On 9/30/22 16:20, Petr Mladek wrote:
On Sat 2022-09-24 02:10:47, John Ogness wrote:From: Thomas Gleixner <tglx@xxxxxxxxxxxxx> Replace the open coded single linked list with a hlist so a conversion to SRCU protected list walks can reuse the existing primitives. --- a/arch/parisc/kernel/pdc_cons.c +++ b/arch/parisc/kernel/pdc_cons.c @@ -272,15 +267,17 @@ void pdc_console_restart(bool hpmc) if (pdc_console_initialized) return; - if (!hpmc && console_drivers) + if (!hpmc && !hlist_empty(&console_list)) return; /* If we've already seen the output, don't bother to print it again */ - if (console_drivers != NULL) + if (!hlist_empty(&console_list)) pdc_cons.flags &= ~CON_PRINTBUFFER; - while ((console = console_drivers) != NULL) - unregister_console(console_drivers); + while (!hlist_empty(&console_list)) { + unregister_console(READ_ONCE(hlist_entry(console_list.first, + struct console, node)));The READ_ONCE() is in a wrong place. This is why it did not compile. It should be: unregister_console(hlist_entry(READ_ONCE(console_list.first), struct console, node)); I know that it is all hope for good. But there is also a race between the hlist_empty() and hlist_entry().
I wonder if pdc_console() is still needed as it is today. When this was written, early_console and such didn't worked for parisc as it should. That's proably why we have this register/unregister in here. Would it make sense, and would we gain something for this printk-series, if I'd try to convert pdc_console to a standard earlycon or earlyprintk device? Helge
