| Message ID | 20200123160031.9853-1-saiprakash.ranjan@codeaurora.org (mailing list archive) |
|---|---|
| State | New, archived |
| Headers | show |
| Series | pstore: Fix printing of duplicate boot messages to console | expand |
Quoting Sai Prakash Ranjan (2020-01-23 08:00:31) > Since commit f92b070f2dc8 ("printk: Do not miss new messages > when replaying the log"), CON_PRINTBUFFER flag causes the > duplicate boot messages to be printed on the console when > PSTORE_CONSOLE and earlycon (boot console) is enabled. > Pstore console registers to boot console when earlycon is > enabled during pstore_register_console as a part of ramoops > initialization in postcore_initcall and the printk core > checks for CON_PRINTBUFFER flag and replays the log buffer > to registered console (in this case pstore console which > just registered to boot console) causing duplicate messages > to be printed. Remove the CON_PRINTBUFFER flag from pstore > console since pstore is not concerned with the printing of > buffer to console but with writing of the buffer to the > backend. > > Console log with earlycon and pstore console enabled: > > [ 0.008342] Console: colour dummy device 80x25 > [ 0.000000] Booting Linux on physical CPU 0x0000000000 [0x51df805e] > ... > [ 1.244049] hw-breakpoint: found 6 breakpoint and 4 watchpoint registers. > [ 0.000000] Booting Linux on physical CPU 0x0000000000 [0x51df805e] > > Fixes: f92b070f2dc8 ("printk: Do not miss new messages when replaying the log") > Reported-by: Douglas Anderson <dianders@chromium.org> > Signed-off-by: Sai Prakash Ranjan <saiprakash.ranjan@codeaurora.org> > --- While I like the idea, it seems that this breaks console-ramoops by removing all the text that is printed in the kernel log before this console is registered. I reboot and see that /sys/fs/pstore/console-ramoops-1 starts like this now: localhost ~ # cat /sys/fs/pstore/console-ramoops-0 [ 0.943472] printk: console [pstore-1] enabled Maybe this console can be "special" and not require anything to be printed out to visible consoles but still get the entire log contents? Or we should just not worry about it.
On 2020-01-24 01:19, Stephen Boyd wrote: > Quoting Sai Prakash Ranjan (2020-01-23 08:00:31) >> Since commit f92b070f2dc8 ("printk: Do not miss new messages >> when replaying the log"), CON_PRINTBUFFER flag causes the >> duplicate boot messages to be printed on the console when >> PSTORE_CONSOLE and earlycon (boot console) is enabled. >> Pstore console registers to boot console when earlycon is >> enabled during pstore_register_console as a part of ramoops >> initialization in postcore_initcall and the printk core >> checks for CON_PRINTBUFFER flag and replays the log buffer >> to registered console (in this case pstore console which >> just registered to boot console) causing duplicate messages >> to be printed. Remove the CON_PRINTBUFFER flag from pstore >> console since pstore is not concerned with the printing of >> buffer to console but with writing of the buffer to the >> backend. >> >> Console log with earlycon and pstore console enabled: >> >> [ 0.008342] Console: colour dummy device 80x25 >> [ 0.000000] Booting Linux on physical CPU 0x0000000000 [0x51df805e] >> ... >> [ 1.244049] hw-breakpoint: found 6 breakpoint and 4 watchpoint >> registers. >> [ 0.000000] Booting Linux on physical CPU 0x0000000000 [0x51df805e] >> >> Fixes: f92b070f2dc8 ("printk: Do not miss new messages when replaying >> the log") >> Reported-by: Douglas Anderson <dianders@chromium.org> >> Signed-off-by: Sai Prakash Ranjan <saiprakash.ranjan@codeaurora.org> >> --- > > While I like the idea, it seems that this breaks console-ramoops by > removing all the text that is printed in the kernel log before this > console is registered. I reboot and see that > /sys/fs/pstore/console-ramoops-1 starts like this now: > > localhost ~ # cat /sys/fs/pstore/console-ramoops-0 > [ 0.943472] printk: console [pstore-1] enabled > > Maybe this console can be "special" and not require anything to be > printed out to visible consoles but still get the entire log contents? > Or we should just not worry about it. Thanks for testing this out Stephen, I should have tested this some more. Lets drop this patch. -Sai
On Thu, Jan 23, 2020 at 09:30:31PM +0530, Sai Prakash Ranjan wrote: > Since commit f92b070f2dc8 ("printk: Do not miss new messages > when replaying the log"), CON_PRINTBUFFER flag causes the > duplicate boot messages to be printed on the console when > PSTORE_CONSOLE and earlycon (boot console) is enabled. > Pstore console registers to boot console when earlycon is > enabled during pstore_register_console as a part of ramoops > initialization in postcore_initcall and the printk core > checks for CON_PRINTBUFFER flag and replays the log buffer > to registered console (in this case pstore console which > just registered to boot console) causing duplicate messages > to be printed. Remove the CON_PRINTBUFFER flag from pstore > console since pstore is not concerned with the printing of > buffer to console but with writing of the buffer to the > backend. I agree this patch isn't the solution, but I'm trying to understand where better logic could be added. Is the issue that printk sees both earlycon and CON_PRINTBUFFER active? Can we add a new CON_* flag that means "not actually printing anything"? (Or maybe a new flag for non-printing to replace CON_PRINTBUFFER that lets pstore still work?) -Kees > > Console log with earlycon and pstore console enabled: > > [ 0.008342] Console: colour dummy device 80x25 > [ 0.000000] Booting Linux on physical CPU 0x0000000000 [0x51df805e] > ... > [ 1.244049] hw-breakpoint: found 6 breakpoint and 4 watchpoint registers. > [ 0.000000] Booting Linux on physical CPU 0x0000000000 [0x51df805e] > > Fixes: f92b070f2dc8 ("printk: Do not miss new messages when replaying the log") > Reported-by: Douglas Anderson <dianders@chromium.org> > Signed-off-by: Sai Prakash Ranjan <saiprakash.ranjan@codeaurora.org> > --- > fs/pstore/platform.c | 2 +- > 1 file changed, 1 insertion(+), 1 deletion(-) > > diff --git a/fs/pstore/platform.c b/fs/pstore/platform.c > index d896457e7c11..271b00db0973 100644 > --- a/fs/pstore/platform.c > +++ b/fs/pstore/platform.c > @@ -505,7 +505,7 @@ static void pstore_console_write(struct console *con, const char *s, unsigned c) > static struct console pstore_console = { > .name = "pstore", > .write = pstore_console_write, > - .flags = CON_PRINTBUFFER | CON_ENABLED | CON_ANYTIME, > + .flags = CON_ENABLED | CON_ANYTIME, > .index = -1, > }; > > -- > QUALCOMM INDIA, on behalf of Qualcomm Innovation Center, Inc. is a member > of Code Aurora Forum, hosted by The Linux Foundation
Quoting Kees Cook (2020-01-27 15:27:33) > On Thu, Jan 23, 2020 at 09:30:31PM +0530, Sai Prakash Ranjan wrote: > > Since commit f92b070f2dc8 ("printk: Do not miss new messages > > when replaying the log"), CON_PRINTBUFFER flag causes the > > duplicate boot messages to be printed on the console when > > PSTORE_CONSOLE and earlycon (boot console) is enabled. > > Pstore console registers to boot console when earlycon is > > enabled during pstore_register_console as a part of ramoops > > initialization in postcore_initcall and the printk core > > checks for CON_PRINTBUFFER flag and replays the log buffer > > to registered console (in this case pstore console which > > just registered to boot console) causing duplicate messages > > to be printed. Remove the CON_PRINTBUFFER flag from pstore > > console since pstore is not concerned with the printing of > > buffer to console but with writing of the buffer to the > > backend. > > I agree this patch isn't the solution, but I'm trying to understand > where better logic could be added. Is the issue that printk sees both > earlycon and CON_PRINTBUFFER active? Can we add a new CON_* flag that > means "not actually printing anything"? (Or maybe a new flag for > non-printing to replace CON_PRINTBUFFER that lets pstore still work?) > This seems to be fixed by commit def97da13651 ("printk: fix exclusive_console replaying"). I think there's nothing to do.
diff --git a/fs/pstore/platform.c b/fs/pstore/platform.c index d896457e7c11..271b00db0973 100644 --- a/fs/pstore/platform.c +++ b/fs/pstore/platform.c @@ -505,7 +505,7 @@ static void pstore_console_write(struct console *con, const char *s, unsigned c) static struct console pstore_console = { .name = "pstore", .write = pstore_console_write, - .flags = CON_PRINTBUFFER | CON_ENABLED | CON_ANYTIME, + .flags = CON_ENABLED | CON_ANYTIME, .index = -1, };
Since commit f92b070f2dc8 ("printk: Do not miss new messages when replaying the log"), CON_PRINTBUFFER flag causes the duplicate boot messages to be printed on the console when PSTORE_CONSOLE and earlycon (boot console) is enabled. Pstore console registers to boot console when earlycon is enabled during pstore_register_console as a part of ramoops initialization in postcore_initcall and the printk core checks for CON_PRINTBUFFER flag and replays the log buffer to registered console (in this case pstore console which just registered to boot console) causing duplicate messages to be printed. Remove the CON_PRINTBUFFER flag from pstore console since pstore is not concerned with the printing of buffer to console but with writing of the buffer to the backend. Console log with earlycon and pstore console enabled: [ 0.008342] Console: colour dummy device 80x25 [ 0.000000] Booting Linux on physical CPU 0x0000000000 [0x51df805e] ... [ 1.244049] hw-breakpoint: found 6 breakpoint and 4 watchpoint registers. [ 0.000000] Booting Linux on physical CPU 0x0000000000 [0x51df805e] Fixes: f92b070f2dc8 ("printk: Do not miss new messages when replaying the log") Reported-by: Douglas Anderson <dianders@chromium.org> Signed-off-by: Sai Prakash Ranjan <saiprakash.ranjan@codeaurora.org> --- fs/pstore/platform.c | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-)