Overlay USB config instead of sed-patching upstream's config/usb.h

[mastacontrola]

Every EFI binary this repository publishes is built without a USB keyboard driver. On ipxe.efi that leaves the machine with no keyboard at all at the FOG boot menu — reported against Lenovo M70t/M80t Gen3 in forums #18213.

buildipxe.sh patched three settings into upstream's config/usb.h with sed. v2.0.0 restructured that file — a space where the pattern wants a tab, and USB_KEYBOARD/USB_EFI moved into a live PLATFORM_efi block instead of being commented-out lines:

#if defined ( PLATFORM_efi )
  #define USB_EFI
  //#define USB_HCD_USBIO
  #undef USB_KEYBOARD	/* Use built-in EFI keyboard driver */
#endif

All three patterns stopped matching. sed exits 0 when nothing matches, so the build stayed green and every EFI binary silently took upstream's defaults.

Upstream's default is coherent on its own terms: leave USB_KEYBOARD out and re-expose the devices iPXE owns through USB_EFI so the firmware's keyboard driver can re-bind. But ipxe.efi carries the native xHCI/EHCI/UHCI drivers and efi_driver_connect_all() binds them, taking the controller off the firmware first. Where that re-bind doesn't happen, no keyboard driver is left anywhere.

Fixed by overlaying config/local/usb.h, which upstream's config/usb.h includes last, so FOG's values win. This restores the configuration FOG shipped for years rather than inventing one. The point isn't only that it works today — an overlaid file can't half-apply the way a sed pattern can, so the next refresh onto a new upstream tag can only break this loudly. Same pattern as the general.h override in 9b69e5b.

Verification

Built config/config_usb.c both ways and read the objects it drags in:

build	before	after
EFI	obj_efi_usb obj_xhci — no keyboard driver	obj_usbio obj_usbkbd obj_xhci
BIOS	obj_usbkbd obj_xhci	obj_usbkbd obj_xhci (unchanged)

Linked both EFI targets to confirm the blast radius is limited to ipxe.efi:

ipxe.efi     ehci.o uhci.o usbblk.o usbhid.o usbio.o usbkbd.o xhci.o
snponly.efi  (none)

snponly.efi links no USB objects either way and is byte-identical with and without the overlay (267776 bytes both). That also confirms the reported asymmetry: ipxe.efi binds the USB controllers away from the firmware and then has nothing to drive the keyboard with, while snponly.efi never touches USB.

Not yet verified on affected hardware.

Scope

Does not address the other half of #18213, the /dev/ram0 kernel panic on snponly.efi. That one is the installer publishing autoexec.ipxe into the TFTP root and is fixed in fogproject.

Refs FOGProject/fogproject#960.

🤖 Generated with Claude Code