awge/dwc_gmac: enable flow control
Avoids recieve FIFO overflows and dropped packets when connected to a switch
supporting flow control.
Gets us from hundeds of retransmits a second and ~450Mbps to zero retransmits
and ~650Mbps on RockPro64 with iperf3 recieve testing.
(To do: figure out why flow control isn't enabled until explicitly setting
'media auto' with ifconfig.)
Revert rev 1.73 and instead insert a delay after the idle transition
on Mac II-style ADB controllers so that the microcontroller properly
recognizes the transition.
Patch from zigzagjoe.
Add 440/460 peripherals (tested only on 460EX).
Featuring:
- PLB-PCIX host bridge
- PLB-PCIE root complex
- TCP/IP Acceleration Hardware support for EMAC (checksumming and TSO for IPv4 and IPv6)
- Synopsys DWC SATA-II, the old non-AHCI variant (DMA is not stable, use DWCSATA_PIO_ONLY)
- USB EHCI and OHCI
- Updated watchdog
- and necessary plumbing for the above for PPC460EX
Add support for 68060 CPUs to mac68k.
This requires a 68040->68060 adapter plus ROM patches needed to boot
MacOS with the 68060 (https://github.com/ZigZagJoe/Macintosh-Q650-68060).
These NetBSD patches provided by zigzagjoe, with a little whitespace
cleanup and corresponding mods to INSTALL and SMALLRAM kernel configs
by me.
eqos: Avoid unnecessary bouncing on RX
Allocate memory for RX buffers based on our DMA constraints to prevent
unnecessary bouncing. With this change and checksum offload enabled, I
can finally get a full 940 Mbps RX on a 8GB PINE64 Quartz64.
kern/60318: reproducable bpfjit crash
Remove unncessary struct cpu_info members that were #if defined(PMAP_MI)
- ci_tlb_info isn't necessary as PMAP_TLB_MAX is 1 on aarch64
- ci_pmap_{lastuser,cur} are debug fields from the aarch32 pmap.
These struct members changed the layout so that modules would have the
wrong offset for ci_cpufuncs