User Details
- User Since
- Aug 24 2025, 12:57 PM (64 w, 2 d)
Today
Reword the five data bit note in the header comment
LGTM, but I'll wait for others to look it over as well.
Thanks. I tested this on amd64 main (7cb4d4dfdd18) with a FreeBSD main guest and two e1000 devices whose drivers were detached. The reproducer from the test plan reads ffffffff without the change and the moved BAR's registers with it. Without the change, one more write after that sequence, pciconf -w pci0:0:9:0 0x10 0xc0040000 to move slot 9's BAR back, also aborts bhyve on assert(error == 0) in modify_bar_registration(), because that BAR was never really registered at 0xc0020000. This change fixes that too, which may be worth saying in the commit message. FreeBSD and Alpine Linux guests with virtio, e1000, ahci, xhci and fbuf devices boot as before.
Yesterday
Fold the frame helpers, require 8 samples per bit as the manual says, fix uc_range units, small cleanups
Use the ns8250 baud rate generator hooks (D61611); the driver now only provides the generator
Pass the softc to the helpers and use UART_BAS_IN_PROBE instead of passing get_divisor
Restore the original order in ns8250_param()
Restore the original order in ns8250_param()
set_baudrate() no longer takes the LCR value: ns8250_param() writes LCR first and restores it if the rate is rejected
set_baudrate() no longer takes the LCR value: ns8250_param() writes LCR first and restores it if the rate is rejected
Sun, Nov 15
Add a MEDIATEK kernel configuration, as the other SoC families have.
List the SoC options in NOTES so that LINT builds the drivers.
Muxes carry MTK_MUX_SETCLR: the status registers are read-only (checked on the board). Reset id list dropped.
Muxes change by read-modify-write plus the update register. top_net2pll_d3_d2 divides by 6 (measured with the SoC's frequency meter).
Muxes change by read-modify-write unless the register is read-only. PLL set_freq sets the rate. Critical clocks use clknode_enable(). Reset ids are bounds-checked instead of listed.
Clocks moved to UART_ATTACH, softc via class size, as discussed
Sat, Nov 14
Fix fiptool version (FIPTOOL_CPPFLAGS, BUILD_STRING); drop -Werror
Drop CONFIG_DEFAULT_FDT_FILE from the fragment; it has no effect on MT7986
Fri, Nov 13
Fix RAX preservation in the kernel scalar IN handler too.
Address review
Sun, Nov 8
Fri, Nov 6
Thu, Nov 5
Add vm_can_snapshot_devices() and pci_can_snapshot() as suggested
Add sponsorship credit