Re: [PATCH v3 0/5] console: Miscellaneous clean-ups, do not use FNTCHARCNT() in fbcon.c

From: Peilin Ye <yepeilin.cs@gmail.com>
To: Daniel Vetter <daniel@ffwll.ch>
Cc: Greg Kroah-Hartman <gregkh@linuxfoundation.org>,
	Daniel Vetter <daniel@ffwll.ch>,
	Jiri Slaby <jirislaby@kernel.org>,
	linux-kernel@vger.kernel.org, dri-devel@lists.freedesktop.org,
	linux-fbdev@vger.kernel.org
Subject: Re: [PATCH v3 0/5] console: Miscellaneous clean-ups, do not use FNTCHARCNT() in fbcon.c
Date: Tue, 17 Nov 2020 01:21:42 -0500	[thread overview]
Message-ID: <20201117062142.GA28766@PWN> (raw)
In-Reply-To: <20201116100949.GA401619@phenom.ffwll.local>

On Mon, Nov 16, 2020 at 11:09:49AM +0100, Daniel Vetter wrote:
> On Sat, Nov 14, 2020 at 07:47:16AM -0500, Peilin Ye wrote:
> > On Sat, Nov 14, 2020 at 01:22:22PM +0100, Greg Kroah-Hartman wrote:
> > > On Sat, Nov 14, 2020 at 01:18:06PM +0100, Greg Kroah-Hartman wrote:
> > > > On Sat, Nov 14, 2020 at 03:10:21AM -0500, Peilin Ye wrote:
> > > > > Thanks for reviewing!  Questions about the last patch [5/5] though, it
> > > > > depends on the following assumption:
> > > > > 
> > > > > """
> > > > > For each console `idx`, `vc_cons[idx].d->vc_font.data` and
> > > > > `fb_display[idx].fontdata` always point to the same buffer.
> > > > > """
> > > > > 
> > > > > Is this true?  I think it is true by grepping for `fontdata`.  I also
> > > > > noticed that fbcon.c is using `vc->vc_font.data` and `p->fontdata`
> > > > > interchangeably, see fbcon_get_requirement():
> > > > > 
> > > > > 		vc = vc_cons[fg_console].d;
> > > > > 		[...]
> > > > > 			p = &fb_display[fg_console];
> > > > > 			caps->x = 1 << (vc->vc_font.width - 1);
> > > > > 					^^^^^^^^^^^
> > > > > 			caps->y = 1 << (vc->vc_font.height - 1);
> > > > > 					^^^^^^^^^^^
> > > > > 			caps->len = (p->userfont) ?
> > > > > 				FNTCHARCNT(p->fontdata) : 256;
> > > > > 					   ^^^^^^^^^^^
> > > > > 
> > > > > If it is true, then what is the point of using `fontdata` in `struct
> > > > > fbcon_display`?  Just for the `userfont` flag?  Should we delete
> > > > > `fontdata`, when we no longer need the `userfont` flag?
> > > > 
> > > > Yes, after a quick look, I think your analysis here is correct.  So if
> > > > you can delete that, it would be nice if possible.
> > 
> > I see, at the moment we still need `userfont` for refcount handling, I
> > will try to delete both `fontdata` and `userfont` after refcount is
> > taken care of.
> 
> +1 on sunsetting fondata. I think there's a bunch of these in fbcon code,
> because the console subsystem is older than the standard "allow drivers to
> embed the subsystem struct into their own private struct" subclassing
> model. So lots of global arrays indexed by the console id :-/

Yeah, I saw your comments about registered_fb[] :(

> We're also trying to start some kind of test suite for fbdev chardev ioctl
> interface in the gpu test suite. fbcon tests are maybe more related to
> tty/vt, and I have no idea whether anything exists there already.
> 
> But if you want to do some automated testcases for fbcon and there's
> absolutely no other home for them, the gpu test suite might be an option
> of last resort too:
> 
> https://dri.freedesktop.org/docs/drm/gpu/drm-uapi.html#testing-and-validation

Oh, I didn't know about igt-gpu-tools, thanks for the pointer!  Now,
since charcount is taken care of, and font_desc now contains all fields
of console_font, I think it is a good time to replace console_font with
font_desc in vc_data.  Then we can get rid of FNTSUM(), FNTSIZE(), then
(finally) REFCOUNT().

I will start working on vc_data, after done enough testing on [5/5],
ofc. Thanks,

Peilin Ye