Skip to content

Commit 55fd7e0

Browse files
rodrigovivijnikula
authored andcommitted
drm/i915: Include asm sources for {ivb, hsw}_clear_kernel.c
Alexandre Oliva has recently removed these files from Linux Libre with concerns that the sources weren't available. The sources are available on IGT repository, and only open source tools are used to generate the {ivb,hsw}_clear_kernel.c files. However, the remaining concern from Alexandre Oliva was around GPL license and the source not been present when distributing the code. So, it looks like 2 alternatives are possible, the use of linux-firmware.git repository to store the blob or making sure that the source is also present in our tree. Since the goal is to limit the i915 firmware to only the micro-controller blobs let's make sure that we do include the asm sources here in our tree. Btw, I tried to have some diligence here and make sure that the asms that these commits are adding are truly the source for the mentioned files: igt$ ./scripts/generate_clear_kernel.sh -g ivb \ -m ~/mesa/build/src/intel/tools/i965_asm Output file not specified - using default file "ivb-cb_assembled" Generating gen7 CB Kernel assembled file "ivb_clear_kernel.c" for i915 driver... igt$ diff ~/i915/drm-tip/drivers/gpu/drm/i915/gt/ivb_clear_kernel.c \ ivb_clear_kernel.c < * Generated by: IGT Gpu Tools on Fri 21 Feb 2020 05:29:32 AM UTC > * Generated by: IGT Gpu Tools on Mon 08 Jun 2020 10:00:54 AM PDT 61c61 < }; > }; \ No newline at end of file igt$ ./scripts/generate_clear_kernel.sh -g hsw \ -m ~/mesa/build/src/intel/tools/i965_asm Output file not specified - using default file "hsw-cb_assembled" Generating gen7.5 CB Kernel assembled file "hsw_clear_kernel.c" for i915 driver... igt$ diff ~/i915/drm-tip/drivers/gpu/drm/i915/gt/hsw_clear_kernel.c \ hsw_clear_kernel.c 5c5 < * Generated by: IGT Gpu Tools on Fri 21 Feb 2020 05:30:13 AM UTC > * Generated by: IGT Gpu Tools on Mon 08 Jun 2020 10:01:42 AM PDT 61c61 < }; > }; \ No newline at end of file Used IGT and Mesa master repositories from Fri Jun 5 2020) IGT: 53e8c878a6fb ("tests/kms_chamelium: Force reprobe after replugging the connector") Mesa: 5d13c7477eb1 ("radv: set keep_statistic_info with RADV_DEBUG=shaderstats") Mesa built with: meson build -D platforms=drm,x11 -D dri-drivers=i965 \ -D gallium-drivers=iris -D prefix=/usr \ -D libdir=/usr/lib64/ -Dtools=intel \ -Dkulkan-drivers=intel && ninja -C build v2: Header clean-up and include build instructions in a readme (Chris) Modified commit message to respect check-patch Reference: http://www.fsfla.org/pipermail/linux-libre/2020-June/003374.html Reference: http://www.fsfla.org/pipermail/linux-libre/2020-June/003375.html Fixes: 47f8253 ("drm/i915/gen7: Clear all EU/L3 residual contexts") Cc: <[email protected]> # v5.7+ Cc: Alexandre Oliva <[email protected]> Cc: Prathap Kumar Valsan <[email protected]> Cc: Akeem G Abodunrin <[email protected]> Cc: Mika Kuoppala <[email protected]> Cc: Chris Wilson <[email protected]> Cc: Jani Nikula <[email protected]> Cc: Joonas Lahtinen <[email protected]> Signed-off-by: Rodrigo Vivi <[email protected]> Reviewed-by: Jon Bloomfield <[email protected]> Link: https://patchwork.freedesktop.org/patch/msgid/[email protected] (cherry picked from commit 5a7eeb8) Signed-off-by: Jani Nikula <[email protected]>
1 parent cd65bbb commit 55fd7e0

File tree

3 files changed

+282
-0
lines changed

3 files changed

+282
-0
lines changed
Lines changed: 46 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -0,0 +1,46 @@
1+
ASM sources for auto generated shaders
2+
======================================
3+
4+
The i915/gt/hsw_clear_kernel.c and i915/gt/ivb_clear_kernel.c files contain
5+
pre-compiled batch chunks that will clear any residual render cache during
6+
context switch.
7+
8+
They are generated from their respective platform ASM files present on
9+
i915/gt/shaders/clear_kernel directory.
10+
11+
The generated .c files should never be modified directly. Instead, any modification
12+
needs to be done on the on their respective ASM files and build instructions below
13+
needes to be followed.
14+
15+
Building
16+
========
17+
18+
Environment
19+
-----------
20+
21+
IGT GPU tool scripts and the Mesa's i965 instruction assembler tool are used
22+
on building.
23+
24+
Please make sure your Mesa tool is compiled with "-Dtools=intel" and
25+
"-Ddri-drivers=i965", and run this script from IGT source root directory"
26+
27+
The instructions bellow assume:
28+
* IGT gpu tools source code is located on your home directory (~) as ~/igt
29+
* Mesa source code is located on your home directory (~) as ~/mesa
30+
and built under the ~/mesa/build directory
31+
* Linux kernel source code is under your home directory (~) as ~/linux
32+
33+
Instructions
34+
------------
35+
36+
~ $ cp ~/linux/drivers/gpu/drm/i915/gt/shaders/clear_kernel/ivb.asm \
37+
~/igt/lib/i915/shaders/clear_kernel/ivb.asm
38+
~ $ cd ~/igt
39+
igt $ ./scripts/generate_clear_kernel.sh -g ivb \
40+
-m ~/mesa/build/src/intel/tools/i965_asm
41+
42+
~ $ cp ~/linux/drivers/gpu/drm/i915/gt/shaders/clear_kernel/hsw.asm \
43+
~/igt/lib/i915/shaders/clear_kernel/hsw.asm
44+
~ $ cd ~/igt
45+
igt $ ./scripts/generate_clear_kernel.sh -g hsw \
46+
-m ~/mesa/build/src/intel/tools/i965_asm
Lines changed: 119 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -0,0 +1,119 @@
1+
// SPDX-License-Identifier: MIT
2+
/*
3+
* Copyright © 2020 Intel Corporation
4+
*/
5+
6+
/*
7+
* Kernel for PAVP buffer clear.
8+
*
9+
* 1. Clear all 64 GRF registers assigned to the kernel with designated value;
10+
* 2. Write 32x16 block of all "0" to render target buffer which indirectly clears
11+
* 512 bytes of Render Cache.
12+
*/
13+
14+
/* Store designated "clear GRF" value */
15+
mov(1) f0.1<1>UW g1.2<0,1,0>UW { align1 1N };
16+
17+
/**
18+
* Curbe Format
19+
*
20+
* DW 1.0 - Block Offset to write Render Cache
21+
* DW 1.1 [15:0] - Clear Word
22+
* DW 1.2 - Delay iterations
23+
* DW 1.3 - Enable Instrumentation (only for debug)
24+
* DW 1.4 - Rsvd (intended for context ID)
25+
* DW 1.5 - [31:16]:SliceCount, [15:0]:SubSlicePerSliceCount
26+
* DW 1.6 - Rsvd MBZ (intended for Enable Wait on Total Thread Count)
27+
* DW 1.7 - Rsvd MBZ (inteded for Total Thread Count)
28+
*
29+
* Binding Table
30+
*
31+
* BTI 0: 2D Surface to help clear L3 (Render/Data Cache)
32+
* BTI 1: Wait/Instrumentation Buffer
33+
* Size : (SliceCount * SubSliceCount * 16 EUs/SubSlice) rows * (16 threads/EU) cols (Format R32_UINT)
34+
* Expected to be initialized to 0 by driver/another kernel
35+
* Layout:
36+
* RowN: Histogram for EU-N: (SliceID*SubSlicePerSliceCount + SSID)*16 + EUID [assume max 16 EUs / SS]
37+
* Col-k[DW-k]: Threads Executed on ThreadID-k for EU-N
38+
*/
39+
add(1) g1.2<1>UD g1.2<0,1,0>UD 0x00000001UD { align1 1N }; /* Loop count to delay kernel: Init to (g1.2 + 1) */
40+
cmp.z.f0.0(1) null<1>UD g1.3<0,1,0>UD 0x00000000UD { align1 1N };
41+
(+f0.0) jmpi(1) 352D { align1 WE_all 1N };
42+
43+
/**
44+
* State Register has info on where this thread is running
45+
* IVB: sr0.0 :: [15:13]: MBZ, 12: HSID (Half-Slice ID), [11:8]EUID, [2:0] ThreadSlotID
46+
* HSW: sr0.0 :: 15: MBZ, [14:13]: SliceID, 12: HSID (Half-Slice ID), [11:8]EUID, [2:0] ThreadSlotID
47+
*/
48+
mov(8) g3<1>UD 0x00000000UD { align1 1Q };
49+
shr(1) g3<1>D sr0<0,1,0>D 12D { align1 1N };
50+
and(1) g3<1>D g3<0,1,0>D 1D { align1 1N }; /* g3 has HSID */
51+
shr(1) g3.1<1>D sr0<0,1,0>D 13D { align1 1N };
52+
and(1) g3.1<1>D g3.1<0,1,0>D 3D { align1 1N }; /* g3.1 has sliceID */
53+
mul(1) g3.5<1>D g3.1<0,1,0>D g1.10<0,1,0>UW { align1 1N };
54+
add(1) g3<1>D g3<0,1,0>D g3.5<0,1,0>D { align1 1N }; /* g3 = sliceID * SubSlicePerSliceCount + HSID */
55+
shr(1) g3.2<1>D sr0<0,1,0>D 8D { align1 1N };
56+
and(1) g3.2<1>D g3.2<0,1,0>D 15D { align1 1N }; /* g3.2 = EUID */
57+
mul(1) g3.4<1>D g3<0,1,0>D 16D { align1 1N };
58+
add(1) g3.2<1>D g3.2<0,1,0>D g3.4<0,1,0>D { align1 1N }; /* g3.2 now points to EU row number (Y-pixel = V address ) in instrumentation surf */
59+
60+
mov(8) g5<1>UD 0x00000000UD { align1 1Q };
61+
and(1) g3.3<1>D sr0<0,1,0>D 7D { align1 1N };
62+
mul(1) g3.3<1>D g3.3<0,1,0>D 4D { align1 1N };
63+
64+
mov(8) g4<1>UD g0<8,8,1>UD { align1 1Q }; /* Initialize message header with g0 */
65+
mov(1) g4<1>UD g3.3<0,1,0>UD { align1 1N }; /* Block offset */
66+
mov(1) g4.1<1>UD g3.2<0,1,0>UD { align1 1N }; /* Block offset */
67+
mov(1) g4.2<1>UD 0x00000003UD { align1 1N }; /* Block size (1 row x 4 bytes) */
68+
and(1) g4.3<1>UD g4.3<0,1,0>UW 0xffffffffUD { align1 1N };
69+
70+
/* Media block read to fetch current value at specified location in instrumentation buffer */
71+
sendc(8) g5<1>UD g4<8,8,1>F 0x02190001
72+
73+
render MsgDesc: media block read MsgCtrl = 0x0 Surface = 1 mlen 1 rlen 1 { align1 1Q };
74+
add(1) g5<1>D g5<0,1,0>D 1D { align1 1N };
75+
76+
/* Media block write for updated value at specified location in instrumentation buffer */
77+
sendc(8) g5<1>UD g4<8,8,1>F 0x040a8001
78+
render MsgDesc: media block write MsgCtrl = 0x0 Surface = 1 mlen 2 rlen 0 { align1 1Q };
79+
80+
/* Delay thread for specified parameter */
81+
add.nz.f0.0(1) g1.2<1>UD g1.2<0,1,0>UD -1D { align1 1N };
82+
(+f0.0) jmpi(1) -32D { align1 WE_all 1N };
83+
84+
/* Store designated "clear GRF" value */
85+
mov(1) f0.1<1>UW g1.2<0,1,0>UW { align1 1N };
86+
87+
/* Initialize looping parameters */
88+
mov(1) a0<1>D 0D { align1 1N }; /* Initialize a0.0:w=0 */
89+
mov(1) a0.4<1>W 127W { align1 1N }; /* Loop count. Each loop contains 16 GRF's */
90+
91+
/* Write 32x16 all "0" block */
92+
mov(8) g2<1>UD g0<8,8,1>UD { align1 1Q };
93+
mov(8) g127<1>UD g0<8,8,1>UD { align1 1Q };
94+
mov(2) g2<1>UD g1<2,2,1>UW { align1 1N };
95+
mov(1) g2.2<1>UD 0x000f000fUD { align1 1N }; /* Block size (16x16) */
96+
and(1) g2.3<1>UD g2.3<0,1,0>UW 0xffffffefUD { align1 1N };
97+
mov(16) g3<1>UD 0x00000000UD { align1 1H };
98+
mov(16) g4<1>UD 0x00000000UD { align1 1H };
99+
mov(16) g5<1>UD 0x00000000UD { align1 1H };
100+
mov(16) g6<1>UD 0x00000000UD { align1 1H };
101+
mov(16) g7<1>UD 0x00000000UD { align1 1H };
102+
mov(16) g8<1>UD 0x00000000UD { align1 1H };
103+
mov(16) g9<1>UD 0x00000000UD { align1 1H };
104+
mov(16) g10<1>UD 0x00000000UD { align1 1H };
105+
sendc(8) null<1>UD g2<8,8,1>F 0x120a8000
106+
render MsgDesc: media block write MsgCtrl = 0x0 Surface = 0 mlen 9 rlen 0 { align1 1Q };
107+
add(1) g2<1>UD g1<0,1,0>UW 0x0010UW { align1 1N };
108+
sendc(8) null<1>UD g2<8,8,1>F 0x120a8000
109+
render MsgDesc: media block write MsgCtrl = 0x0 Surface = 0 mlen 9 rlen 0 { align1 1Q };
110+
111+
/* Now, clear all GRF registers */
112+
add.nz.f0.0(1) a0.4<1>W a0.4<0,1,0>W -1W { align1 1N };
113+
mov(16) g[a0]<1>UW f0.1<0,1,0>UW { align1 1H };
114+
add(1) a0<1>D a0<0,1,0>D 32D { align1 1N };
115+
(+f0.0) jmpi(1) -64D { align1 WE_all 1N };
116+
117+
/* Terminante the thread */
118+
sendc(8) null<1>UD g127<8,8,1>F 0x82000010
119+
thread_spawner MsgDesc: mlen 1 rlen 0 { align1 1Q EOT };
Lines changed: 117 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -0,0 +1,117 @@
1+
// SPDX-License-Identifier: MIT
2+
/*
3+
* Copyright © 2020 Intel Corporation
4+
*/
5+
6+
/*
7+
* Kernel for PAVP buffer clear.
8+
*
9+
* 1. Clear all 64 GRF registers assigned to the kernel with designated value;
10+
* 2. Write 32x16 block of all "0" to render target buffer which indirectly clears
11+
* 512 bytes of Render Cache.
12+
*/
13+
14+
/* Store designated "clear GRF" value */
15+
mov(1) f0.1<1>UW g1.2<0,1,0>UW { align1 1N };
16+
17+
/**
18+
* Curbe Format
19+
*
20+
* DW 1.0 - Block Offset to write Render Cache
21+
* DW 1.1 [15:0] - Clear Word
22+
* DW 1.2 - Delay iterations
23+
* DW 1.3 - Enable Instrumentation (only for debug)
24+
* DW 1.4 - Rsvd (intended for context ID)
25+
* DW 1.5 - [31:16]:SliceCount, [15:0]:SubSlicePerSliceCount
26+
* DW 1.6 - Rsvd MBZ (intended for Enable Wait on Total Thread Count)
27+
* DW 1.7 - Rsvd MBZ (inteded for Total Thread Count)
28+
*
29+
* Binding Table
30+
*
31+
* BTI 0: 2D Surface to help clear L3 (Render/Data Cache)
32+
* BTI 1: Wait/Instrumentation Buffer
33+
* Size : (SliceCount * SubSliceCount * 16 EUs/SubSlice) rows * (16 threads/EU) cols (Format R32_UINT)
34+
* Expected to be initialized to 0 by driver/another kernel
35+
* Layout :
36+
* RowN: Histogram for EU-N: (SliceID*SubSlicePerSliceCount + SSID)*16 + EUID [assume max 16 EUs / SS]
37+
* Col-k[DW-k]: Threads Executed on ThreadID-k for EU-N
38+
*/
39+
add(1) g1.2<1>UD g1.2<0,1,0>UD 0x00000001UD { align1 1N }; /* Loop count to delay kernel: Init to (g1.2 + 1) */
40+
cmp.z.f0.0(1) null<1>UD g1.3<0,1,0>UD 0x00000000UD { align1 1N };
41+
(+f0.0) jmpi(1) 44D { align1 WE_all 1N };
42+
43+
/**
44+
* State Register has info on where this thread is running
45+
* IVB: sr0.0 :: [15:13]: MBZ, 12: HSID (Half-Slice ID), [11:8]EUID, [2:0] ThreadSlotID
46+
* HSW: sr0.0 :: 15: MBZ, [14:13]: SliceID, 12: HSID (Half-Slice ID), [11:8]EUID, [2:0] ThreadSlotID
47+
*/
48+
mov(8) g3<1>UD 0x00000000UD { align1 1Q };
49+
shr(1) g3<1>D sr0<0,1,0>D 12D { align1 1N };
50+
and(1) g3<1>D g3<0,1,0>D 1D { align1 1N }; /* g3 has HSID */
51+
shr(1) g3.1<1>D sr0<0,1,0>D 13D { align1 1N };
52+
and(1) g3.1<1>D g3.1<0,1,0>D 3D { align1 1N }; /* g3.1 has sliceID */
53+
mul(1) g3.5<1>D g3.1<0,1,0>D g1.10<0,1,0>UW { align1 1N };
54+
add(1) g3<1>D g3<0,1,0>D g3.5<0,1,0>D { align1 1N }; /* g3 = sliceID * SubSlicePerSliceCount + HSID */
55+
shr(1) g3.2<1>D sr0<0,1,0>D 8D { align1 1N };
56+
and(1) g3.2<1>D g3.2<0,1,0>D 15D { align1 1N }; /* g3.2 = EUID */
57+
mul(1) g3.4<1>D g3<0,1,0>D 16D { align1 1N };
58+
add(1) g3.2<1>D g3.2<0,1,0>D g3.4<0,1,0>D { align1 1N }; /* g3.2 now points to EU row number (Y-pixel = V address ) in instrumentation surf */
59+
60+
mov(8) g5<1>UD 0x00000000UD { align1 1Q };
61+
and(1) g3.3<1>D sr0<0,1,0>D 7D { align1 1N };
62+
mul(1) g3.3<1>D g3.3<0,1,0>D 4D { align1 1N };
63+
64+
mov(8) g4<1>UD g0<8,8,1>UD { align1 1Q }; /* Initialize message header with g0 */
65+
mov(1) g4<1>UD g3.3<0,1,0>UD { align1 1N }; /* Block offset */
66+
mov(1) g4.1<1>UD g3.2<0,1,0>UD { align1 1N }; /* Block offset */
67+
mov(1) g4.2<1>UD 0x00000003UD { align1 1N }; /* Block size (1 row x 4 bytes) */
68+
and(1) g4.3<1>UD g4.3<0,1,0>UW 0xffffffffUD { align1 1N };
69+
70+
/* Media block read to fetch current value at specified location in instrumentation buffer */
71+
sendc(8) g5<1>UD g4<8,8,1>F 0x02190001
72+
render MsgDesc: media block read MsgCtrl = 0x0 Surface = 1 mlen 1 rlen 1 { align1 1Q };
73+
add(1) g5<1>D g5<0,1,0>D 1D { align1 1N };
74+
75+
/* Media block write for updated value at specified location in instrumentation buffer */
76+
sendc(8) g5<1>UD g4<8,8,1>F 0x040a8001
77+
render MsgDesc: media block write MsgCtrl = 0x0 Surface = 1 mlen 2 rlen 0 { align1 1Q };
78+
/* Delay thread for specified parameter */
79+
add.nz.f0.0(1) g1.2<1>UD g1.2<0,1,0>UD -1D { align1 1N };
80+
(+f0.0) jmpi(1) -4D { align1 WE_all 1N };
81+
82+
/* Store designated "clear GRF" value */
83+
mov(1) f0.1<1>UW g1.2<0,1,0>UW { align1 1N };
84+
85+
/* Initialize looping parameters */
86+
mov(1) a0<1>D 0D { align1 1N }; /* Initialize a0.0:w=0 */
87+
mov(1) a0.4<1>W 127W { align1 1N }; /* Loop count. Each loop contains 16 GRF's */
88+
89+
/* Write 32x16 all "0" block */
90+
mov(8) g2<1>UD g0<8,8,1>UD { align1 1Q };
91+
mov(8) g127<1>UD g0<8,8,1>UD { align1 1Q };
92+
mov(2) g2<1>UD g1<2,2,1>UW { align1 1N };
93+
mov(1) g2.2<1>UD 0x000f000fUD { align1 1N }; /* Block size (16x16) */
94+
and(1) g2.3<1>UD g2.3<0,1,0>UW 0xffffffefUD { align1 1N };
95+
mov(16) g3<1>UD 0x00000000UD { align1 1H };
96+
mov(16) g4<1>UD 0x00000000UD { align1 1H };
97+
mov(16) g5<1>UD 0x00000000UD { align1 1H };
98+
mov(16) g6<1>UD 0x00000000UD { align1 1H };
99+
mov(16) g7<1>UD 0x00000000UD { align1 1H };
100+
mov(16) g8<1>UD 0x00000000UD { align1 1H };
101+
mov(16) g9<1>UD 0x00000000UD { align1 1H };
102+
mov(16) g10<1>UD 0x00000000UD { align1 1H };
103+
sendc(8) null<1>UD g2<8,8,1>F 0x120a8000
104+
render MsgDesc: media block write MsgCtrl = 0x0 Surface = 0 mlen 9 rlen 0 { align1 1Q };
105+
add(1) g2<1>UD g1<0,1,0>UW 0x0010UW { align1 1N };
106+
sendc(8) null<1>UD g2<8,8,1>F 0x120a8000
107+
render MsgDesc: media block write MsgCtrl = 0x0 Surface = 0 mlen 9 rlen 0 { align1 1Q };
108+
109+
/* Now, clear all GRF registers */
110+
add.nz.f0.0(1) a0.4<1>W a0.4<0,1,0>W -1W { align1 1N };
111+
mov(16) g[a0]<1>UW f0.1<0,1,0>UW { align1 1H };
112+
add(1) a0<1>D a0<0,1,0>D 32D { align1 1N };
113+
(+f0.0) jmpi(1) -8D { align1 WE_all 1N };
114+
115+
/* Terminante the thread */
116+
sendc(8) null<1>UD g127<8,8,1>F 0x82000010
117+
thread_spawner MsgDesc: mlen 1 rlen 0 { align1 1Q EOT };

0 commit comments

Comments
 (0)