| lists.openwall.net | lists / announce owl-users owl-dev john-users john-dev passwdqc-users yescrypt popa3d-users / oss-security kernel-hardening musl sabotage tlsify passwords / crypt-dev xvendor / Bugtraq Full-Disclosure linux-kernel linux-netdev linux-ext4 linux-hardening linux-cve-announce PHC | |
|
Open Source and information security mailing list archives
| ||
|
Message-ID: <20250622155026.15b3723c@jic23-huawei>
Date: Sun, 22 Jun 2025 15:50:26 +0100
From: Jonathan Cameron <jic23@...nel.org>
To: David Lechner <dlechner@...libre.com>
Cc: Michael Hennerich <Michael.Hennerich@...log.com>, Nuno Sá
<nuno.sa@...log.com>, Andy Shevchenko <andy@...nel.org>, Rob Herring
<robh@...nel.org>, Krzysztof Kozlowski <krzk+dt@...nel.org>, Conor Dooley
<conor+dt@...nel.org>, Mark Brown <broonie@...nel.org>,
linux-iio@...r.kernel.org, linux-kernel@...r.kernel.org,
devicetree@...r.kernel.org, linux-spi@...r.kernel.org
Subject: Re: [PATCH 5/9] iio: adc: ad_sigma_delta: use
spi_optimize_message()
On Fri, 20 Jun 2025 17:20:11 -0500
David Lechner <dlechner@...libre.com> wrote:
> Use spi_optimize_message() to improve the performance of buffered reads.
>
> By setting up the SPI message and pre-optimizing it in the buffer
> postenable callback, we can reduce overhead during each sample read.
>
> A rough estimate shows that this reduced the CPU usage of the interrupt
> handler thread from 22% to 16% using an EVAL-AD4112ARDZ board on a
> DE10-Nano (measuring a single channel at the default 6.2 kHz sample
> rate).
Nice.
>
> Signed-off-by: David Lechner <dlechner@...libre.com>
A query inline that is really about what the existing code is doing.
Also a request for duplicating the comment about extra padding as that
is unusual!
Jonathan
> ---
> drivers/iio/adc/ad_sigma_delta.c | 74 ++++++++++++++++------------------
> include/linux/iio/adc/ad_sigma_delta.h | 3 ++
> 2 files changed, 38 insertions(+), 39 deletions(-)
>
> diff --git a/drivers/iio/adc/ad_sigma_delta.c b/drivers/iio/adc/ad_sigma_delta.c
> index 883cc02d4099908644af523bcf66c0178fbd0e55..a9b97f5d4107a2e1bb74877d30403445e9b04a44 100644
> --- a/drivers/iio/adc/ad_sigma_delta.c
> +++ b/drivers/iio/adc/ad_sigma_delta.c
> @@ -458,8 +458,10 @@ EXPORT_SYMBOL_NS_GPL(ad_sigma_delta_single_conversion, "IIO_AD_SIGMA_DELTA");
> static int ad_sd_buffer_postenable(struct iio_dev *indio_dev)
> {
> struct ad_sigma_delta *sigma_delta = iio_device_get_drvdata(indio_dev);
> + const struct iio_scan_type *scan_type = &indio_dev->channels[0].scan_type;
> + struct spi_transfer *xfer = sigma_delta->sample_xfer;
> unsigned int i, slot, samples_buf_size;
> - unsigned int channel;
> + unsigned int channel, scan_size;
> u8 *samples_buf;
> int ret;
>
> @@ -493,7 +495,7 @@ static int ad_sd_buffer_postenable(struct iio_dev *indio_dev)
> return ret;
> }
>
> - samples_buf_size = ALIGN(slot * indio_dev->channels[0].scan_type.storagebits, 8);
> + samples_buf_size = ALIGN(slot * scan_type->storagebits, 8);
What is this doing? Storage bits as the name suggests is in bits but this superficially
seems to be treating it like it was in bytes.
The ALIGN() is I think just about the timestamp alignment.
> samples_buf_size += sizeof(int64_t);
> samples_buf = devm_krealloc(&sigma_delta->spi->dev, sigma_delta->samples_buf,
> samples_buf_size, GFP_KERNEL);
> @@ -501,6 +503,27 @@ static int ad_sd_buffer_postenable(struct iio_dev *indio_dev)
> return -ENOMEM;
>
> sigma_delta->samples_buf = samples_buf;
> + scan_size = BITS_TO_BYTES(scan_type->realbits + scan_type->shift);
> + xfer[1].rx_buf = &sigma_delta->rx_buf[scan_size == 3 ? 1 : 0];
This is black magic. Comment needed I think on why the offset in rx_buf.
> + xfer[1].len = scan_size + (sigma_delta->status_appended ? 1 : 0);
> + xfer[1].cs_change = 1;
> +
> + if (sigma_delta->info->has_registers) {
> + xfer[0].tx_buf = &sigma_delta->sample_addr;
> + xfer[0].len = 1;
> +
> + ad_sd_set_read_reg_addr(sigma_delta,
> + sigma_delta->info->data_reg ?: AD_SD_REG_DATA,
> + &sigma_delta->sample_addr);
> + spi_message_init_with_transfers(&sigma_delta->sample_msg, xfer, 2);
> + } else {
> + spi_message_init_with_transfers(&sigma_delta->sample_msg,
> + &xfer[1], 1);
> + }
> +
> + ret = spi_optimize_message(sigma_delta->spi, &sigma_delta->sample_msg);
> + if (ret)
> + return ret;
>
> spi_bus_lock(sigma_delta->spi->controller);
> sigma_delta->bus_locked = true;
> @@ -520,6 +543,7 @@ static int ad_sd_buffer_postenable(struct iio_dev *indio_dev)
>
> err_unlock:
> spi_bus_unlock(sigma_delta->spi->controller);
> + spi_unoptimize_message(&sigma_delta->sample_msg);
>
> return ret;
> }
> @@ -550,51 +577,20 @@ static irqreturn_t ad_sd_trigger_handler(int irq, void *p)
> struct iio_dev *indio_dev = pf->indio_dev;
> struct ad_sigma_delta *sigma_delta = iio_device_get_drvdata(indio_dev);
> u8 *data = sigma_delta->rx_buf;
> - unsigned int transfer_size;
> unsigned int sample_size;
> unsigned int sample_pos;
> unsigned int status_pos;
> unsigned int reg_size;
> - unsigned int data_reg;
> + int ret;
>
> reg_size = BITS_TO_BYTES(indio_dev->channels[0].scan_type.realbits +
> indio_dev->channels[0].scan_type.shift);
> + /* For 24-bit data, there is an extra byte of padding. */
> + status_pos = reg_size + (reg_size == 3 ? 1 : 0);
Ah. I guess this applies above as well. I'd duplicate the comment.
>
> - if (sigma_delta->info->data_reg != 0)
> - data_reg = sigma_delta->info->data_reg;
> - else
> - data_reg = AD_SD_REG_DATA;
> -
> - /* Status word will be appended to the sample during transfer */
> - if (sigma_delta->status_appended)
> - transfer_size = reg_size + 1;
> - else
> - transfer_size = reg_size;
> -
> - switch (reg_size) {
> - case 4:
> - case 2:
> - case 1:
> - status_pos = reg_size;
> - ad_sd_read_reg_raw(sigma_delta, data_reg, transfer_size, &data[0]);
> - break;
> - case 3:
> - /*
> - * Data array after transfer will look like (if status is appended):
> - * data[] = { [0][sample][sample][sample][status] }
> - * Keeping the first byte 0 shifts the status position by 1 byte to the right.
> - */
> - status_pos = reg_size + 1;
> -
> - /* We store 24 bit samples in a 32 bit word. Keep the upper
> - * byte set to zero. */
> - ad_sd_read_reg_raw(sigma_delta, data_reg, transfer_size, &data[1]);
> - break;
> -
> - default:
> - dev_err_ratelimited(&indio_dev->dev, "Unsupported reg_size: %u\n", reg_size);
> + ret = spi_sync_locked(sigma_delta->spi, &sigma_delta->sample_msg);
> + if (ret)
> goto irq_handled;
> - }
>
> /*
> * For devices sampling only one channel at
> diff --git a/include/linux/iio/adc/ad_sigma_delta.h b/include/linux/iio/adc/ad_sigma_delta.h
> index 5056677c9941afadc2383febbcafeb02e23a4f44..2037bb68b44115681ff48f66b580b63f50c2ea9e 100644
> --- a/include/linux/iio/adc/ad_sigma_delta.h
> +++ b/include/linux/iio/adc/ad_sigma_delta.h
> @@ -105,6 +105,8 @@ struct ad_sigma_delta {
> bool status_appended;
> /* map slots to channels in order to know what to expect from devices */
> unsigned int *slots;
> + struct spi_message sample_msg;
> + struct spi_transfer sample_xfer[2];
> u8 *samples_buf;
>
> /*
> @@ -116,6 +118,7 @@ struct ad_sigma_delta {
> */
> u8 tx_buf[4] __aligned(IIO_DMA_MINALIGN);
> u8 rx_buf[16] __aligned(8);
> + u8 sample_addr;
> };
>
> static inline int ad_sigma_delta_set_channel(struct ad_sigma_delta *sd,
>
Powered by blists - more mailing lists