Y2Linux 1.0 Beta 7.11 (Baseline 74)

Status: PUBLISHED as GitHub release baseline-74 (2026-10-17). Release 1.0.0-beta.7.11, Reborn 1.0.0-beta.7.11, kernel 6.18.56-y2linux-1.0.0-beta.7.11, update sequence 6352, Linux 9c3e4f1, Reborn 814cd14. Built with tools/production/release.py in 101 s on top of Baseline 73: container suite 766 OK, all seal checks PASS, no MediaTek firmware in any image. Milestone audit BETA-7.11 and the USB-RX-DMA update in roadmap-gap-audit; design and full results in USB-RX-DMA-MODE1-PLAN.

Image	SHA-256
BOOTIMG	`f90f42c3c72c967042ce41c8e4345ef52f12d1d6b3e845375d64a6d985155ed2`
Y2ROOT	`016ea5681360072d21fd5fc616242a2cd0c2dfa2c911b0dc46af99f6ea366f94`
RECOVERY	`6fdcd8c66c783553583c40f6e39dbd7e05fffd16b4b568df45d1f84382006a1a`
Y2DATA (no SSH key)	`9dc3cba5b60daea0ea7107ca46f08193f19fcb009154f1bb864da991ca906aa4`

Since Beta 7.10

Part Change

Faster copies to the player Multi-packet (mode 1) receive DMA for USB bulk OUT: one DMA program per request (64 KB for MTP) instead of a CPU read of every 512-byte packet. At the transfer's short or zero-length packet the CPU waits until two samples agree (RXCOUNT counts down while the engine unloads a packet), stops the channel and reads only that packet. On by default (rx_dma=2); echo 0 > /sys/devices/platform/11200000.usb/rx_dma returns to PIO, never to mode 0. The stock MediaTek kernel (REA) used mode 1 only for mass storage. (Linux 1f86b4f, b49afaa, 049df5c, 10ec3eb)

Test tool tools/usb/mtpbench.c reads the closing zero-length packet of a download that fills its last read. (64af1b2)

Part	Change
Faster copies to the player	Multi-packet (mode 1) receive DMA for USB bulk OUT: one DMA program per request (64 KB for MTP) instead of a CPU read of every 512-byte packet. At the transfer's short or zero-length packet the CPU waits until two samples agree (RXCOUNT counts down while the engine unloads a packet), stops the channel and reads only that packet. On by default (`rx_dma=2`); `echo 0 > /sys/devices/platform/11200000.usb/rx_dma` returns to PIO, never to mode 0. The stock MediaTek kernel (REA) used mode 1 only for mass storage. (Linux `1f86b4f`, `b49afaa`, `049df5c`, `10ec3eb`)
Test tool	`tools/usb/mtpbench.c` reads the closing zero-length packet of a download that fills its last read. (`64af1b2`)

Device result (owner's player, test builds of this release, every byte SHA-256 verified)

	Beta 7.10 (PIO)	Beta 7.11
Upload, Internal Storage 1 GiB	≈10 MB/s	29.6 MB/s
Upload, SD card 512 MiB	≈9 MB/s	23.3 MB/s
Upload over SSH (USB Ethernet)	6–8 MB/s	10–14 MB/s
Download	37.7 MB/s	37.5–37.9 MB/s
USB interrupts during an upload	≈22 000/s, one core busy	≈1 450/s, CPU ≈80 % idle

Boundary sizes (0 B to 3 MB around every packet, 64 KB and 1 MiB edge) on both storages, 100 random sizes, tools/usb/mtp-file-ops.c 26/26 (switch on and default build), 1.5 GB soak, 300 MB over SSH, an upload aborted by the computer recovers like before, the off switch moves no byte by DMA. Dolphin copies confirmed by the owner. The first test build (sequence 6350) cut one upload after 9.6 MB (a single RXCOUNT sample mid-unload); fixed before 6351.

Known gaps

Not yet run with this release: the cable pulled during a copy, sleep and wake with the faster path on, Windows Explorer. If copying to the player ever misbehaves, the developer switch above restores the Beta 7.10 behaviour.