The branch main has been updated by jrtc27: URL: https://cgit.FreeBSD.org/src/commit/?id=17a4a163df85e2cbe20e4fbdb03c56085e8079e6
commit 17a4a163df85e2cbe20e4fbdb03c56085e8079e6 Author: Jessica Clarke <[email protected]> AuthorDate: 2022-07-23 21:44:19 +0000 Commit: Jessica Clarke <[email protected]> CommitDate: 2022-07-23 21:44:19 +0000 Revert "Fill in cn_name in struct consdev." This reverts commit 82a21151cf1d7a3e9e95b9edbbf74ac10f386d6a. This commit was made to aid debugging before consoles are initialised so that they can more easily be identified from a debugger. However, various consoles (cfecons, ofwcons, mambocons and rcons) all check whether cn_name[0] is non-zero to see they are attached or not, and so this breaks that (perhaps misguided) approach. On RISC-V this results in rcons (the SBI firmware console driver) racing with the real UART driver and so input gets probabilistically lost on the real UART (around 2/3 of the time for me on QEMU). Moreover, the name given to CONSOLE_DRIVER isn't necessarily the same as what eventually gets written to cn_name, such as cfecons vs cfe, rcons vs riscv, ttyj0 vs aju, ttyv0 vs sc, ttyuN vs uart_phyp/opal and all manner of things vs ucom, so in some cases this is in fact misleading as the name will change after attaching. Discussed with: cperciva --- sys/sys/cons.h | 1 - 1 file changed, 1 deletion(-) diff --git a/sys/sys/cons.h b/sys/sys/cons.h index fec7f6c01b84..ad73b258df6d 100644 --- a/sys/sys/cons.h +++ b/sys/sys/cons.h @@ -104,7 +104,6 @@ extern struct tty *constty; /* Temporary virtual console. */ static struct consdev name = { \ .cn_ops = &ops, \ .cn_arg = (arg), \ - .cn_name = #name, \ }; \ DATA_SET(cons_set, name)
