Re: [PATCH v6 4/5] fpga manager: Add cyclone-ps-spi driver for Altera FPGAs

From: Joshua Clayton
Date: Tue Dec 20 2016 - 14:47:47 EST


Uwe,

Thanks so much for your review.

On 12/18/2016 11:23 PM, Uwe Kleine-König wrote:
> On Fri, Dec 16, 2016 at 03:17:53PM -0800, Joshua Clayton wrote:
>> cyclone-ps-spi loads FPGA firmware over spi, using the "passive serial"
>> interface on Altera Cyclone FPGAS.
>>
>> This is one of the simpler ways to set up an FPGA at runtime.
>> The signal interface is close to unidirectional spi with lsb first.
>>
>> Signed-off-by: Joshua Clayton <stillcompiling@xxxxxxxxx>
>> ---
>> drivers/fpga/Kconfig | 7 ++
>> drivers/fpga/Makefile | 1 +
>> drivers/fpga/cyclone-ps-spi.c | 186 ++++++++++++++++++++++++++++++++++++++++++
>> 3 files changed, 194 insertions(+)
>> create mode 100644 drivers/fpga/cyclone-ps-spi.c
>>
>> diff --git a/drivers/fpga/Kconfig b/drivers/fpga/Kconfig
>> index ce861a2..e6c032d 100644
>> --- a/drivers/fpga/Kconfig
>> +++ b/drivers/fpga/Kconfig
>> @@ -20,6 +20,13 @@ config FPGA_REGION
>> FPGA Regions allow loading FPGA images under control of
>> the Device Tree.
>>
>> +config FPGA_MGR_CYCLONE_PS_SPI
>> + tristate "Altera Cyclone FPGA Passive Serial over SPI"
>> + depends on SPI
>> + help
>> + FPGA manager driver support for Altera Cyclone using the
>> + passive serial interface over SPI
>> +
>> config FPGA_MGR_SOCFPGA
>> tristate "Altera SOCFPGA FPGA Manager"
>> depends on ARCH_SOCFPGA || COMPILE_TEST
>> diff --git a/drivers/fpga/Makefile b/drivers/fpga/Makefile
>> index 8df07bc..a112bef 100644
>> --- a/drivers/fpga/Makefile
>> +++ b/drivers/fpga/Makefile
>> @@ -6,6 +6,7 @@
>> obj-$(CONFIG_FPGA) += fpga-mgr.o
>>
>> # FPGA Manager Drivers
>> +obj-$(CONFIG_FPGA_MGR_CYCLONE_PS_SPI) += cyclone-ps-spi.o
>> obj-$(CONFIG_FPGA_MGR_SOCFPGA) += socfpga.o
>> obj-$(CONFIG_FPGA_MGR_SOCFPGA_A10) += socfpga-a10.o
>> obj-$(CONFIG_FPGA_MGR_ZYNQ_FPGA) += zynq-fpga.o
>> diff --git a/drivers/fpga/cyclone-ps-spi.c b/drivers/fpga/cyclone-ps-spi.c
>> new file mode 100644
>> index 0000000..f9126f9
>> --- /dev/null
>> +++ b/drivers/fpga/cyclone-ps-spi.c
>> @@ -0,0 +1,186 @@
>> +/**
>> + * Altera Cyclone Passive Serial SPI Driver
>> + *
>> + * Copyright (c) 2017 United Western Technologies, Corporation
> In which timezone it's already 2017? s/ / /
>
LOL. It will be 2017 long before 4.11 was my thinking.
I guess I've never spent much time on time stamp etiquette for copyright.
It said "2015" in v5, despite still being revised.
>> + *
>> + * Joshua Clayton <stillcompiling@xxxxxxxxx>
>> + *
>> + * Manage Altera FPGA firmware that is loaded over spi using the passive
>> + * serial configuration method.
>> + * Firmware must be in binary "rbf" format.
>> + * Works on Cyclone V. Should work on cyclone series.
>> + * May work on other Altera FPGAs.
> I can test this later on an Arria 10. I'm not sure what the connection
> between "Cyclone" and "Arria" is, but the protocol looks similar.
My guess was it would be.
Would be wonderful to have someone to test.
>> + *
>> + */
>> +
>> +#include <linux/bitrev.h>
>> +#include <linux/delay.h>
>> +#include <linux/fpga/fpga-mgr.h>
>> +#include <linux/gpio/consumer.h>
>> +#include <linux/module.h>
>> +#include <linux/of_gpio.h>
>> +#include <linux/spi/spi.h>
>> +#include <linux/sizes.h>
>> +
>> +#define FPGA_RESET_TIME 50 /* time in usecs to trigger FPGA config */
>> +#define FPGA_MIN_DELAY 50 /* min usecs to wait for config status */
>> +#define FPGA_MAX_DELAY 1000 /* max usecs to wait for config status */
>> +
>> +struct cyclonespi_conf {
>> + struct gpio_desc *config;
>> + struct gpio_desc *status;
>> + struct spi_device *spi;
>> +};
>> +
>> +static const struct of_device_id of_ef_match[] = {
>> + { .compatible = "altr,cyclone-ps-spi-fpga-mgr", },
>> + {}
>> +};
>> +MODULE_DEVICE_TABLE(of, of_ef_match);
> barebox already has such a driver, the binding is available at
>
> https://git.pengutronix.de/cgit/barebox/tree/Documentation/devicetree/bindings/firmware/altr,passive-serial.txt
>
> (This isn't completely accurate because nstat is optional in the driver.)
Interesting.
The binding looks ... like we should synchronize those bindings.
In the case of my hardware, I don't have access to the confd, but do
have access to nstat. I was thinking that using confd to signal done
would be a nice but optional... Ideally you'd have both.
>> +static enum fpga_mgr_states cyclonespi_state(struct fpga_manager *mgr)
>> +{
>> + struct cyclonespi_conf *conf = (struct cyclonespi_conf *)mgr->priv;
>> +
>> + if (gpiod_get_value(conf->status))
>> + return FPGA_MGR_STATE_RESET;
>> +
>> + return FPGA_MGR_STATE_UNKNOWN;
>> +}
>> +
>> +static int cyclonespi_write_init(struct fpga_manager *mgr,
>> + struct fpga_image_info *info,
>> + const char *buf, size_t count)
>> +{
>> + struct cyclonespi_conf *conf = (struct cyclonespi_conf *)mgr->priv;
>> + int i;
>> +
>> + if (info->flags & FPGA_MGR_PARTIAL_RECONFIG) {
>> + dev_err(&mgr->dev, "Partial reconfiguration not supported.\n");
>> + return -EINVAL;
>> + }
>> +
>> + gpiod_set_value(conf->config, 1);
>> + usleep_range(FPGA_RESET_TIME, FPGA_RESET_TIME + 20);
>> + if (!gpiod_get_value(conf->status)) {
>> + dev_err(&mgr->dev, "Status pin should be low.\n");
> You write this when get_value returns 0. There is something fishy.
I'll take a look. These gpios are "active low", so a logical 1 is a
physical low, IIRC. Maybe I should change the wording:
Something like dev_err(&mgr->dev, "Status pin should be in reset.\n");

Perhaps?


>> + return -EIO;
>> + }
>> +
>> + gpiod_set_value(conf->config, 0);
>> + for (i = 0; i < (FPGA_MAX_DELAY / FPGA_MIN_DELAY); i++) {
>> + usleep_range(FPGA_MIN_DELAY, FPGA_MIN_DELAY + 20);
>> + if (!gpiod_get_value(conf->status))
>> + return 0;
>> + }
>> +
>> + dev_err(&mgr->dev, "Status pin not ready.\n");
>> + return -EIO;
> For Arria 10 the documentation has:
>
> To ensure a successful configuration, send the entire
> configuration data to the device. CONF_DONE is released high
> when the device receives all the configuration data
> successfully. After CONF_DONE goes high, send two additional
> falling edges on DCLK to begin initialization and enter user
> mode.
>
> ISTR this is necessary for Arria V, too.
DCLK is the spi clock, yes?
Would sending an extra byte after CONF_DONE is released suffice?
>> +}
>> +
>> +static void rev_buf(void *buf, size_t len)
>> +{
>> + u32 *fw32 = (u32 *)buf;
>> + const u32 *fw_end = (u32 *)(buf + len);
>> +
>> + /* set buffer to lsb first */
>> + while (fw32 < fw_end) {
>> + *fw32 = bitrev8x4(*fw32);
>> + fw32++;
>> + }
> Is the size of the firmware always a multiple of 32 bit? If len isn't a
> multiple of 4 you access data after the end of buf.
The rbf cyclone V bitstream is padded out with extra bytes of FFFFFFFF
and always a multiple of 4 bytes.
I could not find anywhere this is documented.
I guess we should not assume this will always be the case.
I'll add something to handle the tail.
>> +}
>> +
>> +static int cyclonespi_write(struct fpga_manager *mgr, const char *buf,
>> + size_t count)
>> +{
>> + struct cyclonespi_conf *conf = (struct cyclonespi_conf *)mgr->priv;
>> + const char *fw_data = buf;
>> + const char *fw_data_end = fw_data + count;
>> +
>> + while (fw_data < fw_data_end) {
>> + int ret;
>> + size_t stride = min(fw_data_end - fw_data, SZ_4K);
>> +
>> + rev_buf((void *)fw_data, stride);
> This isn't necessary if the spi controller supports SPI_LSB_FIRST. At
> least the mvebu spi core can do this for you. (The driver doesn't
> support it yet, though.)
This is true, but many of them do not.

Moritz Fischer had proposal to add things like this with a flag.
It could then be part of the library, rather than part of the driver

Speaking of which,
I made an unsuccessful attempt to hack generic lsb first SPI
with an extra bounce buffer.
Sending was fine, but I ran into trouble with LSB first rx
(I think) because of dma.

>> + ret = spi_write(conf->spi, fw_data, stride);
>> + if (ret) {
>> + dev_err(&mgr->dev, "spi error in firmware write: %d\n",
>> + ret);
>> + return ret;
>> + }
>> + fw_data += stride;
>> + }
>> +
>> + return 0;
>> +}
>> [...]
>> +static int cyclonespi_probe(struct spi_device *spi)
>> +{
>> + struct cyclonespi_conf *conf = devm_kzalloc(&spi->dev, sizeof(*conf),
>> + GFP_KERNEL);
> please indent to the opening (.
Will fix.
>> +
>> + if (!conf)
>> + return -ENOMEM;
>> +
>> + conf->spi = spi;
>> + conf->config = devm_gpiod_get(&spi->dev, "config", GPIOD_OUT_HIGH);
>> + if (IS_ERR(conf->config)) {
>> + dev_err(&spi->dev, "Failed to get config gpio: %ld\n",
>> + PTR_ERR(conf->config));
>> + return PTR_ERR(conf->config);
>> + }
>> +
>> + conf->status = devm_gpiod_get(&spi->dev, "status", GPIOD_IN);
>> + if (IS_ERR(conf->status)) {
>> + dev_err(&spi->dev, "Failed to get status gpio: %ld\n",
>> + PTR_ERR(conf->status));
>> + return PTR_ERR(conf->status);
>> + }
>> +
>> + return fpga_mgr_register(&spi->dev,
>> + "Altera Cyclone PS SPI FPGA Manager",
>> + &cyclonespi_ops, conf);
>> +}
>> +
>> +static int cyclonespi_remove(struct spi_device *spi)
>> +{
>> + fpga_mgr_unregister(&spi->dev);
>> +
>> + return 0;
>> +}
>> +
>> +static struct spi_driver cyclonespi_driver = {
>> + .driver = {
>> + .name = "cyclone-ps-spi",
>> + .owner = THIS_MODULE,
>> + .of_match_table = of_match_ptr(of_ef_match),
>> + },
>> + .probe = cyclonespi_probe,
>> + .remove = cyclonespi_remove,
>> +};
> I'm not a big fan of aligning the assignment operators. This tends to
> get out of sync or results in bigger than necessary changes in follow up
> patches. Note that it's out of sync already now, so I suggest to use a
> single space before =.
Yes, I can see it your way. Will change.
>> +
>> +module_spi_driver(cyclonespi_driver)
>> +
>> +MODULE_LICENSE("GPL");
>> +MODULE_AUTHOR("Joshua Clayton <stillcompiling@xxxxxxxxx>");
>> +MODULE_DESCRIPTION("Module to load Altera FPGA firmware over spi");
>> --
> Best regards
> Uwe
>
Even bester regards,

Joshua