Let a caller defer the PHY link wait and resolve the link later

[dgarske]

Description

amd_eth_init() previously blocked for up to 10 seconds waiting for copper link establishment (GEM_PHY_ANEG_TIMEOUT_MS + GEM_PHY_LINK_TIMEOUT_MS), preventing callers from running initialization in parallel.

Key Changes

• Non-blocking Negotiation: Setting both timeout budgets to 0 now triggers auto-negotiation and returns immediately. Callers poll via gem_link_resolve() to handle MAC programming and speed/duplex resolution asynchronously.
• Guarded Speed Resolution: Added a BMSR_ANCOMPLETE check before resolving link parameters. Prevents early reads from transient PHY registers (e.g., TI DP83867 PHYSTS defaulting to 10M/HD) from corrupting MAC and reference clock settings.
• Direct MAC Configuration: Updated gem_link_resolve() to overwrite the entire MAC speed/duplex register field rather than bitwise clearing, ensuring idempotency across multiple resolutions.

Hardware / Test Status

• Hardware: Tested on ZCU102 (GEM3/RGMII, DP83867 PHY) with timeouts set to 0.
• Behavior: amd_eth_init() returned immediately while negotiation ran in background. Link resolved ~3.1s later at 1000 Mbps Full Duplex with successful ICMP ping responses.
• Build Targets: Clean builds verified for zcu102, zynq7000, and versal.

[Copilot]

Copilot encountered an error and was unable to review this pull request. You can try again by re-requesting a review.