rasops: fix automagic box chars for stride > 1 on little endian

This makes DEC line drawing (acsc) work correctly for "iso" fonts of
large sizes, e.g. spleen16x32 on x86.

rasops_mapchar: cosmetics, same object code.

Don't hide the important function call inside an if condition. Don't
reuse a variable, changing what it means in the middle of an
expression.

rasops: make autogenerated box drawing chars actually used

Fix mapchar to use PICK_FONT() to match what putchar does. Otherwise
putchar will never get to see the glyph codes for the autogenerated
box drawing chars.

wsfb: Prefer wide fonts when EDID data is available.

To give us a better chance of picking a readable font, prefer fonts that
will render at least 3mm wide instead of picking the font that will be
closest to that size.

rasops_reconfig: Do not abort even if font stride is not supported by
rasops_make_box_chars_*() functions; Stop using ri_optfont instead.

XXX
Add rasops_make_box_chars_24().

Remove pointless cast; dp is already uint32_t *.
No functional changes intended.

Misc style clean up's.
- Introduce and use proper macros.
- Use not ambiguous variable names.
- Unify similar functions as possible as I can.
- G/C unused headers.
- Use #include <dev/rasops/foo.h> instead of "foo.h"
No particular functional changes intended.

Set 4-bpp devcmap in a similar manner to non-RGB case of 8-bpp.
No functional changes since this is not in use (4-bpp is monochrome).

Correctly check whether character is in font in rasops_mapchar().
Also, make sure that in putchar functions for completeness.

Fix color range overflow; we cannot make bright colors more brighter.

When legacy Apple 4-bpp color palette is used, make green dark so that
kernel messages are printed nicely on white background.

Scaling dimensions of underline by font height.

Currently,
- offset of underline is fixed to 1-row from bottom of characters, and
- height of underline is fixed to 1.

Both are good for standard 8x16 fonts. However, it is too thin for
larger fonts, especially when used on display of high resolution.

Also, 1-row offset of underline is ugly for small fonts, e.g.,
spleen5x8.

Therefore, adjust offset and height as,
- no changes for standard 16-height fonts.
- scaling by font height for larger fonts.
- set offset to zero for fonts of height smaller than 16.

If RI_CLEAR is set, do not forget to clear real framebuffer.

Stop allocating ri_buf and ri_stamp dynamically. As commented in
rasops.h, it is not safe to use kmem_alloc(9) in rasops_init();
rasops routines can be used for early putchar, which means that
UVM is not fully initialized.

Should fix a problem reported by macallan:
http://mail-index.netbsd.org/tech-kern/2019/08/02/msg025327.html

Instead of using ri_buf, inline function rasops_memcpy32() is
introduced to fill 32bit data efficiently.

Instead of using ri_stamp (per device stamp), stamp_ri is
introduced to distinguish for which device stamp is calculated.

Simplify rasops_do_cursor():

- Use static masks similar to that used in rasops_bitops.h,
rather than generating them on the fly.
- Use pointer for proper type to avoid unnecessary casts.

Fix a critical bug for rasops_copyrows() introduced in rev. 1.90:
http://cvsweb.netbsd.org/bsdweb.cgi/src/sys/dev/rasops/rasops.c#rev1.90

When src < dst, we have to copy backward.

Depth 2 is monochrome.

IMO, it is impossible to support ANSI colors on 2-bpp display; fore-
and background can be same value even if they are different colors
logically.

Fix black color-attribution for depths 2 and 4.

Simplify calculation for 12-byte alignment.
No functional changes.

Use _KERNEL_OPT.

Real fix for 24-bpp color:
- When centering screen, locate effective base address of framebuffer to
both word and 24-bit color boundary.
- Consistently convert ri_devcmap to ``big endian'' if not RI_BSWAP.

Also, fix possible bug for 15/16-bpp with RI_BSWAP (not tested).

Fix unaligned writes to buffer, that are introduced in 1.105:
http://cvsweb.netbsd.org/bsdweb.cgi/src/sys/dev/rasops/rasops.c#rev1.105

Oops, for rasops_copycols(), we cannot use memmove even if src == dst.
On the other hand, memmove is safe for rasops_copyrows().

Provide buffer capable of single-row pixels in order to make things simpler.

XXX
Bump kernel version for rasops_info later.

G/C ri_delta.

XXX
Bump kernel version after other changes for struct rasops_info.

Switch to per-device stamp, and retire stamp_mutex.

XXX
Bump kernel version after other changes for struct rasops_info.

Misc clean-up's:
- protect private stuff in rasops.h by _RASOPS_PRIVATE
- staticify rasops_copycols() and rasops_isgray[]
- G/C unused extern int cold

branches: 1.101.2;
Try to improve performance when shadow framebuffer is present;
Use block copy from shadow fb to real fb, instead of repeating
the same operations to two fb's.

In rasops_copy{rows,cols}(), if src == dst, we have nothing to do.
Otherwise, we can use memcpy safely instead of memmove.

Fix catastrophe when ri_emustride != ri_stride in rasops_copyrows().

Treat highlighted and reversed text in the same manner to xterm.

Hmmm, color was still strange for 24bpp on little endian machine,
only when font width is 12.

We need to use different devcmap for that case, if we wish to share
codes for other depths/font widths as possible as we can.

XXX
What should we do for big endian? I have no big endian machines with
24bpp framebuffer...

Fix color on 24bpp screen for little endian.

Fix missing underlines on mono screen.
Style.

Cast attr to uint32_t before right shift to avoid undefined behavior.

Also, misc style/cosmetic changes for clarity.

Free kmem_alloc'd memory with kmem_free

Put back byte-wise copy to stop using memcpy, which
does not work for device memory on some platforms.

Pointed out by Jared, many thanks!

Add genfb(4) driver for mac68k grfbus.
1, 2, 4, 8, 15, and 32 color-depths are supported.

ANSI colors on console are functional for depth >= 4.

Graphic applications based on wsfb API should work, provided
they can correctly handle fbi_fboffset and your color depth.

wsfb driver of Xorg 1.20 and mlterm-fb (framebuffer version of
x11/mlterm) work fine for depth == 1 || depth >= 8.

For depth == 8 (256 colors), graphic applications require
colormap callback, which is currently supported for some
internal graphic adapters, and only Civic (found on Quadra
AV series) was tested. Register definition and its usage are
taken from Linux. You can easily support other adapters if
Linux supports that.

Have fun, and any problem reports are welcomed!

Replace manually unrolled loops with memcpy/memmove or simple loops.
Modern compilers are smart enough; there's no measurable changes in
performance even on MC68040 with optimization level -Os.

Also, convert loop of byte-wise copy into memset.

Misc creen up for rasops.c:

- sort headers
- return error value instead of panic
- use uintptr_t for cast pointer to integer
- use macros appropriately
- use __func__
- some consistency check ifndef DEBUG
- try to avoid undefined behaviors related to shift
- convert malloc/free to kmem_alloc/kmem_free
- convert MIN to uimin
- style

Misc cleen up:
- Make 32bit mask unsigned
- DPRINTF --> __nothing ifndef DEBUG_RASOPS
- "#ifdef DIAGNOSTIC if (x) panic(); #endif" --> KASSERT(!x);
- KNF

No functional changes intended.

For kUBSan, avoid undefined behaviors even if harmless.
No functional changes intended.

Requested by uwe.

Oops, revert unintentional changes.

Well, masks do not have to be updated every time in loop.
Calculate them in advance even if it may be bogus.

Simplify logic and tiny clean-up.

Avoid shift-count overflow to fix strange cursor behaviors on amd64
when font width is odd.

Use unsigned integers for binary data storage.
No functional changes intended.

Style:
- u_char --> uint8_t
- u_int*_t --> uint*_t

No functional changes.

Fix cursor movement for ri_xscale = 1, e.g., fontwidth = 8 and bpp = 1.

rasops reused wscons attribute bits for internal control.
- make upper 4 attribute bits available for such use
- use wscons flag names instead of literal constants.

rasops_do_cursor():
- simplify & sanitize the unaligned case
- use only 32bit accesses
... no longer crash with odd sized fonts in 8bit

branches: 1.77.8; 1.77.10;
remove checks for failure after memory allocation calls that cannot fail:

kmem_alloc() with KM_SLEEP
kmem_zalloc() with KM_SLEEP
percpu_alloc()
pserialize_create()
psref_class_create()

all of these paths include an assertion that the allocation has not failed,
so callers should not assert that again.

rasops_reconfig(): don't clear flags we won't set here

branches: 1.75.2;
add RI_PREFER_ALPHA flag, for drivers that can draw such fonts by hardware

reset ri->ri_hwbits too if RI_CFGDONE is set.

prevent ri->ri_hwbits from moving to center every time rasops_reconfig() is
called when RI_CENTER is set.

branches: 1.73.2; 1.73.4;
add "best match" algorithm to wsfont and use this instead of a private
function in rasops.

branches: 1.72.2;
Don't leak f on failure. Noted by maxv@.

Compile-tested only, with zaurus.

branches: 1.71.2; 1.71.14;
add another convenience function:
rasops_get_cmap() which returns either the ANSI map or an R3G3B2 map,
depending on the rasops_info handed to it so drivers don't have to
duplicate this particular code snippet

branches: 1.70.2; 1.70.6; 1.70.8;
make rasops_init()'s font selection code a bit less boneheaded by:
- actually trying to pick a font which gets as close as possible to the
requested terminal size
- accepting 0,0 as 'use system default' which can be changed by
options RASOPS_DEFAULT_WIDTH=100
options RASOPS_DEFAULT_HEIGHT=30
default is 80x25
- putting alpha and bitmap fonts in the same list and making wsfont_find()
aware of wether we support alpha fonts or not
- if supported, prefer alpha fonts over otherwise equally suitable bitmap
fonts

rasops8 expects the colour value replicated in every byte in devcmap[], so
fix that for r3g3b2

add a new flag for ri_flg to allow drivers to request an r3g3b2 devcmap
in 8 bit colour. Drivers are still responsible to generate an appropriate
colour map for the actual hardware.
For use with alpha blending which needs some sort of 'true' colour.

support anti-aliased fonts ( as in, pre-rendered alpha maps ), for now only
in rasops32 although adding support in 15, 16 and 24 would be trivial.
Enabled only if the driver explicitly requests it by setting the
RI_ENABLE_ALPHA flag.

branches: 1.66.8; 1.66.12;
Fix typo in comment.

Put underline properly (not upperline) on CCW screen.
Tested on hpcarm WS003SH.

Introduce a new flag for rasops_info to keep rasops_reconfig() from trying
to allocate memory. Use this when calling rasops_reconfig() before it is safe
to call kmem_alloc().

autogenerate box drawing characters for fonts that don't have them, put them
into an alternate font pointed at by the recently added mappings in wsfont,
adapt all putchar() methods except the rotated ones to use them
XXX no attempt has been made to make this work with rotation

Added Sharp W-ZERO3 series support.

branches: 1.61.2; 1.61.4;
make RI_CENTER and RI_FULLCLEAR work with a shadow framebuffer

ANSIfy another 1261 function definitions.
The only ones left in sys are beyond by sed script!
(or in sys/dist or sys/external)
Mostly they have function pointer parameters.

Change about 4500 of the K&R function definitions to ANSI ones.
There are still about 1600 left, but they have ',' or /* ... */
in the actual variable definitions - which my awk script doesn't handle.
There are also many that need () -> (void).
(The script does handle misordered arguments.)

branches: 1.58.8; 1.58.14;
Remove clause 3 and 4 from TNF licenses

branches: 1.57.10; 1.57.12; 1.57.14;
Merge jmcneill-pm branch.

branches: 1.56.4; 1.56.6; 1.56.14; 1.56.16; 1.56.18;
Implement bounds checking in some places in display driver code to avoid
the possibility of a local user panic.

branches: 1.55.6; 1.55.14;
Updates to allow Zaurus screen to rotate 90 degrees to a usable state with the keyboard. Patch from peter@ copied from OpenBSD. Feedback and OK from matt@

branches: 1.54.2; 1.54.4;
__unused removal on arguments; approved by core.

- sprinkle __unused on function decls.
- fix a couple of unused bugs
- no more -Wno-unused for i386

branches: 1.52.14; 1.52.16;
Fixup a performance problem in rasops_copyrows that I introduced during
the shadowfb merge.

Add shadow framebuffer support.

branches: 1.50.2; 1.50.4;
Make almost everything #include <sys/bswap.h> instead of <machine/bswap.h>
The bswap.h and endian.h files are all rather incestuous, but I want to
get the constant folding stuff into one place - sys/bswap.h

branches: 1.49.2;
merge ktrace-lwp.

branches: 1.48.2;
fixed tab width and formatting

made rasops_allocattr_color() respect WS_DEFAULT_FG and WS_DEFAULT_BG instead of using white on black

branches: 1.46.2;
nuke trailing whitespace

de-__P

branches: 1.44.2; 1.44.6; 1.44.8; 1.44.10; 1.44.12;
For the rasops-drawn cursor the fg/bg indices are bit inverted, so
provide complimentary colors in the upper 16 entries.

branches: 1.43.2;
In rasops_copycols change bcopy to memmove. In that particular place
the src and dst are very likely to overlap, so using bcopy causes
garbage to be displayed

remove trailing \n in panic(). approved perry.

Rename rasops_alloc_cattr and rasops_alloc_mattr to
rasops_allocattr_color and rasops_allocattr_mono, respectively.

alloc_attr -> allocattr

Approved by Matthias Drochner.

branches: 1.39.4;
Reorganise the wsfont stuff slightly so that multiple display adapters
with different bit/byte order requirements can co-exist happily.

don't need <sys/types.h> when including <sys/param.h>

Give initiali[sz]e all the "i"s it deserves.

branches: 1.36.2; 1.36.4; 1.36.6;
Correct the ri_xorigin calculation for the RI_CENTER case again.

Original calculation (bits += (ri_stride - ri_emustride) / 2) was
incorrect because stride may be wider than visible width.

Fix in 1.33 (bits += (ri_width - ri_emustride) / 2) was incorrect
because units do not match; "bits" and "ri_emustride" are in bytes,
but "width" is in pels. Works by accident for 8bpp displays.

Change to bits += ((ri_width * bpp / 8) - ri_emustride) / 2
to correctly account for visible width and bpp.

Now does character remapping depending on font encoding.

Rasops supports 4bit depth.

Fix an error in xoffset calculation. Revealed in the case when
ri_width is less than ri_stride and screen is layouted RI_CENTERed.

Use my proper name.

Let this build on LP64 if DEBUG is defined.

branches: 1.30.2;
* Spell shift counts in decimal and masks in hex
* Optimize numerous array references
* Cleanup whitespace turds

Advance index into rasops_cmap for each color, not per triplet.
This makes highlighing and color possible on truecolor displays.

Fix computation of ri_xorigin in RI_CENTER case; convery from bytes to
pixels, not from bytes to bytes^2/pixel.

Resolve LP64 issues.

replace WSFONT_L2R by WSDISPLAY_FONTORDER_L2R to compile again.

Dispatch another nit.

-use the right namespace for screen capabilities
-make attribute decomposing a bit more friendly if the caller doesn't
care about underlines

-initialize the colormap completely at compile time, to allow drivers
to use it early
-now we can declare it "const" (as "rasops_isgray[]", while we are here)
-don't use the fg/bg colors in ...alloc_attr() if the WSATTR_WSCOLORS
flag was not given - use reasonable defaults instead
-add an opaque "ri_hw" member to "rasops_info", for driver use

Possible typo.

In rasops_do_cursor(), don't pull the mask from ri_devcmap[], just use ~0.
Needed to make cursor DTRT on NetBSD/hpcmips - from takemura.

- New option (RASOPS_SMALL) for the tight-fisted.
- Don't use int32_t/u_int32_t unless we must.
- Remove C++ single line comment delimeters that crept in.
- Remove defs pertaining to byte granularity 'ragged-edge' bitmasks.
- Move all declarations of per-depth initialization functions to rasops.h.
- Other minor cleanup.

branches: 1.19.2; 1.19.4; 1.19.6;
Use proper mask when RI_FORCEMONO is set and XORing cursor.

- Collapse forward and reverse cases in rasops_copyrows() into one
- Pull in opt_rasops.h so we know if clipping is enabled
- Some KNF

Clean up attribute allocation some more.

Small consistency nit.

Make monochrome attribte allocation a bit more obvious.

When clearing the entire display, point to the start of the framebuffer,
not start of console output. Also, do not advance by ri_delta every row.

- Change the semantics of rasops_init slightly
- Re-name rasops_setfont to rasops_reconfig
- Add some new run-time 'features'
- Bring all run-time 'features' under control of rasops_info::ri_flg
- Some cosmetic changes

- Style nits
- Kill some of the dainbramage in variable-depth copycols()

branches: 1.11.2;
- Don't even try to emulate WSATTR_HILIT on mono displays
- WSATTR_REVERSE is a capability of both mono/color

Fix stupid alignment bug. The 'slop' stuff is due for replacement with masks
RSN per the README.

Assign ownership & copyright to TNF. There is probably a procedure for this
that I am unaware of. Also some KNF.

Add diagnostic measures to check for missing font.

wsfont cookies <= 0 are invalid, not just those < 0.

In rasops_erasecols(), try word aligned case before halfword aligned.

Add cookie for wsfont to rasops_info.

Innumerable fixes and improvements. README contains list of remaining bugs
that need to be squashed.

Code cleanup. Add ri_bswap member to 'struct rasops_info' to indicate that
framebuffer endianness differs from CPU (for 15,16,32-bit displays).

Fixed licensing due to a foobar on my behalf.

Initial import of 'rasops', the new raster operations set for wscons/rcons.

wsfb: Prefer wide fonts when EDID data is available.

To give us a better chance of picking a readable font, prefer fonts that
will render at least 3mm wide instead of picking the font that will be
closest to that size.

rasops_reconfig: Do not abort even if font stride is not supported by
rasops_make_box_chars_*() functions; Stop using ri_optfont instead.

XXX
Add rasops_make_box_chars_24().

Remove pointless cast; dp is already uint32_t *.
No functional changes intended.

Misc style clean up's.
- Introduce and use proper macros.
- Use not ambiguous variable names.
- Unify similar functions as possible as I can.
- G/C unused headers.
- Use #include <dev/rasops/foo.h> instead of "foo.h"
No particular functional changes intended.

Set 4-bpp devcmap in a similar manner to non-RGB case of 8-bpp.
No functional changes since this is not in use (4-bpp is monochrome).

Correctly check whether character is in font in rasops_mapchar().
Also, make sure that in putchar functions for completeness.

Fix color range overflow; we cannot make bright colors more brighter.

When legacy Apple 4-bpp color palette is used, make green dark so that
kernel messages are printed nicely on white background.

Scaling dimensions of underline by font height.

Currently,
- offset of underline is fixed to 1-row from bottom of characters, and
- height of underline is fixed to 1.

Both are good for standard 8x16 fonts. However, it is too thin for
larger fonts, especially when used on display of high resolution.

Also, 1-row offset of underline is ugly for small fonts, e.g.,
spleen5x8.

Therefore, adjust offset and height as,
- no changes for standard 16-height fonts.
- scaling by font height for larger fonts.
- set offset to zero for fonts of height smaller than 16.

If RI_CLEAR is set, do not forget to clear real framebuffer.

Stop allocating ri_buf and ri_stamp dynamically. As commented in
rasops.h, it is not safe to use kmem_alloc(9) in rasops_init();
rasops routines can be used for early putchar, which means that
UVM is not fully initialized.

Should fix a problem reported by macallan:
http://mail-index.netbsd.org/tech-kern/2019/08/02/msg025327.html

Instead of using ri_buf, inline function rasops_memcpy32() is
introduced to fill 32bit data efficiently.

Instead of using ri_stamp (per device stamp), stamp_ri is
introduced to distinguish for which device stamp is calculated.

Simplify rasops_do_cursor():

- Use static masks similar to that used in rasops_bitops.h,
rather than generating them on the fly.
- Use pointer for proper type to avoid unnecessary casts.

Fix a critical bug for rasops_copyrows() introduced in rev. 1.90:
http://cvsweb.netbsd.org/bsdweb.cgi/src/sys/dev/rasops/rasops.c#rev1.90

When src < dst, we have to copy backward.

Depth 2 is monochrome.

IMO, it is impossible to support ANSI colors on 2-bpp display; fore-
and background can be same value even if they are different colors
logically.

Fix black color-attribution for depths 2 and 4.

Simplify calculation for 12-byte alignment.
No functional changes.

Use _KERNEL_OPT.

Real fix for 24-bpp color:
- When centering screen, locate effective base address of framebuffer to
both word and 24-bit color boundary.
- Consistently convert ri_devcmap to ``big endian'' if not RI_BSWAP.

Also, fix possible bug for 15/16-bpp with RI_BSWAP (not tested).

Fix unaligned writes to buffer, that are introduced in 1.105:
http://cvsweb.netbsd.org/bsdweb.cgi/src/sys/dev/rasops/rasops.c#rev1.105

Oops, for rasops_copycols(), we cannot use memmove even if src == dst.
On the other hand, memmove is safe for rasops_copyrows().

Provide buffer capable of single-row pixels in order to make things simpler.

XXX
Bump kernel version for rasops_info later.

G/C ri_delta.

XXX
Bump kernel version after other changes for struct rasops_info.

Switch to per-device stamp, and retire stamp_mutex.

XXX
Bump kernel version after other changes for struct rasops_info.

Misc clean-up's:
- protect private stuff in rasops.h by _RASOPS_PRIVATE
- staticify rasops_copycols() and rasops_isgray[]
- G/C unused extern int cold

branches: 1.101.2;
Try to improve performance when shadow framebuffer is present;
Use block copy from shadow fb to real fb, instead of repeating
the same operations to two fb's.

In rasops_copy{rows,cols}(), if src == dst, we have nothing to do.
Otherwise, we can use memcpy safely instead of memmove.

Fix catastrophe when ri_emustride != ri_stride in rasops_copyrows().

Treat highlighted and reversed text in the same manner to xterm.

Hmmm, color was still strange for 24bpp on little endian machine,
only when font width is 12.

We need to use different devcmap for that case, if we wish to share
codes for other depths/font widths as possible as we can.

XXX
What should we do for big endian? I have no big endian machines with
24bpp framebuffer...

Fix color on 24bpp screen for little endian.

Fix missing underlines on mono screen.
Style.

Cast attr to uint32_t before right shift to avoid undefined behavior.

Also, misc style/cosmetic changes for clarity.

Free kmem_alloc'd memory with kmem_free

Put back byte-wise copy to stop using memcpy, which
does not work for device memory on some platforms.

Pointed out by Jared, many thanks!

Add genfb(4) driver for mac68k grfbus.
1, 2, 4, 8, 15, and 32 color-depths are supported.

ANSI colors on console are functional for depth >= 4.

Graphic applications based on wsfb API should work, provided
they can correctly handle fbi_fboffset and your color depth.

wsfb driver of Xorg 1.20 and mlterm-fb (framebuffer version of
x11/mlterm) work fine for depth == 1 || depth >= 8.

For depth == 8 (256 colors), graphic applications require
colormap callback, which is currently supported for some
internal graphic adapters, and only Civic (found on Quadra
AV series) was tested. Register definition and its usage are
taken from Linux. You can easily support other adapters if
Linux supports that.

Have fun, and any problem reports are welcomed!

Replace manually unrolled loops with memcpy/memmove or simple loops.
Modern compilers are smart enough; there's no measurable changes in
performance even on MC68040 with optimization level -Os.

Also, convert loop of byte-wise copy into memset.

Misc creen up for rasops.c:

- sort headers
- return error value instead of panic
- use uintptr_t for cast pointer to integer
- use macros appropriately
- use __func__
- some consistency check ifndef DEBUG
- try to avoid undefined behaviors related to shift
- convert malloc/free to kmem_alloc/kmem_free
- convert MIN to uimin
- style

Misc cleen up:
- Make 32bit mask unsigned
- DPRINTF --> __nothing ifndef DEBUG_RASOPS
- "#ifdef DIAGNOSTIC if (x) panic(); #endif" --> KASSERT(!x);
- KNF

No functional changes intended.

For kUBSan, avoid undefined behaviors even if harmless.
No functional changes intended.

Requested by uwe.

Oops, revert unintentional changes.

Well, masks do not have to be updated every time in loop.
Calculate them in advance even if it may be bogus.

Simplify logic and tiny clean-up.

Avoid shift-count overflow to fix strange cursor behaviors on amd64
when font width is odd.

Use unsigned integers for binary data storage.
No functional changes intended.

Style:
- u_char --> uint8_t
- u_int*_t --> uint*_t

No functional changes.

Fix cursor movement for ri_xscale = 1, e.g., fontwidth = 8 and bpp = 1.

rasops reused wscons attribute bits for internal control.
- make upper 4 attribute bits available for such use
- use wscons flag names instead of literal constants.

rasops_do_cursor():
- simplify & sanitize the unaligned case
- use only 32bit accesses
... no longer crash with odd sized fonts in 8bit

branches: 1.77.8; 1.77.10;
remove checks for failure after memory allocation calls that cannot fail:

kmem_alloc() with KM_SLEEP
kmem_zalloc() with KM_SLEEP
percpu_alloc()
pserialize_create()
psref_class_create()

all of these paths include an assertion that the allocation has not failed,
so callers should not assert that again.

rasops_reconfig(): don't clear flags we won't set here

branches: 1.75.2;
add RI_PREFER_ALPHA flag, for drivers that can draw such fonts by hardware

reset ri->ri_hwbits too if RI_CFGDONE is set.

prevent ri->ri_hwbits from moving to center every time rasops_reconfig() is
called when RI_CENTER is set.

branches: 1.73.2; 1.73.4;
add "best match" algorithm to wsfont and use this instead of a private
function in rasops.

branches: 1.72.2;
Don't leak f on failure. Noted by maxv@.

Compile-tested only, with zaurus.

branches: 1.71.2; 1.71.14;
add another convenience function:
rasops_get_cmap() which returns either the ANSI map or an R3G3B2 map,
depending on the rasops_info handed to it so drivers don't have to
duplicate this particular code snippet

branches: 1.70.2; 1.70.6; 1.70.8;
make rasops_init()'s font selection code a bit less boneheaded by:
- actually trying to pick a font which gets as close as possible to the
requested terminal size
- accepting 0,0 as 'use system default' which can be changed by
options RASOPS_DEFAULT_WIDTH=100
options RASOPS_DEFAULT_HEIGHT=30
default is 80x25
- putting alpha and bitmap fonts in the same list and making wsfont_find()
aware of wether we support alpha fonts or not
- if supported, prefer alpha fonts over otherwise equally suitable bitmap
fonts

rasops8 expects the colour value replicated in every byte in devcmap[], so
fix that for r3g3b2

add a new flag for ri_flg to allow drivers to request an r3g3b2 devcmap
in 8 bit colour. Drivers are still responsible to generate an appropriate
colour map for the actual hardware.
For use with alpha blending which needs some sort of 'true' colour.

support anti-aliased fonts ( as in, pre-rendered alpha maps ), for now only
in rasops32 although adding support in 15, 16 and 24 would be trivial.
Enabled only if the driver explicitly requests it by setting the
RI_ENABLE_ALPHA flag.

branches: 1.66.8; 1.66.12;
Fix typo in comment.

Put underline properly (not upperline) on CCW screen.
Tested on hpcarm WS003SH.

Introduce a new flag for rasops_info to keep rasops_reconfig() from trying
to allocate memory. Use this when calling rasops_reconfig() before it is safe
to call kmem_alloc().

autogenerate box drawing characters for fonts that don't have them, put them
into an alternate font pointed at by the recently added mappings in wsfont,
adapt all putchar() methods except the rotated ones to use them
XXX no attempt has been made to make this work with rotation

Added Sharp W-ZERO3 series support.

branches: 1.61.2; 1.61.4;
make RI_CENTER and RI_FULLCLEAR work with a shadow framebuffer

ANSIfy another 1261 function definitions.
The only ones left in sys are beyond by sed script!
(or in sys/dist or sys/external)
Mostly they have function pointer parameters.

Change about 4500 of the K&R function definitions to ANSI ones.
There are still about 1600 left, but they have ',' or /* ... */
in the actual variable definitions - which my awk script doesn't handle.
There are also many that need () -> (void).
(The script does handle misordered arguments.)

branches: 1.58.8; 1.58.14;
Remove clause 3 and 4 from TNF licenses

branches: 1.57.10; 1.57.12; 1.57.14;
Merge jmcneill-pm branch.

branches: 1.56.4; 1.56.6; 1.56.14; 1.56.16; 1.56.18;
Implement bounds checking in some places in display driver code to avoid
the possibility of a local user panic.

branches: 1.55.6; 1.55.14;
Updates to allow Zaurus screen to rotate 90 degrees to a usable state with the keyboard. Patch from peter@ copied from OpenBSD. Feedback and OK from matt@

branches: 1.54.2; 1.54.4;
__unused removal on arguments; approved by core.

- sprinkle __unused on function decls.
- fix a couple of unused bugs
- no more -Wno-unused for i386

branches: 1.52.14; 1.52.16;
Fixup a performance problem in rasops_copyrows that I introduced during
the shadowfb merge.

Add shadow framebuffer support.

branches: 1.50.2; 1.50.4;
Make almost everything #include <sys/bswap.h> instead of <machine/bswap.h>
The bswap.h and endian.h files are all rather incestuous, but I want to
get the constant folding stuff into one place - sys/bswap.h

branches: 1.49.2;
merge ktrace-lwp.

branches: 1.48.2;
fixed tab width and formatting

made rasops_allocattr_color() respect WS_DEFAULT_FG and WS_DEFAULT_BG instead of using white on black

branches: 1.46.2;
nuke trailing whitespace

de-__P

branches: 1.44.2; 1.44.6; 1.44.8; 1.44.10; 1.44.12;
For the rasops-drawn cursor the fg/bg indices are bit inverted, so
provide complimentary colors in the upper 16 entries.

branches: 1.43.2;
In rasops_copycols change bcopy to memmove. In that particular place
the src and dst are very likely to overlap, so using bcopy causes
garbage to be displayed

remove trailing \n in panic(). approved perry.

Rename rasops_alloc_cattr and rasops_alloc_mattr to
rasops_allocattr_color and rasops_allocattr_mono, respectively.

alloc_attr -> allocattr

Approved by Matthias Drochner.

branches: 1.39.4;
Reorganise the wsfont stuff slightly so that multiple display adapters
with different bit/byte order requirements can co-exist happily.

don't need <sys/types.h> when including <sys/param.h>

Give initiali[sz]e all the "i"s it deserves.

branches: 1.36.2; 1.36.4; 1.36.6;
Correct the ri_xorigin calculation for the RI_CENTER case again.

Original calculation (bits += (ri_stride - ri_emustride) / 2) was
incorrect because stride may be wider than visible width.

Fix in 1.33 (bits += (ri_width - ri_emustride) / 2) was incorrect
because units do not match; "bits" and "ri_emustride" are in bytes,
but "width" is in pels. Works by accident for 8bpp displays.

Change to bits += ((ri_width * bpp / 8) - ri_emustride) / 2
to correctly account for visible width and bpp.

Now does character remapping depending on font encoding.

Rasops supports 4bit depth.

Fix an error in xoffset calculation. Revealed in the case when
ri_width is less than ri_stride and screen is layouted RI_CENTERed.

Use my proper name.

Let this build on LP64 if DEBUG is defined.

branches: 1.30.2;
* Spell shift counts in decimal and masks in hex
* Optimize numerous array references
* Cleanup whitespace turds

Advance index into rasops_cmap for each color, not per triplet.
This makes highlighing and color possible on truecolor displays.

Fix computation of ri_xorigin in RI_CENTER case; convery from bytes to
pixels, not from bytes to bytes^2/pixel.

Resolve LP64 issues.

replace WSFONT_L2R by WSDISPLAY_FONTORDER_L2R to compile again.

Dispatch another nit.

-use the right namespace for screen capabilities
-make attribute decomposing a bit more friendly if the caller doesn't
care about underlines

-initialize the colormap completely at compile time, to allow drivers
to use it early
-now we can declare it "const" (as "rasops_isgray[]", while we are here)
-don't use the fg/bg colors in ...alloc_attr() if the WSATTR_WSCOLORS
flag was not given - use reasonable defaults instead
-add an opaque "ri_hw" member to "rasops_info", for driver use

Possible typo.

In rasops_do_cursor(), don't pull the mask from ri_devcmap[], just use ~0.
Needed to make cursor DTRT on NetBSD/hpcmips - from takemura.

- New option (RASOPS_SMALL) for the tight-fisted.
- Don't use int32_t/u_int32_t unless we must.
- Remove C++ single line comment delimeters that crept in.
- Remove defs pertaining to byte granularity 'ragged-edge' bitmasks.
- Move all declarations of per-depth initialization functions to rasops.h.
- Other minor cleanup.

branches: 1.19.2; 1.19.4; 1.19.6;
Use proper mask when RI_FORCEMONO is set and XORing cursor.

- Collapse forward and reverse cases in rasops_copyrows() into one
- Pull in opt_rasops.h so we know if clipping is enabled
- Some KNF

Clean up attribute allocation some more.

Small consistency nit.

Make monochrome attribte allocation a bit more obvious.

When clearing the entire display, point to the start of the framebuffer,
not start of console output. Also, do not advance by ri_delta every row.

- Change the semantics of rasops_init slightly
- Re-name rasops_setfont to rasops_reconfig
- Add some new run-time 'features'
- Bring all run-time 'features' under control of rasops_info::ri_flg
- Some cosmetic changes

- Style nits
- Kill some of the dainbramage in variable-depth copycols()

branches: 1.11.2;
- Don't even try to emulate WSATTR_HILIT on mono displays
- WSATTR_REVERSE is a capability of both mono/color

Fix stupid alignment bug. The 'slop' stuff is due for replacement with masks
RSN per the README.

Assign ownership & copyright to TNF. There is probably a procedure for this
that I am unaware of. Also some KNF.

Add diagnostic measures to check for missing font.

wsfont cookies <= 0 are invalid, not just those < 0.

In rasops_erasecols(), try word aligned case before halfword aligned.

Add cookie for wsfont to rasops_info.

Innumerable fixes and improvements. README contains list of remaining bugs
that need to be squashed.

Code cleanup. Add ri_bswap member to 'struct rasops_info' to indicate that
framebuffer endianness differs from CPU (for 15,16,32-bit displays).

Fixed licensing due to a foobar on my behalf.

Initial import of 'rasops', the new raster operations set for wscons/rcons.

rasops_reconfig: Do not abort even if font stride is not supported by
rasops_make_box_chars_*() functions; Stop using ri_optfont instead.

XXX
Add rasops_make_box_chars_24().

Remove pointless cast; dp is already uint32_t *.
No functional changes intended.

Misc style clean up's.
- Introduce and use proper macros.
- Use not ambiguous variable names.
- Unify similar functions as possible as I can.
- G/C unused headers.
- Use #include <dev/rasops/foo.h> instead of "foo.h"
No particular functional changes intended.

Set 4-bpp devcmap in a similar manner to non-RGB case of 8-bpp.
No functional changes since this is not in use (4-bpp is monochrome).

Correctly check whether character is in font in rasops_mapchar().
Also, make sure that in putchar functions for completeness.

Fix color range overflow; we cannot make bright colors more brighter.

When legacy Apple 4-bpp color palette is used, make green dark so that
kernel messages are printed nicely on white background.

Scaling dimensions of underline by font height.

Currently,
- offset of underline is fixed to 1-row from bottom of characters, and
- height of underline is fixed to 1.

Both are good for standard 8x16 fonts. However, it is too thin for
larger fonts, especially when used on display of high resolution.

Also, 1-row offset of underline is ugly for small fonts, e.g.,
spleen5x8.

Therefore, adjust offset and height as,
- no changes for standard 16-height fonts.
- scaling by font height for larger fonts.
- set offset to zero for fonts of height smaller than 16.

If RI_CLEAR is set, do not forget to clear real framebuffer.

Stop allocating ri_buf and ri_stamp dynamically. As commented in
rasops.h, it is not safe to use kmem_alloc(9) in rasops_init();
rasops routines can be used for early putchar, which means that
UVM is not fully initialized.

Should fix a problem reported by macallan:
http://mail-index.netbsd.org/tech-kern/2019/08/02/msg025327.html

Instead of using ri_buf, inline function rasops_memcpy32() is
introduced to fill 32bit data efficiently.

Instead of using ri_stamp (per device stamp), stamp_ri is
introduced to distinguish for which device stamp is calculated.

Simplify rasops_do_cursor():

- Use static masks similar to that used in rasops_bitops.h,
rather than generating them on the fly.
- Use pointer for proper type to avoid unnecessary casts.

Fix a critical bug for rasops_copyrows() introduced in rev. 1.90:
http://cvsweb.netbsd.org/bsdweb.cgi/src/sys/dev/rasops/rasops.c#rev1.90

When src < dst, we have to copy backward.

Depth 2 is monochrome.

IMO, it is impossible to support ANSI colors on 2-bpp display; fore-
and background can be same value even if they are different colors
logically.

Fix black color-attribution for depths 2 and 4.

Simplify calculation for 12-byte alignment.
No functional changes.

Use _KERNEL_OPT.

Real fix for 24-bpp color:
- When centering screen, locate effective base address of framebuffer to
both word and 24-bit color boundary.
- Consistently convert ri_devcmap to ``big endian'' if not RI_BSWAP.

Also, fix possible bug for 15/16-bpp with RI_BSWAP (not tested).

Fix unaligned writes to buffer, that are introduced in 1.105:
http://cvsweb.netbsd.org/bsdweb.cgi/src/sys/dev/rasops/rasops.c#rev1.105

Oops, for rasops_copycols(), we cannot use memmove even if src == dst.
On the other hand, memmove is safe for rasops_copyrows().

Provide buffer capable of single-row pixels in order to make things simpler.

XXX
Bump kernel version for rasops_info later.

G/C ri_delta.

XXX
Bump kernel version after other changes for struct rasops_info.

Switch to per-device stamp, and retire stamp_mutex.

XXX
Bump kernel version after other changes for struct rasops_info.

Misc clean-up's:
- protect private stuff in rasops.h by _RASOPS_PRIVATE
- staticify rasops_copycols() and rasops_isgray[]
- G/C unused extern int cold

branches: 1.101.2;
Try to improve performance when shadow framebuffer is present;
Use block copy from shadow fb to real fb, instead of repeating
the same operations to two fb's.

In rasops_copy{rows,cols}(), if src == dst, we have nothing to do.
Otherwise, we can use memcpy safely instead of memmove.

Fix catastrophe when ri_emustride != ri_stride in rasops_copyrows().

Treat highlighted and reversed text in the same manner to xterm.

Hmmm, color was still strange for 24bpp on little endian machine,
only when font width is 12.

We need to use different devcmap for that case, if we wish to share
codes for other depths/font widths as possible as we can.

XXX
What should we do for big endian? I have no big endian machines with
24bpp framebuffer...

Fix color on 24bpp screen for little endian.

Fix missing underlines on mono screen.
Style.

Cast attr to uint32_t before right shift to avoid undefined behavior.

Also, misc style/cosmetic changes for clarity.

Free kmem_alloc'd memory with kmem_free

Put back byte-wise copy to stop using memcpy, which
does not work for device memory on some platforms.

Pointed out by Jared, many thanks!

Add genfb(4) driver for mac68k grfbus.
1, 2, 4, 8, 15, and 32 color-depths are supported.

ANSI colors on console are functional for depth >= 4.

Graphic applications based on wsfb API should work, provided
they can correctly handle fbi_fboffset and your color depth.

wsfb driver of Xorg 1.20 and mlterm-fb (framebuffer version of
x11/mlterm) work fine for depth == 1 || depth >= 8.

For depth == 8 (256 colors), graphic applications require
colormap callback, which is currently supported for some
internal graphic adapters, and only Civic (found on Quadra
AV series) was tested. Register definition and its usage are
taken from Linux. You can easily support other adapters if
Linux supports that.

Have fun, and any problem reports are welcomed!

Replace manually unrolled loops with memcpy/memmove or simple loops.
Modern compilers are smart enough; there's no measurable changes in
performance even on MC68040 with optimization level -Os.

Also, convert loop of byte-wise copy into memset.

Misc creen up for rasops.c:

- sort headers
- return error value instead of panic
- use uintptr_t for cast pointer to integer
- use macros appropriately
- use __func__
- some consistency check ifndef DEBUG
- try to avoid undefined behaviors related to shift
- convert malloc/free to kmem_alloc/kmem_free
- convert MIN to uimin
- style

Misc cleen up:
- Make 32bit mask unsigned
- DPRINTF --> __nothing ifndef DEBUG_RASOPS
- "#ifdef DIAGNOSTIC if (x) panic(); #endif" --> KASSERT(!x);
- KNF

No functional changes intended.

For kUBSan, avoid undefined behaviors even if harmless.
No functional changes intended.

Requested by uwe.

Oops, revert unintentional changes.

Well, masks do not have to be updated every time in loop.
Calculate them in advance even if it may be bogus.

Simplify logic and tiny clean-up.

Avoid shift-count overflow to fix strange cursor behaviors on amd64
when font width is odd.

Use unsigned integers for binary data storage.
No functional changes intended.

Style:
- u_char --> uint8_t
- u_int*_t --> uint*_t

No functional changes.

Fix cursor movement for ri_xscale = 1, e.g., fontwidth = 8 and bpp = 1.

rasops reused wscons attribute bits for internal control.
- make upper 4 attribute bits available for such use
- use wscons flag names instead of literal constants.

rasops_do_cursor():
- simplify & sanitize the unaligned case
- use only 32bit accesses
... no longer crash with odd sized fonts in 8bit

branches: 1.77.8; 1.77.10;
remove checks for failure after memory allocation calls that cannot fail:

kmem_alloc() with KM_SLEEP
kmem_zalloc() with KM_SLEEP
percpu_alloc()
pserialize_create()
psref_class_create()

all of these paths include an assertion that the allocation has not failed,
so callers should not assert that again.

rasops_reconfig(): don't clear flags we won't set here

branches: 1.75.2;
add RI_PREFER_ALPHA flag, for drivers that can draw such fonts by hardware

reset ri->ri_hwbits too if RI_CFGDONE is set.

prevent ri->ri_hwbits from moving to center every time rasops_reconfig() is
called when RI_CENTER is set.

branches: 1.73.2; 1.73.4;
add "best match" algorithm to wsfont and use this instead of a private
function in rasops.

branches: 1.72.2;
Don't leak f on failure. Noted by maxv@.

Compile-tested only, with zaurus.

branches: 1.71.2; 1.71.14;
add another convenience function:
rasops_get_cmap() which returns either the ANSI map or an R3G3B2 map,
depending on the rasops_info handed to it so drivers don't have to
duplicate this particular code snippet

branches: 1.70.2; 1.70.6; 1.70.8;
make rasops_init()'s font selection code a bit less boneheaded by:
- actually trying to pick a font which gets as close as possible to the
requested terminal size
- accepting 0,0 as 'use system default' which can be changed by
options RASOPS_DEFAULT_WIDTH=100
options RASOPS_DEFAULT_HEIGHT=30
default is 80x25
- putting alpha and bitmap fonts in the same list and making wsfont_find()
aware of wether we support alpha fonts or not
- if supported, prefer alpha fonts over otherwise equally suitable bitmap
fonts

rasops8 expects the colour value replicated in every byte in devcmap[], so
fix that for r3g3b2

add a new flag for ri_flg to allow drivers to request an r3g3b2 devcmap
in 8 bit colour. Drivers are still responsible to generate an appropriate
colour map for the actual hardware.
For use with alpha blending which needs some sort of 'true' colour.

support anti-aliased fonts ( as in, pre-rendered alpha maps ), for now only
in rasops32 although adding support in 15, 16 and 24 would be trivial.
Enabled only if the driver explicitly requests it by setting the
RI_ENABLE_ALPHA flag.

branches: 1.66.8; 1.66.12;
Fix typo in comment.

Put underline properly (not upperline) on CCW screen.
Tested on hpcarm WS003SH.

Introduce a new flag for rasops_info to keep rasops_reconfig() from trying
to allocate memory. Use this when calling rasops_reconfig() before it is safe
to call kmem_alloc().

autogenerate box drawing characters for fonts that don't have them, put them
into an alternate font pointed at by the recently added mappings in wsfont,
adapt all putchar() methods except the rotated ones to use them
XXX no attempt has been made to make this work with rotation

Added Sharp W-ZERO3 series support.

branches: 1.61.2; 1.61.4;
make RI_CENTER and RI_FULLCLEAR work with a shadow framebuffer

ANSIfy another 1261 function definitions.
The only ones left in sys are beyond by sed script!
(or in sys/dist or sys/external)
Mostly they have function pointer parameters.

Change about 4500 of the K&R function definitions to ANSI ones.
There are still about 1600 left, but they have ',' or /* ... */
in the actual variable definitions - which my awk script doesn't handle.
There are also many that need () -> (void).
(The script does handle misordered arguments.)

branches: 1.58.8; 1.58.14;
Remove clause 3 and 4 from TNF licenses

branches: 1.57.10; 1.57.12; 1.57.14;
Merge jmcneill-pm branch.

branches: 1.56.4; 1.56.6; 1.56.14; 1.56.16; 1.56.18;
Implement bounds checking in some places in display driver code to avoid
the possibility of a local user panic.

branches: 1.55.6; 1.55.14;
Updates to allow Zaurus screen to rotate 90 degrees to a usable state with the keyboard. Patch from peter@ copied from OpenBSD. Feedback and OK from matt@

branches: 1.54.2; 1.54.4;
__unused removal on arguments; approved by core.

- sprinkle __unused on function decls.
- fix a couple of unused bugs
- no more -Wno-unused for i386

branches: 1.52.14; 1.52.16;
Fixup a performance problem in rasops_copyrows that I introduced during
the shadowfb merge.

Add shadow framebuffer support.

branches: 1.50.2; 1.50.4;
Make almost everything #include <sys/bswap.h> instead of <machine/bswap.h>
The bswap.h and endian.h files are all rather incestuous, but I want to
get the constant folding stuff into one place - sys/bswap.h

branches: 1.49.2;
merge ktrace-lwp.

branches: 1.48.2;
fixed tab width and formatting

made rasops_allocattr_color() respect WS_DEFAULT_FG and WS_DEFAULT_BG instead of using white on black

branches: 1.46.2;
nuke trailing whitespace

de-__P

branches: 1.44.2; 1.44.6; 1.44.8; 1.44.10; 1.44.12;
For the rasops-drawn cursor the fg/bg indices are bit inverted, so
provide complimentary colors in the upper 16 entries.

branches: 1.43.2;
In rasops_copycols change bcopy to memmove. In that particular place
the src and dst are very likely to overlap, so using bcopy causes
garbage to be displayed

remove trailing \n in panic(). approved perry.

Rename rasops_alloc_cattr and rasops_alloc_mattr to
rasops_allocattr_color and rasops_allocattr_mono, respectively.

alloc_attr -> allocattr

Approved by Matthias Drochner.

branches: 1.39.4;
Reorganise the wsfont stuff slightly so that multiple display adapters
with different bit/byte order requirements can co-exist happily.

don't need <sys/types.h> when including <sys/param.h>

Give initiali[sz]e all the "i"s it deserves.

branches: 1.36.2; 1.36.4; 1.36.6;
Correct the ri_xorigin calculation for the RI_CENTER case again.

Original calculation (bits += (ri_stride - ri_emustride) / 2) was
incorrect because stride may be wider than visible width.

Fix in 1.33 (bits += (ri_width - ri_emustride) / 2) was incorrect
because units do not match; "bits" and "ri_emustride" are in bytes,
but "width" is in pels. Works by accident for 8bpp displays.

Change to bits += ((ri_width * bpp / 8) - ri_emustride) / 2
to correctly account for visible width and bpp.

Now does character remapping depending on font encoding.

Rasops supports 4bit depth.

Fix an error in xoffset calculation. Revealed in the case when
ri_width is less than ri_stride and screen is layouted RI_CENTERed.

Use my proper name.

Let this build on LP64 if DEBUG is defined.

branches: 1.30.2;
* Spell shift counts in decimal and masks in hex
* Optimize numerous array references
* Cleanup whitespace turds

Advance index into rasops_cmap for each color, not per triplet.
This makes highlighing and color possible on truecolor displays.

Fix computation of ri_xorigin in RI_CENTER case; convery from bytes to
pixels, not from bytes to bytes^2/pixel.

Resolve LP64 issues.

replace WSFONT_L2R by WSDISPLAY_FONTORDER_L2R to compile again.

Dispatch another nit.

-use the right namespace for screen capabilities
-make attribute decomposing a bit more friendly if the caller doesn't
care about underlines

-initialize the colormap completely at compile time, to allow drivers
to use it early
-now we can declare it "const" (as "rasops_isgray[]", while we are here)
-don't use the fg/bg colors in ...alloc_attr() if the WSATTR_WSCOLORS
flag was not given - use reasonable defaults instead
-add an opaque "ri_hw" member to "rasops_info", for driver use

Possible typo.

In rasops_do_cursor(), don't pull the mask from ri_devcmap[], just use ~0.
Needed to make cursor DTRT on NetBSD/hpcmips - from takemura.

- New option (RASOPS_SMALL) for the tight-fisted.
- Don't use int32_t/u_int32_t unless we must.
- Remove C++ single line comment delimeters that crept in.
- Remove defs pertaining to byte granularity 'ragged-edge' bitmasks.
- Move all declarations of per-depth initialization functions to rasops.h.
- Other minor cleanup.

branches: 1.19.2; 1.19.4; 1.19.6;
Use proper mask when RI_FORCEMONO is set and XORing cursor.

- Collapse forward and reverse cases in rasops_copyrows() into one
- Pull in opt_rasops.h so we know if clipping is enabled
- Some KNF

Clean up attribute allocation some more.

Small consistency nit.

Make monochrome attribte allocation a bit more obvious.

When clearing the entire display, point to the start of the framebuffer,
not start of console output. Also, do not advance by ri_delta every row.

- Change the semantics of rasops_init slightly
- Re-name rasops_setfont to rasops_reconfig
- Add some new run-time 'features'
- Bring all run-time 'features' under control of rasops_info::ri_flg
- Some cosmetic changes

- Style nits
- Kill some of the dainbramage in variable-depth copycols()

branches: 1.11.2;
- Don't even try to emulate WSATTR_HILIT on mono displays
- WSATTR_REVERSE is a capability of both mono/color

Fix stupid alignment bug. The 'slop' stuff is due for replacement with masks
RSN per the README.

Assign ownership & copyright to TNF. There is probably a procedure for this
that I am unaware of. Also some KNF.

Add diagnostic measures to check for missing font.

wsfont cookies <= 0 are invalid, not just those < 0.

In rasops_erasecols(), try word aligned case before halfword aligned.

Add cookie for wsfont to rasops_info.

Innumerable fixes and improvements. README contains list of remaining bugs
that need to be squashed.

Code cleanup. Add ri_bswap member to 'struct rasops_info' to indicate that
framebuffer endianness differs from CPU (for 15,16,32-bit displays).

Fixed licensing due to a foobar on my behalf.

Initial import of 'rasops', the new raster operations set for wscons/rcons.

Remove pointless cast; dp is already uint32_t *.
No functional changes intended.

Misc style clean up's.
- Introduce and use proper macros.
- Use not ambiguous variable names.
- Unify similar functions as possible as I can.
- G/C unused headers.
- Use #include <dev/rasops/foo.h> instead of "foo.h"
No particular functional changes intended.

Set 4-bpp devcmap in a similar manner to non-RGB case of 8-bpp.
No functional changes since this is not in use (4-bpp is monochrome).

Correctly check whether character is in font in rasops_mapchar().
Also, make sure that in putchar functions for completeness.

Fix color range overflow; we cannot make bright colors more brighter.

When legacy Apple 4-bpp color palette is used, make green dark so that
kernel messages are printed nicely on white background.

Scaling dimensions of underline by font height.

Currently,
- offset of underline is fixed to 1-row from bottom of characters, and
- height of underline is fixed to 1.

Both are good for standard 8x16 fonts. However, it is too thin for
larger fonts, especially when used on display of high resolution.

Also, 1-row offset of underline is ugly for small fonts, e.g.,
spleen5x8.

Therefore, adjust offset and height as,
- no changes for standard 16-height fonts.
- scaling by font height for larger fonts.
- set offset to zero for fonts of height smaller than 16.

If RI_CLEAR is set, do not forget to clear real framebuffer.

Stop allocating ri_buf and ri_stamp dynamically. As commented in
rasops.h, it is not safe to use kmem_alloc(9) in rasops_init();
rasops routines can be used for early putchar, which means that
UVM is not fully initialized.

Should fix a problem reported by macallan:
http://mail-index.netbsd.org/tech-kern/2019/08/02/msg025327.html

Instead of using ri_buf, inline function rasops_memcpy32() is
introduced to fill 32bit data efficiently.

Instead of using ri_stamp (per device stamp), stamp_ri is
introduced to distinguish for which device stamp is calculated.

Simplify rasops_do_cursor():

- Use static masks similar to that used in rasops_bitops.h,
rather than generating them on the fly.
- Use pointer for proper type to avoid unnecessary casts.

Fix a critical bug for rasops_copyrows() introduced in rev. 1.90:
http://cvsweb.netbsd.org/bsdweb.cgi/src/sys/dev/rasops/rasops.c#rev1.90

When src < dst, we have to copy backward.

Depth 2 is monochrome.

IMO, it is impossible to support ANSI colors on 2-bpp display; fore-
and background can be same value even if they are different colors
logically.

Fix black color-attribution for depths 2 and 4.

Simplify calculation for 12-byte alignment.
No functional changes.

Use _KERNEL_OPT.

Real fix for 24-bpp color:
- When centering screen, locate effective base address of framebuffer to
both word and 24-bit color boundary.
- Consistently convert ri_devcmap to ``big endian'' if not RI_BSWAP.

Also, fix possible bug for 15/16-bpp with RI_BSWAP (not tested).

Fix unaligned writes to buffer, that are introduced in 1.105:
http://cvsweb.netbsd.org/bsdweb.cgi/src/sys/dev/rasops/rasops.c#rev1.105

Oops, for rasops_copycols(), we cannot use memmove even if src == dst.
On the other hand, memmove is safe for rasops_copyrows().

Provide buffer capable of single-row pixels in order to make things simpler.

XXX
Bump kernel version for rasops_info later.

G/C ri_delta.

XXX
Bump kernel version after other changes for struct rasops_info.

Switch to per-device stamp, and retire stamp_mutex.

XXX
Bump kernel version after other changes for struct rasops_info.

Misc clean-up's:
- protect private stuff in rasops.h by _RASOPS_PRIVATE
- staticify rasops_copycols() and rasops_isgray[]
- G/C unused extern int cold

branches: 1.101.2;
Try to improve performance when shadow framebuffer is present;
Use block copy from shadow fb to real fb, instead of repeating
the same operations to two fb's.

In rasops_copy{rows,cols}(), if src == dst, we have nothing to do.
Otherwise, we can use memcpy safely instead of memmove.

Fix catastrophe when ri_emustride != ri_stride in rasops_copyrows().

Treat highlighted and reversed text in the same manner to xterm.

Hmmm, color was still strange for 24bpp on little endian machine,
only when font width is 12.

We need to use different devcmap for that case, if we wish to share
codes for other depths/font widths as possible as we can.

XXX
What should we do for big endian? I have no big endian machines with
24bpp framebuffer...

Fix color on 24bpp screen for little endian.

Fix missing underlines on mono screen.
Style.

Cast attr to uint32_t before right shift to avoid undefined behavior.

Also, misc style/cosmetic changes for clarity.

Free kmem_alloc'd memory with kmem_free

Put back byte-wise copy to stop using memcpy, which
does not work for device memory on some platforms.

Pointed out by Jared, many thanks!

Add genfb(4) driver for mac68k grfbus.
1, 2, 4, 8, 15, and 32 color-depths are supported.

ANSI colors on console are functional for depth >= 4.

Graphic applications based on wsfb API should work, provided
they can correctly handle fbi_fboffset and your color depth.

wsfb driver of Xorg 1.20 and mlterm-fb (framebuffer version of
x11/mlterm) work fine for depth == 1 || depth >= 8.

For depth == 8 (256 colors), graphic applications require
colormap callback, which is currently supported for some
internal graphic adapters, and only Civic (found on Quadra
AV series) was tested. Register definition and its usage are
taken from Linux. You can easily support other adapters if
Linux supports that.

Have fun, and any problem reports are welcomed!

Replace manually unrolled loops with memcpy/memmove or simple loops.
Modern compilers are smart enough; there's no measurable changes in
performance even on MC68040 with optimization level -Os.

Also, convert loop of byte-wise copy into memset.

Misc creen up for rasops.c:

- sort headers
- return error value instead of panic
- use uintptr_t for cast pointer to integer
- use macros appropriately
- use __func__
- some consistency check ifndef DEBUG
- try to avoid undefined behaviors related to shift
- convert malloc/free to kmem_alloc/kmem_free
- convert MIN to uimin
- style

Misc cleen up:
- Make 32bit mask unsigned
- DPRINTF --> __nothing ifndef DEBUG_RASOPS
- "#ifdef DIAGNOSTIC if (x) panic(); #endif" --> KASSERT(!x);
- KNF

No functional changes intended.

For kUBSan, avoid undefined behaviors even if harmless.
No functional changes intended.

Requested by uwe.

Oops, revert unintentional changes.

Well, masks do not have to be updated every time in loop.
Calculate them in advance even if it may be bogus.

Simplify logic and tiny clean-up.

Avoid shift-count overflow to fix strange cursor behaviors on amd64
when font width is odd.

Use unsigned integers for binary data storage.
No functional changes intended.

Style:
- u_char --> uint8_t
- u_int*_t --> uint*_t

No functional changes.

Fix cursor movement for ri_xscale = 1, e.g., fontwidth = 8 and bpp = 1.

rasops reused wscons attribute bits for internal control.
- make upper 4 attribute bits available for such use
- use wscons flag names instead of literal constants.

rasops_do_cursor():
- simplify & sanitize the unaligned case
- use only 32bit accesses
... no longer crash with odd sized fonts in 8bit

branches: 1.77.8; 1.77.10;
remove checks for failure after memory allocation calls that cannot fail:

kmem_alloc() with KM_SLEEP
kmem_zalloc() with KM_SLEEP
percpu_alloc()
pserialize_create()
psref_class_create()

all of these paths include an assertion that the allocation has not failed,
so callers should not assert that again.

rasops_reconfig(): don't clear flags we won't set here

branches: 1.75.2;
add RI_PREFER_ALPHA flag, for drivers that can draw such fonts by hardware

reset ri->ri_hwbits too if RI_CFGDONE is set.

prevent ri->ri_hwbits from moving to center every time rasops_reconfig() is
called when RI_CENTER is set.

branches: 1.73.2; 1.73.4;
add "best match" algorithm to wsfont and use this instead of a private
function in rasops.

branches: 1.72.2;
Don't leak f on failure. Noted by maxv@.

Compile-tested only, with zaurus.