From mboxrd@z Thu Jan  1 00:00:00 1970
From: Ben Skeggs <skeggsb-Re5JQEeQqe8AvxtiuMwx3w@public.gmane.org>
Subject: Re: [PATCH] Import EVO commands into envytools
Date: Mon, 10 Oct 2011 09:21:23 +1000
Message-ID: <1318202483.2533.17.camel@nisroch>
References: <1318200004-17439-1-git-send-email-maximlevitsky@gmail.com>
	<1318200004-17439-2-git-send-email-maximlevitsky@gmail.com>
Reply-To: skeggsb-Re5JQEeQqe8AvxtiuMwx3w@public.gmane.org
Mime-Version: 1.0
Content-Type: text/plain; charset="us-ascii"
Content-Transfer-Encoding: 7bit
Return-path: <nouveau-bounces+gcfxn-nouveau=m.gmane.org-PD4FTy7X32lNgt0PjOBp9y5qC8QIuHrW@public.gmane.org>
In-Reply-To: <1318200004-17439-2-git-send-email-maximlevitsky-Re5JQEeQqe8AvxtiuMwx3w@public.gmane.org>
List-Unsubscribe: <http://lists.freedesktop.org/mailman/options/nouveau>,
	<mailto:nouveau-request-PD4FTy7X32lNgt0PjOBp9y5qC8QIuHrW@public.gmane.org?subject=unsubscribe>
List-Archive: <http://lists.freedesktop.org/archives/nouveau>
List-Post: <mailto:nouveau-PD4FTy7X32lNgt0PjOBp9y5qC8QIuHrW@public.gmane.org>
List-Help: <mailto:nouveau-request-PD4FTy7X32lNgt0PjOBp9y5qC8QIuHrW@public.gmane.org?subject=help>
List-Subscribe: <http://lists.freedesktop.org/mailman/listinfo/nouveau>,
	<mailto:nouveau-request-PD4FTy7X32lNgt0PjOBp9y5qC8QIuHrW@public.gmane.org?subject=subscribe>
Sender: nouveau-bounces+gcfxn-nouveau=m.gmane.org-PD4FTy7X32lNgt0PjOBp9y5qC8QIuHrW@public.gmane.org
Errors-To: nouveau-bounces+gcfxn-nouveau=m.gmane.org-PD4FTy7X32lNgt0PjOBp9y5qC8QIuHrW@public.gmane.org
To: Maxim Levitsky <maximlevitsky-Re5JQEeQqe8AvxtiuMwx3w@public.gmane.org>
Cc: nouveau-PD4FTy7X32lNgt0PjOBp9y5qC8QIuHrW@public.gmane.org
List-Id: nouveau.vger.kernel.org

On Mon, 2011-10-10 at 00:40 +0200, Maxim Levitsky wrote:
> ---
>  rnndb/nv_evo.xml |  311 ++++++++++++++++++++++++++++++++++++++++++++++++++++++
>  1 files changed, 311 insertions(+), 0 deletions(-)
>  create mode 100644 rnndb/nv_evo.xml
Nice work, comments inlined where appropriate below.

> 
> diff --git a/rnndb/nv_evo.xml b/rnndb/nv_evo.xml
> new file mode 100644
> index 0000000..d776dbd
> --- /dev/null
> +++ b/rnndb/nv_evo.xml
> @@ -0,0 +1,311 @@
> +<?xml version="1.0" encoding="UTF-8"?>
> +<database xmlns="http://nouveau.freedesktop.org/"
> +	  xmlns:xsi="http://www.w3.org/2001/XMLSchema-instance"
> +	  xsi:schemaLocation="http://nouveau.freedesktop.org/ rules-ng.xsd">
> +<import file="copyright.xml"/>
> +<import file="nvchipsets.xml"/>
> +<import file="nv_defs.xml"/>
> +
> +<!-- main EVO channel -->
> +<domain name="NV_EVO_MASTER" bare="yes" prefix="chipset">
> +
> +	<!-- write 0 here executes all evo commands pending in PDISPLAY fifo -->
> +	<reg32 offset="0x80" name="FLUSH_COMMANDS"/>
Perhaps just call it FLUSH?

> +
> +	<reg32 offset="0x84" name="FLUSH2">
> +		<bitfield pos="2" name="FLUSH_CLUT_CRTC0"/>
> +		<bitfield pos="3" name="FLUSH_CLUT_CRTC1"/>
> +		<bitfield pos="30" name="NOTIFY_ENABLED"/> <!--???-->
> +		<bitfield pos="31" name="FLUSH_MAIN"/>
> +	</reg32>
> +
> +	<!-- dma object that PDISPLAY needs when it uses display sync channels
> +	     maybe that stores its state to
> +	     when PDISPLAY executes non-main channel?...-->
> +	<reg32 offset="0x88" name="DMA_NOTIFY"/>
The documentation isn't correct here, but I'm not sure what to write.
Nouveau currently uses it for synchronisation, to know when the entire
set of queued commands is completely finished, nothing technically to do
with the sync channels however.

The binary driver uses this in some other ways too, IIRC some magic with
0x8c fills the buffer with data that they read back.  But, could be
misremembering.  A valgrind trace will show what I mean though.

> +
> +	<array name="DAC" offset="0x400" stride="0x80" length="2">
> +
> +		<reg32 offset="0x0" name="MODE_CTL">
> +			<bitfield pos="0" name="CRTC0"/>
> +			<bitfield pos="1" name="CRTC1"/>
> +			<bitfield pos="6" name="ENABLED"/>
This is *not* ENABLED, nouveau doesn't even set it.

> +
> +			<bitfield low="8" high="13"  name="PROTOCOL">
> +				<value name="VGA"     value="0x0"/>
> +				<value name="NTSC-M"  value="0x1"/>
> +				<value name="NTSC-J"  value="0x2"/>
> +				<value name="PAL"     value="0x3"/>
> +				<value name="PAL-M"   value="0x4"/>
> +				<value name="PAL-N"   value="0x5"/>
> +				<value name="PAL-NC"  value="0x6"/>
> +
> +				<!-- HD480I/HD576I doesn't work with blob,
> +				     maybe hw doesn't support it? -->
> +				<value name="HD480P"  value="0xd"/>
> +				<value name="HD576p"  value="0xe"/>
> +
> +				<value name="HD720p"  value="0x10"/>
> +				<value name="HD1080i" value="0x12"/>
> +				<value name="HD1080p" value="0x3F"/>
> +			</bitfield>
> +
> +			<bitfield pos="14" name="SUB_CARR_FREQ">
> +				<value name="3_57MHZ"  value="0"/>
> +
> +				<!-- PAL and PAL-NC use this -->
> +				<value name="4_42MHZ" value="1"/>
> +			</bitfield>
> +		</reg32>
> +
> +		<reg32 offset="0x04" name="MODE_CTL2">
> +			<bitfield pos="0" name="NEG_HSYNC" type="boolean"/>
> +			<bitfield pos="1" name="NEG_VSYNC" type="boolean"/>
> +		</reg32>
> +
> +		<!-- TODO: always set to 0x10000 on TV-out -->
> +		<reg32 offset="0x20" name="UNK20"/>
> +	</array>
> +
> +	<array name="SOR" offset="0x600" stride="0x40" length="2">
> +
> +		<reg32 offset="0x0" name="MODE_CTL">
> +			<bitfield pos="0" name="CRTC0"/>
> +			<bitfield pos="1" name="CRTC1"/>
> +
> +			<bitfield pos="6" name="ENABLED"/>
Again, can't be ENABLED.

> +
> +			<bitfield low="8" high="11"  name="PROTOCOL">
> +				<value name="LVDS" value="0"/>
> +				<value name="TMDS_SINGLE_A" value="1"/>
> +				<value name="TMDS_SINGLE_B" value="2"/>
> +				<value name="TMDS_DUAL" value="5"/>
> +				<value name="DP_A" value="8"/>
> +				<value name="DP_B" value="9"/>
> +			</bitfield>
> +
> +			<bitfield pos="12" name="NEG_HSYNC" type="boolean"/>
> +			<bitfield pos="13" name="NEG_VSYNC" type="boolean"/>
> +		</reg32>
> +	</array>
> +
> +
> +	<array name="CRTC" offset="0x800" stride="0x400" length="2">
> +
> +		<reg32 offset="0x004" name="CLOCK">
> +			<bitfield low="0" high="22" name="CLK_KHZ" type="uint"/>
Maybe FREQ_KHZ?

> +			<bitfield pos="23" name="ENABLED"/>
> +		</reg32>
> +
> +		<reg32 offset="0x008" name="INTERLACE">
> +			<bitfield pos="1" name="ENABLED"/>
> +		</reg32>
> +
> +		<reg32 offset="0x010" name="BACKGROUND_COLOR">
> +			<bitfield low="0" high="9" name="RED" type="uint"/>
> +			<bitfield low="10" high="19" name="GREEN" type="uint"/>
> +			<bitfield low="20" high="29" name="BLUE" type="uint"/>
> +		</reg32>
BORDER_COLOUR might be more descriptive.

> +
> +		<!-- start of sync pulse is the reference, so: -->
> +
> +		<!-- totat scanlines of display -->
> +		<reg32 offset="0x014" name="DISPLAY_TOTAL" type="wh16"/>
> +
> +		<!-- from start of sync to end of sync -->
> +		<reg32 offset="0x018" name="SYNC_DURATION" type="wh16"/>
> +
> +		<!-- from start of sync to start of visible portion (end of back porch) -->
> +		<reg32 offset="0x01c" name="SYNC_START_TO_DISP_START" type="wh16"/>
> +
> +		<!-- from start of sync to end of visible portion (start of front porch) -->
> +		<reg32 offset="0x020" name="SYNC_START_TO_DISP_END" type="wh16"/>
> +
> +
> +		<reg32 offset="0x024" name="INTERLACE_TIMINGS">
> +			<!-- from start of sync to end of visible portion of next field (start of front porch) -->
> +			<bitfield low="0" high="15" name="VSYNC_TO_END_DISP_NEXTFIELD" type="uint" />
> +			<!-- from start of sync to start of visible portion of the next field (end of back porch) -->
> +			<bitfield low="16" high="31" name="VSYNC_TO_START_DISP_NEXTFIELD" type="uint" />
> +		</reg32>
> +
> +		<!-- TODO: (!) very related to scale/mode. Large values cause flicker on LVDS/DVI/VGA, clock?-->
> +		<reg32 offset="0x028" name="UNK828">
> +			<bitfield low="0" high="11" name="UNK" type="uint" />
> +		</reg32>
> +
> +		<reg32 offset="0x02C" name="UNK82C"/> <!-- set always to 0 -->
> +
> +
> +		<reg32 offset="0x040" name="CLUT">
> +			<bitfield pos="31" name="ENABLED">
> +				<value value="1" name="YES"/>
> +				<value value="0" name="NO"/>
> +			</bitfield>
> +
> +			<!-- in reduced mode, first and last 64 entries of LUT
> +			     are ignored, and middle 128 entries are interpolated
> +			     into 256 entries. Used by blob for 8-bit framebuffers -->
> +			<bitfield pos="30" name="MODE">
> +				<value value="0" name="REDUCED"/>
> +				<value value="1" name="NORMAL"/>
> +			</bitfield>
> +		</reg32>
> +
> +		<reg32 offset="0x044" name="CLUT_ADDRESS"/>
> +		<reg32 offset="0x05c" name="CLUT_DMA_HANDLE" variants="NV84-"/>
> +
> +		<group name="nv50_evo_fb_settings">
> +			<reg32 offset="0x00" name="FB_OFFSET"/>
> +
> +			<reg32 offset="0x08" name="FB_SIZE"/>
> +
> +			<reg32 offset="0x0c" name="FB_CONFIG">
> +				<bitfield pos="20" name="MODE">
> +					<value value="0" name="TILE"/>
> +					<value value="1" name="PITCH"/>
> +				</bitfield>
> +
> +				<bitfield high="19" low="4" shr="4" name="PITCH"/>
> +				<bitfield high="3" low="0" name="TILE_FLAGS"/>
> +			</reg32>
> +			<reg32 offset="0x010" name="FB_DEPTH" >
> +				<bitfield high="15" low="8" name="DEPTH">
> +					<value value="0x1e" name="8BPP"/>
> +					<value value="0xe9" name="15BPP"/>
> +					<value value="0xe8" name="16BPP"/>
> +					<value value="0xcf" name="24BPP"/>
> +					<value value="0xd1" name="30BPP"/>
We could use the same names here as in NV50_SURFACE_FORMAT in rnndb.  I
don't have any idea what formats aside from these that EVO accepts, it
could be worth checking too.

> +				</bitfield>
> +			</reg32>
> +		</group>
> +
> +		<stripe offset="0x060">
> +			<use-group name="nv50_evo_fb_settings" />
> +		</stripe>
> +
> +		<reg32 offset="0x074" name="FB_DMA_HANDLE"/>
> +
> +
> +		<reg32 offset="0x080" name="CURSOR_CTL">
> +			<bitfield pos="31" name="SHOW" type="boolean"/>
> +		</reg32>
> +		<reg32 offset="0x084" name="CURSOR_ADDRESS"/>
> +		<reg32 offset="0x09c" name="CURSOR_DMA_HANDLE" variants="NV84-"/>
> +
> +		<reg32 offset="0x0a0" name="DITHER_CTRL">
> +			<bitfield low="0" high="2" name="DEPTH">
> +				<value value="0" name="DISABLED"/>
> +				<value value="1" name="6BPP"/>
> +				<value value="3" name="8BPP"/>
> +			</bitfield>
> +
> +			<bitfield low="3" high="5" name="MODE">
> +				<value value="0" name="OFF"/>
> +				<value value="3" name="STATIC_2X2"/>
> +				<value value="2" name="DYNAMIC_2X2"/>
Perhaps reorder.  Also, is "4" accepted as a value here on your card?
It's TEMPORAL on NVD9.

> +			</bitfield>
> +		</reg32>
> +
> +		<reg32 offset="0x0a4" name="BLUR_CTRL">
> +			<!-- BLUR_ENABLE + UNK1 = black screen + hung card,
> +			     no way to fix by setting back the old value -->
> +			<bitfield low="0" high="1" name="UNK1"/>
> +			<bitfield pos="2" name="ENABLE"/>
> +			<bitfield pos="3" name="UNK3"/>
> +
> +			<!-- set to same values as BLUR1, but seems to do nothing -->
> +			<bitfield low="16" high="23" name="BLUR2" type="int"/>
> +
> +			<!-- Blur value, (negative == sharpness) -->
> +			<bitfield low="24" high="31" name="BLUR1" type="int"/>
> +		</reg32>
> +
> +		<reg32 offset="0x0a8" name="COLOR_CTL">
> +			<!-- TV_COLORS used on most TV modes,
> +			     applies proper black/white levels -->
> +			<bitfield pos="0" name="TV_COLORS"/>
> +			<!-- ALT_TV_COLORS used in PAL-N and 720P/1080P/1080I modes,
> +			     slightly different colors that with TV_COLORS -->
> +			<bitfield pos="1" name="ALT_TV_COLORS"/>
> +
> +			<!-- draws on left and right of the screen two
> +			     thin/thick violet lines -->
> +			<bitfield pos="3" name="LEFTRIGHT_THIN_BORDER"/>
> +			<bitfield pos="4" name="LEFTRIGHT_THICK_BORDER"/>
> +
> +			<bitfield  low="8" high="19" name="SATURATION" type="int"/>
> +			<bitfield  low="20" high="31" name="TINT" type="int"/>
> +		</reg32>
> +
> +		<!-- offset of CRTC window in framebuffer texture -->
> +		<reg32 offset="0x0c0" name="FB_POS" type="xy16"/>
> +
> +		<!-- size of CRTC window in framebuffer texture -->
> +		<reg32 offset="0x0c8" name="FB_REAL_RES" type="wh16"/>
> +
> +		<!-- offset to move CRTC window before output to physical screen -->
> +		<reg32 offset="0x0d4" name="FB_SCREEN_POS" type="xy16"/>
> +
> +		<!-- size to scale CRTC window (which size is given in FB_REAL_RES)
> +		     before output to physical screen -->
> +		<reg32 offset="0x0d8" name="FB_SCALE_RES1" type="wh16" />
> +		<!-- this value doesn't do anything, but must be same/lower that first
> +			its sort of bottom limit/optimization. 0 seems to work -->
> +		<reg32 offset="0x0dc" name="FB_SCALE_RES2" type="wh16" />
> +
> +		<!-- darktama: required to make display sync channels not hate life.
> +		     UNK900 set to 0, UNK904 to 0x100 by blob -->
> +		<reg32 offset="0x100" name="UNK900"/>
> +		<reg32 offset="0x104" name="UNK904"/>
> +
> +		<!-- blob seeems to set that to 0x10000 -->
> +		<reg32 offset="0x3a8" name="UNK3A8"/>
> +
> +	</array>
> +</domain>
> +
> +<!-- display sync EVO channels, one per CRTC -->
> +<domain name="NV_EVO_SYNC" bare="yes" prefix="chipset">
> +
> +	<!-- same as on main EVO channel -->
> +	<reg32 offset="0x80" name="FLUSH_COMMANDS"/>
Just FLUSH?

> +
> +	<!-- TODO: different values depends on whenever sync semaphore is used -->
> +	<reg32 offset="0x84" name="SYNC_CTRL">
> +		<bitfield pos="4" name="NOT_USE_SEM"/>
This is actually 7:4 (IIRC), and should be called SWAP_INTERVAL

> +		<bitfield pos="8" name="USE_SEM"/>
Not 100% sure what this actually is, but it was required in some way to
completely disable waiting for vblank before flipping.  I'd probably
call it "IMMEDIATE" or something.

> +	</reg32>
> +
> +	<!-- Semaphore support to sync against PGRAPH -->
> +	<reg32 offset="0x88" name="SEM_OFFSET"/>
> +	<reg32 offset="0x8c" name="SEM_WAIT_VALUE"/>
SEM_ACQUIRE

> +	<reg32 offset="0x90" name="SEM_WRITE_VALUE"/>
SEM_RELEASE

> +	<reg32 offset="0x94" name="SEM_DMAOBJ"/>
> +
> +	<!-- 0 written on pageflip-->
> +	<reg32 offset="0xa0" name="UNKA0"/>
> +	<reg32 offset="0xa5" name="UNKA4"/>
> +
> +
> +	<reg32 offset="0xc0" name="FB_DMAOBJ"/>
> +
> +	<!-- darktama: allows gamma somehow, PDISP will bitch at you if
> +		* you don't wait for vblank before changing this..
> +		nouveau write 0x40000000 here-->
> +	<reg32 offset="0xe0" name="UNKe0"/>
> +
> +	<!-- set to 0xfffe0000 on page flip -->
> +	<reg32 offset="0x100" name="UNK100"/>
> +
> +	<!-- 0 written on pageflip -->
> +	<reg32 offset="0x110" name="UNK110"/>
> +	<reg32 offset="0x114" name="UNK114"/>
> +
> +	<!-- new framebuffer settings to apply. Same as in main evo channel-->
> +	<stripe offset="0x800">
> +		<use-group name="nv50_evo_fb_settings" />
> +	</stripe>
> +</domain>
> +
> +</database>