ad5791 offload #2616

ahaslam2 · 2024-10-15T15:04:43Z

spi offload support for ad57xx dac

This adds offload support for the ad57xx dac to be able to send samples up to 1MSPS.

PR Type

Bug fix (a change that fixes an issue)
New feature (a change that adds new functionality)
Breaking change (a change that affects other repos or cause CIs to fail)

PR Checklist

I have conducted a self-review of my own code changes
I have tested the changes on the relevant hardware
I have updated the documentation outside this repo accordingly (if there is the case)

nunojsa

Here it goes my review. Note that I'm expecting some stuff to be sent upstream (like gpio reset)... Also it seems this driver could use a proper refactor if you are willing to do that 😄

nunojsa · 2024-10-16T11:25:55Z

drivers/iio/dac/ad5791.c

+	st->gpio_ldac = devm_gpiod_get_optional(&spi->dev, "ldac",
+						  GPIOD_OUT_LOW);
+	if (IS_ERR(st->gpio_ldac))
+		return PTR_ERR(st->gpio_ldac);


What's the intent of the above two gpio's? I would say, if we're not using them now, don't bother in adding them. Still ok to leave them in the bindings though.

they are actually on the eval board and if i dont drive them they are set as input. we need to make sure they are on the right state: CLR high and LDAC low (datasheet table9 p21) for the output to be set according to the register value when we write on spi xfer. I think we need to keep them.

nunojsa · 2024-10-16T11:28:33Z

drivers/iio/dac/ad5791.c

+	st->gpio_reset = devm_gpiod_get_optional(&spi->dev, "reset",
+						  GPIOD_OUT_LOW);
+	if (IS_ERR(st->gpio_reset))
+		return PTR_ERR(st->gpio_reset);


The above makes sense to have already. But do a reset in case we have it. Typical pattern is:

gpio_reset = devm_gpiod_get_optional(&spi->dev, "reset", GPIOD_OUT_HIGH); if (IS_ERR(gpio_reset)) return PTR_ERR(gpio_reset); if (gpio_reset) { // typically there's some time we need to wait for the reset to happen fseep(x); // get it out of reset gpiod_set_value_cansleep(gpio_reset, 0); }

drivers/iio/dac/ad5791.c

nunojsa · 2024-10-16T11:36:23Z

drivers/iio/dac/ad5791.c

-	st->pwr_down = pwr_down;
-
-	ret = ad5791_spi_write(st, AD5791_ADDR_CTRL, st->ctrl);
+	ret = __ad5791_dac_powerdown(indio_dev, pwr_down);


Unrelated change. Not disagreeing with it but should be in a separate patch. If you want to take it one step further, it even seems that some proper locking is missing... Moreover, we should set st->pwr_down after we actually do ad5791_spi_write(st, AD5791_ADDR_CTRL, st->ctrl)

i will drop this change and only check the power status in pre-enable. as you suggested below.

nunojsa · 2024-10-16T11:38:49Z

drivers/iio/dac/ad5791.c

+
+	if (spi_engine_ex_offload_supported(spi)) {
+		indio_dev->channels =
+			&ad5791_channels[spi_get_device_id(spi)->driver_data];


The above could also be an opportunity for a nice cleanup. Adding a chip_info structure and actual OF support. Thus getting rid of the id enum

nunojsa · 2024-10-16T11:42:19Z

drivers/iio/dac/ad5791.c

+	struct ad5791_state *st = iio_priv(indio_dev);
+	int ret;
+
+	ret = __ad5791_dac_powerdown(indio_dev, false);


I see now it's related but see below...

nunojsa · 2024-10-16T11:42:31Z

drivers/iio/dac/ad5791.c

+	pwm_disable(st->cnv_trigger);
+	spi_bus_unlock(st->spi->master);
+
+	return __ad5791_dac_powerdown(indio_dev, true);


Any reason to assume we want to powerdown the DAC? Since it's a user control I would rather check it and return error or something in ad5791_buffer_preenable() in case it's in a powerdown state. This would also mean that we can't power it down while in buffer mode. Thus, using iio_device_claim_direct_mode()

nunojsa · 2024-10-16T11:43:46Z

drivers/iio/dac/ad5791.c

+	int ret;
+
+	if (!st->cnv_trigger)
+		return -ENODEV;


AFAICT, this is only available for offload support and cnv_trigger is mandatory (otherwise we fail to probe). So, seems like a redundant check.

drivers/iio/dac/ad5791.c

dlech

A couple more suggestions for upstream "cleanup":

Use devm_regulator_get_enable_read_voltage() to simplify the regulator code.
Drop the driver remove callback by using more devm_.

dlech · 2024-10-16T15:41:34Z

Documentation/devicetree/bindings/iio/dac/adi,ad5791.yaml

@@ -31,6 +31,17 @@ properties:
      gain of two configuration.
    type: boolean

+  reset-gpios:


Jonathan also likes to have all of the power supplies in the bindings, so we can add vcc-suppy, vrefp-supply, vrefn-supply and iovcc-supply when we submit this upstream.

arch/arm/boot/dts/socfpga_cyclone5_de10_nano.dtsi

arch/arm/boot/dts/socfpga_cyclone5_de10_nano_ad57xx.dts

Documentation/devicetree/bindings/iio/dac/adi,ad5791.yaml

drivers/iio/dac/ad5791.c

ahaslam2 · 2024-10-21T13:48:27Z

changes in V2:

implement new suggested driver simplifications:

Include chip_info in device match tables / remove chip id enum
use devm_regulator_get_enable_read_voltage on probe.

Fixes related to comments:

drop pwm DT binding
use devm_spi_optimize_message
dont power_up/down on enable
remove redundant trigger check
us hw pin reset if available
factor out offload setup from probe.
rebase after de10nano DT changes from uwe.

arch/arm/boot/dts/socfpga_cyclone5_de10_nano_ad57xx.dts

drivers/iio/dac/ad5791.c

dlech · 2024-10-21T21:08:15Z

drivers/iio/dac/ad5791.c

@@ -430,30 +412,9 @@ static int ad5791_probe(struct spi_device *spi)
 	indio_dev->name = st->chip_info->name;
 	ret = iio_device_register(indio_dev);
 	if (ret)
-		goto error_disable_reg_neg;
+		return dev_err_probe(&spi->dev, ret, "unable to register iio device\n");


Most drivers will just return devm_iio_device_register() directly, so don't usually have the dev_err_probe() here.

arch/arm/boot/dts/socfpga_cyclone5_de10_nano_hps.dtsi

ahaslam2 · 2024-10-22T13:48:07Z

changes in V3:

add patch to remove altr,gpio-bank-width in dt nodes
remove adi,channels node
fix macro style.
use fsleep
remove spi_set_drvdata
use devm_iio_device_register and delete remove callback.

drivers/iio/dac/ad5791.c

arch/arm/boot/dts/socfpga_cyclone5_de10_nano_ad57xx.dts

drivers/iio/dac/ad5791.c

Depending on board layout, the ad57xx may need control of reset, clear, and ldac pins by the host driver. Add optional bindings for these gpios. Signed-off-by: Axel Haslam <ahaslam@baylibre.com>

Include a chip info struct in device SPI and device OF match tables to provide channel definitions for each particular ADC model and eliminate device enum. Suggested-by: Nuno Sa <nuno.sa@analog.com> Signed-off-by: Axel Haslam <ahaslam@baylibre.com>

The ad7591 has reset, clr and ldac gpios. For the DAC to output data writen to the data register the state of these gpios has to be: RESET=1, CLR=1, LDAC=0 Add these gpios to the driver making them optional in case they are fixed on the pcb. Signed-off-by: Axel Haslam <ahaslam@baylibre.com>

Simplify probe by using of the devm_regulator_get_enable_read_voltage. Suggested-by: David Lechner <dlechner@baylibre.com> Signed-off-by: Axel Haslam <ahaslam@baylibre.com>

Use devm_iio_device_register to automatically free resources. Suggested-by: David Lechner <dlechner@baylibre.com> Signed-off-by: Axel Haslam <ahaslam@baylibre.com>

This property is not defined on the binidngs, and is not used by the driver. Remove it Suggested-by: David Lechner <dlechner@baylibre.com> Signed-off-by: Axel Haslam <ahaslam@baylibre.com>

This adds the bi-directional gpio block to the d10 nano Signed-off-by: Axel Haslam <ahaslam@baylibre.com>

Add devicetree for ad57xx on de10nano. Signed-off-by: Axel Haslam <ahaslam@baylibre.com>

To allow offload to stream a "tx" type of xfer we need the write flag set but no messages on the SDO memory of the offload engine. but the current code relies on the tx_buf pointer for both of these operations, so there is no way to separate the two. The offload patches currently on development for upstream add flags on core spi structures to flag an offload xfer as a streaming xfer. While those patches get review and are ready to backport to the adi tree, Workaround this limitation by checking if the tx_buf address is -1 to flag it as "streaming type". Signed-off-by: Axel Haslam <ahaslam@baylibre.com>

Add spi offload support to stream tx buffers using dma. This allows loading samples to the dac with a rate of 1 MSPS. Signed-off-by: Axel Haslam <ahaslam@baylibre.com>

ahaslam2 · 2024-10-23T15:31:18Z

Changes V4

remove ref_clk from dts
ldac gpio is active low
use spi->dev instead of indio_dev->dev.parent
more links in dts.

dlech

LGTM. I think it would be best to upstream the first 5 patches first before merging this in case there are additional changes there. And I still think it would be a good idea to also add the power supplies to the DT bindings since we are touching this anyway.

ukleinek · 2024-10-24T10:27:32Z

arch/arm/boot/dts/socfpga_cyclone5_de10_nano_ad57xx.dts

+	adc_trigger: pwm@0x00050000 {
+		compatible = "adi,axi-pwmgen-2.00.a";
+		reg = <0x00050000 0x1000>;
+		#pwm-cells = <2>;


Using #pwm-cells = <3> would be nice, see https://lore.kernel.org/linux-pwm/20241024102554.711689-2-u.kleine-koenig@baylibre.com

ukleinek · 2024-10-24T14:52:32Z

arch/arm/boot/dts/socfpga_cyclone5_de10_nano_ad57xx.dts

+
+			dmas = <&tx_dma 0>;
+			dma-names = "tx";
+			pwms = <&adc_trigger 0 0>;


How unusual to have the default period 0. I'd rather expect to see a period that matches the max freq for that chip here (or something bigger).

ok. just for the record, i could not clearly make out a max frequency clearly mentioned on the datasheet.
only the spi speed of 35Mbps is mentioned. if we calculate 24 bits, i think a sample rate of 1Mbps is "reasonable"

ukleinek · 2024-10-24T14:59:03Z

drivers/iio/dac/ad5791.c

+static int ad5791_get_sampling_freq(struct ad5791_state *st)
+{
+	return DIV_ROUND_CLOSEST_ULL(NSEC_PER_SEC,
+				     pwm_get_period(st->cnv_trigger));
+}
+
+static int __ad5791_set_sampling_freq(struct ad5791_state *st, int freq)
+{
+	struct pwm_state cnv_state;
+	u32 rem;
+
+	pwm_init_state(st->cnv_trigger, &cnv_state);
+	cnv_state.period = div_u64_rem((u64)DIV_ROUND_UP(st->ref_clk_rate, freq) * NSEC_PER_SEC,
+				       st->ref_clk_rate, &rem);
+	if (rem)
+		cnv_state.period += 1;
+
+	cnv_state.duty_cycle = DIV_ROUND_UP(NSEC_PER_SEC, st->ref_clk_rate);
+
+	return pwm_apply_state(st->cnv_trigger, &cnv_state);
+}


It's strange to use up-rounding when calculating the period from the frequency, but round-closest for the other direction.

The calculation freq -> period isn't very intuitive, a comment would be nice to explain the details. (I hope you didn't copy that formula from a driver where I introduced that without a comment :-))

ok. i copied were you had the comment. i just did not copy the comment itself. setting pwm in the driver should go away when the offload stuff is backpacked to mainline, thats why i did not bother. but ill add it.

ukleinek · 2024-10-24T15:00:56Z

drivers/iio/dac/ad5791.c

+		return PTR_ERR(st->cnv_trigger);
+
+	ret = devm_add_action_or_reset(&spi->dev, ad5791_pwm_diasble,
+				       st->cnv_trigger);


Typo: s/diasble/disable/

Registering that devm action before enabling the PWM is wrong.

ahaslam2 force-pushed the axelh/ad5791_offload branch 5 times, most recently from 802e9b1 to 3976d47 Compare October 15, 2024 15:55

ahaslam2 changed the title ~~Axelh/ad5791 offload~~ ad5791 offload Oct 15, 2024

ahaslam2 force-pushed the axelh/ad5791_offload branch 2 times, most recently from 03d271e to 65e5a18 Compare October 16, 2024 10:14

ahaslam2 marked this pull request as ready for review October 16, 2024 10:16

ahaslam2 requested review from nunojsa, mhennerich, dbogdan, btogorean, chegbeli, buha, amiclaus, rbolboac, danmois, CiprianRegus and machschmitt as code owners October 16, 2024 10:16

ahaslam2 requested a review from dlech October 16, 2024 10:17

nunojsa reviewed Oct 16, 2024

View reviewed changes

dlech reviewed Oct 16, 2024

View reviewed changes

ahaslam2 force-pushed the axelh/ad5791_offload branch from 65e5a18 to 5be52c8 Compare October 21, 2024 13:46

dlech reviewed Oct 21, 2024

View reviewed changes

ahaslam2 force-pushed the axelh/ad5791_offload branch 2 times, most recently from 0c0d193 to 9c53b59 Compare October 22, 2024 13:47

dlech reviewed Oct 22, 2024

View reviewed changes

drivers/iio/dac/ad5791.c Outdated Show resolved Hide resolved

ahaslam2 added 10 commits October 23, 2024 10:44

dt-bindings: adi,ad5791.yaml: Add support optional gpios

bcad791

Depending on board layout, the ad57xx may need control of reset, clear, and ldac pins by the host driver. Add optional bindings for these gpios. Signed-off-by: Axel Haslam <ahaslam@baylibre.com>

iio: dac: ad5791: Use devm_regulator_get_enable_read_voltage

5c9afcd

Simplify probe by using of the devm_regulator_get_enable_read_voltage. Suggested-by: David Lechner <dlechner@baylibre.com> Signed-off-by: Axel Haslam <ahaslam@baylibre.com>

iio: dac: ad5791: Use devm_iio_device_register

ac02938

Use devm_iio_device_register to automatically free resources. Suggested-by: David Lechner <dlechner@baylibre.com> Signed-off-by: Axel Haslam <ahaslam@baylibre.com>

arm: dts: socfpga: Remove reference to gpio-bank-width

52de110

This property is not defined on the binidngs, and is not used by the driver. Remove it Suggested-by: David Lechner <dlechner@baylibre.com> Signed-off-by: Axel Haslam <ahaslam@baylibre.com>

arm: dts: de10_nano: Add bi-directional gpios

bac3e06

This adds the bi-directional gpio block to the d10 nano Signed-off-by: Axel Haslam <ahaslam@baylibre.com>

arm: dts: de10_nano: Add support for ad57xx.

86c20ff

Add devicetree for ad57xx on de10nano. Signed-off-by: Axel Haslam <ahaslam@baylibre.com>

iio: dac: ad5791: Add offload support

e9af1e7

Add spi offload support to stream tx buffers using dma. This allows loading samples to the dac with a rate of 1 MSPS. Signed-off-by: Axel Haslam <ahaslam@baylibre.com>

ahaslam2 force-pushed the axelh/ad5791_offload branch from 9c53b59 to e9af1e7 Compare October 23, 2024 15:28

dlech approved these changes Oct 23, 2024

View reviewed changes

ukleinek reviewed Oct 24, 2024

View reviewed changes

ukleinek suggested changes Oct 24, 2024

View reviewed changes

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

ad5791 offload #2616

ad5791 offload #2616

ahaslam2 commented Oct 15, 2024

nunojsa left a comment

nunojsa Oct 16, 2024

ahaslam2 Oct 16, 2024

nunojsa Oct 16, 2024

ahaslam2 Oct 16, 2024

nunojsa Oct 16, 2024

ahaslam2 Oct 16, 2024

nunojsa Oct 16, 2024 •

edited

Loading

ahaslam2 Oct 16, 2024

nunojsa Oct 16, 2024

nunojsa Oct 16, 2024

ahaslam2 Oct 16, 2024

nunojsa Oct 16, 2024

dlech left a comment

dlech Oct 16, 2024

ahaslam2 commented Oct 21, 2024

dlech Oct 21, 2024

ahaslam2 Oct 22, 2024

ahaslam2 commented Oct 22, 2024

ahaslam2 commented Oct 23, 2024

dlech left a comment

ukleinek Oct 24, 2024

ahaslam2 Oct 25, 2024

ukleinek Oct 24, 2024

ahaslam2 Oct 25, 2024

ukleinek Oct 24, 2024

ahaslam2 Oct 25, 2024

ukleinek Oct 24, 2024

ahaslam2 Oct 25, 2024

ad5791 offload #2616

Are you sure you want to change the base?

ad5791 offload #2616

Conversation

ahaslam2 commented Oct 15, 2024

spi offload support for ad57xx dac

PR Type

PR Checklist

nunojsa 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

nunojsa Oct 16, 2024 • 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

Choose a reason for hiding this comment

Choose a reason for hiding this comment

dlech left a comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

ahaslam2 commented Oct 21, 2024

Choose a reason for hiding this comment

Choose a reason for hiding this comment

ahaslam2 commented Oct 22, 2024

ahaslam2 commented Oct 23, 2024

dlech 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

nunojsa Oct 16, 2024 •

edited

Loading