Re: [RFC PATCH v1 13/25] printk: track seq per console

[John Ogness]

On 2019-02-25, Petr Mladek <pmladek@xxxxxxxx> wrote:
>> diff --git a/kernel/printk/printk.c b/kernel/printk/printk.c
>> index ece54c24ea0d..ebd9aac06323 100644
>> --- a/kernel/printk/printk.c
>> +++ b/kernel/printk/printk.c
>> @@ -1504,6 +1514,19 @@ static void call_console_drivers(const char *ext_text, size_t ext_len,
>>  		if (!cpu_online(raw_smp_processor_id()) &&
>>  		    !(con->flags & CON_ANYTIME))
>>  			continue;
>> +		if (con->printk_seq >= seq)
>> +			continue;
>> +
>> +		con->printk_seq++;
>> +		if (con->printk_seq < seq) {
>> +			print_console_dropped(con, seq - con->printk_seq);
>> +			con->printk_seq = seq;
>
> It would be great to print this message only when the real one
> is not superseded.

You mean if there was some function to check if "seq" is the newest
entry. And only in that situation would any dropped information be
presented?

> The suppressed messages can be proceed quickly. They allow consoles
> to catch up with the message producers. But if the spend time
> with printing the warning, we just risk loosing more messages
> again and again.

So instead of:

message A
message B
3 messages dropped
message F
message G
2 messages dropped
message J
message K

you would prefer to see:

message A
message B
message F
message G
message J
message K
5 messages dropped

... with the hope that maybe we are able to print messages H and/or I by
not spending time on the first dropped message?

If there are a lof printk's coming (sysrq-z) then probably there will be
many dropped during output. With your proposal, it wouldn't be seen that
so many intermediate messages are dropped. Only at the end of the output
the user is presented with some huge dropped number.

With my implementation if you see 2 printk lines together you can be
sure that nothing was dropped between those two lines. I think that is
more valuable than having the possibility of maybe losing a few less
messages in a flood situation.

John Ogness