Enable cursor for the frame buffer console

[frno7]

Do

• display the cursor by XOR;
• have the cursor blink, if /sys/class/graphics/fbcon/cursor_blink is nonzero;
• figure out whether the cursor can attain shapes other than FB_TILE_CURSOR_UNDERLINE by some console command or other reasonable means;
• figure out whether foreground and background cursor colours are supposed to do anything.

For details, see #10 (comment).

[frno7]

We’ve now got a XOR cursor in the console. @rickgaiser, setting up the ALPHA register was simple:

        GIF_PACKAGE_AD(package) {
                .addr = gs_addr_alpha_1,
                .data.alpha_1 = {
                        .a = gs_alpha_a_cs,
                        .b = gs_alpha_b_cd,
                        .c = gs_alpha_c_fix,
                        .d = gs_alpha_d_0,
                        .fix = GS_ALPHA_ONE
                }
        };

Then it was only a matter of sending four GS packages for a sprite with the abe field set to gs_blendning_on:

        GIF_PACKAGE_TAG(package) {
                .flg = gif_reglist_mode,
                .reg0 = gif_reg_prim,
                .reg1 = gif_reg_rgbaq,
                .reg2 = gif_reg_xyz2,
                .reg3 = gif_reg_xyz2,
                .nreg = 4,
                .nloop = 1,
                .eop = 1
        };
        GIF_PACKAGE_REG(package) {
                .lo.prim = {
                        .abe = gs_blendning_on,
                        .prim = gs_sprite
                },
                .hi.rgbaq = c->fg_rgbaq
        };
        GIF_PACKAGE_REG(package) {
                .lo.xyz2 = {
                        .x = gs_fbcs_to_pcs(c->sx),
                        .y = gs_fbcs_to_pcs(c->sy + c->th - c->dy)
                },
                .hi.xyz2 = {
                        .x = gs_fbcs_to_pcs(c->sx + c->tw),
                        .y = gs_fbcs_to_pcs(c->sy + c->th)
                }
        };

The cursor shape FB_TILE_CURSOR_UNDERLINE seems hardwired by the kernel, but all shapes are available in ps2fb.c:

	const u32 ch =
		cursor->shape == FB_TILE_CURSOR_NONE        ? 0 :
		cursor->shape == FB_TILE_CURSOR_UNDERLINE   ? 1 :
		cursor->shape == FB_TILE_CURSOR_LOWER_THIRD ? th / 3 :
		cursor->shape == FB_TILE_CURSOR_LOWER_HALF  ? th / 2 :
		cursor->shape == FB_TILE_CURSOR_TWO_THIRDS  ? (2 * th) / 3 :
		cursor->shape == FB_TILE_CURSOR_BLOCK       ? th : th;

Also, the cursor isn’t blinking but it remains unclear by which logic it would. Anyhow, a reasonable cursor is displayed now. :-)

[frno7]

@rickgaiser, in addition, I think the alpha blending function, with similarities to a blitter, will be highly useful for a fully GS-accelerated X window system, eventually. :-)

[Arch91]

@rickgaiser, thank you very much for that! @frno7, way to go!
@frno7, how about also to fit the GNU linux cursor tune standard? As I previously noted in the issue #10, there (in GNU standardized linux) is possible to change the background color of the cursor, the speed of it's flashing and the times after those flashing stops - on the way, by directing the value into some parameter which is/are mounted in /sys/ folder. As an example what I mean, for the first one, the cursor color, it can be changed using this command:
echo N > /sys/module/vt/parameters/cur_default
where N is the integer value from 0 to 15. And each this value corresponds to the appropriate color. Those colors should be the same as in GNU standard.
And same parameter tunes availability for the speed of flashing/blinking and for the 'times after those flashings/blinkings stops'.

[frno7]

@Arch91, I figured out how to manipulate the shape of the cursor. Amusingly, it works already: try echo -e "\e[?6c", with 6 meaning a CUR_BLOCK shape as per

linux/include/linux/console_struct.h

Lines 161 to 167 in 59a11ab

	#define CUR_DEF 0
	#define CUR_NONE 1
	#define CUR_UNDERLINE 2
	#define CUR_LOWER_THIRD 3
	#define CUR_LOWER_HALF 4
	#define CUR_TWO_THIRDS 5
	#define CUR_BLOCK 6

[frno7]

Haha, try echo -e '\033[?17;0;64c' for a big red block of a cursor! :-) See some details in Documentation/admin-guide/vga-softcursor.rst.

[frno7]

@Arch91, a cursor blinking like a Christmas tree could be a real crowd-pleaser. We should make a test suite to cover the combinations of settings. It seems most if not all settings can be applied using the echo command with various escape sequences.

[Arch91]

We should make a test suite to cover the combinations of settings. It seems most if not all settings can be applied using the echo command with various escape sequences.

Yes, and there should be a tutotial which describes the all available values for the all parameters, with an examples (4 high bits - background, 4 low bits - foreground, and for the both - the first cipher turns blinking/highliting, and the last three ciphers sets the color code (what are high/low bits, where they are in the examples, where to get the color code combos appropriate the colors)). Though, I see there is not described what need to be done to certainly achieve the blinking cursor, to set the times it will blink and after that times and moments of blinkings passed it will not blinks until cursor moving, and the speed between the moments of those blinkings.
From the other side... it might be that blinking times/moments related things I described are available in Xes, not in the main terminal/console...

[frno7]

The article Cursor Appearance in the Linux Console is apparently an interpretation of Documentation/admin-guide/vga-softcursor.rst. The Wikipedia article on VGA text mode also says some things about the cursor, for example that

the VGA standard does not provide a way to alter the blink rate, although common workarounds involve hiding the cursor and using a normal character glyph to provide a so-called software cursor.

I could imagine that the kernel console would emulate blinking in software, either by periodically alternating the shape of the cursor with FB_TILE_CURSOR_NONE (to have it temporarily vanish) as in

linux/drivers/video/fbdev/ps2fb.c

Line 1339 in 7455188

cursor->shape == FB_TILE_CURSOR_NONE ? 0 :

or by setting mode to 0 meaning erase as in

linux/include/linux/fb.h

Line 349 in 7455188

__u32 mode; /* 0 = erase, 1 = draw */

but the kernel console doesn’t seem to do any of those two alternatives for a blinking cursor effect.

Also, the ps2fb driver already makes use of foreground fg cursor colour in

linux/drivers/video/fbdev/ps2fb.c

Line 1352 in 7455188

.fg_rgbaq = console_pseudo_palette(cursor->fg, par)

(which can be observed with echo -e '\033[?17;0;64c' for a red block cursor), but there’s also an background bg cursor colour setting in

linux/include/linux/fb.h

Line 352 in 7455188

__u32 bg; /* background color */

that’s maybe the colour of the upper part of the cursor?

[frno7]

Cursor blinking is on a timer function fb_flashcursor in

linux/drivers/video/fbdev/core/fbcon.c

Lines 388 to 420 in 59a11ab

	static void fb_flashcursor(struct work_struct *work)
	{
	struct fb_info *info = container_of(work, struct fb_info, queue);
	struct fbcon_ops *ops = info->fbcon_par;
	struct vc_data *vc = NULL;
	int c;
	int mode;
	int ret;
	/* FIXME: we should sort out the unbind locking instead */
	/* instead we just fail to flash the cursor if we can't get
	* the lock instead of blocking fbcon deinit */
	ret = console_trylock();
	if (ret == 0)
	return;
	if (ops && ops->currcon != -1)
	vc = vc_cons[ops->currcon].d;
	if (!vc || !con_is_visible(vc) ||
	registered_fb[con2fb_map[vc->vc_num]] != info ||
	vc->vc_deccm != 1) {
	console_unlock();
	return;
	}
	c = scr_readw((u16 *) vc->vc_pos);
	mode = (!ops->cursor_flash || ops->cursor_state.enable) ?
	CM_ERASE : CM_DRAW;
	ops->cursor(vc, info, mode, softback_lines, get_color(vc, info, c, 1),
	get_color(vc, info, c, 0));
	console_unlock();
	}

and it calls tile_cursor in

linux/drivers/video/fbdev/core/tileblit.c

Lines 82 to 117 in 59a11ab

	static void tile_cursor(struct vc_data *vc, struct fb_info *info, int mode,
	int softback_lines, int fg, int bg)
	{
	struct fb_tilecursor cursor;
	int use_sw = (vc->vc_cursor_type & 0x10);
	cursor.sx = vc->vc_x;
	cursor.sy = vc->vc_y;
	cursor.mode = (mode == CM_ERASE || use_sw) ? 0 : 1;
	cursor.fg = fg;
	cursor.bg = bg;
	switch (vc->vc_cursor_type & 0x0f) {
	case CUR_NONE:
	cursor.shape = FB_TILE_CURSOR_NONE;
	break;
	case CUR_UNDERLINE:
	cursor.shape = FB_TILE_CURSOR_UNDERLINE;
	break;
	case CUR_LOWER_THIRD:
	cursor.shape = FB_TILE_CURSOR_LOWER_THIRD;
	break;
	case CUR_LOWER_HALF:
	cursor.shape = FB_TILE_CURSOR_LOWER_HALF;
	break;
	case CUR_TWO_THIRDS:
	cursor.shape = FB_TILE_CURSOR_TWO_THIRDS;
	break;
	case CUR_BLOCK:
	default:
	cursor.shape = FB_TILE_CURSOR_BLOCK;
	break;
	}
	info->tileops->fb_tilecursor(info, &cursor);
	}

but it’s not entirely clear what, if anything, is controlling a blinking effect.

[frno7]

@Arch91, if you’re satisfied with a dirty blinker, try this 3-line piece of hack (having the new blink variable as indicated by <<<----) for the write_cb_tilecursor function:

        static bool blink;    // <<<----
        union package * const base_package = par->package.buffer;
        union package *package = base_package;
        const u32 tw = par->cb.tile.width;
        const u32 th = par->cb.tile.height;
        const u32 dy =
                blink                                       ? 0 :    // <<<----
                cursor->shape == FB_TILE_CURSOR_NONE        ? 0 :
                cursor->shape == FB_TILE_CURSOR_UNDERLINE   ? 1 :
                cursor->shape == FB_TILE_CURSOR_LOWER_THIRD ? th / 3 :
                cursor->shape == FB_TILE_CURSOR_LOWER_HALF  ? th / 2 :
                cursor->shape == FB_TILE_CURSOR_TWO_THIRDS  ? (2 * th) / 3 :
                cursor->shape == FB_TILE_CURSOR_BLOCK       ? th : th;
        const struct cb_cursor c = {
                .sx = (var->xoffset + tw * cursor->sx) % var->xres_virtual,
                .sy = (var->yoffset + th * cursor->sy) % var->yres_virtual,
                .dy = dy,
                .tw = tw,
                .th = th,
                .draw = cursor->mode,
                .fg_rgbaq = console_pseudo_palette(cursor->fg, par)
        };

        blink = !blink;    // <<<----

Then do echo -e "\e[?6c", as previously mentioned, for a big block of blinking awesomeness. :-)

[Arch91]

The nearest time I only was reading your posts, guys, but today I redownloaded the source code from the main branch of the linux repository, recompiled the kernel and updated the modules in the appropriate places. Also changed to modprobe ps2fb instead of gs (the new cursor codes are not applied to the drm driver yet?..) in the initramfs init script.
@frno7, you forced me to search the place where to apply those blink inlines!) That's the kernel_source/drivers/video/fbdev/ps2fb.c file.
I wrote those three places you marked with <<<---- by looking to your post. I loaded the kernel and obtained that the cursor is not blinking. Isn't there a chance that it is blinking SO FAST so I can't notice the blinking? I used mode_option=1920x1080p@50 mode_margin=+13+0 addition for the modprobe ps2fb inline, if that matters as to be noted for my videomode choosed.

Then do echo -e "\e[?6c"

I saw just a white rectangle/block. I tried them all from 0 to 6)

try echo -e '\033[?17;0;64c' for a big red block of a cursor!

Red. Nice.

satisfied with a dirty blinker

Well, I described the perfect) Adjustable cursor blink times and after these times and moments of blinkings passed it will not blinks until cursor moving; adjustable speed between the moments of those blinkings - both of GNU standards, which means that if there could be an app which tweaks those things, then that app will be working without any edits/additions in it's source code.
But as I noted, maybe those things are not related to main terminal console. And in the console there is a non-blinking cursor only. I'm not sure about that.

And a side question. Is ps2fb driver ready for to be launched through Xvfb X-server?

[frno7]

(the new cursor codes are not applied to the non-drm driver yet?..)

No, the DRM driver is likely to become deprecated.

I loaded the kernel and obtained that the cursor is not blinking.

Did you rerun make modules_install and so on after adding the blinker?

I loaded the kernel and obtained that the cursor is not blinking. Isn't there a chance that it is blinking SO FAST so I can't notice the blinking?

Probably not too fast, no.

Then do echo -e "\e[?6c"

I saw just a white rectangle/block. I tried them all from 0 to 6)

This one should blink. Check that /sys/class/graphics/fbcon/cursor_blink is 1 also.

try echo -e '\033[?17;0;64c' for a big red block of a cursor!

Red. Nice.

This one turns off blinking, though...

Well, I described the perfect) Adjustable cursor blink times and after these times and moments of blinkings passed it will not blinks until cursor moving; adjustable speed between the moments of those blinkings - both of GNU standards, which means that if there could be an app which tweaks those things, then that app will be working without any edits/additions in it's source code. But as I noted, maybe those things are not related to main terminal console. And in the console there is a non-blinking cursor only. I'm not sure about that.

Moving the cursor should not be necessary. It should blink by itself, and at least it does for me.

And a side question. Is ps2fb driver ready for to be launched through Xvfb X-server?

Not yet, it only does text at the moment. But it's pretty effective at that. :-)

[Arch91]

Did you rerun make modules_install and so on after adding the blinker?

Well, yes, that was the place where I made a mistake - my old modules were placed in initramfs(and in my target system too)/lib/modules/5.4.169 while the modules from the new kernel are now placed in initramfs/lib/modules/5.4.169+ (with plus). I removed that 5.4.169 folder before the kernel recompiling. And that blinds my eyes, I thought that after the ps2fb.c is edited, the kernel rebuilding is enough. I did not mind about that modules must be rebuilt either. Exactly that ps2fb.c related edits are for the appropriate module.
Now I saw the blinking. About 2.5-3 times in second.

By the way, if to change the cursor shape to rectangle with echo -e "\e[?6c" and then to use a nano text editor with a further quit from the editing screen - the cursor shape changes back to the underline. In different from the vi text editor -
the cursor stays unchanged before, while and after the vi using.

Moving the cursor should not be necessary. It should blink by itself, and at least it does for me.

after these times and moments of blinkings passed it will not blinks until cursor moving

I mean... other. Here is the example of the cursor behaviour in the terminal with Xes in my Fedora 30:
The time before the blinkings is 1 second - visible->half_of_sec_passed->not_visible->half_of_sec_passed->visible->...
If to do nothing/not doing the cursor moves - it blinks 10 times and becomes constantly visible UNTIL you'll push some keyboard button or escape-return to the window of that terminal.

Not yet

I dare myself an unwary question of inquire - what is insufficient in ps2fb to become a serious linux videodriver?) I'm just in curious, I beleive whatever the answer I can't influence for that driver upgradation in any imaginable by me way :/