8250/16550 UARTs and SoC/PCIe serial port controllers
Linux's main serial port subsystem, covering the classic 8250/16550 UART found on PC serial ports for decades plus countless SoC, PCIe, and I2C/SPI-bridge UART variants in embedded systems, industrial gear, and server BMC consoles. It powers /dev/ttyS* and the serial consoles still used for headless setup and debugging in 2025.
recommendation
It should stay because serial ports remain everywhere: the directory saw roughly 1,400 commits from over 270 authors in the last five years and supports hardware still being sold new, including NXP's SC16IS7xx I2C/SPI UART bridges and Microchip's PCI11101C / PCI1xxxx PCIe UARTs that vendors list as in production with 2025 datasheets. There is no successor subsystem and no plausible replacement; this is core, actively maintained infrastructure.
repository signals
sources
- git.kernel.org
Upstream Kconfig describes this subtree as covering standard 8250/16550 serial support plus many PCI, platform, and bridge-specific variants, indicating it is a broad active serial-driver subsystem rather than a single legacy device driver.
- nxp.com
NXP lists SC16IS740/750/760 UART bridge parts as active and shows a 2025 datasheet, demonstrating hardware covered by this directory is still sold for new designs.
- nxp.com
NXP lists SC16IS752/762 as active dual-UART bridge products, supporting ongoing embedded and industrial deployments for drivers under this subtree.
- microchip.com
Microchip lists PCI11101C as 'In Production'; the directory includes `8250_pci1xxxx.c`, so at least one modern PCIe UART-capable family served here remains current.
- microchip.com
Microchip's 2025 AN5213 states the PCI1xxxx family includes UART peripherals, reinforcing that contemporary PCIe devices still rely on Linux serial support in this subtree.
codex reasoning notes (technical)
Static prompt data already shows very high current upstream activity: 1429 substantive commits in 5y, 274 authors, and most recent touch on 2026-03-12, which is incompatible with deprecation/removal. I attempted lore-first verification via `list_mcp_resources` for `lore-http` and shell `lei`, but the MCP server was unavailable and `lei` was not installed, so no direct lore thread URL could be obtained in this environment. URLs were obtained via `web.search_query` for NXP/Microchip product evidence, plus one canonical-recall kernel.org source-tree URL for subsystem scope. Because this directory covers mainstream 8250/16550 serial, many SoC UARTs, and active bridge/PCIe parts still sold for industrial and embedded use, there is no natural single replacement driver and the correct recommendation is to keep it.