linux_dsm_epyc7002/Documentation/fpga/fpga-mgr.txt
Jason Gunthorpe baa6d39663 fpga: Add scatterlist based programming
Requiring contiguous kernel memory is not a good idea, this is a limited
resource and allocation can fail under normal work loads.

This introduces a .write_sg op that supporting drivers can provide
to DMA directly from dis-contiguous memory and a new entry point
fpga_mgr_buf_load_sg that users can call to directly provide page
lists.

The full matrix of compatibility is provided, either the linear or sg
interface can be used by the user with a driver supporting either
interface.

A notable change for drivers is that the .write op can now be called
multiple times.

Signed-off-by: Jason Gunthorpe <jgunthorpe@obsidianresearch.com>
Acked-by: Alan Tull <atull@opensource.altera.com>
Acked-by: Moritz Fischer <moritz.fischer@ettus.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
2017-02-10 15:20:44 +01:00

200 lines
6.3 KiB
Plaintext

FPGA Manager Core
Alan Tull 2015
Overview
========
The FPGA manager core exports a set of functions for programming an FPGA with
an image. The API is manufacturer agnostic. All manufacturer specifics are
hidden away in a low level driver which registers a set of ops with the core.
The FPGA image data itself is very manufacturer specific, but for our purposes
it's just binary data. The FPGA manager core won't parse it.
API Functions:
==============
To program the FPGA from a file or from a buffer:
-------------------------------------------------
int fpga_mgr_buf_load(struct fpga_manager *mgr,
struct fpga_image_info *info,
const char *buf, size_t count);
Load the FPGA from an image which exists as a contiguous buffer in
memory. Allocating contiguous kernel memory for the buffer should be avoided,
users are encouraged to use the _sg interface instead of this.
int fpga_mgr_buf_load_sg(struct fpga_manager *mgr,
struct fpga_image_info *info,
struct sg_table *sgt);
Load the FPGA from an image from non-contiguous in memory. Callers can
construct a sg_table using alloc_page backed memory.
int fpga_mgr_firmware_load(struct fpga_manager *mgr,
struct fpga_image_info *info,
const char *image_name);
Load the FPGA from an image which exists as a file. The image file must be on
the firmware search path (see the firmware class documentation). If successful,
the FPGA ends up in operating mode. Return 0 on success or a negative error
code.
A FPGA design contained in a FPGA image file will likely have particulars that
affect how the image is programmed to the FPGA. These are contained in struct
fpga_image_info. Currently the only such particular is a single flag bit
indicating whether the image is for full or partial reconfiguration.
To get/put a reference to a FPGA manager:
-----------------------------------------
struct fpga_manager *of_fpga_mgr_get(struct device_node *node);
struct fpga_manager *fpga_mgr_get(struct device *dev);
Given a DT node or device, get an exclusive reference to a FPGA manager.
void fpga_mgr_put(struct fpga_manager *mgr);
Release the reference.
To register or unregister the low level FPGA-specific driver:
-------------------------------------------------------------
int fpga_mgr_register(struct device *dev, const char *name,
const struct fpga_manager_ops *mops,
void *priv);
void fpga_mgr_unregister(struct device *dev);
Use of these two functions is described below in "How To Support a new FPGA
device."
How to write an image buffer to a supported FPGA
================================================
/* Include to get the API */
#include <linux/fpga/fpga-mgr.h>
/* device node that specifies the FPGA manager to use */
struct device_node *mgr_node = ...
/* FPGA image is in this buffer. count is size of the buffer. */
char *buf = ...
int count = ...
/* struct with information about the FPGA image to program. */
struct fpga_image_info info;
/* flags indicates whether to do full or partial reconfiguration */
info.flags = 0;
int ret;
/* Get exclusive control of FPGA manager */
struct fpga_manager *mgr = of_fpga_mgr_get(mgr_node);
/* Load the buffer to the FPGA */
ret = fpga_mgr_buf_load(mgr, &info, buf, count);
/* Release the FPGA manager */
fpga_mgr_put(mgr);
How to write an image file to a supported FPGA
==============================================
/* Include to get the API */
#include <linux/fpga/fpga-mgr.h>
/* device node that specifies the FPGA manager to use */
struct device_node *mgr_node = ...
/* FPGA image is in this file which is in the firmware search path */
const char *path = "fpga-image-9.rbf"
/* struct with information about the FPGA image to program. */
struct fpga_image_info info;
/* flags indicates whether to do full or partial reconfiguration */
info.flags = 0;
int ret;
/* Get exclusive control of FPGA manager */
struct fpga_manager *mgr = of_fpga_mgr_get(mgr_node);
/* Get the firmware image (path) and load it to the FPGA */
ret = fpga_mgr_firmware_load(mgr, &info, path);
/* Release the FPGA manager */
fpga_mgr_put(mgr);
How to support a new FPGA device
================================
To add another FPGA manager, write a driver that implements a set of ops. The
probe function calls fpga_mgr_register(), such as:
static const struct fpga_manager_ops socfpga_fpga_ops = {
.write_init = socfpga_fpga_ops_configure_init,
.write = socfpga_fpga_ops_configure_write,
.write_complete = socfpga_fpga_ops_configure_complete,
.state = socfpga_fpga_ops_state,
};
static int socfpga_fpga_probe(struct platform_device *pdev)
{
struct device *dev = &pdev->dev;
struct socfpga_fpga_priv *priv;
int ret;
priv = devm_kzalloc(dev, sizeof(*priv), GFP_KERNEL);
if (!priv)
return -ENOMEM;
/* ... do ioremaps, get interrupts, etc. and save
them in priv... */
return fpga_mgr_register(dev, "Altera SOCFPGA FPGA Manager",
&socfpga_fpga_ops, priv);
}
static int socfpga_fpga_remove(struct platform_device *pdev)
{
fpga_mgr_unregister(&pdev->dev);
return 0;
}
The ops will implement whatever device specific register writes are needed to
do the programming sequence for this particular FPGA. These ops return 0 for
success or negative error codes otherwise.
The programming sequence is:
1. .write_init
2. .write or .write_sg (may be called once or multiple times)
3. .write_complete
The .write_init function will prepare the FPGA to receive the image data. The
buffer passed into .write_init will be atmost .initial_header_size bytes long,
if the whole bitstream is not immediately available then the core code will
buffer up at least this much before starting.
The .write function writes a buffer to the FPGA. The buffer may be contain the
whole FPGA image or may be a smaller chunk of an FPGA image. In the latter
case, this function is called multiple times for successive chunks. This interface
is suitable for drivers which use PIO.
The .write_sg version behaves the same as .write except the input is a sg_table
scatter list. This interface is suitable for drivers which use DMA.
The .write_complete function is called after all the image has been written
to put the FPGA into operating mode.
The ops include a .state function which will read the hardware FPGA manager and
return a code of type enum fpga_mgr_states. It doesn't result in a change in
hardware state.