Re: [PATCH v5 1/2] fpga: dfl: add the userspace I/O device support for DFL devices

[Tom Rix <trix@redhat.com>]

On 1/10/21 10:16 PM, Xu Yilun wrote:
> On Sun, Jan 10, 2021 at 12:11:17PM -0800, Moritz Fischer wrote:
>> On Sat, Jan 02, 2021 at 11:13:01AM +0800, Xu Yilun wrote:
>>> This patch supports the DFL drivers be written in userspace. This is
>>> realized by exposing the userspace I/O device interfaces.
>>>
>>> The driver leverages the uio_pdrv_genirq, it adds the uio_pdrv_genirq
>>> platform device with the DFL device's resources, and let the generic UIO
>>> platform device driver provide support to userspace access to kernel
>>> interrupts and memory locations.
>>>
>>> The driver now supports the ether group feature. To support a new DFL
>>> feature been directly accessed via UIO, its feature id should be added to
>>> the driver's id_table.
>>>
>>> Signed-off-by: Xu Yilun <yilun.xu@intel.com>
>>> Reviewed-by: Tom Rix <trix@redhat.com>
>>> ---
>>> v2: switch to the new matching algorithem. It matches DFL devices which
>>>      could not be handled by other DFL drivers.
>>>     refacor the code about device resources filling.
>>>     fix some comments.
>>> v3: split the dfl.c changes out of this patch.
>>>     some minor fixes
>>> v4: drop the idea of a generic matching algorithem, instead we specify
>>>      each matching device in id_table.
>>>     to make clear that only one irq is supported, the irq handling code
>>>      is refactored.
>>> v5: refactor the irq resource code.
>>> ---
>>>  drivers/fpga/Kconfig        | 10 +++++
>>>  drivers/fpga/Makefile       |  1 +
>>>  drivers/fpga/dfl-uio-pdev.c | 91 +++++++++++++++++++++++++++++++++++++++++++++
>>>  3 files changed, 102 insertions(+)
>>>  create mode 100644 drivers/fpga/dfl-uio-pdev.c
>>>
>>> diff --git a/drivers/fpga/Kconfig b/drivers/fpga/Kconfig
>>> index 5ff9438..61445be 100644
>>> --- a/drivers/fpga/Kconfig
>>> +++ b/drivers/fpga/Kconfig
>>> @@ -203,6 +203,16 @@ config FPGA_DFL_NIOS_INTEL_PAC_N3000
>>>  	  the card. It also instantiates the SPI master (spi-altera) for
>>>  	  the card's BMC (Board Management Controller).
>>>  
>>> +config FPGA_DFL_UIO_PDEV
>>> +	tristate "FPGA DFL Driver for Userspace I/O platform devices"
>>> +	depends on FPGA_DFL && UIO_PDRV_GENIRQ
>>> +	help
>>> +	  Enable this to allow some DFL drivers be written in userspace. It
>>> +	  adds the uio_pdrv_genirq platform device with the DFL feature's
>>> +	  resources, and lets the generic UIO platform device driver provide
>>> +	  support for userspace access to kernel interrupts and memory
>>> +	  locations.
>>> +
>>>  config FPGA_DFL_PCI
>>>  	tristate "FPGA DFL PCIe Device Driver"
>>>  	depends on PCI && FPGA_DFL
>>> diff --git a/drivers/fpga/Makefile b/drivers/fpga/Makefile
>>> index 18dc9885..8847fe0 100644
>>> --- a/drivers/fpga/Makefile
>>> +++ b/drivers/fpga/Makefile
>>> @@ -45,6 +45,7 @@ dfl-afu-objs := dfl-afu-main.o dfl-afu-region.o dfl-afu-dma-region.o
>>>  dfl-afu-objs += dfl-afu-error.o
>>>  
>>>  obj-$(CONFIG_FPGA_DFL_NIOS_INTEL_PAC_N3000)	+= dfl-n3000-nios.o
>>> +obj-$(CONFIG_FPGA_DFL_UIO_PDEV)		+= dfl-uio-pdev.o
>>>  
>>>  # Drivers for FPGAs which implement DFL
>>>  obj-$(CONFIG_FPGA_DFL_PCI)		+= dfl-pci.o
>>> diff --git a/drivers/fpga/dfl-uio-pdev.c b/drivers/fpga/dfl-uio-pdev.c
>>> new file mode 100644
>>> index 0000000..a4cd581
>>> --- /dev/null
>>> +++ b/drivers/fpga/dfl-uio-pdev.c
>>> @@ -0,0 +1,91 @@
>>> +// SPDX-License-Identifier: GPL-2.0
>>> +/*
>>> + * DFL driver for Userspace I/O platform devices
>>> + *
>>> + * Copyright (C) 2020 Intel Corporation, Inc.
>>> + */
>>> +#include <linux/dfl.h>
>>> +#include <linux/errno.h>
>>> +#include <linux/kernel.h>
>>> +#include <linux/module.h>
>>> +#include <linux/platform_device.h>
>>> +#include <linux/slab.h>
>>> +#include <linux/uio_driver.h>
>>> +
>>> +#define DRIVER_NAME "dfl-uio-pdev"
>>> +
>>> +static int dfl_uio_pdev_probe(struct dfl_device *ddev)
>>> +{
>>> +	struct platform_device_info pdevinfo = { 0 };
>>> +	struct uio_info uio_pdata = { 0 };
>>> +	struct platform_device *uio_pdev;
>>> +	struct device *dev = &ddev->dev;
>>> +	unsigned int num_res = 1;
>>> +	struct resource res[2];
>>> +
>>> +	res[0].parent = &ddev->mmio_res;
>>> +	res[0].flags = IORESOURCE_MEM;
>>> +	res[0].start = ddev->mmio_res.start;
>>> +	res[0].end = ddev->mmio_res.end;
>>> +
>>> +	if (ddev->num_irqs) {
>>> +		if (ddev->num_irqs > 1)
>>> +			dev_warn(&ddev->dev,
>>> +				 "%d irqs for %s, but UIO only supports the first one\n",
>>> +				 ddev->num_irqs, dev_name(&ddev->dev));
>>> +
>>> +		res[1].flags = IORESOURCE_IRQ;
>>> +		res[1].start = ddev->irqs[0];
>>> +		res[1].end = ddev->irqs[0];
>>> +		num_res++;
>>> +	}
>>> +
>>> +	uio_pdata.name = DRIVER_NAME;
>>> +	uio_pdata.version = "0";
>>> +
>>> +	pdevinfo.name = "uio_pdrv_genirq";
>>> +	pdevinfo.res = res;
>>> +	pdevinfo.num_res = num_res;
>>> +	pdevinfo.parent = &ddev->dev;
>>> +	pdevinfo.id = PLATFORM_DEVID_AUTO;
>>> +	pdevinfo.data = &uio_pdata;
>>> +	pdevinfo.size_data = sizeof(uio_pdata);
>>> +
>>> +	uio_pdev = platform_device_register_full(&pdevinfo);
>>> +	if (!IS_ERR(uio_pdev))
>>> +		dev_set_drvdata(dev, uio_pdev);
>> I'm not sure if this is more readable than:
>>
>>    	uio_pdev = platform_device_register_full(&pdevinfo);
>>    	if (IS_ERR(uio_pdev))
>>    		return PTR_ERR(uio_pdev);
>>
>> 	dev_set_drvdata(dev, uio_pdev);
>> 	return 0;
>>
>> No strong preference, though ... :)
> I think your version is more readable, I'll change it.
>
>>> +
>>> +	return PTR_ERR_OR_ZERO(uio_pdev);
>>> +}
>>> +
>>> +static void dfl_uio_pdev_remove(struct dfl_device *ddev)
>>> +{
>>> +	struct platform_device *uio_pdev = dev_get_drvdata(&ddev->dev);
>>> +
>>> +	platform_device_unregister(uio_pdev);
>>> +}
>>> +
>>> +#define FME_FEATURE_ID_ETH_GROUP	0x10
>>> +
>>> +static const struct dfl_device_id dfl_uio_pdev_ids[] = {
>>> +	{ FME_ID, FME_FEATURE_ID_ETH_GROUP },
>> Will you want to always bind FME_FEATURE_ID_ETH_GROUP? If not I'd suggest not
>> to add it here.
> Actually this is not the most preferable to me. I'm always looking for a
> generic way to bind the uio driver to user assigned dfl devices. But there
> is concern that userspace should not be responsible for the device driver
> matching in previous mail thread:
>
> https://lore.kernel.org/linux-fpga/1602828151-24784-2-git-send-email-yilun.xu@intel.com/
>
> But TBH I still didn't figure out why driver_override is not OK in this
> case.
>
>> If you want to provide an option to somewhat non-ABI fixed bind things
>> you could look at what vfio-pci does (provide a module parameter),
>> otherwise use sysfs 'new_id' or 'bind'.
> I would like to have a "new_id" for dfl bus driver. It is not generic to
> all drivers, I need to add the attr for dfl drivers like pci do.
>
> My concern is how the module param or "new_id" is different from
> "driver_override", seems userspace is also taking part in the device
> matching.
>
> But since we've restarted the discussion, I'm very much willing to have
> a try on the "new_id".

I don't believe there is any problem with the basic platform uio driver.

Can we split it out and work the new_id change in parallel ?

Tom

>
> Thanks,
> Yilun
>