mbox series

[v7,0/8] Introduce on-chip interconnect API

Message ID 20180731161340.13000-1-georgi.djakov@linaro.org
Headers show
Series Introduce on-chip interconnect API | expand

Message

Georgi Djakov July 31, 2018, 4:13 p.m. UTC
Modern SoCs have multiple processors and various dedicated cores (video, gpu,
graphics, modem). These cores are talking to each other and can generate a
lot of data flowing through the on-chip interconnects. These interconnect
buses could form different topologies such as crossbar, point to point buses,
hierarchical buses or use the network-on-chip concept.

These buses have been sized usually to handle use cases with high data
throughput but it is not necessary all the time and consume a lot of power.
Furthermore, the priority between masters can vary depending on the running
use case like video playback or CPU intensive tasks.

Having an API to control the requirement of the system in terms of bandwidth
and QoS, so we can adapt the interconnect configuration to match those by
scaling the frequencies, setting link priority and tuning QoS parameters.
This configuration can be a static, one-time operation done at boot for some
platforms or a dynamic set of operations that happen at run-time.

This patchset introduce a new API to get the requirement and configure the
interconnect buses across the entire chipset to fit with the current demand.
The API is NOT for changing the performance of the endpoint devices, but only
the interconnect path in between them.

The API is using a consumer/provider-based model, where the providers are
the interconnect buses and the consumers could be various drivers.
The consumers request interconnect resources (path) to an endpoint and set
the desired constraints on this data flow path. The provider(s) receive
requests from consumers and aggregate these requests for all master-slave
pairs on that path. Then the providers configure each participating in the
topology node according to the requested data flow path, physical links and
constraints. The topology could be complicated and multi-tiered and is SoC
specific.

Below is a simplified diagram of a real-world SoC topology. The interconnect
providers are the NoCs.

+----------------+    +----------------+
| HW Accelerator |--->|      M NoC     |<---------------+
+----------------+    +----------------+                |
                        |      |                    +------------+
 +-----+  +-------------+      V       +------+     |            |
 | DDR |  |                +--------+  | PCIe |     |            |
 +-----+  |                | Slaves |  +------+     |            |
   ^ ^    |                +--------+     |         |   C NoC    |
   | |    V                               V         |            |
+------------------+   +------------------------+   |            |   +-----+
|                  |-->|                        |-->|            |-->| CPU |
|                  |-->|                        |<--|            |   +-----+
|     Mem NoC      |   |         S NoC          |   +------------+
|                  |<--|                        |---------+    |
|                  |<--|                        |<------+ |    |   +--------+
+------------------+   +------------------------+       | |    +-->| Slaves |
  ^  ^    ^    ^          ^                             | |        +--------+
  |  |    |    |          |                             | V
+------+  |  +-----+   +-----+  +---------+   +----------------+   +--------+
| CPUs |  |  | GPU |   | DSP |  | Masters |-->|       P NoC    |-->| Slaves |
+------+  |  +-----+   +-----+  +---------+   +----------------+   +--------+
          |
      +-------+
      | Modem |
      +-------+

TODO:
* Create icc_set_extended() to handle parameters such as latency and other
  QoS values.
* Convert from using global node identifiers to local per provider ids.
* Cache the path between the nodes instead of walking the graph on each get().
* Sync interconnect requests with the idle state of the device.

Changes since patchset v6 (https://lkml.org/lkml/2018/7/9/698)
* [patches 1,6]: Move the aggregation within the provider from the framework to
  the platform driver's set() callback, as the aggregation point could be SoC
  specific.
* [patch 1]: Include missing header, reset state only of the traversed nodes,
  move more code into path_init(), add more asserts, move misplaced mutex,
  simplify icc_link_destroy() (Evan)
* [patch 1]: Fix the order of requests to go from source to destination. (Alex)
* [patch 7]: Use better wording in the documentation. (Evan)
* [patch 6]: Reorder struct members, sort nodes alphabetically, improve naming
  of variables , add missing clk_disable_unprepare() in error paths. (Matthias)
* [patch 6]: Remove redundant NULL pointer check in msm8916 driver. (Alex)
* [patch 6]: Add missing depend on QCOM_SMD_RPM in Kconfig. (Evan)
* [patch 3]: Don't check for errors on debugfs calls, remove debugfs directory
  when module is unloaded (Greg)

Changes since patchset v5 (https://lkml.org/lkml/2018/6/20/453)
* Fix the modular build, make rpm-smd driver a module.
* Optimize locking and move to higher level. (Evan)
* Code cleanups. Fix typos. (Evan, Matthias)
* Add the source node to the path. (Evan)
* Rename path_allocate() to path_init() with minor refactoring. (Evan)
* Rename *_remove() functions to *_destroy().
* Return fixed errors in icc_link_destroy(). (Evan)
* Fix krealloc() usage in icc_link_destroy(). (Evan)
* Add missing kfree() in icc_node_create(). (Matthias)
* Make icc_node_add() return void. (Matthias)
* Change mutex_init to mutex_lock in icc_provider_add(). (Matthias)
* Add new icc_node_del() function to delete nodes from provider.
* Fix the header guard to reflect the path in smd-rpm.h. (Evan)
* Check for errors returned by qcom_icc_rpm_smd_send(). (Evan)
* Propagate the error of icc_provider_del(). (Evan)

Changes since patchset v4 (https://lkml.org/lkml/2018/3/9/856)
* Simplified locking by using a single global mutex. (Evan)
* Changed the aggregation function interface.
* Implemented functions for node, link, provider removal. (Evan)
* Naming changes on variables and functions, removed redundant code. (Evan)
* Fixes and clarifications in the docs. (Matthias, Evan, Amit, Alexandre)
* Removed mandatory reg DT property, made interconnect-names optional. (Bjorn)
* Made interconnect-cells property required to align with other bindings. (Neil)
* Moved msm8916 specific bindings into a separate file and patch. (Bjorn)
* Use the names, instead of the hardcoded ids for topology. (Matthias)
* Init the node before creating the links. (Evan)
* Added icc_units_to_bps macro. (Amit)

Changes since patchset v3 (https://lkml.org/lkml/2017/9/8/544)
* Refactored the constraints aggregation.
* Use the IDR API.
* Split the provider and consumer bindings into separate patches and propose
  new bindings for consumers, which allows to specify the local source port.
* Adopted the icc_ prefix for API functions.
* Introduced separate API functions for creating interconnect nodes and links.
* Added DT lookup support in addition to platform data.
* Dropped the event tracing patch for now.
* Added a patch to provide summary via debugfs.
* Use macro for the list of topology definitions in the platform driver.
* Various minor changes.

Changes since patchset v2 (https://lkml.org/lkml/2017/7/20/825)
* Split the aggregation into per node and per provider. Cache the
  aggregated values.
* Various small refactorings and cleanups in the framework.
* Added a patch introducing basic tracepoint support for monitoring
  the time required to update the interconnect nodes.

Changes since patchset v1 (https://lkml.org/lkml/2017/6/27/890)
* Updates in the documentation.
* Changes in request aggregation, locking.
* Dropped the aggregate() callback and use the default as it currently
  sufficient for the single vendor driver. Will add it later when needed.
* Dropped the dt-bindings draft patch for now.

Changes since RFC v2 (https://lkml.org/lkml/2017/6/12/316)
* Converted documentation to rst format.
* Fixed an incorrect call to mutex_lock. Renamed max_bw to peak_bw.

Changes since RFC v1 (https://lkml.org/lkml/2017/5/15/605)
* Refactored code into shorter functions.
* Added a new aggregate() API function.
* Rearranged some structs to reduce padding bytes.

Changes since RFC v0 (https://lkml.org/lkml/2017/3/1/599)
* Removed DT support and added optional Patch 3 with new bindings proposal.
* Converted the topology into internal driver data.
* Made the framework modular.
* interconnect_get() now takes (src and dst ports as arguments).
* Removed public declarations of some structs.
* Now passing prev/next nodes to the vendor driver.
* Properly remove requests on _put().
* Added refcounting.
* Updated documentation.
* Changed struct interconnect_path to use array instead of linked list.

Georgi Djakov (8):
  interconnect: Add generic on-chip interconnect API
  dt-bindings: Introduce interconnect provider bindings
  interconnect: Add debugfs support
  interconnect: qcom: Add RPM communication
  dt-bindings: interconnect: Document qcom,msm8916 NoC bindings
  interconnect: qcom: Add msm8916 interconnect provider driver
  dt-bindings: Introduce interconnect consumers bindings
  interconnect: Allow endpoints translation via DT

 .../bindings/interconnect/interconnect.txt    |  60 ++
 .../bindings/interconnect/qcom-msm8916.txt    |  39 +
 .../bindings/interconnect/qcom-smd.txt        |  32 +
 Documentation/interconnect/interconnect.rst   |  96 +++
 drivers/Kconfig                               |   2 +
 drivers/Makefile                              |   1 +
 drivers/interconnect/Kconfig                  |  15 +
 drivers/interconnect/Makefile                 |   3 +
 drivers/interconnect/core.c                   | 709 ++++++++++++++++++
 drivers/interconnect/qcom/Kconfig             |  23 +
 drivers/interconnect/qcom/Makefile            |   4 +
 drivers/interconnect/qcom/msm8916.c           | 510 +++++++++++++
 drivers/interconnect/qcom/smd-rpm.c           |  91 +++
 drivers/interconnect/qcom/smd-rpm.h           |  16 +
 include/dt-bindings/interconnect/qcom.h       | 187 +++++
 include/linux/interconnect-provider.h         | 125 +++
 include/linux/interconnect.h                  |  49 ++
 17 files changed, 1962 insertions(+)
 create mode 100644 Documentation/devicetree/bindings/interconnect/interconnect.txt
 create mode 100644 Documentation/devicetree/bindings/interconnect/qcom-msm8916.txt
 create mode 100644 Documentation/devicetree/bindings/interconnect/qcom-smd.txt
 create mode 100644 Documentation/interconnect/interconnect.rst
 create mode 100644 drivers/interconnect/Kconfig
 create mode 100644 drivers/interconnect/Makefile
 create mode 100644 drivers/interconnect/core.c
 create mode 100644 drivers/interconnect/qcom/Kconfig
 create mode 100644 drivers/interconnect/qcom/Makefile
 create mode 100644 drivers/interconnect/qcom/msm8916.c
 create mode 100644 drivers/interconnect/qcom/smd-rpm.c
 create mode 100644 drivers/interconnect/qcom/smd-rpm.h
 create mode 100644 include/dt-bindings/interconnect/qcom.h
 create mode 100644 include/linux/interconnect-provider.h
 create mode 100644 include/linux/interconnect.h

--
To unsubscribe from this list: send the line "unsubscribe linux-arm-msm" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html

Comments

Saravana Kannan Aug. 1, 2018, 10:57 p.m. UTC | #1
On 2018-07-31 09:13, Georgi Djakov wrote:
> Currently we support only platform data for specifying the interconnect

> endpoints. As now the endpoints are hard-coded into the consumer driver

> this may lead to complications when a single driver is used by multiple

> SoCs, which may have different interconnect topology.

> To avoid cluttering the consumer drivers, introduce a translation 

> function

> to help us get the board specific interconnect data from device-tree.

> 

> Signed-off-by: Georgi Djakov <georgi.djakov@linaro.org>

> ---

>  drivers/interconnect/core.c  | 62 ++++++++++++++++++++++++++++++++++++

>  include/linux/interconnect.h |  7 ++++

>  2 files changed, 69 insertions(+)

> 

> diff --git a/drivers/interconnect/core.c b/drivers/interconnect/core.c

> index 9fef180cf77e..d1b6adff0a3d 100644

> --- a/drivers/interconnect/core.c

> +++ b/drivers/interconnect/core.c

> @@ -16,6 +16,7 @@

>  #include <linux/module.h>

>  #include <linux/mutex.h>

>  #include <linux/slab.h>

> +#include <linux/of.h>

>  #include <linux/overflow.h>

>  #include <linux/uaccess.h>

> 

> @@ -251,6 +252,67 @@ static int apply_constraints(struct icc_path 

> *path)

>  	return ret;

>  }

> 

> +struct icc_path *of_icc_get(struct device *dev, const char *name)

> +{

> +	struct device_node *np = NULL;

> +	struct of_phandle_args src_args, dst_args;

> +	u32 src_id, dst_id;

> +	int idx = 0;

> +	int ret;

> +

> +	if (!dev || !dev->of_node)

> +		return ERR_PTR(-ENODEV);

> +

> +	np = dev->of_node;

> +

> +	/*

> +	 * When the consumer DT node do not have "interconnects" property

> +	 * return a NULL path to skip setting constraints.

> +	 */

> +	if (!of_find_property(np, "interconnects", NULL))

> +		return NULL;

> +

> +	/*

> +	 * We use a combination of phandle and specifier for endpoint. For 

> now

> +	 * lets support only global ids and extend this is the future if 

> needed

> +	 * without breaking DT compatibility.

> +	 */

> +	if (name) {

> +		idx = of_property_match_string(np, "interconnect-names", name);

> +		if (idx < 0)

> +			return ERR_PTR(idx);

> +	}

> +

> +	ret = of_parse_phandle_with_args(np, "interconnects",

> +					 "#interconnect-cells", idx * 2,

> +					 &src_args);

> +	if (ret)

> +		return ERR_PTR(ret);

> +

> +	of_node_put(src_args.np);

> +

> +	if (!src_args.args_count || src_args.args_count > 1)

> +		return ERR_PTR(-EINVAL);

> +

> +	src_id = src_args.args[0];

> +

> +	ret = of_parse_phandle_with_args(np, "interconnects",

> +					 "#interconnect-cells", idx * 2 + 1,

> +					 &dst_args);

> +	if (ret)

> +		return ERR_PTR(ret);

> +

> +	of_node_put(dst_args.np);

> +

> +	if (!dst_args.args_count || dst_args.args_count > 1)

> +		return ERR_PTR(-EINVAL);

> +

> +	dst_id = dst_args.args[0];

> +

> +	return icc_get(dev, src_id, dst_id);

> +}

> +EXPORT_SYMBOL_GPL(of_icc_get);

> +

>  /**

>   * icc_set() - set constraints on an interconnect path between two 

> endpoints

>   * @path: reference to the path returned by icc_get()

> diff --git a/include/linux/interconnect.h 

> b/include/linux/interconnect.h

> index 593215371fd6..ae6744da9bc2 100644

> --- a/include/linux/interconnect.h

> +++ b/include/linux/interconnect.h

> @@ -17,6 +17,7 @@ struct device;

> 

>  struct icc_path *icc_get(struct device *dev, const int src_id,

>  			 const int dst_id);

> +struct icc_path *of_icc_get(struct device *dev, const char *name);

>  void icc_put(struct icc_path *path);

>  int icc_set(struct icc_path *path, u32 avg_bw, u32 peak_bw);

> 

> @@ -28,6 +29,12 @@ static inline struct icc_path *icc_get(struct

> device *dev, const int src_id,

>  	return NULL;

>  }

> 

> +static inline struct icc_path *of_icc_get(struct device *dev,

> +					  const char *name)

> +{

> +	return NULL;

> +}

> +


Might want to return PTR(-ENODEV) or some error code so that client 
doesn't have to do NULL check AND an error check?

-Saravana
--
To unsubscribe from this list: send the line "unsubscribe linux-arm-msm" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Randy Dunlap Aug. 2, 2018, 12:05 a.m. UTC | #2
On 07/31/2018 09:13 AM, Georgi Djakov wrote:
> This patch introduces a new API to get requirements and configure the

> interconnect buses across the entire chipset to fit with the current

> demand.

> 

> The API is using a consumer/provider-based model, where the providers are

> the interconnect buses and the consumers could be various drivers.

> The consumers request interconnect resources (path) between endpoints and

> set the desired constraints on this data flow path. The providers receive

> requests from consumers and aggregate these requests for all master-slave

> pairs on that path. Then the providers configure each participating in the

> topology node according to the requested data flow path, physical links and

> constraints. The topology could be complicated and multi-tiered and is SoC

> specific.

> 

> Signed-off-by: Georgi Djakov <georgi.djakov@linaro.org>

> ---

>  Documentation/interconnect/interconnect.rst |  96 ++++

>  drivers/Kconfig                             |   2 +

>  drivers/Makefile                            |   1 +

>  drivers/interconnect/Kconfig                |  10 +

>  drivers/interconnect/Makefile               |   2 +

>  drivers/interconnect/core.c                 | 569 ++++++++++++++++++++

>  include/linux/interconnect-provider.h       | 125 +++++

>  include/linux/interconnect.h                |  42 ++

>  8 files changed, 847 insertions(+)

>  create mode 100644 Documentation/interconnect/interconnect.rst

>  create mode 100644 drivers/interconnect/Kconfig

>  create mode 100644 drivers/interconnect/Makefile

>  create mode 100644 drivers/interconnect/core.c

>  create mode 100644 include/linux/interconnect-provider.h

>  create mode 100644 include/linux/interconnect.h

> 

> diff --git a/Documentation/interconnect/interconnect.rst b/Documentation/interconnect/interconnect.rst

> new file mode 100644

> index 000000000000..e628881ee218

> --- /dev/null

> +++ b/Documentation/interconnect/interconnect.rst

> @@ -0,0 +1,96 @@

> +.. SPDX-License-Identifier: GPL-2.0

> +

> +=====================================

> +GENERIC SYSTEM INTERCONNECT SUBSYSTEM

> +=====================================

> +

> +Introduction

> +------------

> +

> +This framework is designed to provide a standard kernel interface to control

> +the settings of the interconnects on a SoC. These settings can be throughput,


I would say:                         on an SoC.

Do you pronounce that as "sock" or the letters S.O.C.?


> +latency and priority between multiple interconnected devices or functional

> +blocks. This can be controlled dynamically in order to save power or provide

> +maximum performance.

> +

> +The interconnect bus is a hardware with configurable parameters, which can be


                    bus is hardware

> +set on a data path according to the requests received from various drivers.

> +An example of interconnect buses are the interconnects between various

> +components or functional blocks in chipsets. There can be multiple interconnects

> +on a SoC that can be multi-tiered.


      an SoC

> +

> +Below is a simplified diagram of a real-world SoC interconnect bus topology.

> +

> +::

> +

> + +----------------+    +----------------+

> + | HW Accelerator |--->|      M NoC     |<---------------+

> + +----------------+    +----------------+                |

> +                         |      |                    +------------+

> +  +-----+  +-------------+      V       +------+     |            |

> +  | DDR |  |                +--------+  | PCIe |     |            |

> +  +-----+  |                | Slaves |  +------+     |            |

> +    ^ ^    |                +--------+     |         |   C NoC    |

> +    | |    V                               V         |            |

> + +------------------+   +------------------------+   |            |   +-----+

> + |                  |-->|                        |-->|            |-->| CPU |

> + |                  |-->|                        |<--|            |   +-----+

> + |     Mem NoC      |   |         S NoC          |   +------------+

> + |                  |<--|                        |---------+    |

> + |                  |<--|                        |<------+ |    |   +--------+

> + +------------------+   +------------------------+       | |    +-->| Slaves |

> +   ^  ^    ^    ^          ^                             | |        +--------+

> +   |  |    |    |          |                             | V

> + +------+  |  +-----+   +-----+  +---------+   +----------------+   +--------+

> + | CPUs |  |  | GPU |   | DSP |  | Masters |-->|       P NoC    |-->| Slaves |

> + +------+  |  +-----+   +-----+  +---------+   +----------------+   +--------+

> +           |

> +       +-------+

> +       | Modem |

> +       +-------+

> +

> +Terminology

> +-----------

> +

> +Interconnect provider is the software definition of the interconnect hardware.

> +The interconnect providers on the above diagram are M NoC, S NoC, C NoC, P NoC

> +and Mem NoC.

> +

> +Interconnect node is the software definition of the interconnect hardware

> +port. Each interconnect provider consists of multiple interconnect nodes,

> +which are connected to other SoC components including other interconnect

> +providers. The point on the diagram where the CPUs connect to the memory is

> +called an interconnect node, which belongs to the Mem NoC interconnect provider.

> +

> +Interconnect endpoints are the first or the last element of the path. Every

> +endpoint is a node, but not every node is an endpoint.

> +

> +Interconnect path is everything between two endpoints including all the nodes

> +that have to be traversed to reach from a source to destination node. It may

> +include multiple master-slave pairs across several interconnect providers.

> +

> +Interconnect consumers are the entities which make use of the data paths exposed

> +by the providers. The consumers send requests to providers requesting various

> +throughput, latency and priority. Usually the consumers are device drivers, that

> +send request based on their needs. An example for a consumer is a video decoder

> +that supports various formats and image sizes.

> +

> +Interconnect providers

> +----------------------

> +

> +Interconnect provider is an entity that implements methods to initialize and

> +configure a interconnect bus hardware. The interconnect provider drivers should


   configure interconnect bus hardware.
(i.e., drop the "a")

> +be registered with the interconnect provider core.

> +

> +The interconnect framework provider API functions are documented in

> +.. kernel-doc:: include/linux/interconnect-provider.h


What do you want that to do?  and does that happen?

The .. kernel-doc:: line won't be printed.  It will just be expanded to the
contents of that header file, so the preceding sentence fragment will look/sound
odd.

> +

> +Interconnect consumers

> +----------------------

> +

> +Interconnect consumers are the clients which use the interconnect APIs to

> +get paths between endpoints and set their bandwidth/latency/QoS requirements

> +for these interconnect paths.

> +

> +The interconnect framework consumer API functions are documented in

> +.. kernel-doc:: include/linux/interconnect.h


same as above.


-- 
~Randy
--
To unsubscribe from this list: send the line "unsubscribe linux-arm-msm" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Saravana Kannan Aug. 2, 2018, 7:12 p.m. UTC | #3
On 2018-08-02 05:07, Georgi Djakov wrote:
> Hi Saravana,

> 

> On 08/02/2018 01:57 AM, skannan@codeaurora.org wrote:

>> On 2018-07-31 09:13, Georgi Djakov wrote:

>>> Currently we support only platform data for specifying the 

>>> interconnect

>>> endpoints. As now the endpoints are hard-coded into the consumer 

>>> driver

>>> this may lead to complications when a single driver is used by 

>>> multiple

>>> SoCs, which may have different interconnect topology.

>>> To avoid cluttering the consumer drivers, introduce a translation

>>> function

>>> to help us get the board specific interconnect data from device-tree.

>>> 

>>> Signed-off-by: Georgi Djakov <georgi.djakov@linaro.org>

>>> ---

>>>  drivers/interconnect/core.c  | 62 

>>> ++++++++++++++++++++++++++++++++++++

>>>  include/linux/interconnect.h |  7 ++++

>>>  2 files changed, 69 insertions(+)

>>> 

>>> diff --git a/drivers/interconnect/core.c 

>>> b/drivers/interconnect/core.c

>>> index 9fef180cf77e..d1b6adff0a3d 100644

>>> --- a/drivers/interconnect/core.c

>>> +++ b/drivers/interconnect/core.c

> [..]

>>> --- a/include/linux/interconnect.h

>>> +++ b/include/linux/interconnect.h

>>> @@ -17,6 +17,7 @@ struct device;

>>> 

>>>  struct icc_path *icc_get(struct device *dev, const int src_id,

>>>               const int dst_id);

>>> +struct icc_path *of_icc_get(struct device *dev, const char *name);

>>>  void icc_put(struct icc_path *path);

>>>  int icc_set(struct icc_path *path, u32 avg_bw, u32 peak_bw);

>>> 

>>> @@ -28,6 +29,12 @@ static inline struct icc_path *icc_get(struct

>>> device *dev, const int src_id,

>>>      return NULL;

>>>  }

>>> 

>>> +static inline struct icc_path *of_icc_get(struct device *dev,

>>> +                      const char *name)

>>> +{

>>> +    return NULL;

>>> +}

>>> +

>> 

>> Might want to return PTR(-ENODEV) or some error code so that client

>> doesn't have to do NULL check AND an error check?

>> 

>> -Saravana

> 

> NULL is returned when CONFIG_INTERCONNECT=n. Configuration of

> interconnects by consumer drivers could be optional and that's why null

> is returned instead of an error. The consumer drivers decide how to

> proceed in this case and if there is a hard requirement for 

> interconnect

> support, then i would suggest to express it as a dependency in Kconfig.


Ehh... you could make the same argument with an error. If it's not 
mandatory for functioning, they can ignore a specific error and 
continue? At a minimum, these stub functions returning NULL doesn't 
match with the documentation that says these APIs will only ever return 
ERR_PTR().

-Saravana
--
To unsubscribe from this list: send the line "unsubscribe linux-arm-msm" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Georgi Djakov Aug. 24, 2018, 2:51 p.m. UTC | #4
Hi Maxime,

On 08/20/2018 06:32 PM, Maxime Ripard wrote:
> Hi Georgi,

> 

> On Tue, Aug 07, 2018 at 05:54:38PM +0300, Georgi Djakov wrote:

>>> There is also a patch series from Maxime Ripard that's addressing the

>>> same general area. See "dt-bindings: Add a dma-parent property". We

>>> don't need multiple ways to address describing the device to memory

>>> paths, so you all had better work out a common solution.

>>

>> Looks like this fits exactly into the interconnect API concept. I see

>> MBUS as interconnect provider and display/camera as consumers, that

>> report their bandwidth needs. I am also planning to add support for

>> priority.

> 

> Thanks for working on this. After looking at your serie, the one thing

> I'm a bit uncertain about (and the most important one to us) is how we

> would be able to tell through which interconnect the DMA are done.

> 

> This is important to us since our topology is actually quite simple as

> you've seen, but the RAM is not mapped on that bus and on the CPU's,

> so we need to apply an offset to each buffer being DMA'd.


Ok, i see - your problem is not about bandwidth scaling but about using
different memory ranges by the driver to access the same location. So
this is not really the same and your problem is different. Also the
interconnect bindings are describing a path and endpoints. However i am
open to any ideas.

Thanks,
Georgi
Rob Herring (Arm) Aug. 24, 2018, 3:35 p.m. UTC | #5
On Fri, Aug 24, 2018 at 9:51 AM Georgi Djakov <georgi.djakov@linaro.org> wrote:
>

> Hi Maxime,

>

> On 08/20/2018 06:32 PM, Maxime Ripard wrote:

> > Hi Georgi,

> >

> > On Tue, Aug 07, 2018 at 05:54:38PM +0300, Georgi Djakov wrote:

> >>> There is also a patch series from Maxime Ripard that's addressing the

> >>> same general area. See "dt-bindings: Add a dma-parent property". We

> >>> don't need multiple ways to address describing the device to memory

> >>> paths, so you all had better work out a common solution.

> >>

> >> Looks like this fits exactly into the interconnect API concept. I see

> >> MBUS as interconnect provider and display/camera as consumers, that

> >> report their bandwidth needs. I am also planning to add support for

> >> priority.

> >

> > Thanks for working on this. After looking at your serie, the one thing

> > I'm a bit uncertain about (and the most important one to us) is how we

> > would be able to tell through which interconnect the DMA are done.

> >

> > This is important to us since our topology is actually quite simple as

> > you've seen, but the RAM is not mapped on that bus and on the CPU's,

> > so we need to apply an offset to each buffer being DMA'd.

>

> Ok, i see - your problem is not about bandwidth scaling but about using

> different memory ranges by the driver to access the same location. So

> this is not really the same and your problem is different. Also the

> interconnect bindings are describing a path and endpoints. However i am

> open to any ideas.


It may be different things you need, but both are related to the path
between a bus master and memory. We can't have each 'problem'
described in a different way. Well, we could as long as each platform
has different problems, but that's unlikely.

It could turn out that the only commonality is property naming
convention, but that's still better than 2 independent solutions.

I know you each want to just fix your issues, but the fact that DT
doesn't model the DMA side of the bus structure has been an issue at
least since the start of DT on ARM. Either we should address this in a
flexible way or we can just continue to manage without. So I'm not
inclined to take something that only addresses one SoC family.

Rob
Georgi Djakov Aug. 29, 2018, 12:33 p.m. UTC | #6
Hi Rob and Maxime,

On 08/27/2018 06:11 PM, Maxime Ripard wrote:
> On Fri, Aug 24, 2018 at 10:35:23AM -0500, Rob Herring wrote:

>> On Fri, Aug 24, 2018 at 9:51 AM Georgi Djakov <georgi.djakov@linaro.org> wrote:

>>>

>>> Hi Maxime,

>>>

>>> On 08/20/2018 06:32 PM, Maxime Ripard wrote:

>>>> Hi Georgi,

>>>>

>>>> On Tue, Aug 07, 2018 at 05:54:38PM +0300, Georgi Djakov wrote:

>>>>>> There is also a patch series from Maxime Ripard that's addressing the

>>>>>> same general area. See "dt-bindings: Add a dma-parent property". We

>>>>>> don't need multiple ways to address describing the device to memory

>>>>>> paths, so you all had better work out a common solution.

>>>>>

>>>>> Looks like this fits exactly into the interconnect API concept. I see

>>>>> MBUS as interconnect provider and display/camera as consumers, that

>>>>> report their bandwidth needs. I am also planning to add support for

>>>>> priority.

>>>>

>>>> Thanks for working on this. After looking at your serie, the one thing

>>>> I'm a bit uncertain about (and the most important one to us) is how we

>>>> would be able to tell through which interconnect the DMA are done.

>>>>

>>>> This is important to us since our topology is actually quite simple as

>>>> you've seen, but the RAM is not mapped on that bus and on the CPU's,

>>>> so we need to apply an offset to each buffer being DMA'd.

>>>

>>> Ok, i see - your problem is not about bandwidth scaling but about using

>>> different memory ranges by the driver to access the same location. So

>>> this is not really the same and your problem is different. Also the

>>> interconnect bindings are describing a path and endpoints. However i am

>>> open to any ideas.

>>

>> It may be different things you need, but both are related to the path

>> between a bus master and memory. We can't have each 'problem'

>> described in a different way. Well, we could as long as each platform

>> has different problems, but that's unlikely.

>>

>> It could turn out that the only commonality is property naming

>> convention, but that's still better than 2 independent solutions.

> 

> Yeah, I really don't think the two issues are unrelated. Can we maybe

> have a particular interconnect-names value to mark the interconnect

> being used to perform DMA?


We can call one of the paths "dma" and use it to perform DMA for the
current device. I don't see a problem with this. The name of the path is
descriptive and makes sense. And by doing we avoid adding more DT
properties, which would be an other option.

This also makes me think that it might be a good idea to have a standard
name for the path to memory as i expect some people will call it "mem",
others "ddr" etc.

Thanks,
Georgi

>> I know you each want to just fix your issues, but the fact that DT

>> doesn't model the DMA side of the bus structure has been an issue at

>> least since the start of DT on ARM. Either we should address this in a

>> flexible way or we can just continue to manage without. So I'm not

>> inclined to take something that only addresses one SoC family.

> 

> I'd really like to have it addressed. We're getting bit by this, and

> the hacks we have don't work well anymore.
Maxime Ripard Aug. 30, 2018, 7:47 a.m. UTC | #7
Hi,

On Wed, Aug 29, 2018 at 03:33:29PM +0300, Georgi Djakov wrote:
> On 08/27/2018 06:11 PM, Maxime Ripard wrote:

> > On Fri, Aug 24, 2018 at 10:35:23AM -0500, Rob Herring wrote:

> >> On Fri, Aug 24, 2018 at 9:51 AM Georgi Djakov <georgi.djakov@linaro.org> wrote:

> >>>

> >>> Hi Maxime,

> >>>

> >>> On 08/20/2018 06:32 PM, Maxime Ripard wrote:

> >>>> Hi Georgi,

> >>>>

> >>>> On Tue, Aug 07, 2018 at 05:54:38PM +0300, Georgi Djakov wrote:

> >>>>>> There is also a patch series from Maxime Ripard that's addressing the

> >>>>>> same general area. See "dt-bindings: Add a dma-parent property". We

> >>>>>> don't need multiple ways to address describing the device to memory

> >>>>>> paths, so you all had better work out a common solution.

> >>>>>

> >>>>> Looks like this fits exactly into the interconnect API concept. I see

> >>>>> MBUS as interconnect provider and display/camera as consumers, that

> >>>>> report their bandwidth needs. I am also planning to add support for

> >>>>> priority.

> >>>>

> >>>> Thanks for working on this. After looking at your serie, the one thing

> >>>> I'm a bit uncertain about (and the most important one to us) is how we

> >>>> would be able to tell through which interconnect the DMA are done.

> >>>>

> >>>> This is important to us since our topology is actually quite simple as

> >>>> you've seen, but the RAM is not mapped on that bus and on the CPU's,

> >>>> so we need to apply an offset to each buffer being DMA'd.

> >>>

> >>> Ok, i see - your problem is not about bandwidth scaling but about using

> >>> different memory ranges by the driver to access the same location. So

> >>> this is not really the same and your problem is different. Also the

> >>> interconnect bindings are describing a path and endpoints. However i am

> >>> open to any ideas.

> >>

> >> It may be different things you need, but both are related to the path

> >> between a bus master and memory. We can't have each 'problem'

> >> described in a different way. Well, we could as long as each platform

> >> has different problems, but that's unlikely.

> >>

> >> It could turn out that the only commonality is property naming

> >> convention, but that's still better than 2 independent solutions.

> > 

> > Yeah, I really don't think the two issues are unrelated. Can we maybe

> > have a particular interconnect-names value to mark the interconnect

> > being used to perform DMA?

> 

> We can call one of the paths "dma" and use it to perform DMA for the

> current device. I don't see a problem with this. The name of the path is

> descriptive and makes sense. And by doing we avoid adding more DT

> properties, which would be an other option.


That works for me. If Rob is fine with it too, I'll send an updated
version of my serie based on yours.

Thanks!
Maxime



-- 
Maxime Ripard, Bootlin
Embedded Linux and Kernel engineering
https://bootlin.com