changed mcp2515 driver to use nonblocking spi #817

kikass13 · 2022-02-01T11:18:25Z

change mcp2515 to use proper modm::can driver api
change mcp2515 to use the nonblocking spi api
- mostly, initialization routines are still blocking but i guess tahts fine
add example
utilize attachConfigurationHandler for switching device spi types
test example on a nucleo ~~or discovery board~~
optional:
- utilize attachConfigurationHandler for switching device clock speeds

… standard modm::can api

…king, which should be fine because it is controlled via interrupt

rleh

Nice!
Thanks for taking care of this!

rleh · 2022-02-01T13:25:40Z

src/modm/driver/can/mcp2515_impl.hpp

 	return true;
 }

+template <typename SPI, typename CS, typename INT>
+modm::ResumableResult<bool>


Shouldn't update() return modm::ResumableResult<void>?
Because nobody (in the main loop) will handle the return value...

it's just the return value of sendMessage indicating if a message was sent this cycle ... not really useful but i thought why not :D ... iwill remove it

it is void now

it is void now

No it's not.

src/modm/driver/can/mcp2515_impl.hpp

src/modm/driver/can/mcp2515.hpp

examples/stm32f4_discovery/can_m2515/main.cpp

src/modm/driver/can/mcp2515_impl.hpp

… as well; fixed some things in example

src/modm/driver/can/mcp2515_impl.hpp

rleh · 2022-02-01T22:47:37Z

Another addition: This driver thould use the SpiDevice abstraction and should make use of void attachConfigurationHandler(Spi::ConfigurationHandler handler)!
@kikass13: We need this, because we have multiple MCP2515 and one ADIS16470 connected to the same SPI.

Off-topic discussion: In the case mentioned above the SPI clock has to be limited to 1MHz or 2MHz (ADIS16470 limitation), but for the MCP2515 a higher SPI clock would be desirable. This probably affects many (SPI) drivers, what would be the most elegant way to solve this in modm?

chris-durand · 2022-02-01T23:12:48Z

Off-topic discussion: In the case mentioned above the SPI clock has to be limited to 1MHz or 2MHz (ADIS16470 limitation), but for the MCP2515 a higher SPI clock would be desirable. This probably affects many (SPI) drivers, what would be the most elegant way to solve this in modm?

That requires to reconfigure SPI timings when the SpiDevice acquires the SpiMaster. A method similar to the configuration handler could be used but this time it is provided externally by the user and not the driver.
The handler function pointer could be put into the SpiDevice class. It would be called inside of the SpiMaster implementation. If not used the runtime overhead of the solution would be a not taken branch with a nullptr check inside of the SpiMaster::acquire() method.

chris-durand · 2022-02-01T23:50:13Z

@rleh I have quickly hacked together a sub-optimal implementation for switching the SPI baudrate with multiple devices. This is just a rough proof of concept for STM32 only and makes changes to the SpiMaster API.

https://github.com/chris-durand/modm/tree/wip_spi_user_conf_handler

kikass13 · 2022-02-01T23:51:21Z

Another addition: This driver thould use the SpiDevice abstraction and should make use of void attachConfigurationHandler(Spi::ConfigurationHandler handler)! @kikass13: We need this, because we have multiple MCP2515 and one ADIS16470 connected to the same SPI.

Off-topic discussion: In the case mentioned above the SPI clock has to be limited to 1MHz or 2MHz (ADIS16470 limitation), but for the MCP2515 a higher SPI clock would be desirable. This probably affects many (SPI) drivers, what would be the most elegant way to solve this in modm?

I don't know what that is bit I'll look into it tomorrow (after our work meeting stuff)

chris-durand · 2022-02-01T23:58:35Z

I don't know what that is bit I'll look into it tomorrow (after our work meeting stuff)

The configuration handler is actually quite simple. If you have multiple devices on one bus that require different SPI modes you'll need to reconfigure the mode whenever you run a transfer on the next device. You can define a callback that configures the required parameters and modm will call it automatically when needed. Here is an example.

EDIT: if you mean the SpiDevice with your comment, it's a class that manages shared access to one SpiMaster by many device drivers. You basically inherit your driver from SpiDevice<SpiMaster> and put RF_WAIT_UNTIL(this->acquireMaster()); before you do anything with the SPI and call releaseMaster() when you are done. That way it's made sure only one driver uses the bus at the same time. You can look at basically any other SPI based driver to see how it is used.

kikass13 · 2022-02-02T00:32:57Z

I don't know what that is bit I'll look into it tomorrow (after our work meeting stuff)

The configuration handler is actually quite simple. If you have multiple devices on one bus that require different SPI modes you'll need to reconfigure the mode whenever you run a transfer on the next device. You can define a callback that configures the required parameters and modm will call it automatically when needed. Here is an example.

EDIT: if you mean the SpiDevice with your comment, it's a class that manages shared access to one SpiMaster by many device drivers. You basically inherit your driver from SpiDevice<SpiMaster> and put RF_WAIT_UNTIL(this->acquireMaster()); before you do anything with the SPI and call releaseMaster() when you are done. That way it's made sure only one driver uses the bus at the same time. You can look at basically any other SPI based driver to see how it is used.

Yeah the second one .. you explained what it does quite well, thanks :)

…evice is now used internally for aquiring / releasing the bus whenever accessed

…he spi clock frequency when master is acquired

…ationHandler in example; changes canbdaudrate to 500kbps

kikass13 · 2022-02-04T10:24:57Z

@rleh I have quickly hacked together a sub-optimal implementation for switching the SPI baudrate with multiple devices. This is just a rough proof of concept for STM32 only and makes changes to the SpiMaster API.

https://github.com/chris-durand/modm/tree/wip_spi_user_conf_handler

@rleh @chris-durand
i tested this and seems to work (one would have to check with an oscilloscope if it actually happened, and I don't have one at home :P)

kikass13 · 2022-02-04T14:55:08Z

SpiDevice test seems to be broken? Or did I do something wrong?

https://github.com/modm-io/modm/runs/5065030473?check_suite_focus=true

examples/nucleo_f439zi/can_m2515/main.cpp

chris-durand · 2022-02-04T15:03:01Z

SpiDevice test seems to be broken? Or did I do something wrong?
* https://github.com/modm-io/modm/runs/5065030473?check_suite_focus=true

That is because of my experimental commit that was cherry-picked. It is only a proof of concept for STM32. Remove it and everything will be fine :)

rleh · 2022-02-08T15:41:09Z

this whole discussion is now geared towwards changing the whole device / driver api ... as I see it that is a necessary step because (as I said earlier) the api seems to be a little messy. But does that mean this has to be done in this specific pr?

This does not have to happen in this PR, in my view it is even nicer to do it in a separate PR.

chris-durand · 2022-02-08T15:45:22Z

I guess @kikass13 might be asking if we should change the API of the MCP2515 driver, not the SpiDevice. Using SpiDevice requires an instantiated object but the driver had only static methods. Now half of them are static. This is messy. I would be in favor of changing the MCP2515 API to make it more consistent.

kikass13 · 2022-02-08T15:55:05Z

@chris-durand currently i like the fact that this thing is static. I dislike the "mix" but i can live with it ... I have some software running this thing currently (successfully), and I really depend on the fact, that the can interfaces can be used statically.

i.e. just provide it as a template argument and call some functions on it (without pushing / holding) references to everything and so on

At least the current implementation works for me

after fixing some of the random ci errors for avr

rleh · 2022-02-08T16:05:56Z

I really depend on the fact, that the can interfaces can be used statically

Can't you in the class where you passed the static class as a template argument instantiate an object of the template argument? Then it works with both static and non-static classes.

kikass13 · 2022-02-08T16:10:33Z

I really depend on the fact, that the can interfaces can be used statically

Can't you in the class where you passed the static class as a template argument instantiate an object of the template argument? Then it works with both static and non-static classes.

sure, i was just trying to explain that the old "design" could still be valid. @rleh you blocked my initial changes because I changed the API, so changing it back again now seems kind of redundant :D

…vel intact

…dify()

…t (aka NOT attempting to send it) if the mcp is not readyToSend() instead of waiting for it to be ready, which defeats the point of buffering packets indefiniteley

examples/nucleo_f439zi/can_m2515/main.cpp

src/modm/driver/can/mcp2515.lb

rleh · 2022-02-24T00:48:04Z

src/modm/driver/can/mcp2515_impl.hpp

@@ -16,24 +16,13 @@
 #ifndef MODM_MCP2515_HPP
 	#error	"Don't include this file directly, use 'mcp2515.hpp' instead!"
 #endif
-


Unnecessary whitespace change.

Not resolved.

src/modm/driver/can/mcp2515_impl.hpp

rleh · 2022-02-24T00:55:14Z

src/modm/driver/can/mcp2515_impl.hpp

+			if(not modm_assert_continue_ignore(rxQueue.push(messageBuffer_), "mcp2515.can.tx",
+			"CAN transmit software buffer overflowed!", 1)){
+				/// ignore
+			}


Remove the if statement if the body is empty.

well it was there before ... it shows intent .. so i kept it for now

…code; minor style changes to mcp2515 driver impl

kikass13 · 2022-02-24T21:23:18Z

@rleh i have changed all the points you made (hopefully to your satisfaction) :)

…d unused project xml things; cleaned up unused headers; changed mcp2515 driver class attributes to not have _ underscore postfix

…rescaler

rleh

Inside initializeWithPrescaler() SPI access is performed without locking the bus (using SpiDevice::acquireMaster()/::releaseMaster())
writeRegister(), readRegister() and bitModify() also perform SPI access without using SpiDevice::acquireMaster()/::releaseMaster() and are not resumable

You have to switch from the update() resumable function to a protothread, and inside the protothread you can then do the initialization stuff, triggered by a flag set from the initialize() function.

rleh · 2022-03-16T14:59:27Z

examples/nucleo_f439zi/can_m2515/project.xml

+    <option name="modm:driver:mcp2515:buffer.tx">32</option>
+    <option name="modm:driver:mcp2515:buffer.rx">32</option>


32 is already the default value. Remove?

not a good idea, this shows intent .. nobody would know that this exists otherwise

Everybody that reads the documentation or uses lbuild discover-options will know about it...

src/modm/driver/can/mcp2515_options.hpp.in

rleh · 2022-03-16T15:04:57Z

src/modm/driver/can/mcp2515_options.hpp.in

+namespace mcp2515{
+namespace options{
+
+%% if options["buffer.tx"] > 0


Suggested change

%% if options["buffer.tx"] > 0

buffer.tx is enforced to be within range 1..(2^16-2) by the lbuild module.

(Same for rx buffer)

rleh · 2022-03-16T15:06:03Z

src/modm/driver/can/mcp2515_impl.hpp

@@ -16,24 +16,13 @@
 #ifndef MODM_MCP2515_HPP
 	#error	"Don't include this file directly, use 'mcp2515.hpp' instead!"
 #endif
-


Not resolved.

src/modm/driver/can/mcp2515_impl.hpp

rleh · 2022-03-16T15:45:20Z

src/modm/driver/can/mcp2515_impl.hpp

 	return true;
 }

+template <typename SPI, typename CS, typename INT>
+modm::ResumableResult<bool>


it is void now

No it's not.

rleh · 2022-03-16T15:46:55Z

src/modm/driver/can/mcp2515_impl.hpp

+template <typename SPI, typename CS, typename INT>
+modm::ResumableResult<bool>
+modm::Mcp2515<SPI, CS, INT>::update(){
+	using namespace mcp2515;
+
+	RF_BEGIN();


Shouldn't that rather be a Protothread and not a resumable function?

dunno what you mean. As the api is right now (static), there exists a protothread in the user code (which you told me in a prior version of this driver update, that it should be consistent to the general driver api) ... which one is it?

Hint:

Suggested change

template <typename SPI, typename CS, typename INT>

modm::ResumableResult<bool>

modm::Mcp2515<SPI, CS, INT>::update(){

using namespace mcp2515;

RF_BEGIN();

template <typename SPI, typename CS, typename INT>

void

modm::Mcp2515<SPI, CS, INT>::run()

{

PT_BEGIN();

//...

then the user hjas two protothreads runnign inside each other?
why should the thread be in the driver instead of the user deciding when or in which context update is called?

Hint:

not making a point clear / answering the implicity question of what do you actually want to do and why does not help me :D

rleh · 2022-03-16T15:47:43Z

src/modm/driver/can/mcp2515_impl.hpp

-	if ((readStatus(READ_STATUS) & (TXB2CNTRL_TXREQ | TXB1CNTRL_TXREQ | TXB0CNTRL_TXREQ)) ==
+	RF_BEGIN();
+
+	tempS = true;


Meaningful variable name or explanation please...

rleh · 2022-03-16T15:49:10Z

src/modm/driver/can/mcp2515_impl.hpp

+		ready = false;
 	}
+	return ready;


Suggested change

ready = false;

}

return ready;

return false;

}

return true;

rleh · 2022-03-16T16:11:28Z

src/modm/driver/can/mcp2515_impl.hpp

@@ -313,11 +398,9 @@ void
 modm::Mcp2515<SPI, CS, INT>::writeRegister(uint8_t address, uint8_t data)
 {
 	chipSelect.reset();
-


Still a lot of unnecessary whitespace change in this part of the file.

Co-authored-by: Raphael Lehmann <[email protected]>

rleh · 2022-03-16T16:31:38Z

@kikass13 Please stop marking my change requests which are not resolved as resolved.

changed mcp2515 driver to use rx and tx buffers as well as uphold the…

148e788

… standard modm::can api

kikass13 changed the title ~~changed mcp2515 driver to use rx and tx buffers as well as uphold the…~~ changed mcp2515 driver to use nonblocking spi Feb 1, 2022

kikass13 added 2 commits February 1, 2022 12:23

added example; small line cleanup

10b0e46

added update function to example, change build out dir

33db784

chris-durand added enhancement 🌈 example 🔑 labels Feb 1, 2022

added nonblocking stuff for can send api part; the read is still bloc…

21e4036

…king, which should be fine because it is controlled via interrupt

rleh reviewed Feb 1, 2022

View reviewed changes

rleh added the advanced 🤯 label Feb 1, 2022

chris-durand reviewed Feb 1, 2022

View reviewed changes

src/modm/driver/can/mcp2515_impl.hpp Outdated Show resolved Hide resolved

changed read/receive part so that it is polled in the update function…

03639b4

… as well; fixed some things in example

kikass13 force-pushed the mcp2515_protothread_update_3 branch from dc57d6f to 03639b4 Compare February 1, 2022 20:26

chris-durand reviewed Feb 1, 2022

View reviewed changes

src/modm/driver/can/mcp2515_impl.hpp Outdated Show resolved Hide resolved

kikass13 and others added 6 commits February 3, 2022 14:48

the should now be non static (the rx/tx queues as well); a modm::SpiD…

bf7b0fa

…evice is now used internally for aquiring / releasing the bus whenever accessed

WIP: Add SPI user configuration handler for speed reconfiguration

9584f22

added configurationHanlder for spi device, which should also change t…

8e6a52f

…he spi clock frequency when master is acquired

moved example to nucleo_f439; fixed example to work properly

bda5887

re-enable debug messages in main

7c9065d

re-enable @chris-durand hack for changing bus clock speed in configur…

5357bc2

…ationHandler in example; changes canbdaudrate to 500kbps

whitespaces

c177db3

chris-durand reviewed Feb 4, 2022

View reviewed changes

examples/nucleo_f439zi/can_m2515/main.cpp Outdated Show resolved Hide resolved

merge other modm stuff

6c79bc2

kikass13 and others added 3 commits February 8, 2022 17:17

removed debug log because ci will fail for avr if we leave the log le…

6e6e359

…vel intact

small acquire/release spi master fix for blocking setFilter and bitMo…

49091cf

…dify()

small send rearrange, The send function will now skip sending a packe…

dadd55f

…t (aka NOT attempting to send it) if the mcp is not readyToSend() instead of waiting for it to be ready, which defeats the point of buffering packets indefiniteley

rleh reviewed Feb 24, 2022

View reviewed changes

examples/nucleo_f439zi/can_m2515/main.cpp Outdated Show resolved Hide resolved

rleh requested changes Feb 24, 2022

View reviewed changes

fixed some things related to comments from @rleh; cleanup of exmaple …

5ac11f4

…code; minor style changes to mcp2515 driver impl

kikass13 added 2 commits February 26, 2022 14:43

@rleh: added rx & tx queue options to lbuild + mcp2515 driver; remove…

2b5b355

…d unused project xml things; cleaned up unused headers; changed mcp2515 driver class attributes to not have _ underscore postfix

remove useless / harmful modm log level overwrite + debug logs from p…

5aa143e

…rescaler

kikass13 force-pushed the mcp2515_protothread_update_3 branch from 814c603 to 5aa143e Compare February 27, 2022 16:42

Merge branch 'develop' into mcp2515_protothread_update_3

4d7df65

rleh requested changes Mar 16, 2022

View reviewed changes

kikass13 and others added 2 commits March 16, 2022 17:26

Update src/modm/driver/can/mcp2515_options.hpp.in

2e4ad3b

Co-authored-by: Raphael Lehmann <[email protected]>

Update src/modm/driver/can/mcp2515_impl.hpp

5906e17

Co-authored-by: Raphael Lehmann <[email protected]>

fixed some minor pedantic things

0ff3eef

rleh added this to the 2022q1 milestone Mar 16, 2022

rleh mentioned this pull request Mar 28, 2022

2022q1 release #842

Merged

30 tasks

rleh modified the milestones: 2022q1, 2022q2 Mar 30, 2022

rleh modified the milestones: 2022q2, 2022q3 Jul 1, 2022

salkinium removed this from the 2022q3 milestone Oct 1, 2022

rleh added the stale ♾ label Apr 9, 2023

		<option name="modm:driver:mcp2515:buffer.tx">32</option>
		<option name="modm:driver:mcp2515:buffer.rx">32</option>

changed mcp2515 driver to use nonblocking spi #817

Are you sure you want to change the base?

changed mcp2515 driver to use nonblocking spi #817

Conversation

kikass13 commented Feb 1, 2022 • edited Loading

rleh left a comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

rleh commented Feb 1, 2022

chris-durand commented Feb 1, 2022

chris-durand commented Feb 1, 2022

kikass13 commented Feb 1, 2022

chris-durand commented Feb 1, 2022 • edited Loading

kikass13 commented Feb 2, 2022

kikass13 commented Feb 4, 2022 • edited Loading

kikass13 commented Feb 4, 2022

chris-durand commented Feb 4, 2022

rleh commented Feb 8, 2022

chris-durand commented Feb 8, 2022

kikass13 commented Feb 8, 2022

rleh commented Feb 8, 2022

kikass13 commented Feb 8, 2022

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

kikass13 commented Feb 24, 2022

rleh left a comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

kikass13 Mar 18, 2022 • edited Loading

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

rleh commented Mar 16, 2022 • edited Loading

kikass13 commented Feb 1, 2022 •

edited

Loading

chris-durand commented Feb 1, 2022 •

edited

Loading

kikass13 commented Feb 4, 2022 •

edited

Loading

kikass13 Mar 18, 2022 •

edited

Loading

rleh commented Mar 16, 2022 •

edited

Loading