Adding John into Cc. He is working on introducing printk kthreads that might allow to handle consoles asynchronously. On Tue 2022-10-11 11:03:39, Paul Menzel wrote: > Dear Linux folks, > > > We boot our systems generally with the serial console enabled > `console=ttyS0,115200n8`, and noticed that initializing the console takes at > least one second. > > $ dmesg | grep -e 'DMI:' -e 'printk: console' > [ 0.000000] DMI: Dell Inc. PowerEdge R7525/05Y13N, BIOS 2.7.3 > 03/30/2022 > [ 2.691432] printk: console [tty0] enabled > [ 5.414384] printk: console [ttyS0] enabled > > $ dmesg | grep -e 'DMI:' -e 'printk:' > [ 0.000000] DMI: Dell Inc. Precision Tower 3620/0MWYPT, BIOS 2.22.0 > 07/13/2022 > [ 0.146953] printk: console [tty0] enabled > [ 1.374382] printk: console [ttyS0] enabled > > $ dmesg | grep -e 'DMI:' -e 'printk: console' > [ 0.000000] DMI: Dell Inc. PowerEdge R7425/08V001, BIOS 1.6.7 > 10/29/2018 > [ 1.589543] printk: console [tty0] enabled > [ 3.057770] printk: console [ttyS0] enabled > > $ dmesg | grep -e 'DMI:' -e 'printk: console' > [ 0.000000] DMI: HPE ProLiant DL385 Gen10 Plus/ProLiant DL385 Gen10 > Plus, BIOS A42 12/03/2021 > [ 5.171202] printk: console [tty0] enabled > [ 8.066602] printk: console [ttyS0] enabled > > `initcall_debug` shows: > > [ 0.190491] calling univ8250_console_init+0x0/0x2b @ 0 > [ 1.488645] printk: console [ttyS0] enabled > [ 1.492945] initcall univ8250_console_init+0x0/0x2b returned 0 after > 0 usecs > > (Note, that the “after time” by initcall does not match the timestamp at the > beginning.) The message "printk: console [ttyS0] enabled" is printed almost at the end of register_console(). It happens after all previously printed messages reach the console. It might take a long time to show all messages on slow serial consoles. > So it looks to me like, that the already recorded messages are printed out > over the serial console first, before it continues. The documentation of > `register_console()` in `kernel/printk/printk.c` confirms that: > > > > * The console driver calls this routine during kernel initialization > > * to register the console printing procedure with printk() and to > > * print any messages that were printed by the kernel before the > > * console driver was initialized. Most people want to see all messages. Any previous warning or error might the reason why the system crashed later. If you want faster boot, you might consider using "quiet" or loglevel=<some low number> on the kernel command line. It will show only the most important messages on the console. > Is there an option to enable the serial console but send the messages to it > asynchronously so it does not hold up the boot? We have been trying to get this into upstream for years. It is not that easy. Asynchronous handling increases the risk that a console will not show the last important messages when things go wrong. Anyway, we are still trying. One puzzle of this effort is just being reviewed in this thread, https://lore.kernel.org/r/20221107141638.3790965-1-john.ogness@xxxxxxxxxxxxx Best Regards, Petr
