Device Drivers

Overview

RVComp provides two out-of-tree Linux kernel drivers as Buildroot packages. Both drivers are licensed under the GNU General Public License version 2 (GPL-2.0).

Package	Module	Role
`rvcomp-ethernet`	`rvcomp_ethernet`	Linux network interface for the RVComp Ethernet controller
`rvcomp-mmc`	`rvcomp_mmc`	Linux block device backed by sdcram

rvcomp-ethernet

Source: RVComp/buildroot/package/rvcomp-ethernet/src/rvcomp_ethernet.c
Module name: rvcomp_ethernet
Device tree compatible: isct,rvcomp-ethernet
License: GPL-2.0

This driver exposes the RVComp Ethernet controller as a standard Linux network interface (eth0). It maps the CSR region and the RX/TX MMIO buffer regions at probe time.

Receive Flow

PHY → [hardware: MAC filter, FCS check, write to RX buffer, advance ADDR_END]
    → driver detects ADDR_END != ADDR_START
    → allocates skb
    → reads frame bytes from RX buffer into skb
    → passes skb to Linux network stack
    → advances ADDR_START

Transmit Flow

Linux network stack calls ndo_start_xmit
    → driver writes 32-bit frame-length header + payload into TX buffer
    → driver advances TX_BUFFER_END
    → [hardware: prepend preamble/SFD, append FCS, transmit via PHY]

Hardware handles destination-MAC filtering and FCS processing; the driver does not need to calculate or strip the FCS.

rvcomp-mmc

Source: RVComp/buildroot/package/rvcomp-mmc/src/rvcomp_mmc.c
Module name: rvcomp_mmc
Device tree compatible: isct,rvcomp-mmc
License: GPL-2.0

This driver exposes sdcram-backed storage as the block device /dev/mmcblk0. It does not use the Linux MMC host framework; it is a block driver that services requests through the sdcram MMIO window.

The driver reads rootfs-offset from the device tree so that the block device address space starts at the root filesystem region on the card, not at offset 0.

Read Flow

Linux block layer issues read request (512-byte sectors)
    → driver writes target page address (upper 29 bits) to CSR_ADDR29
    → driver reads 32-bit words through the 4 KiB MMIO window
    → copies data into the request buffer
    → completes the block request

Write Flow

Linux block layer issues write request (512-byte sectors)
    → driver writes target page address (upper 29 bits) to CSR_ADDR29
    → driver writes through the 4 KiB MMIO window
    → completes the block request (writes buffered in write-back cache)

Flush is not issued on every write. It is triggered only when the block layer issues an explicit REQ_OP_FLUSH (e.g. on fsync or unmount):

Linux block layer issues flush request (REQ_OP_FLUSH)
    → driver writes 1 to CSR_FLUSH
    → driver polls CSR_FLUSH_DONE until flush completes
    → driver clears FLUSH_DONE latch (CSR_FLUSH_DONE_CLR)
    → completes the flush request

In-Kernel Support

PLIC, CLINT, and the serial console are handled by standard in-kernel drivers configured through the RVComp kernel defconfig. No out-of-tree modules are required for these peripherals.