From: Philipp Zabel <p.zabel@pengutronix.de> Date: 2014-02-27 17:22:45
Hi,
this version of the OF graph helper move series addresses a few of
Grant's and Tomi's comments.
Changes since v4:
- Moved graph helpers into drivers/of/base.c
- Fixed endpoint parsing patch to update users
- Improved documentation, emphasizing features that differentiate
the graph bindings from simple phandle graphs. Made it clear that
this is not necessarily specific to data connections
- Added cleanups to of_graph_get_next_endpoint routine
- Added simplified binding for single port devices
regards
Philipp
Philipp Zabel (7):
[media] of: move graph helpers from drivers/media/v4l2-core to
drivers/of
Documentation: of: Document graph bindings
of: Warn if of_graph_get_next_endpoint is called with the root node
of: Reduce indentation in of_graph_get_next_endpoint
[media] of: move common endpoint parsing to drivers/of
of: Implement simplified graph binding for single port devices
of: Document simplified graph binding for single port devices
Documentation/devicetree/bindings/graph.txt | 137 +++++++++++++++++++++
drivers/media/i2c/adv7343.c | 4 +-
drivers/media/i2c/mt9p031.c | 4 +-
drivers/media/i2c/s5k5baf.c | 3 +-
drivers/media/i2c/tvp514x.c | 3 +-
drivers/media/i2c/tvp7002.c | 3 +-
drivers/media/platform/exynos4-is/fimc-is.c | 6 +-
drivers/media/platform/exynos4-is/media-dev.c | 13 +-
drivers/media/platform/exynos4-is/mipi-csis.c | 5 +-
drivers/media/v4l2-core/v4l2-of.c | 133 +--------------------
drivers/of/base.c | 165 ++++++++++++++++++++++++++
include/linux/of_graph.h | 66 +++++++++++
include/media/v4l2-of.h | 33 +-----
13 files changed, 397 insertions(+), 178 deletions(-)
create mode 100644 Documentation/devicetree/bindings/graph.txt
create mode 100644 include/linux/of_graph.h
--
1.8.5.3
From: Philipp Zabel <p.zabel@pengutronix.de> Date: 2014-02-27 17:22:17
A 'return endpoint;' at the end of the (!prev) case allows to
reduce the indentation level of the (prev) case.
Signed-off-by: Philipp Zabel <p.zabel@pengutronix.de>
---
drivers/of/base.c | 42 ++++++++++++++++++++++--------------------
1 file changed, 22 insertions(+), 20 deletions(-)
@@ -2026,32 +2026,34 @@ struct device_node *of_graph_get_next_endpoint(const struct device_node *parent,pr_err("%s(): no endpoint nodes specified for %s\n",__func__,parent->full_name);of_node_put(node);-}else{-port=of_get_parent(prev);-if(WARN_ONCE(!port,"%s(): endpoint has no parent node\n",-__func__))-returnNULL;-/* Avoid dropping prev node refcount to 0. */-of_node_get(prev);-endpoint=of_get_next_child(port,prev);-if(endpoint){-of_node_put(port);-returnendpoint;-}+returnendpoint;+}-/* No more endpoints under this port, try the next one. */-do{-port=of_get_next_child(parent,port);-if(!port)-returnNULL;-}while(of_node_cmp(port->name,"port"));+port=of_get_parent(prev);+if(WARN_ONCE(!port,"%s(): endpoint has no parent node\n",+__func__))+returnNULL;-/* Pick up the first endpoint in this port. */-endpoint=of_get_next_child(port,NULL);+/* Avoid dropping prev node refcount to 0. */+of_node_get(prev);+endpoint=of_get_next_child(port,prev);+if(endpoint){of_node_put(port);+returnendpoint;}+/* No more endpoints under this port, try the next one. */+do{+port=of_get_next_child(parent,port);+if(!port)+returnNULL;+}while(of_node_cmp(port->name,"port"));++/* Pick up the first endpoint in this port. */+endpoint=of_get_next_child(port,NULL);+of_node_put(port);+returnendpoint;}EXPORT_SYMBOL(of_graph_get_next_endpoint);
From: Philipp Zabel <p.zabel@pengutronix.de> Date: 2014-02-27 17:22:19
If of_graph_get_next_endpoint is given a parentless node instead of an
endpoint node, it is clearly a bug.
Signed-off-by: Philipp Zabel <p.zabel@pengutronix.de>
---
drivers/of/base.c | 4 ++--
1 file changed, 2 insertions(+), 2 deletions(-)
@@ -2028,8 +2028,8 @@ struct device_node *of_graph_get_next_endpoint(const struct device_node *parent,of_node_put(node);}else{port=of_get_parent(prev);-if(!port)-/* Hm, has someone given us the root node ?... */+if(WARN_ONCE(!port,"%s(): endpoint has no parent node\n",+__func__))returnNULL;/* Avoid dropping prev node refcount to 0. */
From: Philipp Zabel <p.zabel@pengutronix.de> Date: 2014-02-27 17:22:25
For simple devices with only one port, it can be made implicit.
The endpoint node can be a direct child of the device node.
Signed-off-by: Philipp Zabel <p.zabel@pengutronix.de>
---
Documentation/devicetree/bindings/graph.txt | 8 ++++++++
1 file changed, 8 insertions(+)
@@ -84,6 +84,14 @@ device { }; };+For devices with only a single port and a single endpoint, this can be further+simplified by making the port implicit, and adding the endpoint node as a direct+child of the device node.++device {+ endpoint { ... };+};+ Links between endpoints -----------------------
From: Philipp Zabel <p.zabel@pengutronix.de> Date: 2014-02-27 17:22:32
This patch adds a new struct of_endpoint which is then embedded in struct
v4l2_of_endpoint and contains the endpoint properties that are not V4L2
(or even media) specific: the port number, endpoint id, local device tree
node and remote endpoint phandle. of_graph_parse_endpoint parses those
properties and is used by v4l2_of_parse_endpoint, which just adds the
V4L2 MBUS information to the containing v4l2_of_endpoint structure.
Signed-off-by: Philipp Zabel <p.zabel@pengutronix.de>
---
Changes since v4:
- Fixed users of struct v4l2_of_endpoint
- Removed left-over #include <media/of_graph.h> from v4l2-of.h
---
drivers/media/platform/exynos4-is/media-dev.c | 10 ++++-----
drivers/media/platform/exynos4-is/mipi-csis.c | 2 +-
drivers/media/v4l2-core/v4l2-of.c | 16 +++-----------
drivers/of/base.c | 31 +++++++++++++++++++++++++++
include/linux/of_graph.h | 20 +++++++++++++++++
include/media/v4l2-of.h | 8 ++-----
6 files changed, 62 insertions(+), 25 deletions(-)
@@ -469,10 +469,10 @@ static int fimc_md_parse_port_node(struct fimc_md *fmd,return0;v4l2_of_parse_endpoint(ep,&endpoint);-if(WARN_ON(endpoint.port==0)||index>=FIMC_MAX_SENSORS)+if(WARN_ON(endpoint.base.port==0)||index>=FIMC_MAX_SENSORS)return-EINVAL;-pd->mux_id=(endpoint.port-1)&0x1;+pd->mux_id=(endpoint.base.port-1)&0x1;rem=of_graph_get_remote_port_parent(ep);of_node_put(ep);
@@ -494,13 +494,13 @@ static int fimc_md_parse_port_node(struct fimc_md *fmd,return-EINVAL;}-if(fimc_input_is_parallel(endpoint.port)){+if(fimc_input_is_parallel(endpoint.base.port)){if(endpoint.bus_type==V4L2_MBUS_PARALLEL)pd->sensor_bus_type=FIMC_BUS_TYPE_ITU_601;elsepd->sensor_bus_type=FIMC_BUS_TYPE_ITU_656;pd->flags=endpoint.bus.parallel.flags;-}elseif(fimc_input_is_mipi_csi(endpoint.port)){+}elseif(fimc_input_is_mipi_csi(endpoint.base.port)){/**MIPICSI-2:onlyinputmuxselectionand*thesensor'sclockfrequencyisneeded.
@@ -508,7 +508,7 @@ static int fimc_md_parse_port_node(struct fimc_md *fmd,pd->sensor_bus_type=FIMC_BUS_TYPE_MIPI_CSI2;}else{v4l2_err(&fmd->v4l2_dev,"Wrong port id (%u) at node %s\n",-endpoint.port,rem->full_name);+endpoint.base.port,rem->full_name);}/**ForFIMC-IShandledsensors,thatareplacedunderi2c-ispdevice
@@ -772,7 +772,7 @@ static int s5pcsis_parse_dt(struct platform_device *pdev,/* Get port node and validate MIPI-CSI channel id. */v4l2_of_parse_endpoint(node,&endpoint);-state->index=endpoint.port-FIMC_INPUT_MIPI_CSI2_0;+state->index=endpoint.base.port-FIMC_INPUT_MIPI_CSI2_0;if(state->index<0||state->index>=CSIS_MAX_ENTITIES)return-ENXIO;
From: Philipp Zabel <p.zabel@pengutronix.de> Date: 2014-02-27 17:22:43
This patch moves the parsing helpers used to parse connected graphs
in the device tree, like the video interface bindings documented in
Documentation/devicetree/bindings/media/video-interfaces.txt, from
drivers/media/v4l2-core/v4l2-of.c into drivers/of/base.c.
This allows to reuse the same parser code from outside the V4L2
framework, most importantly from display drivers.
The functions v4l2_of_get_next_endpoint, v4l2_of_get_remote_port,
and v4l2_of_get_remote_port_parent are moved. They are renamed to
of_graph_get_next_endpoint, of_graph_get_remote_port, and
of_graph_get_remote_port_parent, respectively.
Since there are not that many current users yet, switch all of
them to the new functions right away.
Signed-off-by: Philipp Zabel <p.zabel@pengutronix.de>
---
Changes since v4:
- Moved into drivers/of/base.c instead of creating of_graph.c
---
drivers/media/i2c/adv7343.c | 4 +-
drivers/media/i2c/mt9p031.c | 4 +-
drivers/media/i2c/s5k5baf.c | 3 +-
drivers/media/i2c/tvp514x.c | 3 +-
drivers/media/i2c/tvp7002.c | 3 +-
drivers/media/platform/exynos4-is/fimc-is.c | 6 +-
drivers/media/platform/exynos4-is/media-dev.c | 3 +-
drivers/media/platform/exynos4-is/mipi-csis.c | 3 +-
drivers/media/v4l2-core/v4l2-of.c | 117 -------------------------
drivers/of/base.c | 118 ++++++++++++++++++++++++++
include/linux/of_graph.h | 46 ++++++++++
include/media/v4l2-of.h | 25 +-----
12 files changed, 182 insertions(+), 153 deletions(-)
create mode 100644 include/linux/of_graph.h
@@ -473,7 +474,7 @@ static int fimc_md_parse_port_node(struct fimc_md *fmd,pd->mux_id=(endpoint.port-1)&0x1;-rem=v4l2_of_get_remote_port_parent(ep);+rem=of_graph_get_remote_port_parent(ep);of_node_put(ep);if(rem==NULL){v4l2_info(&fmd->v4l2_dev,"Remote device at %s not found\n",
@@ -762,7 +763,7 @@ static int s5pcsis_parse_dt(struct platform_device *pdev,&state->max_num_lanes))return-EINVAL;-node=v4l2_of_get_next_endpoint(node,NULL);+node=of_graph_get_next_endpoint(node,NULL);if(!node){dev_err(&pdev->dev,"No port node at %s\n",pdev->dev.of_node->full_name);
@@ -152,120 +152,3 @@ int v4l2_of_parse_endpoint(const struct device_node *node,return0;}EXPORT_SYMBOL(v4l2_of_parse_endpoint);--/**-*v4l2_of_get_next_endpoint()-getnextendpointnode-*@parent:pointertotheparentdevicenode-*@prev:previousendpointnode,orNULLtogetfirst-*-*Return:An'endpoint'nodepointerwithrefcountincremented.Refcount-*ofthepassed@prevnodeisnotdecremented,thecallerhavetouse-*of_node_put()onitwhendone.-*/-structdevice_node*v4l2_of_get_next_endpoint(conststructdevice_node*parent,-structdevice_node*prev)-{-structdevice_node*endpoint;-structdevice_node*port=NULL;--if(!parent)-returnNULL;--if(!prev){-structdevice_node*node;-/*-*It'sthefirstcall,wehavetofindaportsubnode-*withinthisnodeorwithinanoptional'ports'node.-*/-node=of_get_child_by_name(parent,"ports");-if(node)-parent=node;--port=of_get_child_by_name(parent,"port");--if(port){-/* Found a port, get an endpoint. */-endpoint=of_get_next_child(port,NULL);-of_node_put(port);-}else{-endpoint=NULL;-}--if(!endpoint)-pr_err("%s(): no endpoint nodes specified for %s\n",-__func__,parent->full_name);-of_node_put(node);-}else{-port=of_get_parent(prev);-if(!port)-/* Hm, has someone given us the root node ?... */-returnNULL;--/* Avoid dropping prev node refcount to 0. */-of_node_get(prev);-endpoint=of_get_next_child(port,prev);-if(endpoint){-of_node_put(port);-returnendpoint;-}--/* No more endpoints under this port, try the next one. */-do{-port=of_get_next_child(parent,port);-if(!port)-returnNULL;-}while(of_node_cmp(port->name,"port"));--/* Pick up the first endpoint in this port. */-endpoint=of_get_next_child(port,NULL);-of_node_put(port);-}--returnendpoint;-}-EXPORT_SYMBOL(v4l2_of_get_next_endpoint);--/**-*v4l2_of_get_remote_port_parent()-getremoteport'sparentnode-*@node:pointertoalocalendpointdevice_node-*-*Return:Remotedevicenodeassociatedwithremoteendpointnodelinked-*to@node.Useof_node_put()onitwhendone.-*/-structdevice_node*v4l2_of_get_remote_port_parent(-conststructdevice_node*node)-{-structdevice_node*np;-unsignedintdepth;--/* Get remote endpoint node. */-np=of_parse_phandle(node,"remote-endpoint",0);--/* Walk 3 levels up only if there is 'ports' node. */-for(depth=3;depth&&np;depth--){-np=of_get_next_parent(np);-if(depth==2&&of_node_cmp(np->name,"ports"))-break;-}-returnnp;-}-EXPORT_SYMBOL(v4l2_of_get_remote_port_parent);--/**-*v4l2_of_get_remote_port()-getremoteportnode-*@node:pointertoalocalendpointdevice_node-*-*Return:Remoteportnodeassociatedwithremoteendpointnodelinked-*to@node.Useof_node_put()onitwhendone.-*/-structdevice_node*v4l2_of_get_remote_port(conststructdevice_node*node)-{-structdevice_node*np;--/* Get remote endpoint node. */-np=of_parse_phandle(node,"remote-endpoint",0);-if(!np)-returnNULL;-returnof_get_next_parent(np);-}-EXPORT_SYMBOL(v4l2_of_get_remote_port);
@@ -1982,3 +1983,120 @@ struct device_node *of_find_next_cache_node(const struct device_node *np)returnNULL;}++/**+*of_graph_get_next_endpoint()-getnextendpointnode+*@parent:pointertotheparentdevicenode+*@prev:previousendpointnode,orNULLtogetfirst+*+*Return:An'endpoint'nodepointerwithrefcountincremented.Refcount+*ofthepassed@prevnodeisnotdecremented,thecallerhavetouse+*of_node_put()onitwhendone.+*/+structdevice_node*of_graph_get_next_endpoint(conststructdevice_node*parent,+structdevice_node*prev)+{+structdevice_node*endpoint;+structdevice_node*port=NULL;++if(!parent)+returnNULL;++if(!prev){+structdevice_node*node;+/*+*It'sthefirstcall,wehavetofindaportsubnode+*withinthisnodeorwithinanoptional'ports'node.+*/+node=of_get_child_by_name(parent,"ports");+if(node)+parent=node;++port=of_get_child_by_name(parent,"port");++if(port){+/* Found a port, get an endpoint. */+endpoint=of_get_next_child(port,NULL);+of_node_put(port);+}else{+endpoint=NULL;+}++if(!endpoint)+pr_err("%s(): no endpoint nodes specified for %s\n",+__func__,parent->full_name);+of_node_put(node);+}else{+port=of_get_parent(prev);+if(!port)+/* Hm, has someone given us the root node ?... */+returnNULL;++/* Avoid dropping prev node refcount to 0. */+of_node_get(prev);+endpoint=of_get_next_child(port,prev);+if(endpoint){+of_node_put(port);+returnendpoint;+}++/* No more endpoints under this port, try the next one. */+do{+port=of_get_next_child(parent,port);+if(!port)+returnNULL;+}while(of_node_cmp(port->name,"port"));++/* Pick up the first endpoint in this port. */+endpoint=of_get_next_child(port,NULL);+of_node_put(port);+}++returnendpoint;+}+EXPORT_SYMBOL(of_graph_get_next_endpoint);++/**+*of_graph_get_remote_port_parent()-getremoteport'sparentnode+*@node:pointertoalocalendpointdevice_node+*+*Return:Remotedevicenodeassociatedwithremoteendpointnodelinked+*to@node.Useof_node_put()onitwhendone.+*/+structdevice_node*of_graph_get_remote_port_parent(+conststructdevice_node*node)+{+structdevice_node*np;+unsignedintdepth;++/* Get remote endpoint node. */+np=of_parse_phandle(node,"remote-endpoint",0);++/* Walk 3 levels up only if there is 'ports' node. */+for(depth=3;depth&&np;depth--){+np=of_get_next_parent(np);+if(depth==2&&of_node_cmp(np->name,"ports"))+break;+}+returnnp;+}+EXPORT_SYMBOL(of_graph_get_remote_port_parent);++/**+*of_graph_get_remote_port()-getremoteportnode+*@node:pointertoalocalendpointdevice_node+*+*Return:Remoteportnodeassociatedwithremoteendpointnodelinked+*to@node.Useof_node_put()onitwhendone.+*/+structdevice_node*of_graph_get_remote_port(conststructdevice_node*node)+{+structdevice_node*np;++/* Get remote endpoint node. */+np=of_parse_phandle(node,"remote-endpoint",0);+if(!np)+returnNULL;+returnof_get_next_parent(np);+}+EXPORT_SYMBOL(of_graph_get_remote_port);
From: Philipp Zabel <p.zabel@pengutronix.de> Date: 2014-02-27 17:25:56
For simple devices with only one port, it can be made implicit.
The endpoint node can be a direct child of the device node.
Signed-off-by: Philipp Zabel <p.zabel@pengutronix.de>
---
drivers/of/base.c | 24 +++++++++++++++++++-----
1 file changed, 19 insertions(+), 5 deletions(-)
@@ -2049,8 +2054,6 @@ struct device_node *of_graph_get_next_endpoint(const struct device_node *parent,/* Found a port, get an endpoint. */endpoint=of_get_next_child(port,NULL);of_node_put(port);-}else{-endpoint=NULL;}if(!endpoint)
@@ -2065,6 +2068,10 @@ struct device_node *of_graph_get_next_endpoint(const struct device_node *parent,if(WARN_ONCE(!port,"%s(): endpoint has no parent node\n",__func__))returnNULL;+if(port==parent){+of_node_put(port);+returnNULL;+}/* Avoid dropping prev node refcount to 0. */of_node_get(prev);
@@ -2105,9 +2112,11 @@ struct device_node *of_graph_get_remote_port_parent(/* Get remote endpoint node. */np=of_parse_phandle(node,"remote-endpoint",0);-/* Walk 3 levels up only if there is 'ports' node. */+/* Walk 3 levels up only if there is 'ports' node */for(depth=3;depth&&np;depth--){np=of_get_next_parent(np);+if(depth==3&&of_node_cmp(np->name,"port"))+break;if(depth==2&&of_node_cmp(np->name,"ports"))break;}
From: Philipp Zabel <p.zabel@pengutronix.de> Date: 2014-02-27 17:26:50
The device tree graph bindings as used by V4L2 and documented in
Documentation/device-tree/bindings/media/video-interfaces.txt contain
generic parts that are not media specific but could be useful for any
subsystem with data flow between multiple devices. This document
describe the generic bindings.
Signed-off-by: Philipp Zabel <redacted>
---
Changes since v4:
- Differentiate from graphs made by simple phandle links
- Do not mention data flow except in video-interfaces example
-
---
Documentation/devicetree/bindings/graph.txt | 129 ++++++++++++++++++++++++++++
1 file changed, 129 insertions(+)
create mode 100644 Documentation/devicetree/bindings/graph.txt
@@ -0,0 +1,129 @@+Common bindings for device graphs++General concept+---------------++The hierarchical organisation of the device tree is well suited to describe+control flow to devices, but there can be more complex connections between+devices that work together to form a logical compound device, following an+arbitrarily complex graph.+There already is a simple directed graph between devices tree nodes using+phandle properties pointing to other nodes to describe connections that+can not be inferred from device tree parent-child relationships. The device+tree graph bindings described herein abstract more complex devices that can+have multiple specifiable ports, each of which can be linked to one or more+ports of other devices.++These common bindings do not contain any information about the direction of+type of the connections, they just map their existence. Specific properties+may be described by specialized bindings depending on the type of connection.++To see how this binding applies to video pipelines, for example, see+Documentation/device-tree/bindings/media/video-interfaces.txt.+Here the ports describe data interfaces, and the links between them are+the connecting data buses. A single port with multiple connections can+correspond to multiple devices being connected to the same physical bus.++Organisation of ports and endpoints+-----------------------------------++Ports are described by child 'port' nodes contained in the device node.+Each port node contains an 'endpoint' subnode for each remote device port+connected to this port. If a single port is connected to more than one+remote device, an 'endpoint' child node must be provided for each link.+If more than one port is present in a device node or there is more than one+endpoint at a port, or a port node needs to be associated with a selected+hardware interface, a common scheme using '#address-cells', '#size-cells'+and 'reg' properties is used number the nodes.++device {+ ...+ #address-cells = <1>;+ #size-cells = <0>;++ port@0 {+ #address-cells = <1>;+ #size-cells = <0>;+ reg = <0>;++ endpoint@0 {+ reg = <0>;+ ...+ };+ endpoint@1 {+ reg = <1>;+ ...+ };+ };++ port@1 {+ reg = <1>;++ endpoint { ... };+ };+};++All 'port' nodes can be grouped under an optional 'ports' node, which+allows to specify #address-cells, #size-cells properties for the 'port'+nodes independently from any other child device nodes a device might+have.++device {+ ...+ ports {+ #address-cells = <1>;+ #size-cells = <0>;++ port@0 {+ ...+ endpoint@0 { ... };+ endpoint@1 { ... };+ };++ port@1 { ... };+ };+};++Links between endpoints+-----------------------++Each endpoint should contain a 'remote-endpoint' phandle property that points+to the corresponding endpoint in the port of the remote device. In turn, the+remote endpoint should contain a 'remote-endpoint' property. If it has one,+it must not point to another than the local endpoint. Two endpoints with their+'remote-endpoint' phandles pointing at each other form a link between the+containing ports.++device_1 {+ port {+ device_1_output: endpoint {+ remote-endpoint = <&device_2_input>;+ };+ };+};++device_1 {+ port {+ device_2_input: endpoint {+ remote-endpoint = <&device_1_output>;+ };+ };+};+++Required properties+-------------------++If there is more than one 'port' or more than one 'endpoint' node or 'reg'+property is present in port and/or endpoint nodes the following properties+are required in a relevant parent node:++ - #address-cells : number of cells required to define port/endpoint+ identifier, should be 1.+ - #size-cells : should be zero.++Optional endpoint properties+----------------------------++- remote-endpoint: phandle to an 'endpoint' subnode of a remote device node.+
--
1.8.5.3
--
To unsubscribe from this list: send the line "unsubscribe devicetree" in
the body of a message to majordomo-u79uwXL29TY76Z2rM5mHXA@public.gmane.org
More majordomo info at http://vger.kernel.org/majordomo-info.html
Hi Philipp,
Just couple minor comments...
On 02/27/2014 06:35 PM, Philipp Zabel wrote:
The device tree graph bindings as used by V4L2 and documented in
Documentation/device-tree/bindings/media/video-interfaces.txt contain
generic parts that are not media specific but could be useful for any
subsystem with data flow between multiple devices. This document
describe the generic bindings.
s/describe/describes/
quoted hunk
Signed-off-by: Philipp Zabel<p.zabel@pengutronix.de>
---
Changes since v4:
- Differentiate from graphs made by simple phandle links
- Do not mention data flow except in video-interfaces example
-
---
Documentation/devicetree/bindings/graph.txt | 129 ++++++++++++++++++++++++++++
1 file changed, 129 insertions(+)
create mode 100644 Documentation/devicetree/bindings/graph.txt
@@ -0,0 +1,129 @@+Common bindings for device graphs++General concept+---------------++The hierarchical organisation of the device tree is well suited to describe+control flow to devices, but there can be more complex connections between+devices that work together to form a logical compound device, following an+arbitrarily complex graph.+There already is a simple directed graph between devices tree nodes using+phandle properties pointing to other nodes to describe connections that+can not be inferred from device tree parent-child relationships. The device+tree graph bindings described herein abstract more complex devices that can+have multiple specifiable ports, each of which can be linked to one or more+ports of other devices.++These common bindings do not contain any information about the direction of
s/of/or/ ?
+type of the connections, they just map their existence. Specific properties
+may be described by specialized bindings depending on the type of connection.
+
+To see how this binding applies to video pipelines, for example, see
+Documentation/device-tree/bindings/media/video-interfaces.txt.
+Here the ports describe data interfaces, and the links between them are
+the connecting data buses. A single port with multiple connections can
+correspond to multiple devices being connected to the same physical bus.
+
+Organisation of ports and endpoints
+-----------------------------------
+
+Ports are described by child 'port' nodes contained in the device node.
+Each port node contains an 'endpoint' subnode for each remote device port
+connected to this port. If a single port is connected to more than one
+remote device, an 'endpoint' child node must be provided for each link.
+If more than one port is present in a device node or there is more than one
+endpoint at a port, or a port node needs to be associated with a selected
+hardware interface, a common scheme using '#address-cells', '#size-cells'
+and 'reg' properties is used number the nodes.
+
+device {
+ ...
+ #address-cells =<1>;
+ #size-cells =<0>;
+
+ port@0 {
+ #address-cells =<1>;
+ #size-cells =<0>;
+ reg =<0>;
+
+ endpoint@0 {
+ reg =<0>;
+ ...
+ };
+ endpoint@1 {
+ reg =<1>;
+ ...
+ };
+ };
+
+ port@1 {
+ reg =<1>;
+
+ endpoint { ... };
+ };
+};
+
+All 'port' nodes can be grouped under an optional 'ports' node, which
+allows to specify #address-cells, #size-cells properties for the 'port'
+nodes independently from any other child device nodes a device might
+have.
+
+device {
+ ...
+ ports {
+ #address-cells =<1>;
+ #size-cells =<0>;
+
+ port@0 {
+ ...
+ endpoint@0 { ... };
+ endpoint@1 { ... };
+ };
+
+ port@1 { ... };
+ };
+};
+
+Links between endpoints
+-----------------------
+
+Each endpoint should contain a 'remote-endpoint' phandle property that points
+to the corresponding endpoint in the port of the remote device. In turn, the
+remote endpoint should contain a 'remote-endpoint' property. If it has one,
+it must not point to another than the local endpoint. Two endpoints with their
+'remote-endpoint' phandles pointing at each other form a link between the
+containing ports.
+
+device_1 {
+ port {
+ device_1_output: endpoint {
+ remote-endpoint =<&device_2_input>;
+ };
+ };
+};
+
+device_1 {
s/device_1/device_2/
But it might be better to use dashes instead of underscores
for the node names.
+ port {
+ device_2_input: endpoint {
+ remote-endpoint =<&device_1_output>;
+ };
+ };
+};
+
+
+Required properties
+-------------------
+
+If there is more than one 'port' or more than one 'endpoint' node or 'reg'
+property is present in port and/or endpoint nodes the following properties
+are required in a relevant parent node:
+
+ - #address-cells : number of cells required to define port/endpoint
+ identifier, should be 1.
+ - #size-cells : should be zero.
+
+Optional endpoint properties
+----------------------------
+
+- remote-endpoint: phandle to an 'endpoint' subnode of a remote device node.
+
If of_graph_get_next_endpoint is given a parentless node instead of an
endpoint node, it is clearly a bug.
Signed-off-by: Philipp Zabel<p.zabel@pengutronix.de>
---
drivers/of/base.c | 4 ++--
1 file changed, 2 insertions(+), 2 deletions(-)
@@ -2028,8 +2028,8 @@ struct device_node *of_graph_get_next_endpoint(const struct device_node *parent,of_node_put(node);}else{port=of_get_parent(prev);-if(!port)-/* Hm, has someone given us the root node ?... */+if(WARN_ONCE(!port,"%s(): endpoint has no parent node\n",+__func__))
Perhaps we can add more information to this error log, e.g.
if (WARN_ONCE(!port, "%s(): endpoint %s has no parent node\n",
__func__, prev->full_name))
?
This patch adds a new struct of_endpoint which is then embedded in struct
v4l2_of_endpoint and contains the endpoint properties that are not V4L2
(or even media) specific: the port number, endpoint id, local device tree
node and remote endpoint phandle. of_graph_parse_endpoint parses those
properties and is used by v4l2_of_parse_endpoint, which just adds the
V4L2 MBUS information to the containing v4l2_of_endpoint structure.
Signed-off-by: Philipp Zabel<p.zabel@pengutronix.de>
---
Changes since v4:
- Fixed users of struct v4l2_of_endpoint
- Removed left-over #include<media/of_graph.h> from v4l2-of.h
---
drivers/media/platform/exynos4-is/media-dev.c | 10 ++++-----
drivers/media/platform/exynos4-is/mipi-csis.c | 2 +-
drivers/media/v4l2-core/v4l2-of.c | 16 +++-----------
drivers/of/base.c | 31 +++++++++++++++++++++++++++
include/linux/of_graph.h | 20 +++++++++++++++++
include/media/v4l2-of.h | 8 ++-----
6 files changed, 62 insertions(+), 25 deletions(-)
@@ -469,10 +469,10 @@ static int fimc_md_parse_port_node(struct fimc_md *fmd,return0;v4l2_of_parse_endpoint(ep,&endpoint);-if(WARN_ON(endpoint.port==0)||index>=FIMC_MAX_SENSORS)+if(WARN_ON(endpoint.base.port==0)||index>=FIMC_MAX_SENSORS)return-EINVAL;-pd->mux_id=(endpoint.port-1)&0x1;+pd->mux_id=(endpoint.base.port-1)&0x1;rem=of_graph_get_remote_port_parent(ep);of_node_put(ep);
@@ -494,13 +494,13 @@ static int fimc_md_parse_port_node(struct fimc_md *fmd,return-EINVAL;}-if(fimc_input_is_parallel(endpoint.port)){+if(fimc_input_is_parallel(endpoint.base.port)){if(endpoint.bus_type==V4L2_MBUS_PARALLEL)pd->sensor_bus_type=FIMC_BUS_TYPE_ITU_601;elsepd->sensor_bus_type=FIMC_BUS_TYPE_ITU_656;pd->flags=endpoint.bus.parallel.flags;-}elseif(fimc_input_is_mipi_csi(endpoint.port)){+}elseif(fimc_input_is_mipi_csi(endpoint.base.port)){/**MIPICSI-2:onlyinputmuxselectionand*thesensor'sclockfrequencyisneeded.
@@ -508,7 +508,7 @@ static int fimc_md_parse_port_node(struct fimc_md *fmd,pd->sensor_bus_type=FIMC_BUS_TYPE_MIPI_CSI2;}else{v4l2_err(&fmd->v4l2_dev,"Wrong port id (%u) at node %s\n",-endpoint.port,rem->full_name);+endpoint.base.port,rem->full_name);}/**ForFIMC-IShandledsensors,thatareplacedunderi2c-ispdevice
@@ -772,7 +772,7 @@ static int s5pcsis_parse_dt(struct platform_device *pdev,/* Get port node and validate MIPI-CSI channel id. */v4l2_of_parse_endpoint(node,&endpoint);-state->index=endpoint.port-FIMC_INPUT_MIPI_CSI2_0;+state->index=endpoint.base.port-FIMC_INPUT_MIPI_CSI2_0;if(state->index<0||state->index>=CSIS_MAX_ENTITIES)return-ENXIO;
I don't think these two sentences are needed, it's all described in the
DT binding documentation. And struct of_endpoint doesn't contain any
"flags" field.
quoted hunk
+ * The caller should hold a reference to @node.
+ */
+int of_graph_parse_endpoint(const struct device_node *node,
+ struct of_endpoint *endpoint)
+{
+ struct device_node *port_node = of_get_parent(node);
+
+ memset(endpoint, 0, sizeof(*endpoint));
+
+ endpoint->local_node = node;
+ /*
+ * It doesn't matter whether the two calls below succeed.
+ * If they don't then the default value 0 is used.
+ */
+ of_property_read_u32(port_node, "reg",&endpoint->port);
+ of_property_read_u32(node, "reg",&endpoint->id);
+
+ of_node_put(port_node);
+
+ return 0;
+}
+EXPORT_SYMBOL(of_graph_parse_endpoint);
+
+/**
* of_graph_get_next_endpoint() - get next endpoint node
* @parent: pointer to the parent device node
* @prev: previous endpoint node, or NULL to get first
From: Tomi Valkeinen <hidden> Date: 2014-03-04 08:58:53
Hi Philipp,
On 27/02/14 19:35, Philipp Zabel wrote:
This patch adds a new struct of_endpoint which is then embedded in struct
v4l2_of_endpoint and contains the endpoint properties that are not V4L2
(or even media) specific: the port number, endpoint id, local device tree
node and remote endpoint phandle. of_graph_parse_endpoint parses those
properties and is used by v4l2_of_parse_endpoint, which just adds the
V4L2 MBUS information to the containing v4l2_of_endpoint structure.
Can port_node be NULL? Probably only if something is quite wrong, but
maybe it's safer to return error in that case.
+ memset(endpoint, 0, sizeof(*endpoint));
+
+ endpoint->local_node = node;
+ /*
+ * It doesn't matter whether the two calls below succeed.
+ * If they don't then the default value 0 is used.
+ */
+ of_property_read_u32(port_node, "reg", &endpoint->port);
+ of_property_read_u32(node, "reg", &endpoint->id);
If the endpoint does not have 'port' as parent (i.e. the shortened
format), the above will return the 'reg' of the device node (with
'device node' I mean the main node, with 'compatible' property).
And generally speaking, if struct of_endpoint is needed, maybe it would
be better to return the struct of_endpoint when iterating the ports and
endpoints. That way there's no need to do parsing "afterwards", trying
to figure out if there's a parent port node, but the information is
already at hand.
quoted hunk
+
+ of_node_put(port_node);
+
+ return 0;
+}
+EXPORT_SYMBOL(of_graph_parse_endpoint);
+
+/**
* of_graph_get_next_endpoint() - get next endpoint node
* @parent: pointer to the parent device node
* @prev: previous endpoint node, or NULL to get first
A few thoughts about the iteration, and the API in general.
In the omapdss version I separated iterating ports and endpoints, for
the two reasons:
1) I think there are cases where you may want to have properties in the
port node, for things that are common for all the port's endpoints.
2) if there are multiple ports, I think the driver code is cleaner if
you first take the port, decide what port that is and maybe call a
sub-function, and then iterate the endpoints for that port only.
Both of those are possible with the API in the series, but not very cleanly.
Also, if you just want to iterate the endpoints, it's easy to implement
a helper using the separate port and endpoint iterations.
Then, about the get_remote functions: I think there should be only one
function for that purpose, one that returns the device node that
contains the remote endpoint.
My reasoning is that the ports and endpoints, and their contents, should
be private to the device. So the only use to get the remote is to get
the actual device, to see if it's been probed, or maybe get some video
API for that device.
If the driver model used has some kind of master-driver, which goes
through all the display entities, I think the above is still valid. When
the master-driver follows the remote-link, it still needs to first get
the main device node, as the ports and endpoints make no sense without
the context of the main device node.
Tomi
From: Tomi Valkeinen <hidden> Date: 2014-03-04 09:07:07
On 27/02/14 19:35, Philipp Zabel wrote:
For simple devices with only one port, it can be made implicit.
The endpoint node can be a direct child of the device node.
<snip>
quoted hunk
@@ -2105,9 +2112,11 @@ struct device_node *of_graph_get_remote_port_parent( /* Get remote endpoint node. */ np = of_parse_phandle(node, "remote-endpoint", 0);- /* Walk 3 levels up only if there is 'ports' node. */+ /* Walk 3 levels up only if there is 'ports' node */ for (depth = 3; depth && np; depth--) { np = of_get_next_parent(np);+ if (depth == 3 && of_node_cmp(np->name, "port"))+ break; if (depth == 2 && of_node_cmp(np->name, "ports")) break; }
This function becomes a bit funny. Would it be clearer just to do
something like:
np = of_parse_phandle(node, "remote-endpoint", 0);
np = of_get_next_parent(np);
if (of_node_cmp(np->name, "port") != 0)
return np;
np = of_get_next_parent(np);
if (of_node_cmp(np->name, "ports") != 0)
return np;
np = of_get_next_parent(np);
return np;
Tomi
From: Philipp Zabel <p.zabel@pengutronix.de> Date: 2014-03-04 10:04:43
Am Dienstag, den 04.03.2014, 11:06 +0200 schrieb Tomi Valkeinen:
On 27/02/14 19:35, Philipp Zabel wrote:
quoted
For simple devices with only one port, it can be made implicit.
The endpoint node can be a direct child of the device node.
<snip>
quoted
@@ -2105,9 +2112,11 @@ struct device_node *of_graph_get_remote_port_parent( /* Get remote endpoint node. */ np = of_parse_phandle(node, "remote-endpoint", 0);- /* Walk 3 levels up only if there is 'ports' node. */+ /* Walk 3 levels up only if there is 'ports' node */ for (depth = 3; depth && np; depth--) { np = of_get_next_parent(np);+ if (depth == 3 && of_node_cmp(np->name, "port"))+ break; if (depth == 2 && of_node_cmp(np->name, "ports")) break; }
This function becomes a bit funny. Would it be clearer just to do
something like:
np = of_parse_phandle(node, "remote-endpoint", 0);
np = of_get_next_parent(np);
if (of_node_cmp(np->name, "port") != 0)
return np;
np = of_get_next_parent(np);
if (of_node_cmp(np->name, "ports") != 0)
return np;
np = of_get_next_parent(np);
return np;
I'm not sure if this was initially crafted to reduce the number of
function calls, but rolled out it certainly is easier to read. I'll
change this as you suggest.
thanks
Philipp
I don't think these two sentences are needed, it's all described in the
DT binding documentation. And struct of_endpoint doesn't contain any
"flags" field.
From: Philipp Zabel <p.zabel@pengutronix.de> Date: 2014-03-04 10:12:40
Am Freitag, den 28.02.2014, 22:09 +0100 schrieb Sylwester Nawrocki:
On 02/27/2014 06:35 PM, Philipp Zabel wrote:
quoted
If of_graph_get_next_endpoint is given a parentless node instead of an
endpoint node, it is clearly a bug.
Signed-off-by: Philipp Zabel<p.zabel@pengutronix.de>
---
drivers/of/base.c | 4 ++--
1 file changed, 2 insertions(+), 2 deletions(-)
@@ -2028,8 +2028,8 @@ struct device_node *of_graph_get_next_endpoint(const struct device_node *parent,of_node_put(node);}else{port=of_get_parent(prev);-if(!port)-/* Hm, has someone given us the root node ?... */+if(WARN_ONCE(!port,"%s(): endpoint has no parent node\n",+__func__))
Perhaps we can add more information to this error log, e.g.
if (WARN_ONCE(!port, "%s(): endpoint %s has no parent node\n",
__func__, prev->full_name))
?
Yes, I'll include this change next time.
thanks
Philipp
From: Philipp Zabel <p.zabel@pengutronix.de> Date: 2014-03-04 10:17:18
Hi Sylwester,
Am Freitag, den 28.02.2014, 22:08 +0100 schrieb Sylwester Nawrocki:
Hi Philipp,
Just couple minor comments...
Thanks, I'll fix all of those.
On 02/27/2014 06:35 PM, Philipp Zabel wrote:
quoted
The device tree graph bindings as used by V4L2 and documented in
Documentation/device-tree/bindings/media/video-interfaces.txt contain
generic parts that are not media specific but could be useful for any
subsystem with data flow between multiple devices. This document
describe the generic bindings.
s/describe/describes/
[...]
quoted
+These common bindings do not contain any information about the direction of
Can port_node be NULL? Probably only if something is quite wrong, but
maybe it's safer to return error in that case.
both of_property_read_u32 and of_node_put can handle port_node == NULL.
I'll add a WARN_ONCE here as for of_graph_get_next_endpoint and continue
on.
quoted
+ memset(endpoint, 0, sizeof(*endpoint));
+
+ endpoint->local_node = node;
+ /*
+ * It doesn't matter whether the two calls below succeed.
+ * If they don't then the default value 0 is used.
+ */
+ of_property_read_u32(port_node, "reg", &endpoint->port);
+ of_property_read_u32(node, "reg", &endpoint->id);
If the endpoint does not have 'port' as parent (i.e. the shortened
format), the above will return the 'reg' of the device node (with
'device node' I mean the main node, with 'compatible' property).
Yes, a check for the port_node name is in order.
And generally speaking, if struct of_endpoint is needed, maybe it would
be better to return the struct of_endpoint when iterating the ports and
endpoints. That way there's no need to do parsing "afterwards", trying
to figure out if there's a parent port node, but the information is
already at hand.
I'd like to keep the iteration separate from parsing so we can
eventually introduce a for_each_endpoint_of_node helper macro around
of_graph_get_next_endpoint.
[...]
A few thoughts about the iteration, and the API in general.
In the omapdss version I separated iterating ports and endpoints, for
the two reasons:
1) I think there are cases where you may want to have properties in the
port node, for things that are common for all the port's endpoints.
2) if there are multiple ports, I think the driver code is cleaner if
you first take the port, decide what port that is and maybe call a
sub-function, and then iterate the endpoints for that port only.
It depends a bit on whether you are actually iterating over individual
ports, or if you are just walking the whole endpoint graph to find
remote devices that have to be added to the component master's waiting
list, for example.
Both of those are possible with the API in the series, but not very cleanly.
Also, if you just want to iterate the endpoints, it's easy to implement
a helper using the separate port and endpoint iterations.
I started out to move an existing (albeit lightly used) API to a common
place so others can use it and improve upon it, too. I'm happy to pile
on fixes directly in this series, but could we separate the improvement
step from the move, for the bigger modifications?
I had no immediate use for the port iteration, so I have taken no steps
to add a function for this. I see no problem to add this later when
somebody needs it, or even rewrite of_graph_get_next_endpoint to use it
if it is feasible. Iterating over endpoints on a given port needs no
helper, as you can just use for_each_child_of_node.
Then, about the get_remote functions: I think there should be only one
function for that purpose, one that returns the device node that
contains the remote endpoint.
My reasoning is that the ports and endpoints, and their contents, should
be private to the device. So the only use to get the remote is to get
the actual device, to see if it's been probed, or maybe get some video
API for that device.
of_graph_get_remote_port currently is used in the exynos4-is/fimc-is.c
v4l2 driver to get the mipi-csi channel id from the remote port, and
I've started using it in imx-drm-core.c for two cases:
- given an endpoint on the encoder, find the remote port connected to
it, get the associated drm_crtc, to obtain its the drm_crtc_mask
for encoder->possible_crtcs.
- given an encoder and a connected drm_crtc, walk all endpoints to find
the remote port associated with the drm_crtc, and then use the local
endpoint parent port to determine multiplexer settings.
If the driver model used has some kind of master-driver, which goes
through all the display entities, I think the above is still valid. When
the master-driver follows the remote-link, it still needs to first get
the main device node, as the ports and endpoints make no sense without
the context of the main device node.
I'm not sure about this. I might just need the remote port node
associated with a remote drm_crtc or drm_encoder structure to find out
which local endpoint I should look at to retrieve configuration.
regards
Philipp
Can port_node be NULL? Probably only if something is quite wrong, but
maybe it's safer to return error in that case.
both of_property_read_u32 and of_node_put can handle port_node == NULL.
I'll add a WARN_ONCE here as for of_graph_get_next_endpoint and continue
on.
Isn't it better to return an error?
quoted
And generally speaking, if struct of_endpoint is needed, maybe it would
be better to return the struct of_endpoint when iterating the ports and
endpoints. That way there's no need to do parsing "afterwards", trying
to figure out if there's a parent port node, but the information is
already at hand.
I'd like to keep the iteration separate from parsing so we can
eventually introduce a for_each_endpoint_of_node helper macro around
of_graph_get_next_endpoint.
[...]
quoted
A few thoughts about the iteration, and the API in general.
In the omapdss version I separated iterating ports and endpoints, for
the two reasons:
1) I think there are cases where you may want to have properties in the
port node, for things that are common for all the port's endpoints.
2) if there are multiple ports, I think the driver code is cleaner if
you first take the port, decide what port that is and maybe call a
sub-function, and then iterate the endpoints for that port only.
It depends a bit on whether you are actually iterating over individual
ports, or if you are just walking the whole endpoint graph to find
remote devices that have to be added to the component master's waiting
list, for example.
True, but the latter is easily implemented using the separate
port/endpoint iteration. So I see it as a more powerful API.
quoted
Both of those are possible with the API in the series, but not very cleanly.
Also, if you just want to iterate the endpoints, it's easy to implement
a helper using the separate port and endpoint iterations.
I started out to move an existing (albeit lightly used) API to a common
place so others can use it and improve upon it, too. I'm happy to pile
on fixes directly in this series, but could we separate the improvement
step from the move, for the bigger modifications?
Yes, I understand that. What I wonder is that which is easier: make it a
public API now, more or less as it was in v4l2, or make it a public API
only when all the improvements we can think of have been made.
So my fear is that the API is now made public, and you and others start
to use it. But I can't use it, as I need things like separate
port/endpoint iteration. I need to add those, which also means that I
need to change all the users of the API, making the task more difficult
than I'd like.
However, this is more of "thinking out loud" than "I don't like the
series". It's a good series =).
I had no immediate use for the port iteration, so I have taken no steps
to add a function for this. I see no problem to add this later when
somebody needs it, or even rewrite of_graph_get_next_endpoint to use it
if it is feasible. Iterating over endpoints on a given port needs no
helper, as you can just use for_each_child_of_node.
I would have a helper, which should do some sanity checks, like that the
node names are "endpoint".
quoted
Then, about the get_remote functions: I think there should be only one
function for that purpose, one that returns the device node that
contains the remote endpoint.
My reasoning is that the ports and endpoints, and their contents, should
be private to the device. So the only use to get the remote is to get
the actual device, to see if it's been probed, or maybe get some video
API for that device.
of_graph_get_remote_port currently is used in the exynos4-is/fimc-is.c
v4l2 driver to get the mipi-csi channel id from the remote port, and
I've started using it in imx-drm-core.c for two cases:
- given an endpoint on the encoder, find the remote port connected to
it, get the associated drm_crtc, to obtain its the drm_crtc_mask
for encoder->possible_crtcs.
- given an encoder and a connected drm_crtc, walk all endpoints to find
the remote port associated with the drm_crtc, and then use the local
endpoint parent port to determine multiplexer settings.
Ok.
In omapdss each driver handles only the ports and endpoints defined for
its device, and they can be considered private to that device. The only
reason to look for the remote endpoint is to find the remote device. To
me the omapdss model makes sense, and feels logical and sane =). So I
have to say I'm not really familiar with the model you're using.
Of course, the different get_remove_* funcs do no harm, even if we have
a bunch of them. My point was only about enforcing the correct use of
the model, where "correct" is of course subjective =).
quoted
If the driver model used has some kind of master-driver, which goes
through all the display entities, I think the above is still valid. When
the master-driver follows the remote-link, it still needs to first get
the main device node, as the ports and endpoints make no sense without
the context of the main device node.
I'm not sure about this. I might just need the remote port node
associated with a remote drm_crtc or drm_encoder structure to find out
which local endpoint I should look at to retrieve configuration.
Ok. I guess if you have a fixed model for the video pipeline elements it
works out. For omapdss we don't know what's there on the remote side, as
we can have arbitrary amount of video pipeline elements.
Tomi
From: Philipp Zabel <p.zabel@pengutronix.de> Date: 2014-03-04 15:47:52
Am Dienstag, den 04.03.2014, 14:21 +0200 schrieb Tomi Valkeinen:
On 04/03/14 13:36, Philipp Zabel wrote:
[...]
quoted
quoted
Can port_node be NULL? Probably only if something is quite wrong, but
maybe it's safer to return error in that case.
both of_property_read_u32 and of_node_put can handle port_node == NULL.
I'll add a WARN_ONCE here as for of_graph_get_next_endpoint and continue
on.
Isn't it better to return an error?
I am not sure. We can still correctly parse the endpoint properties of a
parentless node. All current users ignore the return value anyway. So as
long as we still do the memset and and set local_node and id, returning
an error effectively won't change the current behaviour.
[...]
quoted
It depends a bit on whether you are actually iterating over individual
ports, or if you are just walking the whole endpoint graph to find
remote devices that have to be added to the component master's waiting
list, for example.
True, but the latter is easily implemented using the separate
port/endpoint iteration. So I see it as a more powerful API.
Indeed. I see no problem in adding an of_graph_get_next_port function.
But I'd like to keep the current of_graph_get_next_endpoint function
iterating over all endpoints of the device.
quoted
quoted
Both of those are possible with the API in the series, but not very cleanly.
Also, if you just want to iterate the endpoints, it's easy to implement
a helper using the separate port and endpoint iterations.
I started out to move an existing (albeit lightly used) API to a common
place so others can use it and improve upon it, too. I'm happy to pile
on fixes directly in this series, but could we separate the improvement
step from the move, for the bigger modifications?
Yes, I understand that. What I wonder is that which is easier: make it a
public API now, more or less as it was in v4l2, or make it a public API
only when all the improvements we can think of have been made.
So my fear is that the API is now made public, and you and others start
to use it.
And I fear that this series might outgrow maintainer attention spans if
I keep adding to it.
But I can't use it, as I need things like separate
port/endpoint iteration. I need to add those, which also means that I
need to change all the users of the API, making the task more difficult
than I'd like.
However, this is more of "thinking out loud" than "I don't like the
series". It's a good series =).
Thanks. How about I follow this up with a split port/endpoint parsing
helpers after I get some acks?
quoted
I had no immediate use for the port iteration, so I have taken no steps
to add a function for this. I see no problem to add this later when
somebody needs it, or even rewrite of_graph_get_next_endpoint to use it
if it is feasible. Iterating over endpoints on a given port needs no
helper, as you can just use for_each_child_of_node.
I would have a helper, which should do some sanity checks, like that the
node names are "endpoint".
I'd prefer this to be a generic function of_get_next_child_by_name and
possibly a macro for_each_named_child_of_node wrapping that.
[...]
In omapdss each driver handles only the ports and endpoints defined for
its device, and they can be considered private to that device. The only
reason to look for the remote endpoint is to find the remote device. To
me the omapdss model makes sense, and feels logical and sane =). So I
have to say I'm not really familiar with the model you're using.
The main difference I see is that a single IPU device will have two port
nodes handled by the DRM driver and two port nodes handled by the V4L2
driver, so we can't go back up to the IPU device tree node and iterate
over all its ports in either the DRM or V4L2 drivers.
You could argue that all the device tree parsing should be done from the
IPU drivers, and the DRM and V4L2 drivers should use preparsed internal
graph structures. But then we are getting into using struct media_entity
in DRM drivers territory, and rather not go there right now.
regards
Philipp
From: Tomi Valkeinen <hidden> Date: 2014-03-05 10:06:05
On 04/03/14 17:47, Philipp Zabel wrote:
Am Dienstag, den 04.03.2014, 14:21 +0200 schrieb Tomi Valkeinen:
quoted
On 04/03/14 13:36, Philipp Zabel wrote:
[...]
quoted
quoted
quoted
Can port_node be NULL? Probably only if something is quite wrong, but
maybe it's safer to return error in that case.
both of_property_read_u32 and of_node_put can handle port_node == NULL.
I'll add a WARN_ONCE here as for of_graph_get_next_endpoint and continue
on.
Isn't it better to return an error?
I am not sure. We can still correctly parse the endpoint properties of a
parentless node. All current users ignore the return value anyway. So as
long as we still do the memset and and set local_node and id, returning
an error effectively won't change the current behaviour.
Is the parentless node case an error or not? If it's not, we can handle
it silently and return 0, no WARN needed. If it is an error, we should
return an error, even if nobody is currently handling that (which is a
bug in itself, as the function does return an error value, and callers
should handle it).
Tomi
Hi Tomi,
On Tuesday 04 March 2014 14:21:09 Tomi Valkeinen wrote:
On 04/03/14 13:36, Philipp Zabel wrote:
quoted
Am Dienstag, den 04.03.2014, 10:58 +0200 schrieb Tomi Valkeinen:
[...]
[snip]
quoted
quoted
Then, about the get_remote functions: I think there should be only one
function for that purpose, one that returns the device node that
contains the remote endpoint.
My reasoning is that the ports and endpoints, and their contents, should
be private to the device. So the only use to get the remote is to get
the actual device, to see if it's been probed, or maybe get some video
API for that device.
of_graph_get_remote_port currently is used in the exynos4-is/fimc-is.c
v4l2 driver to get the mipi-csi channel id from the remote port, and
I've started using it in imx-drm-core.c for two cases:
- given an endpoint on the encoder, find the remote port connected to
it, get the associated drm_crtc, to obtain its the drm_crtc_mask
for encoder->possible_crtcs.
- given an encoder and a connected drm_crtc, walk all endpoints to find
the remote port associated with the drm_crtc, and then use the local
endpoint parent port to determine multiplexer settings.
Ok.
In omapdss each driver handles only the ports and endpoints defined for
its device, and they can be considered private to that device. The only
reason to look for the remote endpoint is to find the remote device. To
me the omapdss model makes sense, and feels logical and sane =). So I
have to say I'm not really familiar with the model you're using.
I agree with you that most of the content of the remote endpoint should be
considered private to the remote device and not accessed by the local device
driver. There is, however, one piece of information from the remote endpoint
useful to the local device driver, it's the remote port identifier. This can
be expressed by a phandle, a remote port number, a media entity pad pointer,
or any other mean, but the information is useful for the local device driver
to communicate with the remote device driver. For instance a driver could use
it to ask its video stream source to start the video stream on a given port.
Of course, the different get_remove_* funcs do no harm, even if we have
a bunch of them. My point was only about enforcing the correct use of
the model, where "correct" is of course subjective =).
From: Eric Boxer <hidden> Date: 2014-03-07 00:06:29
Eric Boxer liked your message with Boxer. On March 6, 2014 at 5:59:17 PM CST, Laurent Pinchart wrote:Hi Tomi,On Tuesday 04 March 2014 14:21:09 Tomi Valkeinen wrote:> On 04/03/14 13:36, Philipp Zabel wrote:> > Am Dienstag, den 04.03.2014, 10:58 +0200 schrieb Tomi Valkeinen:> > [...][snip]> >> Then, about the get_remote functions: I think there should be only one> >> function for that purpose, one that returns the device node that> >> contains the remote endpoint.> >> > >> My reasoning is that the ports and endpoints, and their contents, should> >> be private to the device. So the only use to get the remote is to get> >> the actual device, to see if it's been probed, or maybe get some video> >> API for that device.> > > > of_graph_get_remote_port currently is used in the exynos4-is/fimc-is.c> > v4l2 driver to get the mipi-csi channel id from the remote port, and> > I've started using it in imx-drm-core.c for two cases:> > - given an endpoint on the encoder, find the remote port connected to> > it, get the associated drm_crtc, to obtain its the drm_crtc_mask> > for encoder->possible_crtcs.> > > > - given an encoder and a connected drm_crtc, walk all endpoints to find> > the remote port associated with the drm_crtc, and then use the local> > endpoint parent port to determine multiplexer settings.> > Ok.> > In omapdss each driver handles only the ports and endpoints defined for> its device, and they can be considered private to that device. The only> reason to look for the remote endpoint is to find the remote device. To> me the omapdss model makes sense, and feels logical and sane =). So I> have to say I'm not really familiar with the model you're using.I agree with you that most of the content of the remote endpoint should be considered private to the remote device and not accessed by the local device driver. There is, however, one piece of information from the remote endpoint useful to the local device driver, it's the remote port identifier. This can be expressed by a phandle, a remote port number, a media entity pad pointer, or any other mean, but the information is useful for the local device driver to communicate with the remote device driver. For instance a driver could use it to ask its video stream source to start the video stream on a given port.> Of course, the different get_remove_* funcs do no harm, even if we have> a bunch of them. My point was only about enforcing the correct use of> the model, where "correct" is of course subjective =).-- Regards,Laurent Pinchart