zkkkk12 opened a new pull request, #21485:
URL: https://github.com/apache/nuttx/pull/21485
## Summary
This series adds a PHY abstraction layer (PHYLIB) to NuttX's netdev
lower-half framework, plus two Realtek PHY drivers built on it.
Motivation: Ethernet MAC drivers currently each re-implement their own PHY
handling (link detection, autoneg, register polling). This duplicates code
and makes it hard to reuse a PHY across different MACs. PHYLIB provides a
shared PHY device model and state machine so MAC drivers only wire up the
MDIO bus and let the framework manage the PHY.
The series is split into 5 self-contained commits (each builds on its own):
1. **net/mdio: extend MDIO bus with Clause 45 access and reset level
control**
- Adds `mdio_c45_read()/mdio_c45_write()` for Clause 45 MMD register
access, using native C45 ops when available and falling back to
indirect access over C22 registers 13/14.
- Changes the reset op to level control (assert/deassert by value) with
idempotency tracked in `reset_state`.
2. **net/phy: add PHY abstraction layer (PHYLIB) framework**
- PHY device lifecycle (create/free/scan), connect/disconnect, and a
poll/interrupt-driven state machine.
- Generic C22/C45 PHY drivers as fallbacks, register access helpers, and
upper-half `ethtool`/ioctl integration via `phy_ioctl()`.
- New Kconfig: `CONFIG_NET_PHYLIB` (default n) with poll interval and
max driver/device limits.
3. **net/phy: add Realtek RTL8221B PHY driver** — 2.5Gbps multi-gig PHY,
including firmware ramcode tables and paged register handling.
4. **net/phy: 100BASE-T1 modes and paged register access** — adds
100/1000BASE-T1 link-mode bits, paged register helpers
(`phy_read_paged`/`phy_write_paged`/`phy_modify_paged`) with bus locking,
and lockless MDIO primitives.
5. **net/phy: add RTL9000B 100BASE-T1 PHY driver** — automotive single-pair
Ethernet PHY, with DSP/config verification tables.
### Advantages
As a generic PHY framework (an independent subsystem rather than a
lightweight PHY hook embedded in a single MAC driver), PHYLIB provides:
| Dimension | PHYLIB capability |
|-----------|-------------------|
| Positioning | Independent subsystem (`CONFIG_NET_PHYLIB`), not tied to any
specific MAC driver |
| PHY driver callbacks | 8 callbacks: `config_init` / `config_aneg` /
`read_status` / `config_power` / `config_intr` / `handle_interrupt` /
`read_page` / `write_page` |
| Negotiation / speed resolution | Done in the PHY driver's `read_status`,
MAC-independent |
| Cross-MAC reuse | Write a PHY driver once; any MAC on an MDIO bus can
reuse it without re-implementing negotiation |
| Link state machine | 7 states: `DOWN / READY / UP / RUNNING / NOLINK /
HALTED / ERROR` |
| Lifecycle | Full `connect` / `disconnect` / `start` / `stop` |
| Interrupt modes | Three modes: `PHY_POLL` / `PHY_MAC_INTERRUPT` (MAC owns
the line) / real PHY IRQ (framework attaches its own ISR), with automatic
fallback to polling if IRQ attach fails |
| Link-loss detection latency | Immediate in interrupt mode; bounded by
`CONFIG_NET_PHYLIB_POLL_INTERVAL` (default 1000 ms) in polling mode |
| Paged register access | `phy_read_paged` / `phy_write_paged` /
`phy_modify_paged` hold `mdio_lock` across the whole "select page -> access ->
restore page" sequence, atomic against MII ioctl, with restore failures
propagated |
| PHY identification | Automatic `phy_id & phy_id_mask` matching with
generic C22/C45 fallback; no board-level hardcoded IDs |
| C45 / MMD | Full `phy_read_mmd` etc., native C45 preferred with indirect
access over C22 regs 13/14 as fallback |
| Multi-gig PHYs | Supports 2.5G/5G/10G-class PHYs (RTL8221B 2.5G included) |
| ethtool integration | `ETHTOOL_GLINKSETTINGS` / `SLINKSETTINGS` /
`NWAY_RST` / `GLINK` |
| MII ioctl | Unified `phy_ioctl` in the framework; MACs need not each
implement it |
| PHY reset line | `mdio_ops.reset` callback + idempotent `phy_reset` (level
control, assert/deassert) |
| RGMII delay | Expressed via `phydev->interface` (`RGMII_ID` / `TXID` /
`RXID` etc.) rather than hardcoded per driver |
| Power management | `config_power` (PHY PDOWN bit) |
| Concurrency / SMP | Per-device lock, explicitly handling the
ioctl-vs-state-machine AB-BA deadlock and slot-reuse race |
| Vendor driver examples | RTL8221B (firmware patch / SerDes / C45) and
RTL9000B (100BASE-T1) |
### Limitations
| Dimension | Notes |
|-----------|-------|
| Code size per PHY | Larger than an embedded lightweight hook (~130 lines,
zero deps); requires understanding the subsystem |
| Learning curve | Requires understanding the `phy_driver_s` callback model
and MDIO lower-half integration |
| ethtool coverage | Only the commands needed by current drivers; still a
subset of the full ethtool command set |
| Single-word link-mode mask | Each link-mode mask currently uses one 32-bit
word, limiting the number of representable modes |
| Existing drivers not migrated | In-tree MAC drivers have not yet been
migrated to PHYLIB |
| Static capacity limits | PHY driver/device counts bounded by
`CONFIG_NET_PHYLIB_MAX_DRIVERS` / `MAX_PHYS` (default 8 each) |
| Vendor driver count | Only RTL8221B / RTL9000B so far |
## Impact
- **Users**: No impact by default. All new code is gated behind
`CONFIG_NET_PHYLIB=n` and the per-driver `CONFIG_NET_PHY_REALTEK_*`
options,
all defaulting to `n`. Existing MAC drivers are unaffected.
- **Build**: New sources under `drivers/net/phy/` only compile when
`CONFIG_NET_PHYLIB=y`. Both Make and CMake build systems updated.
- **API**: Adds a new public header section in `include/nuttx/net/phy.h` and
`include/nuttx/net/realtek_phy.h`. Extends `struct netdev_lowerhalf_s` and
`struct ethtool_ops_s`. The MDIO `reset` op signature changed (now takes a
level value) — affects any in-tree MDIO lower-half (updated here).
- **Hardware**: Enables RTL8221B (2.5G) and RTL9000B (100BASE-T1) PHYs.
- **Security / Compatibility**: No security impact. Backward compatible when
the feature is disabled.
## Testing
Verified on hardware that the PHYLIB state machine correctly tracks the
physical link. With the Ethernet cable unplugged the interface stays in the
`UP` (administratively up, no carrier) state; once the cable is plugged in,
the RTL8221B PHY reports link up and the interface transitions to `RUNNING`
(carrier present). TX counters advance after the link comes up.
**Host machine:** Ubuntu 22.04 x86_64
**Target board:** Infineon AURIX TC4 platform
**PHY:** Realtek RTL8221B (`CONFIG_NET_PHY_REALTEK_RTL8221B=y`,
`CONFIG_NET_PHYLIB=y`)
**How tested:** Ran `ifconfig` with the cable unplugged and then plugged in,
and observed the interface state change from `UP` to `RUNNING`, confirming
the PHYLIB link-state machine detects carrier via the RTL8221B driver.
**Logs** (`ifconfig`, cable unplugged → plugged, state `UP` → `RUNNING`):
`eth0.45 Link encap:Ethernet HWaddr 02:04:00:00:04:11 at UP mtu 1500 inet
addr:172.16.45.139 DRaddr:0.0.0.0 Mask:255.255.255.0 RX: Received Fragment
Errors Bytes 00000000 00000000 00000000 0 IPv4 ARP Dropped 00000000 00000000
00000000 TX: Queued Sent Errors Timeouts Bytes 00000002 00000050 00000000
00000000 5c Total Errors: 00000000 IPv4 TCP UDP ICMP Received 0000 0000 0000
0000 Dropped 0000 0000 0000 0000 IPv4 VHL: 0000 Frg: 0000 Checksum 0000 0000
0000 0000 TCP ACK: 0000 SYN: 0000 RST: 0000 0000 HDR: 0000 Type 0000 ---- ----
0000 Sent 0034 0000 0032 0000 Rexmit ---- 0000 ---- ----`
`eth0.45 Link encap:Ethernet HWaddr 02:04:00:00:04:11 at RUNNING mtu 1500
inet addr:172.16.45.139 DRaddr:0.0.0.0 Mask:255.255.255.0 RX: Received Fragment
Errors Bytes 00000000 00000000 00000000 0 IPv4 ARP Dropped 00000000 00000000
00000000 TX: Queued Sent Errors Timeouts Bytes 00000002 00000080 00000000
00000000 5c Total Errors: 00000000 IPv4 TCP UDP ICMP Received 0000 0000 0000
0000 Dropped 0000 0000 0000 0000 IPv4 VHL: 0000 Frg: 0000 Checksum 0000 0000
0000 0000 TCP ACK: 0000 SYN: 0000 RST: 0000 0000 HDR: 0000 Type 0000 ---- ----
0000 Sent 003f 0000 003d 0000 Rexmit ---- ---- ---- ----`
--
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.
To unsubscribe, e-mail: [email protected]
For queries about this service, please contact Infrastructure at:
[email protected]
