mbox series

[v20,0/2] Lattice sysCONFIG SPI FPGA manager

Message ID 20221025053947.2737-1-i.bornyakov@metrotek.ru (mailing list archive)
Headers show
Series Lattice sysCONFIG SPI FPGA manager | expand

Message

Ivan Bornyakov Oct. 25, 2022, 5:39 a.m. UTC
Add support to the FPGA manager for programming Lattice ECP5 FPGA over
slave SPI sysCONFIG interface.

ChangeLog:
  v1 -> v2:
    * remove "spi" from compatible string
    * reword description in dt-bindings doc
    * add reference to spi-peripheral-props.yaml in dt-binding doc
    * fix DTS example in dt-bindings doc: 4-spaces indentations, no
      undersores in node names.
  v2 -> v3:
    * fix typo "##size-cells" -> "#size-cells" in dt-bindings example
  v3 -> v4:
    * dt-bindings: reword description
    * dt-bindings: revert props order
  v4 -> v5:
    * dt-bindings: remove trailing dot from title
    * dt-bindings: reword description to avoid driver reference
    * dt-bindings: add "Reviewed-by: Krzysztof Kozlowski" tag
  v5 -> v6:
    * ecp5-spi: lock SPI bus for exclusive usage in
      ecp5_ops_write_init(), release in ecp5_ops_write_complete()
      or on error
  v6 -> v7:
    * ecp5-spi.c -> lattice-sysconfig-spi.c. Reworked to represent
      generalized sysCONFIG port with implementations for ECP5 and
      MachXO2
    * lattice,ecp5-fpga-mgr.yaml -> lattice,sysconfig.yaml. Reworked to
      document both ECP5 and MachXO2 sysCONFIG.
    * dt-bindings: remove "Reviewed-by: Krzysztof Kozlowski" tag as doc
      was rewritten by a considerable amount.
  v7 -> v8:
    * dt-bindings: move "program-gpios", "init-gpios" and "done-gpios"
      to top-level properties and disallow them for MachXO2 variant.
  v8 -> v9:
    * dt-bindings: "program-gpios", "init-gpios" and "done-gpios" are
      now optional for both ECP5 and MachXO2
    * lattice-sysconfig-spi.c -> sysconfig-spi.c + sysconfig.c +
      sysconfig.h
        ** reworked to be one sysCONFIG FPGA Manager rather than two
	   distinct ECP5 and MachXO2 managers
	** splitted to port type agnostic sysconfig.c and SPI-specific
	   sysconfig-spi.c
	** command transfer function moved to callback for ease of
	   adding another port type, such as I2C
  v9 -> v10:
    * split sysconfig_transfer() callback into separate command_write()
      and command_write_then_read(). There are too many transfers
      without readback.
    * add command_write_with_data() callback which performs single
      transfer of command + data. It's needed for better abstraction of
      paged bitstream write routine.
    * move sysconfig_lsc_burst_init() to bitstream_burst_write_init()
      callback to break dependence of sysconfig.c from sysconfig-spi.c
    * move sysconfig_lsc_burst_complete() to bitstream_burst_write_complete()
      callback to break dependence of sysconfig.c from sysconfig-spi.c
    * add bitstream_burst_write() to abstract fpga_manager_ops->write()
      from bus type
    * remove struct spi_device from struct sysconfig_priv, use
      to_spi_device()
    * move fpga_manager_ops initialization to sysconfig.c
  v10 -> v11:
    * rename sysconfig_lsc_burst_init() to sysconfig_spi_lsc_burst_init()
    * rename sysconfig_bitstream_burst_write() to
      sysconfig_spi_bitstream_burst_write()
    * rename sysconfig_lsc_burst_complete() to
      sysconfig_spi_lsc_burst_complete()
    * rename "ecp5-fpga-mgr" to "sysconfig-ecp5"
    * rename "machxo2-fpga-mgr" to "sysconfig-machxo2"
    * move spi_max_speed_hz from struct sysconfig_fpga_priv to
      struct sysconfig_spi_fpga_priv, which is local to sysconfig-spi.c
    * remove SPI bus unlock on write error form
      sysconfig_spi_bitstream_burst_write(), call
      sysconfig_burst_write_complete() on error in
      sysconfig_bitstream_burst_write() instead.
  v11 -> v12:
    * build sysconfig core as separate module to prevent duplication of
      common code segments across different binaries
    * rename sysconfig.c to lattice-sysconfig.c
    * rename sysconfig.h to lattice-sysconfig.h
    * rename sysconfig-spi.c to lattice-sysconfig-spi.c
    * rename sysconfig_spi_cmd_write_then_read() to
      sysconfig_spi_cmd_read()
    * rename command_write_then_read() callback to command_read()
    * rename sysconfig_cmd_write_then_read() to sysconfig_cmd_read()
    * rename sysconfig_spi_lsc_burst_init() to
      sysconfig_spi_bitstream_burst_init()
    * rename sysconfig_spi_lsc_burst_complete() to
      sysconfig_spi_bitstream_burst_complete()
    * remove excessive !spi check from sysconfig_spi_cmd_write(),
      sysconfig_spi_cmd_read(), sysconfig_spi_bitstream_burst_init(),
      sysconfig_spi_bitstream_burst_write() and
      sysconfig_spi_bitstream_burst_complete()
    * drop MachXO2 support
        ** drop struct sysconfig_fpga_priv
        ** drop paged write
        ** drop command_write_with_data() and friends
        ** drop ISC_PROGRAM_DONE routine
        ** drop refresh from sysconfig_isc_finish()
        ** sysconfig_isc_erase() only erase SRAM
	** drop MachXO2 mentions from DT bindings doc
  v12 -> v13:
    * use device_get_match_data() instead of of_device_get_match_data()
      and drop of_device.h inclusion
    * in polling routines use time_before(jiffies, timeout) instead of
      retries count
    * add usleep_range() to gpio polling routine
    * check fail status of command in more pronounced way
    * check definition of sysconfig_priv callbacks at probe
    * (BIT(23) | BIT(24) | BIT(25)) -> GENMASK(25, 23)
  v13 -> v14:
    * return -ETIMEDOUT instead of -EBUSY from sysconfig_poll_busy() to
      align with sysconfig_poll_gpio()
  v14 -> v15:
    * move sysCONFIG commands from stack to heap, as spi-summary doc
      says: "I/O buffers use the usual Linux rules, and must be
      DMA-safe. [...] Don't use the stack, or anything that's declared
      "static"."
        ** add u8 cmd_tx_buf[4] and u8 cmd_rx_buf[4] to
	   struct sysconfig_priv
	** copy commands to cmd_tx_buf before sending to the device
	** use cmd_rx_buf for commands readback
	** change command_write() and command_read() signatures, as
	   command buffers are now embedded to sysconfig_priv
	** combine sysconfig_read_busy() with sysconfig_poll_busy() to
	   avoid same memcpy in the loop
  v15 -> v16:
    * revert to v14
    * combine command_write() and command_read() to command_transfer()
      which uses spi_write_then_read() underhood. spi_write_then_read()
      is dma-safe for on-stack buffers.
    * in sysconfig_spi_bitstream_burst_init() bounce on-stack buffer
      with LSC_BITSTREAM_BURST command to the heap. Now everything
      should be dma-safe.
  v16 -> v17:
    * return dev_err_probe() from sysconfig_probe() if
      devm_gpiod_get_optional() fails
  v17 -> v18:
    * use read_poll_timeout() in sysconfig_poll_busy() and
      sysconfig_poll_gpio()
    * combine checks for callbacks presence in sysconfig_probe() into
      one if-statement, add missing check for command_transfer()
  v18 -> v19:
    * sysconfig_poll_gpio() improvements:
        ** replace '(val && is_active) || (!val && !is_active)' with
	   simpler expression '!!val == is_active'
	** unfold ternary return
  v19 -> v20:
    * replace kmalloc() + memcpy() with kmemdup()

Ivan Bornyakov (2):
  fpga: lattice-sysconfig-spi: add Lattice sysCONFIG FPGA manager
  dt-bindings: fpga: document Lattice sysCONFIG FPGA manager

 .../bindings/fpga/lattice,sysconfig.yaml      |  81 ++++
 drivers/fpga/Kconfig                          |  11 +
 drivers/fpga/Makefile                         |   2 +
 drivers/fpga/lattice-sysconfig-spi.c          | 153 +++++++
 drivers/fpga/lattice-sysconfig.c              | 397 ++++++++++++++++++
 drivers/fpga/lattice-sysconfig.h              |  39 ++
 6 files changed, 683 insertions(+)
 create mode 100644 Documentation/devicetree/bindings/fpga/lattice,sysconfig.yaml
 create mode 100644 drivers/fpga/lattice-sysconfig-spi.c
 create mode 100644 drivers/fpga/lattice-sysconfig.c
 create mode 100644 drivers/fpga/lattice-sysconfig.h