems: 1 - description: | - Describes the physical address space of IOMMU maps to memory. - mediatek,vpu: $ref: /schemas/types.yaml#/definitions/phandle description:
Documentation/devicetree/bindings/media/mediatek,vcodec-subdev-decoder.yaml +0 -7
@@ -76,11 +76,6 @@ properties: The node of system control processor (SCP), using the remoteproc & rpmsg framework. - dma-ranges: - maxItems: 1 - description: | - Describes the physical address space of IOMMU maps to memory. - "#address-cells": const: 2
@@ -203,7 +198,6 @@ required: - reg - iommus - mediatek,scp - - dma-ranges - ranges if:
@@ -236,7 +230,6 @@ examples: compatible = "mediatek,mt8192-vcodec-dec"; mediatek,scp = <&scp>; iommus = <&iommu0 M4U_PORT_L4_VDEC_MC_EXT>; - dma-ranges = <0x1 0x0 0x0 0x40000000 0x0 0xfff00000>; #address-cells = <2>; #size-cells = <2>; ranges = <0 0 0 0x16000000 0 0x40000>;
Signature · 1 line
2.18.0
Re: [PATCH v2 01/10] dt-bindings: media: mediatek,vcodec: Remove dma-ranges property
Quoted 25 lines · Yong Wu
After commit
f1ad5338a4d5("of: Fix "dma-ranges" handling for bus controllers"), the dma-ranges of the leaf node doesn't work. Remove it for vcodec here.1) For mediatek,vcodec-decoder.yaml and mediatek,vcodec-encoder.yaml, this property is in the leaf node, it is invalid as the above comment.
Currently there is only mt8195 VENC node has this property in upstream. Indeed, VENC is affected, but it is not a fatal issue. Originally it expects its iova range locate at 4GB-8GB. However after that commit, its expectation doesn't come true, it will fall back to 0-4GB iova and also could work well.
2) For mediatek,vcodec-subdev-decoder.yaml, It already uses parent-child node, this property works. Instead, MediaTek iommu will control the masters' iova ranges by the master's larb/port id internally, then the dma-ranges property is unnecessary for the master's node.
Cc Tiffany Lin <[email protected]>Cc Andrew-CT Chen <[email protected]>Cc Yunfei Dong <[email protected]>Cc Mauro Carvalho Chehab <[email protected]>Cc Matthias Brugger <[email protected]>Signed-off-by Yong Wu <[email protected]>Reviewed-by AngeloGioacchino Del Regno <[email protected]>
Quoted 6 lines
--- .../devicetree/bindings/media/mediatek,vcodec-decoder.yaml | 5 ----- .../devicetree/bindings/media/mediatek,vcodec-encoder.yaml | 5 ----- .../bindings/media/mediatek,vcodec-subdev-decoder.yaml | 7 ------- 3 files changed, 17 deletions(-)
Yong Wu
maintainerPATCH v2 2/10 dt-bindings: media: mediatek,jpeg: Remove dma-ranges property
In reply to Yong WuAfter commit f1ad5338a4d5 ("of: Fix "dma-ranges" handling for bus controllers"), the dma-ranges of the leaf node doesn't work. Remove it for jpeg here.
Currently there is only mt8195 jpeg node has this property in upstream, and it already uses parent-child node, this property did work. But instead, MediaTek iommu will control the masters' iova ranges by the master's larb/port id internally, then this property is unnecessary.
--- .../devicetree/bindings/media/mediatek,mt8195-jpegdec.yaml | 7 ------- .../devicetree/bindings/media/mediatek,mt8195-jpegenc.yaml | 7 ------- .../devicetree/bindings/media/mediatek-jpeg-encoder.yaml | 5 ----- 3 files changed, 19 deletions(-)
Documentation/devicetree/bindings/media/mediatek,mt8195-jpegdec.yaml +0 -7
@@ -26,11 +26,6 @@ properties: Documentation/devicetree/bindings/iommu/mediatek,iommu.yaml for details. Ports are according to the HW. - dma-ranges: - maxItems: 1 - description: | - Describes the physical address space of IOMMU maps to memory. - "#address-cells": const: 2
@@ -89,7 +84,6 @@ required: - compatible - power-domains - iommus - - dma-ranges - ranges additionalProperties: false
@@ -115,7 +109,6 @@ examples: <&iommu_vpp M4U_PORT_L19_JPGDEC_BSDMA1>, <&iommu_vpp M4U_PORT_L19_JPGDEC_BUFF_OFFSET1>, <&iommu_vpp M4U_PORT_L19_JPGDEC_BUFF_OFFSET0>; - dma-ranges = <0x1 0x0 0x0 0x40000000 0x0 0xfff00000>; #address-cells = <2>; #size-cells = <2>; ranges;
Documentation/devicetree/bindings/media/mediatek,mt8195-jpegenc.yaml +0 -7
@@ -26,11 +26,6 @@ properties: Documentation/devicetree/bindings/iommu/mediatek,iommu.yaml for details. Ports are according to the HW. - dma-ranges: - maxItems: 1 - description: | - Describes the physical address space of IOMMU maps to memory. - "#address-cells": const: 2
@@ -89,7 +84,6 @@ required: - compatible - power-domains - iommus - - dma-ranges - ranges additionalProperties: false
@@ -113,7 +107,6 @@ examples: <&iommu_vpp M4U_PORT_L20_JPGENC_C_RDMA>, <&iommu_vpp M4U_PORT_L20_JPGENC_Q_TABLE>, <&iommu_vpp M4U_PORT_L20_JPGENC_BSDMA>; - dma-ranges = <0x1 0x0 0x0 0x40000000 0x0 0xfff00000>; #address-cells = <2>; #size-cells = <2>; ranges;
Documentation/devicetree/bindings/media/mediatek-jpeg-encoder.yaml +0 -5
@@ -44,11 +44,6 @@ properties: Documentation/devicetree/bindings/iommu/mediatek,iommu.yaml for details. Ports are according to the HW. - dma-ranges: - maxItems: 1 - description: | - Describes the physical address space of IOMMU maps to memory. - required: - compatible - reg
Signature · 1 line
2.18.0
Re: [PATCH v2 02/10] dt-bindings: media: mediatek,jpeg: Remove dma-ranges property
Quoted 16 lines · Yong Wu
After commit
f1ad5338a4d5("of: Fix "dma-ranges" handling for bus controllers"), the dma-ranges of the leaf node doesn't work. Remove it for jpeg here.Currently there is only mt8195 jpeg node has this property in upstream, and it already uses parent-child node, this property did work. But instead, MediaTek iommu will control the masters' iova ranges by the master's larb/port id internally, then this property is unnecessary.
Cc Mauro Carvalho Chehab <[email protected]>Cc Matthias Brugger <[email protected]>Cc Bin Liu <[email protected]>Cc kyrie wu <[email protected]>Cc Xia Jiang <[email protected]>Signed-off-by Yong Wu <[email protected]>Reviewed-by AngeloGioacchino Del Regno <[email protected]>
Quoted 6 lines
--- .../devicetree/bindings/media/mediatek,mt8195-jpegdec.yaml | 7 ------- .../devicetree/bindings/media/mediatek,mt8195-jpegenc.yaml | 7 ------- .../devicetree/bindings/media/mediatek-jpeg-encoder.yaml | 5 ----- 3 files changed, 19 deletions(-)
Yong Wu
maintainerPATCH v2 3/10 iommu/mediatek: Get regionid from larb/port id
In reply to Yong WuAfter commit f1ad5338a4d5 ("of: Fix "dma-ranges" handling for bus controllers"), the dma-ranges is not allowed for dts leaf node. but we still would like to separate to different masters into different iova regions.
Thus we have to separate it by the HW larbid and portid. For example, larb1/2 are in region2 and larb3 is in region3. The problem is that some ports inside a larb are in region4 while some ports inside this larb are in region5. Therefore I define a "larb_region_msk" to help record the information for each a port. Take a example for a larb:
[1] = ~0: means all ports in this larb are in region1;
[2] = BIT(3) | BIT(4): means port3/4 in this larb are region2;
[3] = ~(BIT(3) | BIT(4)): means all the other ports except port3/4 in this larb are region3.
This method also avoids the users forget/abuse the iova regions.
--- drivers/iommu/mtk_iommu.c | 43 +++++++++++++++++++++------------------ 1 file changed, 23 insertions(+), 20 deletions(-)
drivers/iommu/mtk_iommu.c +23 -20
@@ -8,7 +8,6 @@#include <linux/clk.h> #include <linux/component.h> #include <linux/device.h> -#include <linux/dma-direct.h> #include <linux/err.h> #include <linux/interrupt.h> #include <linux/io.h>
@@ -194,6 +193,7 @@ struct mtk_iommu_plat_data {enum mtk_iommu_plat m4u_plat;u32 flags;u32 inv_sel_reg; +const u32 (*larb_region_msk)[32 ];char *pericfg_comp_str;struct list_head *hw_list;
@@ -508,30 +508,33 @@ static unsigned int mtk_iommu_get_bank_id(struct device *dev,static int mtk_iommu_get_iova_region_id (struct device *dev,const struct mtk_iommu_plat_data *plat_data) { -const struct mtk_iommu_iova_region *rgn = plat_data->iova_region; -const struct bus_dma_region *dma_rgn = dev->dma_range_map; -int i, candidate =-1 ; -dma_addr_t dma_end; +struct iommu_fwspec *fwspec =dev_iommu_fwspec_get (dev); +unsigned int portidmsk =0 , larbid; +const u32 *rgn_larb_msk; +int i, region_id =-1 ; -if (!dma_rgn || plat_data->iova_region_nr ==1 ) +if (plat_data->iova_region_nr ==1 )return 0 ; - dma_end = dma_rgn->dma_start + dma_rgn->size -1 ; -for (i =0 ; i < plat_data->iova_region_nr; i++, rgn++) { -/* Best fit. */ -if (dma_rgn->dma_start == rgn->iova_base && - dma_end == rgn->iova_base + rgn->size -1 ) -return i; -/* ok if it is inside this region. */ -if (dma_rgn->dma_start >= rgn->iova_base && - dma_end < rgn->iova_base + rgn->size) - candidate = i; + larbid =MTK_M4U_TO_LARB (fwspec->ids[0 ]); +for (i =0 ; i < fwspec->num_ids; i++) + portidmsk |=BIT (MTK_M4U_TO_PORT (fwspec->ids[i])); + +for (i =0 ; i < plat_data->iova_region_nr; i++) { + rgn_larb_msk = plat_data->larb_region_msk[i]; +if (!rgn_larb_msk) +continue ; + +if ((rgn_larb_msk[larbid] & portidmsk) == portidmsk) { + region_id = i; +break ; + } } -if (candidate >=0 ) -return candidate; -dev_err (dev,"Can NOT find the iommu domain id(%pad 0x%llx).\n" , - &dma_rgn->dma_start, dma_rgn->size); +if (region_id >=0 ) +return region_id; +dev_err (dev,"Can NOT find the region for larb(%d-%x).\n" , + larbid, portidmsk);return -EINVAL ; }
Signature · 1 line
2.18.0
AngeloGioacchino Del Regno
maintainerRe: [PATCH v2 03/10] iommu/mediatek: Get regionid from larb/port id
Il 08/02/23 06:36, Yong Wu ha scritto:
Quoted 35 lines
After commit
f1ad5338a4d5("of: Fix "dma-ranges" handling for bus controllers"), the dma-ranges is not allowed for dts leaf node. but we still would like to separate to different masters into different iova regions.Thus we have to separate it by the HW larbid and portid. For example, larb1/2 are in region2 and larb3 is in region3. The problem is that some ports inside a larb are in region4 while some ports inside this larb are in region5. Therefore I define a "
larb_region_msk" to help record the information for each a port. Take a example for a larb:
[1] = ~0: means all ports in this larb are in region1;
[2] = BIT(3) | BIT(4): means port3/4 in this larb are region2;
[3] = ~(BIT(3) | BIT(4)): means all the other ports except port3/4 in this larb are region3.This method also avoids the users forget/abuse the iova regions.
Signed-off-by Yong Wu <[email protected]>--- drivers/iommu/mtk_iommu.c | 43 +++++++++++++++++++++------------------ 1 file changed, 23 insertions(+), 20 deletions(-)drivers/iommu/mtk_iommu.c +0
-1@@ -8,7 +8,6 @@#include <linux/clk.h> #include <linux/component.h> #include <linux/device.h> -#include <linux/dma-direct.h> #include <linux/err.h> #include <linux/interrupt.h> #include <linux/io.h>
drivers/iommu/mtk_iommu.c struct mtk_iommu_plat_data { · line 196
Quoted 4 lines
drivers/iommu/mtk_iommu.c +1
-0@@ -194,6 +193,7 @@ struct mtk_iommu_plat_data {enum mtk_iommu_plat m4u_plat;u32 flags;u32 inv_sel_reg; +const u32 (*larb_region_msk)[32 ];
Can you please document this larb region mask in code, other than the commit description?
I can see this being essential for the next person reading this driver's code without digging through the commit history. At least some comment on top of the pointer, or on top of the struct declaration... and perhaps also describe briefly that the array is "indexed by region" (so 1 = region 1; 2 = region 2) and that the region index corresponds to the same index as mtk_iommu_iova_region.
Before doing that, I'd like to check if anyone else has a better solution for that... because when looking at data for one of the SoCs in here, it looks a bit intimidating!
Copy-paste from patch [04/10] of this series for the reader's commodity:
static const unsigned int mt8195_larb_region_msk[][32] = {
[0] = {~0, ~0, ~0, ~0}, /* Region0: all ports for larb0/1/2/3 */
[1] = {0, 0, 0, 0, 0, 0, 0, 0,
0, 0, 0, 0, 0, 0, 0, 0,
0, 0, 0, ~0, ~0, ~0, ~0, ~0, /* Region1: larb19/20/21/22/23/24 */
~0},
[2] = {0, 0, 0, 0, ~0, ~0, ~0, ~0, /* Region2: the other larbs. */
~0, ~0, ~0, ~0, ~0, ~0, ~0, ~0,
~0, ~0, 0, 0, 0, 0, 0, 0,
0, ~0, ~0, ~0, ~0},
[3] = {0},
[4] = {[18] = BIT(0) | BIT(1)}, /* Only larb18 port0/1 */
[5] = {[18] = BIT(2) | BIT(3)}, /* Only larb18 port2/3 */
};^^^^ That's what I actually mean by "intimidating"... :-P
It's just looks though, there's nothing much complicated here.
Regards,
Angelo
Yong Wu (吴勇)
maintainerRe: [PATCH v2 03/10] iommu/mediatek: Get regionid from larb/port id
Quoted 49 lines · AngeloGioacchino Del Regno
Il 08/02/23 06:36, Yong Wu ha scritto:
Quoted 39 lines
After commit
f1ad5338a4d5("of: Fix "dma-ranges" handling for bus controllers"), the dma-ranges is not allowed for dts leaf node. but we still would like to separate to different masters into different iova regions.Thus we have to separate it by the HW larbid and portid. For example,
larb1/2 are in region2 and larb3 is in region3. The problem is that some ports inside a larb are in region4 while some ports inside this
larb are in region5. Therefore I define a "larb_region_msk" to help record the information for each a port. Take a example for a larb:
[1] = ~0: means all ports in this larb are in region1;
[2] = BIT(3) | BIT(4): means port3/4 in this larb are region2;
[3] = ~(BIT(3) | BIT(4)): means all the other ports except
port3/4
in this larb are region3.This method also avoids the users forget/abuse the iova regions.
Signed-off-by Yong Wu <[email protected]>--- drivers/iommu/mtk_iommu.c | 43 +++++++++++++++++++++------------- ----- 1 file changed, 23 insertions(+), 20 deletions(-)drivers/iommu/mtk_iommu.c +0
-1@@ -8,7 +8,6 @@#include <linux/clk.h> #include <linux/component.h> #include <linux/device.h> -#include <linux/dma-direct.h> #include <linux/err.h> #include <linux/interrupt.h> #include <linux/io.h> drivers/iommu/mtk_iommu.c struct mtk_iommu_plat_data { · line 196
Quoted 4 lines
drivers/iommu/mtk_iommu.c +1
-0@@ -194,6 +193,7 @@ struct mtk_iommu_plat_data {enum mtk_iommu_plat m4u_plat;u32 flags;u32 inv_sel_reg; +const u32 (*larb_region_msk)[32 ];Can you please document this larb region mask in code, other than the commit
description?
Quoted 10 lines
I can see this being essential for the next person reading this driver's code
without digging through the commit history. At least some comment on top of
the pointer, or on top of the struct declaration... and perhaps also describe
briefly that the array is "indexed by region" (so 1 = region 1; 2 = region 2)
and that the region index corresponds to the same index asmtk_iommu_iova_region.
Thanks for this suggestion. I will comment these in the code in next version.
Quoted 27 lines
Before doing that, I'd like to check if anyone else has a better solution for
that... because when looking at data for one of the SoCs in here, it looks a bit intimidating!Copy-paste from patch [04/10] of this series for the reader's commodity:
static const unsigned int mt8195_larb_region_msk[][32] = { [0] = {~0, ~0, ~0, ~0}, /* Region0: all ports for larb0/1/2/3 */ [1] = {0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, ~0, ~0, ~0, ~0, ~0, /* Region1: larb19/20/21/22/23/24 */ ~0}, [2] = {0, 0, 0, 0, ~0, ~0, ~0, ~0, /* Region2: the other larbs. */ ~0, ~0, ~0, ~0, ~0, ~0, ~0, ~0, ~0, ~0, 0, 0, 0, 0, 0, 0, 0, ~0, ~0, ~0, ~0}, [3] = {0}, [4] = {[18] = BIT(0) | BIT(1)}, /* Only larb18 port0/1 */ [5] = {[18] = BIT(2) | BIT(3)}, /* Only larb18 port2/3 */ };
Quoted 3 lines
^^^^ That's what I actually mean by "intimidating"... :-P
It's just looks though, there's nothing much complicated here.
Thanks.
Quoted 5 lines
Regards,
Angelo
Yong Wu
maintainerPATCH v2 4/10 iommu/mediatek: mt8195: Add larb_region_msk
In reply to Yong WuAdd larb_region_msk for mt8195. We separate the 16GB iova regions by each device's larbid/portid.
Refer to include/dt-bindings/memory/mt8195-memory-port.h
--- drivers/iommu/mtk_iommu.c | 17 +++++++++++++++++ 1 file changed, 17 insertions(+)
drivers/iommu/mtk_iommu.c +17 -0
@@ -1537,6 +1537,21 @@ static const struct mtk_iommu_plat_data mt8195_data_infra = { .iova_region_nr = ARRAY_SIZE(single_domain), }; +static const unsigned int mt8195_larb_region_msk[][32 ] = { + [0 ] = {~0 , ~0 , ~0 , ~0 },/* Region0: all ports for larb0/1/2/3 */ + [1 ] = {0 ,0 ,0 ,0 ,0 ,0 ,0 ,0 , +0 ,0 ,0 ,0 ,0 ,0 ,0 ,0 , +0 ,0 ,0 , ~0 , ~0 , ~0 , ~0 , ~0 ,/* Region1: larb19/20/21/22/23/24 */ + ~0 }, + [2 ] = {0 ,0 ,0 ,0 , ~0 , ~0 , ~0 , ~0 ,/* Region2: the other larbs. */ + ~0 , ~0 , ~0 , ~0 , ~0 , ~0 , ~0 , ~0 , + ~0 , ~0 ,0 ,0 ,0 ,0 ,0 ,0 , +0 , ~0 , ~0 , ~0 , ~0 }, + [3 ] = {0 }, + [4 ] = {[18 ] =BIT (0 ) |BIT (1 )},/* Only larb18 port0/1 */ + [5 ] = {[18 ] =BIT (2 ) |BIT (3 )},/* Only larb18 port2/3 */ +}; +static const struct mtk_iommu_plat_data mt8195_data_vdo = { .m4u_plat =M4U_MT8195 , .flags = HAS_BCLK | HAS_SUB_COMM_2BITS | OUT_ORDER_WR_EN |
@@ -1547,6 +1562,7 @@ static const struct mtk_iommu_plat_data mt8195_data_vdo = { .banks_enable = {true}, .iova_region = mt8192_multi_dom, .iova_region_nr = ARRAY_SIZE(mt8192_multi_dom), + .larb_region_msk = mt8195_larb_region_msk, .larbid_remap = {{2, 0}, {21}, {24}, {7}, {19}, {9, 10, 11}, {13, 17, 15/* 17b */, 25}, {5}}, };
@@ -1561,6 +1577,7 @@ static const struct mtk_iommu_plat_data mt8195_data_vpp = { .banks_enable = {true}, .iova_region = mt8192_multi_dom, .iova_region_nr = ARRAY_SIZE(mt8192_multi_dom), + .larb_region_msk = mt8195_larb_region_msk, .larbid_remap = {{1}, {3}, {22, MTK_INVALID_LARBID, MTK_INVALID_LARBID, MTK_INVALID_LARBID, 23}, {8}, {20}, {12},
Signature · 1 line
2.18.0
Yong Wu
maintainerPATCH v2 5/10 iommu/mediatek: mt8186: add larb_region_msk
In reply to Yong WuAdd larb_region_msk for mt8186. We separate the 16GB iova regions by each device's larbid/portid.
Note: larb5/6/10/12/14/15/18 connect nothing in this SoC. Refer to include/dt-bindings/memory/mt8186-memory-port.h
--- drivers/iommu/mtk_iommu.c | 13 +++++++++++++ 1 file changed, 13 insertions(+)
drivers/iommu/mtk_iommu.c +13 -0
@@ -1495,6 +1495,18 @@ static const struct mtk_iommu_plat_data mt8183_data = { .larbid_remap = {{0}, {4}, {5}, {6}, {7}, {2}, {3}, {1}}, }; +static const unsigned int mt8186_larb_region_msk[][32 ] = { + [0 ] = {~0 , ~0 , ~0 },/* Region0: all ports for larb0/1/2 */ + [1 ] = {0 ,0 ,0 ,0 , ~0 ,0 ,0 , ~0 },/* Region1: larb4/7 */ + [2 ] = {0 ,0 ,0 ,0 ,0 ,0 ,0 ,0 ,/* Region2: larb8/9/11/13/16/17/19/20 */ + ~0 , ~0 ,0 , ~0 ,0 , ~(u32)(BIT (9 ) |BIT (10 )),0 ,0 , +/* larb13: the other ports except port9/10 */ + ~0 , ~0 ,0 , ~0 , ~0 }, + [3 ] = {0 }, + [4 ] = {[13 ] =BIT (9 ) |BIT (10 )},/* larb13 port9/10 */ + [5 ] = {[14 ] = ~0 },/* larb14 */ +}; +static const struct mtk_iommu_plat_data mt8186_data_mm = { .m4u_plat =M4U_MT8186 , .flags = HAS_BCLK | HAS_SUB_COMM_2BITS | OUT_ORDER_WR_EN |
@@ -1507,6 +1519,7 @@ static const struct mtk_iommu_plat_data mt8186_data_mm = { .banks_enable = {true}, .iova_region = mt8192_multi_dom, .iova_region_nr = ARRAY_SIZE(mt8192_multi_dom), + .larb_region_msk = mt8186_larb_region_msk, }; static const struct mtk_iommu_plat_data mt8192_data = {
Signature · 1 line
2.18.0
Yong Wu
maintainerPATCH v2 6/10 iommu/mediatek: mt8192: add larb_region_msk
In reply to Yong WuAdd larb_region_msk for mt8192. We separate the 16GB iova regions by each device's larbid/portid.
Note: larb3/6/8/10/12/15 connect nothing in this SoC. Refer to the comment in include/dt-bindings/memory/mt8192-larb-port.h
--- drivers/iommu/mtk_iommu.c | 12 ++++++++++++ 1 file changed, 12 insertions(+)
drivers/iommu/mtk_iommu.c +12 -0
@@ -1522,6 +1522,17 @@ static const struct mtk_iommu_plat_data mt8186_data_mm = { .larb_region_msk = mt8186_larb_region_msk, }; +static const unsigned int mt8192_larb_region_msk[][32 ] = { + [0 ] = {~0 , ~0 },/* Region0: larb0/1 */ + [1 ] = {0 ,0 ,0 ,0 , ~0 , ~0 ,0 , ~0 },/* Region1: larb4/5/7 */ + [2 ] = {0 ,0 , ~0 ,0 ,0 ,0 ,0 ,0 ,/* Region2: larb2/9/11/13/14/16/17/18/19/20 */ +0 , ~0 ,0 , ~0 ,0 , ~(u32)(BIT (9 ) |BIT (10 )), ~(u32)(BIT (4 ) |BIT (5 )),0 , + ~0 , ~0 , ~0 , ~0 , ~0 }, + [3 ] = {0 }, + [4 ] = {[13 ] =BIT (9 ) |BIT (10 )},/* larb13 port9/10 */ + [5 ] = {[14 ] =BIT (4 ) |BIT (5 )},/* larb14 port4/5 */ +}; +static const struct mtk_iommu_plat_data mt8192_data = { .m4u_plat =M4U_MT8192 , .flags = HAS_BCLK | HAS_SUB_COMM_2BITS | OUT_ORDER_WR_EN |
@@ -1531,6 +1542,7 @@ static const struct mtk_iommu_plat_data mt8192_data = { .banks_enable = {true}, .iova_region = mt8192_multi_dom, .iova_region_nr = ARRAY_SIZE(mt8192_multi_dom), + .larb_region_msk = mt8192_larb_region_msk, .larbid_remap = {{0}, {1}, {4, 5}, {7}, {2}, {9, 11, 19, 20}, {0, 14, 16}, {0, 13, 18, 17}}, };
Signature · 1 line
2.18.0
Yong Wu
maintainerPATCH v2 7/10 iommu/mediatek: Add a gap for the iova regions
In reply to Yong WuAs the removed property in the vcodec dt-binding, the property is:
dma-ranges = <0x1 0x0 0x0 0x40000000 0x0 0xfff00000>;
The length is 0xfff0_0000 rather than 0x1_0000_0000, this means it requires 1M as a gap. This is because the end address for some vcodec HW is (address + size). If the size is 4G, the end address may be 0x2_0000_0000, and the width for vcodec register only is 32, then the HW may get the ZERO address.
Currently the consumer's dma-ranges property doesn't work, IOMMU have to consider this problem. Add a bigger gap(8M) for all the regions to avoid it.
--- drivers/iommu/mtk_iommu.c | 10 +++++----- 1 file changed, 5 insertions(+), 5 deletions(-)
drivers/iommu/mtk_iommu.c +5 -5
@@ -304,15 +304,15 @@ static LIST_HEAD(m4ulist); /* List all the M4U HWs */#define for_each_m4u (data, head) list_for_each_entry(data, head, list)static const struct mtk_iommu_iova_region single_domain[] = { - {.iova_base =0 , .size =SZ_4G }, + {.iova_base =0 , .size =SZ_4G -SZ_8M }, };static const struct mtk_iommu_iova_region mt8192_multi_dom[] = { - { .iova_base =0x0 , .size =SZ_4G },/* 0 ~ 4G */ + { .iova_base =0x0 , .size =SZ_4G -SZ_8M },/* 0 ~ 4G, 8M as a gap. */ #if IS_ENABLED(CONFIG_ARCH_DMA_ADDR_T_64BIT) - { .iova_base =SZ_4G , .size =SZ_4G },/* 4G ~ 8G */ - { .iova_base =SZ_4G *2 , .size =SZ_4G },/* 8G ~ 12G */ - { .iova_base =SZ_4G *3 , .size =SZ_4G },/* 12G ~ 16G */ + { .iova_base =SZ_4G , .size =SZ_4G -SZ_8M },/* 4G ~ 8G */ + { .iova_base =SZ_4G *2 , .size =SZ_4G -SZ_8M },/* 8G ~ 12G */ + { .iova_base =SZ_4G *3 , .size =SZ_4G -SZ_8M },/* 12G ~ 16G */ { .iova_base =0x240000000ULL , .size =0x4000000 },/* CCU0 */ { .iova_base = 0x244000000ULL, .size = 0x4000000}, /* CCU1 */
Signature · 1 line
2.18.0
Yong Wu
maintainerPATCH v2 8/10 arm64: dts: mt8195: Add dma-ranges for the parent "soc" node
In reply to Yong WuAfter commit f1ad5338a4d5 ("of: Fix "dma-ranges" handling for bus controllers"), the dma-ranges property is not allowed for the leaf node. But our iommu/dma-ranges is 16GB, we still expect separate the 16GB dma-range like:
a) display is in 0 - 4GB;
b) vcodec is in 4GB - 8GB;
c) camera is in 8GB - 12GB.
We can not expect all the masters add a parent node for them, especial for the existed drivers/nodes.
Thus, we add whole the 16GB dma-ranges in the parent "soc" node.
--- arch/arm64/boot/dts/mediatek/mt8195.dtsi | 1 + 1 file changed, 1 insertion(+)
arch/arm64/boot/dts/mediatek/mt8195.dtsi +1 -0
@@ -364,6 +364,7 @@ #size-cells = <2>; compatible = "simple-bus"; ranges; + dma-ranges = <0x0 0x0 0x0 0x0 0x4 0x0>; gic: interrupt-controller@c000000 { compatible = "arm,gic-v3";
Signature · 1 line
2.18.0
Yong Wu
maintainerPATCH v2 9/10 arm64: dts: mt8195: Remove the unnecessary dma-ranges
In reply to Yong WuAfter we add the dma-ranges in the parent "soc" node, this property is unnecessary for the leaf node.
--- arch/arm64/boot/dts/mediatek/mt8195.dtsi | 3 --- 1 file changed, 3 deletions(-)
arch/arm64/boot/dts/mediatek/mt8195.dtsi +0 -3
@@ -2300,7 +2300,6 @@ power-domains = <&spm MT8195_POWER_DOMAIN_VENC>; #address-cells = <2>; #size-cells = <2>; - dma-ranges = <0x1 0x0 0x0 0x40000000 0x0 0xfff00000>; }; jpgdec-master {
@@ -2312,7 +2311,6 @@ <&iommu_vdo M4U_PORT_L19_JPGDEC_BSDMA1>, <&iommu_vdo M4U_PORT_L19_JPGDEC_BUFF_OFFSET1>, <&iommu_vdo M4U_PORT_L19_JPGDEC_BUFF_OFFSET0>; - dma-ranges = <0x1 0x0 0x0 0x40000000 0x0 0xfff00000>; #address-cells = <2>; #size-cells = <2>; ranges;
@@ -2384,7 +2382,6 @@ <&iommu_vpp M4U_PORT_L20_JPGENC_C_RDMA>, <&iommu_vpp M4U_PORT_L20_JPGENC_Q_TABLE>, <&iommu_vpp M4U_PORT_L20_JPGENC_BSDMA>; - dma-ranges = <0x1 0x0 0x0 0x40000000 0x0 0xfff00000>; #address-cells = <2>; #size-cells = <2>; ranges;
Signature · 1 line
2.18.0
Yong Wu
maintainerPATCH v2 10/10 arm64: dts: mt8186: Add dma-ranges for the parent "soc" node
In reply to Yong WuPrepare for the MM nodes whose dma-ranges(iova range) is 16GB.
--- arch/arm64/boot/dts/mediatek/mt8186.dtsi | 1 + 1 file changed, 1 insertion(+)
arch/arm64/boot/dts/mediatek/mt8186.dtsi +1 -0
@@ -324,6 +324,7 @@ #address-cells = <2>; #size-cells = <2>; compatible = "simple-bus"; + dma-ranges = <0x0 0x0 0x0 0x0 0x4 0x0>; ranges; gic: interrupt-controller@c000000 {
Signature · 1 line
2.18.0