diff mbox series

fbcon: Disable accelerated scrolling

Message ID 20201029132229.4068359-1-daniel.vetter@ffwll.ch (mailing list archive)
State New, archived
Headers show
Series fbcon: Disable accelerated scrolling | expand

Commit Message

Daniel Vetter Oct. 29, 2020, 1:22 p.m. UTC
So ever since syzbot discovered fbcon, we have solid proof that it's
full of bugs. And often the solution is to just delete code and remove
features, e.g.  50145474f6ef ("fbcon: remove soft scrollback code").

Now the problem is that most modern-ish drivers really only treat
fbcon as an dumb kernel console until userspace takes over, and Oops
printer for some emergencies. Looking at drm drivers and the basic
vesa/efi fbdev drivers shows that only 3 drivers support any kind of
acceleration:

- nouveau, seems to be enabled by default
- omapdrm, when a DMM remapper exists using remapper rewriting for
  y/xpanning
- gma500, but that is getting deleted now for the GTT remapper trick,
  and the accelerated copyarea never set the FBINFO_HWACCEL_COPYAREA
  flag, so unused (and could be deleted already I think).

No other driver supportes accelerated fbcon. And fbcon is the only
user of this accel code (it's not exposed as uapi through ioctls),
which means we could garbage collect fairly enormous amounts of code
if we kill this.

Plus because syzbot only runs on virtual hardware, and none of the
drivers for that have acceleration, we'd remove a huge gap in testing.
And there's no other even remotely comprehensive testing aside from
syzbot.

This patch here just disables the acceleration code by always
redrawing when scrolling. The plan is that once this has been merged
for well over a year in released kernels, we can start to go around
and delete a lot of code.

v2:
- Drop a few more unused local variables, somehow I missed the
compiler warnings (Sam)
- Fix typo in comment (Jiri)
- add a todo entry for the cleanup (Thomas)

v3: Remove more unused variables (0day)

Reviewed-by: Thomas Zimmermann <tzimmermann@suse.de>
Reviewed-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
Acked-by: Sam Ravnborg <sam@ravnborg.org>
Cc: Jiri Slaby <jirislaby@kernel.org>
Cc: Bartlomiej Zolnierkiewicz <b.zolnierkie@samsung.com>
Cc: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
Cc: Linus Torvalds <torvalds@linux-foundation.org>
Cc: Ben Skeggs <bskeggs@redhat.com>
Cc: nouveau@lists.freedesktop.org
Cc: Tomi Valkeinen <tomi.valkeinen@ti.com>
Cc: Daniel Vetter <daniel.vetter@ffwll.ch>
Cc: Jiri Slaby <jirislaby@kernel.org>
Cc: "Gustavo A. R. Silva" <gustavoars@kernel.org>
Cc: Tetsuo Handa <penguin-kernel@I-love.SAKURA.ne.jp>
Cc: Peilin Ye <yepeilin.cs@gmail.com>
Cc: George Kennedy <george.kennedy@oracle.com>
Cc: Nathan Chancellor <natechancellor@gmail.com>
Cc: Peter Rosin <peda@axentia.se>
Signed-off-by: Daniel Vetter <daniel.vetter@intel.com>
---
 Documentation/gpu/todo.rst       | 18 +++++++++++++
 drivers/video/fbdev/core/fbcon.c | 45 ++++++--------------------------
 2 files changed, 26 insertions(+), 37 deletions(-)

Comments

Tomi Valkeinen Oct. 30, 2020, 8:30 a.m. UTC | #1
On 29/10/2020 15:22, Daniel Vetter wrote:
> So ever since syzbot discovered fbcon, we have solid proof that it's
> full of bugs. And often the solution is to just delete code and remove
> features, e.g.  50145474f6ef ("fbcon: remove soft scrollback code").
> 
> Now the problem is that most modern-ish drivers really only treat
> fbcon as an dumb kernel console until userspace takes over, and Oops
> printer for some emergencies. Looking at drm drivers and the basic
> vesa/efi fbdev drivers shows that only 3 drivers support any kind of
> acceleration:
> 
> - nouveau, seems to be enabled by default
> - omapdrm, when a DMM remapper exists using remapper rewriting for
>   y/xpanning
> - gma500, but that is getting deleted now for the GTT remapper trick,
>   and the accelerated copyarea never set the FBINFO_HWACCEL_COPYAREA
>   flag, so unused (and could be deleted already I think).
> 
> No other driver supportes accelerated fbcon. And fbcon is the only
> user of this accel code (it's not exposed as uapi through ioctls),
> which means we could garbage collect fairly enormous amounts of code
> if we kill this.
> 
> Plus because syzbot only runs on virtual hardware, and none of the
> drivers for that have acceleration, we'd remove a huge gap in testing.
> And there's no other even remotely comprehensive testing aside from
> syzbot.
> 
> This patch here just disables the acceleration code by always
> redrawing when scrolling. The plan is that once this has been merged
> for well over a year in released kernels, we can start to go around
> and delete a lot of code.
> 
> v2:
> - Drop a few more unused local variables, somehow I missed the
> compiler warnings (Sam)
> - Fix typo in comment (Jiri)
> - add a todo entry for the cleanup (Thomas)
> 
> v3: Remove more unused variables (0day)
> 
> Reviewed-by: Thomas Zimmermann <tzimmermann@suse.de>
> Reviewed-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
> Acked-by: Sam Ravnborg <sam@ravnborg.org>
> Cc: Jiri Slaby <jirislaby@kernel.org>
> Cc: Bartlomiej Zolnierkiewicz <b.zolnierkie@samsung.com>
> Cc: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
> Cc: Linus Torvalds <torvalds@linux-foundation.org>
> Cc: Ben Skeggs <bskeggs@redhat.com>
> Cc: nouveau@lists.freedesktop.org
> Cc: Tomi Valkeinen <tomi.valkeinen@ti.com>
> Cc: Daniel Vetter <daniel.vetter@ffwll.ch>
> Cc: Jiri Slaby <jirislaby@kernel.org>
> Cc: "Gustavo A. R. Silva" <gustavoars@kernel.org>
> Cc: Tetsuo Handa <penguin-kernel@I-love.SAKURA.ne.jp>
> Cc: Peilin Ye <yepeilin.cs@gmail.com>
> Cc: George Kennedy <george.kennedy@oracle.com>
> Cc: Nathan Chancellor <natechancellor@gmail.com>
> Cc: Peter Rosin <peda@axentia.se>
> Signed-off-by: Daniel Vetter <daniel.vetter@intel.com>
> ---
>  Documentation/gpu/todo.rst       | 18 +++++++++++++
>  drivers/video/fbdev/core/fbcon.c | 45 ++++++--------------------------
>  2 files changed, 26 insertions(+), 37 deletions(-)
> 
> diff --git a/Documentation/gpu/todo.rst b/Documentation/gpu/todo.rst
> index 6b224ef14455..bec99341a904 100644
> --- a/Documentation/gpu/todo.rst
> +++ b/Documentation/gpu/todo.rst
> @@ -277,6 +277,24 @@ Contact: Daniel Vetter, Noralf Tronnes
>  
>  Level: Advanced
>  
> +Garbage collect fbdev scrolling acceleration
> +--------------------------------------------
> +
> +Scroll acceleration is disabled in fbcon by hard-wiring p->scrollmode =
> +SCROLL_REDRAW. There's a ton of code this will allow us to remove:
> +- lots of code in fbcon.c
> +- a bunch of the hooks in fbcon_ops, maybe the remaining hooks could be called
> +  directly instead of the function table (with a switch on p->rotate)
> +- fb_copyarea is unused after this, and can be deleted from all drivers
> +
> +Note that not all acceleration code can be deleted, since clearing and cursor
> +support is still accelerated, which might be good candidates for further
> +deletion projects.

Apparently omapdrm's accelerated panning has been broken for some time, and no one has noticed. It does:

strcmp(fbi->fix.id, MODULE_NAME), which is a comparison of omapdrmdrmfb == omapdrm and always fails.

Fixing that, and applying this patch, things work fine (unaccelerated, of course). I did notice a
single call to omap_fbdev_pan_display() when loading the drivers. This comes from fbcon_switch ->
bit_update_start -> fb_pan_display. Maybe this is from the clearing you mention above?

Reviewed-by: Tomi Valkeinen <tomi.valkeinen@ti.com>

 Tomi
Daniel Vetter Oct. 30, 2020, 8:52 a.m. UTC | #2
On Fri, Oct 30, 2020 at 9:30 AM Tomi Valkeinen <tomi.valkeinen@ti.com> wrote:
>
> On 29/10/2020 15:22, Daniel Vetter wrote:
> > So ever since syzbot discovered fbcon, we have solid proof that it's
> > full of bugs. And often the solution is to just delete code and remove
> > features, e.g.  50145474f6ef ("fbcon: remove soft scrollback code").
> >
> > Now the problem is that most modern-ish drivers really only treat
> > fbcon as an dumb kernel console until userspace takes over, and Oops
> > printer for some emergencies. Looking at drm drivers and the basic
> > vesa/efi fbdev drivers shows that only 3 drivers support any kind of
> > acceleration:
> >
> > - nouveau, seems to be enabled by default
> > - omapdrm, when a DMM remapper exists using remapper rewriting for
> >   y/xpanning
> > - gma500, but that is getting deleted now for the GTT remapper trick,
> >   and the accelerated copyarea never set the FBINFO_HWACCEL_COPYAREA
> >   flag, so unused (and could be deleted already I think).
> >
> > No other driver supportes accelerated fbcon. And fbcon is the only
> > user of this accel code (it's not exposed as uapi through ioctls),
> > which means we could garbage collect fairly enormous amounts of code
> > if we kill this.
> >
> > Plus because syzbot only runs on virtual hardware, and none of the
> > drivers for that have acceleration, we'd remove a huge gap in testing.
> > And there's no other even remotely comprehensive testing aside from
> > syzbot.
> >
> > This patch here just disables the acceleration code by always
> > redrawing when scrolling. The plan is that once this has been merged
> > for well over a year in released kernels, we can start to go around
> > and delete a lot of code.
> >
> > v2:
> > - Drop a few more unused local variables, somehow I missed the
> > compiler warnings (Sam)
> > - Fix typo in comment (Jiri)
> > - add a todo entry for the cleanup (Thomas)
> >
> > v3: Remove more unused variables (0day)
> >
> > Reviewed-by: Thomas Zimmermann <tzimmermann@suse.de>
> > Reviewed-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
> > Acked-by: Sam Ravnborg <sam@ravnborg.org>
> > Cc: Jiri Slaby <jirislaby@kernel.org>
> > Cc: Bartlomiej Zolnierkiewicz <b.zolnierkie@samsung.com>
> > Cc: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
> > Cc: Linus Torvalds <torvalds@linux-foundation.org>
> > Cc: Ben Skeggs <bskeggs@redhat.com>
> > Cc: nouveau@lists.freedesktop.org
> > Cc: Tomi Valkeinen <tomi.valkeinen@ti.com>
> > Cc: Daniel Vetter <daniel.vetter@ffwll.ch>
> > Cc: Jiri Slaby <jirislaby@kernel.org>
> > Cc: "Gustavo A. R. Silva" <gustavoars@kernel.org>
> > Cc: Tetsuo Handa <penguin-kernel@I-love.SAKURA.ne.jp>
> > Cc: Peilin Ye <yepeilin.cs@gmail.com>
> > Cc: George Kennedy <george.kennedy@oracle.com>
> > Cc: Nathan Chancellor <natechancellor@gmail.com>
> > Cc: Peter Rosin <peda@axentia.se>
> > Signed-off-by: Daniel Vetter <daniel.vetter@intel.com>
> > ---
> >  Documentation/gpu/todo.rst       | 18 +++++++++++++
> >  drivers/video/fbdev/core/fbcon.c | 45 ++++++--------------------------
> >  2 files changed, 26 insertions(+), 37 deletions(-)
> >
> > diff --git a/Documentation/gpu/todo.rst b/Documentation/gpu/todo.rst
> > index 6b224ef14455..bec99341a904 100644
> > --- a/Documentation/gpu/todo.rst
> > +++ b/Documentation/gpu/todo.rst
> > @@ -277,6 +277,24 @@ Contact: Daniel Vetter, Noralf Tronnes
> >
> >  Level: Advanced
> >
> > +Garbage collect fbdev scrolling acceleration
> > +--------------------------------------------
> > +
> > +Scroll acceleration is disabled in fbcon by hard-wiring p->scrollmode =
> > +SCROLL_REDRAW. There's a ton of code this will allow us to remove:
> > +- lots of code in fbcon.c
> > +- a bunch of the hooks in fbcon_ops, maybe the remaining hooks could be called
> > +  directly instead of the function table (with a switch on p->rotate)
> > +- fb_copyarea is unused after this, and can be deleted from all drivers
> > +
> > +Note that not all acceleration code can be deleted, since clearing and cursor
> > +support is still accelerated, which might be good candidates for further
> > +deletion projects.
>
> Apparently omapdrm's accelerated panning has been broken for some time, and no one has noticed. It does:
>
> strcmp(fbi->fix.id, MODULE_NAME), which is a comparison of omapdrmdrmfb == omapdrm and always fails.
>
> Fixing that, and applying this patch, things work fine (unaccelerated, of course). I did notice a
> single call to omap_fbdev_pan_display() when loading the drivers. This comes from fbcon_switch ->
> bit_update_start -> fb_pan_display. Maybe this is from the clearing you mention above?

The accel left is through fb_fillrect and fb_imageblt, plus there's
still fb_cursor (but not much use of that even in fbdev drivers).

I'm honestly not sure why there's the pan call in there, maybe just to
reset to default state in case an fbdev chardev user moved the origin
around. Aside from the accel code there's a call to this in
fbcon_switch and fbcon_modechanged, so I think it's just that.
-Daniel
>
> Reviewed-by: Tomi Valkeinen <tomi.valkeinen@ti.com>
>
>  Tomi
>
> --
> Texas Instruments Finland Oy, Porkkalankatu 22, 00180 Helsinki.
> Y-tunnus/Business ID: 0615521-4. Kotipaikka/Domicile: Helsinki
Geert Uytterhoeven Oct. 31, 2020, 10:27 a.m. UTC | #3
Hi Daniel,

CC linux-fbdev

Thanks for your patch!

On Thu, 29 Oct 2020, Daniel Vetter wrote:
> So ever since syzbot discovered fbcon, we have solid proof that it's
> full of bugs. And often the solution is to just delete code and remove
> features, e.g.  50145474f6ef ("fbcon: remove soft scrollback code").
>
> Now the problem is that most modern-ish drivers really only treat
> fbcon as an dumb kernel console until userspace takes over, and Oops
> printer for some emergencies. Looking at drm drivers and the basic
> vesa/efi fbdev drivers shows that only 3 drivers support any kind of
> acceleration:
>
> - nouveau, seems to be enabled by default
> - omapdrm, when a DMM remapper exists using remapper rewriting for
>  y/xpanning
> - gma500, but that is getting deleted now for the GTT remapper trick,
>  and the accelerated copyarea never set the FBINFO_HWACCEL_COPYAREA
>  flag, so unused (and could be deleted already I think).
>
> No other driver supportes accelerated fbcon. And fbcon is the only
> user of this accel code (it's not exposed as uapi through ioctls),
> which means we could garbage collect fairly enormous amounts of code
> if we kill this.

"git grep FBINFO_HWACCEL_COPYAREA" shows me there are 32 more drivers
using acceleration under drivers/video/fbdev/.

> Plus because syzbot only runs on virtual hardware, and none of the
> drivers for that have acceleration, we'd remove a huge gap in testing.
> And there's no other even remotely comprehensive testing aside from
> syzbot.

That sounds like a great argument to remove all hardware drivers from
the kernel ;-)
Seriously, how hard can it be to add "software-accelerated" acceleration
hooks to drivers/video/fbdev/vfb.c, to enable syzbot to exercise the
core acceleration code paths?

> This patch here just disables the acceleration code by always
> redrawing when scrolling. The plan is that once this has been merged
> for well over a year in released kernels, we can start to go around
> and delete a lot of code.

Have you benchmarked the performance impact on traditional fbdev
drivers?

Thanks!

> v2:
> - Drop a few more unused local variables, somehow I missed the
> compiler warnings (Sam)
> - Fix typo in comment (Jiri)
> - add a todo entry for the cleanup (Thomas)
>
> v3: Remove more unused variables (0day)
>
> Reviewed-by: Thomas Zimmermann <tzimmermann@suse.de>
> Reviewed-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
> Acked-by: Sam Ravnborg <sam@ravnborg.org>
> Cc: Jiri Slaby <jirislaby@kernel.org>
> Cc: Bartlomiej Zolnierkiewicz <b.zolnierkie@samsung.com>
> Cc: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
> Cc: Linus Torvalds <torvalds@linux-foundation.org>
> Cc: Ben Skeggs <bskeggs@redhat.com>
> Cc: nouveau@lists.freedesktop.org
> Cc: Tomi Valkeinen <tomi.valkeinen@ti.com>
> Cc: Daniel Vetter <daniel.vetter@ffwll.ch>
> Cc: Jiri Slaby <jirislaby@kernel.org>
> Cc: "Gustavo A. R. Silva" <gustavoars@kernel.org>
> Cc: Tetsuo Handa <penguin-kernel@I-love.SAKURA.ne.jp>
> Cc: Peilin Ye <yepeilin.cs@gmail.com>
> Cc: George Kennedy <george.kennedy@oracle.com>
> Cc: Nathan Chancellor <natechancellor@gmail.com>
> Cc: Peter Rosin <peda@axentia.se>
> Signed-off-by: Daniel Vetter <daniel.vetter@intel.com>
> ---
> Documentation/gpu/todo.rst       | 18 +++++++++++++
> drivers/video/fbdev/core/fbcon.c | 45 ++++++--------------------------
> 2 files changed, 26 insertions(+), 37 deletions(-)
>
> diff --git a/Documentation/gpu/todo.rst b/Documentation/gpu/todo.rst
> index 6b224ef14455..bec99341a904 100644
> --- a/Documentation/gpu/todo.rst
> +++ b/Documentation/gpu/todo.rst
> @@ -277,6 +277,24 @@ Contact: Daniel Vetter, Noralf Tronnes
>
> Level: Advanced
>
> +Garbage collect fbdev scrolling acceleration
> +--------------------------------------------
> +
> +Scroll acceleration is disabled in fbcon by hard-wiring p->scrollmode =
> +SCROLL_REDRAW. There's a ton of code this will allow us to remove:
> +- lots of code in fbcon.c
> +- a bunch of the hooks in fbcon_ops, maybe the remaining hooks could be called
> +  directly instead of the function table (with a switch on p->rotate)
> +- fb_copyarea is unused after this, and can be deleted from all drivers
> +
> +Note that not all acceleration code can be deleted, since clearing and cursor
> +support is still accelerated, which might be good candidates for further
> +deletion projects.
> +
> +Contact: Daniel Vetter
> +
> +Level: Intermediate
> +
> idr_init_base()
> ---------------
>
> diff --git a/drivers/video/fbdev/core/fbcon.c b/drivers/video/fbdev/core/fbcon.c
> index cef437817b0d..8d1ae973041a 100644
> --- a/drivers/video/fbdev/core/fbcon.c
> +++ b/drivers/video/fbdev/core/fbcon.c
> @@ -1033,7 +1033,7 @@ static void fbcon_init(struct vc_data *vc, int init)
> 	struct vc_data *svc = *default_mode;
> 	struct fbcon_display *t, *p = &fb_display[vc->vc_num];
> 	int logo = 1, new_rows, new_cols, rows, cols, charcnt = 256;
> -	int cap, ret;
> +	int ret;
>
> 	if (WARN_ON(info_idx == -1))
> 	    return;
> @@ -1042,7 +1042,6 @@ static void fbcon_init(struct vc_data *vc, int init)
> 		con2fb_map[vc->vc_num] = info_idx;
>
> 	info = registered_fb[con2fb_map[vc->vc_num]];
> -	cap = info->flags;
>
> 	if (logo_shown < 0 && console_loglevel <= CONSOLE_LOGLEVEL_QUIET)
> 		logo_shown = FBCON_LOGO_DONTSHOW;
> @@ -1147,11 +1146,13 @@ static void fbcon_init(struct vc_data *vc, int init)
>
> 	ops->graphics = 0;
>
> -	if ((cap & FBINFO_HWACCEL_COPYAREA) &&
> -	    !(cap & FBINFO_HWACCEL_DISABLED))
> -		p->scrollmode = SCROLL_MOVE;
> -	else /* default to something safe */
> -		p->scrollmode = SCROLL_REDRAW;
> +	/*
> +	 * No more hw acceleration for fbcon.
> +	 *
> +	 * FIXME: Garbage collect all the now dead code after sufficient time
> +	 * has passed.
> +	 */
> +	p->scrollmode = SCROLL_REDRAW;
>
> 	/*
> 	 *  ++guenther: console.c:vc_allocate() relies on initializing
> @@ -1961,45 +1962,15 @@ static void updatescrollmode(struct fbcon_display *p,
> {
> 	struct fbcon_ops *ops = info->fbcon_par;
> 	int fh = vc->vc_font.height;
> -	int cap = info->flags;
> -	u16 t = 0;
> -	int ypan = FBCON_SWAP(ops->rotate, info->fix.ypanstep,
> -				  info->fix.xpanstep);
> -	int ywrap = FBCON_SWAP(ops->rotate, info->fix.ywrapstep, t);
> 	int yres = FBCON_SWAP(ops->rotate, info->var.yres, info->var.xres);
> 	int vyres = FBCON_SWAP(ops->rotate, info->var.yres_virtual,
> 				   info->var.xres_virtual);
> -	int good_pan = (cap & FBINFO_HWACCEL_YPAN) &&
> -		divides(ypan, vc->vc_font.height) && vyres > yres;
> -	int good_wrap = (cap & FBINFO_HWACCEL_YWRAP) &&
> -		divides(ywrap, vc->vc_font.height) &&
> -		divides(vc->vc_font.height, vyres) &&
> -		divides(vc->vc_font.height, yres);
> -	int reading_fast = cap & FBINFO_READS_FAST;
> -	int fast_copyarea = (cap & FBINFO_HWACCEL_COPYAREA) &&
> -		!(cap & FBINFO_HWACCEL_DISABLED);
> -	int fast_imageblit = (cap & FBINFO_HWACCEL_IMAGEBLIT) &&
> -		!(cap & FBINFO_HWACCEL_DISABLED);
>
> 	p->vrows = vyres/fh;
> 	if (yres > (fh * (vc->vc_rows + 1)))
> 		p->vrows -= (yres - (fh * vc->vc_rows)) / fh;
> 	if ((yres % fh) && (vyres % fh < yres % fh))
> 		p->vrows--;
> -
> -	if (good_wrap || good_pan) {
> -		if (reading_fast || fast_copyarea)
> -			p->scrollmode = good_wrap ?
> -				SCROLL_WRAP_MOVE : SCROLL_PAN_MOVE;
> -		else
> -			p->scrollmode = good_wrap ? SCROLL_REDRAW :
> -				SCROLL_PAN_REDRAW;
> -	} else {
> -		if (reading_fast || (fast_copyarea && !fast_imageblit))
> -			p->scrollmode = SCROLL_MOVE;
> -		else
> -			p->scrollmode = SCROLL_REDRAW;
> -	}
> }
>
> #define PITCH(w) (((w) + 7) >> 3)
> -- 
> 2.28.0
>
> _______________________________________________
> dri-devel mailing list
> dri-devel@lists.freedesktop.org
> https://lists.freedesktop.org/mailman/listinfo/dri-devel

Gr{oetje,eeting}s,

 						Geert

--
Geert Uytterhoeven -- There's lots of Linux beyond ia32 -- geert@linux-m68k.org

In personal conversations with technical people, I call myself a hacker. But
when I'm talking to journalists I just say "programmer" or something like that.
 							    -- Linus Torvalds
Daniel Vetter Oct. 31, 2020, 2:17 p.m. UTC | #4
On Sat, Oct 31, 2020 at 11:28 AM Geert Uytterhoeven
<geert@linux-m68k.org> wrote:
>
>         Hi Daniel,
>
> CC linux-fbdev
>
> Thanks for your patch!
>
> On Thu, 29 Oct 2020, Daniel Vetter wrote:
> > So ever since syzbot discovered fbcon, we have solid proof that it's
> > full of bugs. And often the solution is to just delete code and remove
> > features, e.g.  50145474f6ef ("fbcon: remove soft scrollback code").
> >
> > Now the problem is that most modern-ish drivers really only treat
> > fbcon as an dumb kernel console until userspace takes over, and Oops
> > printer for some emergencies. Looking at drm drivers and the basic
> > vesa/efi fbdev drivers shows that only 3 drivers support any kind of
> > acceleration:
> >
> > - nouveau, seems to be enabled by default
> > - omapdrm, when a DMM remapper exists using remapper rewriting for
> >  y/xpanning
> > - gma500, but that is getting deleted now for the GTT remapper trick,
> >  and the accelerated copyarea never set the FBINFO_HWACCEL_COPYAREA
> >  flag, so unused (and could be deleted already I think).
> >
> > No other driver supportes accelerated fbcon. And fbcon is the only
> > user of this accel code (it's not exposed as uapi through ioctls),
> > which means we could garbage collect fairly enormous amounts of code
> > if we kill this.
>
> "git grep FBINFO_HWACCEL_COPYAREA" shows me there are 32 more drivers
> using acceleration under drivers/video/fbdev/.
>
> > Plus because syzbot only runs on virtual hardware, and none of the
> > drivers for that have acceleration, we'd remove a huge gap in testing.
> > And there's no other even remotely comprehensive testing aside from
> > syzbot.
>
> That sounds like a great argument to remove all hardware drivers from
> the kernel ;-)

fbdev is unmaintained, has no one volunteering to put in the work (and
there's huge amounts of work needed), and there's no test suite. No,
fbtest.c doesn't can't, that's not even close. We're not going to
delete everything in the kernel, but slowly sunsetting stuff that's
just costing and not bringing in up is a good idea.

> Seriously, how hard can it be to add "software-accelerated" acceleration
> hooks to drivers/video/fbdev/vfb.c, to enable syzbot to exercise the
> core acceleration code paths?

Just this one is 5 combinations, which means I'd need to convince
syzbot to test 5 different machine setups.

Plus we're still lacking a test suite, and judging from how much time
it took to get something basic going for kms, that's about 2 engineer
years of effort that no one is even close to willing to spend.

> > This patch here just disables the acceleration code by always
> > redrawing when scrolling. The plan is that once this has been merged
> > for well over a year in released kernels, we can start to go around
> > and delete a lot of code.
>
> Have you benchmarked the performance impact on traditional fbdev
> drivers?

There's still some acceleration if you have an image blit engine for
redrawing the screen. But the complexity is contained in the old
drivers that no one cares about.

For anything I have access to the difference is 0.

Also note that for anything remotely modern the fbcon acceleration
framework is pretty badly designed, because it does not allow
sufficient pipelining and queuing of operations. We've had an fbcon
acceleration prototype for i915 10 years ago or so, it's just not
worth the bother.

And again, no one is volunteering to create an fbcon accel framework
that doesn't just suck, despite that this has been discussed in
various places for years. I've done a summary of the sorry state of 2d
acceleration 2 years ago because it's come up so many times:
https://blog.ffwll.ch/2018/08/no-2d-in-drm.html Nothing at all
happened since then.

Reality is that fbdev is just there nowadays for Oops printing and
emergency usage, and it's plenty good enough for that. If there's
anyone who cares beyond that, they're most definitely not able to put
in time for upstream work.
-Daniel

> Thanks!
>
> > v2:
> > - Drop a few more unused local variables, somehow I missed the
> > compiler warnings (Sam)
> > - Fix typo in comment (Jiri)
> > - add a todo entry for the cleanup (Thomas)
> >
> > v3: Remove more unused variables (0day)
> >
> > Reviewed-by: Thomas Zimmermann <tzimmermann@suse.de>
> > Reviewed-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
> > Acked-by: Sam Ravnborg <sam@ravnborg.org>
> > Cc: Jiri Slaby <jirislaby@kernel.org>
> > Cc: Bartlomiej Zolnierkiewicz <b.zolnierkie@samsung.com>
> > Cc: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
> > Cc: Linus Torvalds <torvalds@linux-foundation.org>
> > Cc: Ben Skeggs <bskeggs@redhat.com>
> > Cc: nouveau@lists.freedesktop.org
> > Cc: Tomi Valkeinen <tomi.valkeinen@ti.com>
> > Cc: Daniel Vetter <daniel.vetter@ffwll.ch>
> > Cc: Jiri Slaby <jirislaby@kernel.org>
> > Cc: "Gustavo A. R. Silva" <gustavoars@kernel.org>
> > Cc: Tetsuo Handa <penguin-kernel@I-love.SAKURA.ne.jp>
> > Cc: Peilin Ye <yepeilin.cs@gmail.com>
> > Cc: George Kennedy <george.kennedy@oracle.com>
> > Cc: Nathan Chancellor <natechancellor@gmail.com>
> > Cc: Peter Rosin <peda@axentia.se>
> > Signed-off-by: Daniel Vetter <daniel.vetter@intel.com>
> > ---
> > Documentation/gpu/todo.rst       | 18 +++++++++++++
> > drivers/video/fbdev/core/fbcon.c | 45 ++++++--------------------------
> > 2 files changed, 26 insertions(+), 37 deletions(-)
> >
> > diff --git a/Documentation/gpu/todo.rst b/Documentation/gpu/todo.rst
> > index 6b224ef14455..bec99341a904 100644
> > --- a/Documentation/gpu/todo.rst
> > +++ b/Documentation/gpu/todo.rst
> > @@ -277,6 +277,24 @@ Contact: Daniel Vetter, Noralf Tronnes
> >
> > Level: Advanced
> >
> > +Garbage collect fbdev scrolling acceleration
> > +--------------------------------------------
> > +
> > +Scroll acceleration is disabled in fbcon by hard-wiring p->scrollmode =
> > +SCROLL_REDRAW. There's a ton of code this will allow us to remove:
> > +- lots of code in fbcon.c
> > +- a bunch of the hooks in fbcon_ops, maybe the remaining hooks could be called
> > +  directly instead of the function table (with a switch on p->rotate)
> > +- fb_copyarea is unused after this, and can be deleted from all drivers
> > +
> > +Note that not all acceleration code can be deleted, since clearing and cursor
> > +support is still accelerated, which might be good candidates for further
> > +deletion projects.
> > +
> > +Contact: Daniel Vetter
> > +
> > +Level: Intermediate
> > +
> > idr_init_base()
> > ---------------
> >
> > diff --git a/drivers/video/fbdev/core/fbcon.c b/drivers/video/fbdev/core/fbcon.c
> > index cef437817b0d..8d1ae973041a 100644
> > --- a/drivers/video/fbdev/core/fbcon.c
> > +++ b/drivers/video/fbdev/core/fbcon.c
> > @@ -1033,7 +1033,7 @@ static void fbcon_init(struct vc_data *vc, int init)
> >       struct vc_data *svc = *default_mode;
> >       struct fbcon_display *t, *p = &fb_display[vc->vc_num];
> >       int logo = 1, new_rows, new_cols, rows, cols, charcnt = 256;
> > -     int cap, ret;
> > +     int ret;
> >
> >       if (WARN_ON(info_idx == -1))
> >           return;
> > @@ -1042,7 +1042,6 @@ static void fbcon_init(struct vc_data *vc, int init)
> >               con2fb_map[vc->vc_num] = info_idx;
> >
> >       info = registered_fb[con2fb_map[vc->vc_num]];
> > -     cap = info->flags;
> >
> >       if (logo_shown < 0 && console_loglevel <= CONSOLE_LOGLEVEL_QUIET)
> >               logo_shown = FBCON_LOGO_DONTSHOW;
> > @@ -1147,11 +1146,13 @@ static void fbcon_init(struct vc_data *vc, int init)
> >
> >       ops->graphics = 0;
> >
> > -     if ((cap & FBINFO_HWACCEL_COPYAREA) &&
> > -         !(cap & FBINFO_HWACCEL_DISABLED))
> > -             p->scrollmode = SCROLL_MOVE;
> > -     else /* default to something safe */
> > -             p->scrollmode = SCROLL_REDRAW;
> > +     /*
> > +      * No more hw acceleration for fbcon.
> > +      *
> > +      * FIXME: Garbage collect all the now dead code after sufficient time
> > +      * has passed.
> > +      */
> > +     p->scrollmode = SCROLL_REDRAW;
> >
> >       /*
> >        *  ++guenther: console.c:vc_allocate() relies on initializing
> > @@ -1961,45 +1962,15 @@ static void updatescrollmode(struct fbcon_display *p,
> > {
> >       struct fbcon_ops *ops = info->fbcon_par;
> >       int fh = vc->vc_font.height;
> > -     int cap = info->flags;
> > -     u16 t = 0;
> > -     int ypan = FBCON_SWAP(ops->rotate, info->fix.ypanstep,
> > -                               info->fix.xpanstep);
> > -     int ywrap = FBCON_SWAP(ops->rotate, info->fix.ywrapstep, t);
> >       int yres = FBCON_SWAP(ops->rotate, info->var.yres, info->var.xres);
> >       int vyres = FBCON_SWAP(ops->rotate, info->var.yres_virtual,
> >                                  info->var.xres_virtual);
> > -     int good_pan = (cap & FBINFO_HWACCEL_YPAN) &&
> > -             divides(ypan, vc->vc_font.height) && vyres > yres;
> > -     int good_wrap = (cap & FBINFO_HWACCEL_YWRAP) &&
> > -             divides(ywrap, vc->vc_font.height) &&
> > -             divides(vc->vc_font.height, vyres) &&
> > -             divides(vc->vc_font.height, yres);
> > -     int reading_fast = cap & FBINFO_READS_FAST;
> > -     int fast_copyarea = (cap & FBINFO_HWACCEL_COPYAREA) &&
> > -             !(cap & FBINFO_HWACCEL_DISABLED);
> > -     int fast_imageblit = (cap & FBINFO_HWACCEL_IMAGEBLIT) &&
> > -             !(cap & FBINFO_HWACCEL_DISABLED);
> >
> >       p->vrows = vyres/fh;
> >       if (yres > (fh * (vc->vc_rows + 1)))
> >               p->vrows -= (yres - (fh * vc->vc_rows)) / fh;
> >       if ((yres % fh) && (vyres % fh < yres % fh))
> >               p->vrows--;
> > -
> > -     if (good_wrap || good_pan) {
> > -             if (reading_fast || fast_copyarea)
> > -                     p->scrollmode = good_wrap ?
> > -                             SCROLL_WRAP_MOVE : SCROLL_PAN_MOVE;
> > -             else
> > -                     p->scrollmode = good_wrap ? SCROLL_REDRAW :
> > -                             SCROLL_PAN_REDRAW;
> > -     } else {
> > -             if (reading_fast || (fast_copyarea && !fast_imageblit))
> > -                     p->scrollmode = SCROLL_MOVE;
> > -             else
> > -                     p->scrollmode = SCROLL_REDRAW;
> > -     }
> > }
> >
> > #define PITCH(w) (((w) + 7) >> 3)
> > --
> > 2.28.0
> >
> > _______________________________________________
> > dri-devel mailing list
> > dri-devel@lists.freedesktop.org
> > https://lists.freedesktop.org/mailman/listinfo/dri-devel
>
> Gr{oetje,eeting}s,
>
>                                                 Geert
>
> --
> Geert Uytterhoeven -- There's lots of Linux beyond ia32 -- geert@linux-m68k.org
>
> In personal conversations with technical people, I call myself a hacker. But
> when I'm talking to journalists I just say "programmer" or something like that.
>                                                             -- Linus Torvalds
Geert Uytterhoeven Nov. 18, 2020, 9:21 a.m. UTC | #5
Hi Daniel,

Replying "early" (see below), as this was applied to
drm-misc/for-linux-next.

On Sat, Oct 31, 2020 at 3:17 PM Daniel Vetter <daniel.vetter@ffwll.ch> wrote:
> On Sat, Oct 31, 2020 at 11:28 AM Geert Uytterhoeven
> <geert@linux-m68k.org> wrote:
> > On Thu, 29 Oct 2020, Daniel Vetter wrote:
> > > So ever since syzbot discovered fbcon, we have solid proof that it's
> > > full of bugs. And often the solution is to just delete code and remove
> > > features, e.g.  50145474f6ef ("fbcon: remove soft scrollback code").
> > >
> > > Now the problem is that most modern-ish drivers really only treat
> > > fbcon as an dumb kernel console until userspace takes over, and Oops
> > > printer for some emergencies. Looking at drm drivers and the basic
> > > vesa/efi fbdev drivers shows that only 3 drivers support any kind of
> > > acceleration:
> > >
> > > - nouveau, seems to be enabled by default
> > > - omapdrm, when a DMM remapper exists using remapper rewriting for
> > >  y/xpanning
> > > - gma500, but that is getting deleted now for the GTT remapper trick,
> > >  and the accelerated copyarea never set the FBINFO_HWACCEL_COPYAREA
> > >  flag, so unused (and could be deleted already I think).
> > >
> > > No other driver supportes accelerated fbcon. And fbcon is the only
> > > user of this accel code (it's not exposed as uapi through ioctls),
> > > which means we could garbage collect fairly enormous amounts of code
> > > if we kill this.
> >
> > "git grep FBINFO_HWACCEL_COPYAREA" shows me there are 32 more drivers
> > using acceleration under drivers/video/fbdev/.
> >
> > > Plus because syzbot only runs on virtual hardware, and none of the
> > > drivers for that have acceleration, we'd remove a huge gap in testing.
> > > And there's no other even remotely comprehensive testing aside from
> > > syzbot.
> >
> > That sounds like a great argument to remove all hardware drivers from
> > the kernel ;-)
>
> fbdev is unmaintained, has no one volunteering to put in the work (and
> there's huge amounts of work needed), and there's no test suite. No,
> fbtest.c doesn't can't, that's not even close. We're not going to
> delete everything in the kernel, but slowly sunsetting stuff that's
> just costing and not bringing in up is a good idea.

The fbcon acceleration code is indeed not tested by fbset, and it is
purely in-kernel acceleration for the console.

> > Seriously, how hard can it be to add "software-accelerated" acceleration
> > hooks to drivers/video/fbdev/vfb.c, to enable syzbot to exercise the
> > core acceleration code paths?
>
> Just this one is 5 combinations, which means I'd need to convince
> syzbot to test 5 different machine setups.

Why 5 combinations?
Enable vfb (which can be a module) and be done with it?

> Plus we're still lacking a test suite, and judging from how much time
> it took to get something basic going for kms, that's about 2 engineer
> years of effort that no one is even close to willing to spend.

Sure, writing test suites is hard, and takes time.

> > > This patch here just disables the acceleration code by always
> > > redrawing when scrolling. The plan is that once this has been merged
> > > for well over a year in released kernels, we can start to go around
> > > and delete a lot of code.
> >
> > Have you benchmarked the performance impact on traditional fbdev
> > drivers?
>
> There's still some acceleration if you have an image blit engine for
> redrawing the screen. But the complexity is contained in the old
> drivers that no one cares about.
>
> For anything I have access to the difference is 0.

Sure, you're doing DRM drivers ;-)

> Reality is that fbdev is just there nowadays for Oops printing and
> emergency usage, and it's plenty good enough for that. If there's

That's true for systems that are supported by a DRM driver.

> anyone who cares beyond that, they're most definitely not able to put
> in time for upstream work.

There exist actual products using out-of-tree fbdev drivers that never
got the chance of being merged upstream due to the moratorium on new
fbdev drivers.

BTW, I'm trying to convert an old fbdev driver to DRM, but don't have it
working yet. I had hoped to get something working before replying to
this email, so I could provide more detailed feedback.

Gr{oetje,eeting}s,

                        Geert
diff mbox series

Patch

diff --git a/Documentation/gpu/todo.rst b/Documentation/gpu/todo.rst
index 6b224ef14455..bec99341a904 100644
--- a/Documentation/gpu/todo.rst
+++ b/Documentation/gpu/todo.rst
@@ -277,6 +277,24 @@  Contact: Daniel Vetter, Noralf Tronnes
 
 Level: Advanced
 
+Garbage collect fbdev scrolling acceleration
+--------------------------------------------
+
+Scroll acceleration is disabled in fbcon by hard-wiring p->scrollmode =
+SCROLL_REDRAW. There's a ton of code this will allow us to remove:
+- lots of code in fbcon.c
+- a bunch of the hooks in fbcon_ops, maybe the remaining hooks could be called
+  directly instead of the function table (with a switch on p->rotate)
+- fb_copyarea is unused after this, and can be deleted from all drivers
+
+Note that not all acceleration code can be deleted, since clearing and cursor
+support is still accelerated, which might be good candidates for further
+deletion projects.
+
+Contact: Daniel Vetter
+
+Level: Intermediate
+
 idr_init_base()
 ---------------
 
diff --git a/drivers/video/fbdev/core/fbcon.c b/drivers/video/fbdev/core/fbcon.c
index cef437817b0d..8d1ae973041a 100644
--- a/drivers/video/fbdev/core/fbcon.c
+++ b/drivers/video/fbdev/core/fbcon.c
@@ -1033,7 +1033,7 @@  static void fbcon_init(struct vc_data *vc, int init)
 	struct vc_data *svc = *default_mode;
 	struct fbcon_display *t, *p = &fb_display[vc->vc_num];
 	int logo = 1, new_rows, new_cols, rows, cols, charcnt = 256;
-	int cap, ret;
+	int ret;
 
 	if (WARN_ON(info_idx == -1))
 	    return;
@@ -1042,7 +1042,6 @@  static void fbcon_init(struct vc_data *vc, int init)
 		con2fb_map[vc->vc_num] = info_idx;
 
 	info = registered_fb[con2fb_map[vc->vc_num]];
-	cap = info->flags;
 
 	if (logo_shown < 0 && console_loglevel <= CONSOLE_LOGLEVEL_QUIET)
 		logo_shown = FBCON_LOGO_DONTSHOW;
@@ -1147,11 +1146,13 @@  static void fbcon_init(struct vc_data *vc, int init)
 
 	ops->graphics = 0;
 
-	if ((cap & FBINFO_HWACCEL_COPYAREA) &&
-	    !(cap & FBINFO_HWACCEL_DISABLED))
-		p->scrollmode = SCROLL_MOVE;
-	else /* default to something safe */
-		p->scrollmode = SCROLL_REDRAW;
+	/*
+	 * No more hw acceleration for fbcon.
+	 *
+	 * FIXME: Garbage collect all the now dead code after sufficient time
+	 * has passed.
+	 */
+	p->scrollmode = SCROLL_REDRAW;
 
 	/*
 	 *  ++guenther: console.c:vc_allocate() relies on initializing
@@ -1961,45 +1962,15 @@  static void updatescrollmode(struct fbcon_display *p,
 {
 	struct fbcon_ops *ops = info->fbcon_par;
 	int fh = vc->vc_font.height;
-	int cap = info->flags;
-	u16 t = 0;
-	int ypan = FBCON_SWAP(ops->rotate, info->fix.ypanstep,
-				  info->fix.xpanstep);
-	int ywrap = FBCON_SWAP(ops->rotate, info->fix.ywrapstep, t);
 	int yres = FBCON_SWAP(ops->rotate, info->var.yres, info->var.xres);
 	int vyres = FBCON_SWAP(ops->rotate, info->var.yres_virtual,
 				   info->var.xres_virtual);
-	int good_pan = (cap & FBINFO_HWACCEL_YPAN) &&
-		divides(ypan, vc->vc_font.height) && vyres > yres;
-	int good_wrap = (cap & FBINFO_HWACCEL_YWRAP) &&
-		divides(ywrap, vc->vc_font.height) &&
-		divides(vc->vc_font.height, vyres) &&
-		divides(vc->vc_font.height, yres);
-	int reading_fast = cap & FBINFO_READS_FAST;
-	int fast_copyarea = (cap & FBINFO_HWACCEL_COPYAREA) &&
-		!(cap & FBINFO_HWACCEL_DISABLED);
-	int fast_imageblit = (cap & FBINFO_HWACCEL_IMAGEBLIT) &&
-		!(cap & FBINFO_HWACCEL_DISABLED);
 
 	p->vrows = vyres/fh;
 	if (yres > (fh * (vc->vc_rows + 1)))
 		p->vrows -= (yres - (fh * vc->vc_rows)) / fh;
 	if ((yres % fh) && (vyres % fh < yres % fh))
 		p->vrows--;
-
-	if (good_wrap || good_pan) {
-		if (reading_fast || fast_copyarea)
-			p->scrollmode = good_wrap ?
-				SCROLL_WRAP_MOVE : SCROLL_PAN_MOVE;
-		else
-			p->scrollmode = good_wrap ? SCROLL_REDRAW :
-				SCROLL_PAN_REDRAW;
-	} else {
-		if (reading_fast || (fast_copyarea && !fast_imageblit))
-			p->scrollmode = SCROLL_MOVE;
-		else
-			p->scrollmode = SCROLL_REDRAW;
-	}
 }
 
 #define PITCH(w) (((w) + 7) >> 3)