On Tue, 29 Aug 2017 10:12:22 -0700
Linus Torvalds <torva...@linux-foundation.org> wrote:
> On Tue, Aug 29, 2017 at 10:00 AM, Linus Torvalds
> <torva...@linux-foundation.org> wrote:
> >
> > I refuse to help those things. We mis-designed things
>
> Actually, let me rephrase that:
>
> It might actually be a good idea to help those things, by making
> helper functions available that do the marshalling.
>
> So not calling "printk()" directly, but having a set of simple
> "buffer_print()" functions where each user has its own buffer, and
> then the "buffer_print()" functions will help people do nicely output
> data.
>
> So if the issue is that people want to print (for example) hex dumps
> one character at a time, but don't want to have each character show up
> on a line of their own, I think we might well add a few functions to
> help dop that.
>
> But they wouldn't be "printk". They would be the buffering functions
> that then call printk when tyhey have buffered a line.
>
> That avoids the whole nasty issue with printk - printk wants to show
> stuff early (because _maybe_ it's critical) and printk wants to make
> log records with timestamps and loglevels. And printk has serious
> locking issues that are really nasty and fundamental.
>
> A private buffer has none of those issues.
What about using the seq_buf*() then?
struct seq_buf s;
buf = kmalloc(mysize);
seq_buf_init(&s, buf, mysize);
seq_printf(&s,"blah blah %d", bah_blah);
[...]
seq_printf(&s, "my last print\n");
printk("%.*s", s.len, s.buffer);
kfree(buf);
This is what the NMI "safe" printks basically do.
-- Steve
