From: "Manszewski, Christoph" <christoph.manszewski@intel.com>
To: Gwan-gyeong Mun <gwan-gyeong.mun@intel.com>,
igt-dev@lists.freedesktop.org
Cc: jonathan.cavitt@intel.com, mika.kuoppala@intel.com,
dominik.grzegorzek@intel.com
Subject: Re: [PATCH i-g-t 2/4] lib/gpgpu_shader: Add causing a read pagefault from the eu thread
Date: Tue, 19 Nov 2024 12:38:58 +0100 [thread overview]
Message-ID: <9b4d1bd7-9918-48aa-85cf-3ad6c1cb90ac@intel.com> (raw)
In-Reply-To: <20241115141132.866838-3-gwan-gyeong.mun@intel.com>
Hi Gwan-gyeong,
On 15.11.2024 15:11, Gwan-gyeong Mun wrote:
> Create a function that causing a read pagefault using the eu thread load
> instruction. If the given ppgtt address points to an unallocated ppgtt
> virtual address, this shader can cause a read pagefault.
> To directly use a 64-bit address as an argument, use the
> Untyped 2D Block Array Load Instruction.
>
> Signed-off-by: Gwan-gyeong Mun <gwan-gyeong.mun@intel.com>
> ---
> lib/gpgpu_shader.c | 92 +++++++++++++++++++++++++++++++++++++
> lib/gpgpu_shader.h | 1 +
> lib/iga64_generated_codes.c | 21 ++++++++-
> 3 files changed, 113 insertions(+), 1 deletion(-)
>
> diff --git a/lib/gpgpu_shader.c b/lib/gpgpu_shader.c
> index 7a2f0d28d..d7c47be80 100644
> --- a/lib/gpgpu_shader.c
> +++ b/lib/gpgpu_shader.c
> @@ -912,3 +912,95 @@ void gpgpu_shader__end_system_routine_step_if_eq(struct gpgpu_shader *shdr,
> ", 0x807fffff, /* leave breakpoint exception */
> y_offset, value, 0x7fffff /* clear all exceptions */ );
> }
> +
> +/**
> + * gpgpu_shader__read_page_fault:
> + * @shdr: shader to be modified
> + * @ppgtt_addr: ppgtt virtual address to raise pagefault
> + *
> + * For a given arbitrary ppgtt virtual address, it raises a pagefault using
> + * the eu thread load instruction.
> + */
> +void gpgpu_shader__read_page_fault(struct gpgpu_shader *shdr, uint64_t ppgtt_addr)
I admit I haven't read much into the assembly however a question about
the naming and intention of this function comes to mind - why does this
function carry the "read_page_fault" suffix? Isn't this supposed to be a
complementary "read" function to the previously introduced
"gpgpu_shader__write_offset" function?
It seems as if they both should just allow to read/write a specified
ppgtt address and if it isn't bound then a page fault will occur. If
that is the case I would suggest to use some matching naming, like
"__write_ppgtt_offset/__read_ppgtt_offset" or some variation of what
Andrzej suggested.
Thanks,
Christoph
> +{
> + /* pagefault ppgtt virtual address */
> + uint64_t addr = CANONICAL(ppgtt_addr);
> +
> + igt_assert_f((addr & 0x3) == 0, "address must be aligned to DWord!\n");
> +
> + emit_iga64_code(shdr, read_page_fault, " \n\
> +#if GEN_VER >= 2000 \n\
> +// Unyped 2D Block Array Load \n\
> +// Instruction_Load2DBlockArray \n\
> +// bspec: 63972 \n\
> +// src0 address payload (Untyped2DBLOCKAddressPayload) specifies both \n\
> +// the block parameters and the 2D Surface parameters. \n\
> +// Untyped2DBLOCKAddressPayload \n\
> +// bspec: 63986 \n\
> +// [243:240] Array Length: 0 (length is 1) \n\
> +// [239:232] Block Height: 0 (height is 1) \n\
> +// [231:224] Block Width: 0xf (width is 16) \n\
> +// [223:192] Block Start Y: 0 \n\
> +// [191:160] Block Start X: 0 \n\
> +// [159:128] Untyped 2D Surface Pitch: 0x3f (pitch is 64 bytes) \n\
> +// [127:96] Untyped 2D Surface Height: 0 (height is 1) \n\
> +// [95:64] Untyped 2D Surface Width: 0x3f (width is 64 bytes) \n\
> +// [63:0] Untyped 2D Surface Base Address \n\
> +// initialize register \n\
> +(W) mov (8) r30.0<1>:uq 0x0:uq \n\
> +// [0:31] Untyped 2D Surface Base Address low \n\
> +(W) mov (1) r30.0<1>:ud ARG(0):ud \n\
> +// [32:63] Untyped 2D Surface Base Address high \n\
> +(W) mov (1) r30.1<1>:ud ARG(1):ud \n\
> +// [95:64] Untyped 2D Surface Width: 0x3f \n\
> +// (Width minus 1 (in bytes) of the 2D surface, it represents 64) \n\
> +(W) mov (1) r30.2<1>:ud 0x3f:ud \n\
> +// [127:96] Untyped 2D Surface Height: 0x0 \n\
> +// (Height minus 1 (in number of data elements) of \n\
> +// the Untyped 2D surface, it represents 1) \n\
> +(W) mov (1) r30.3<1>:ud 0x0:ud \n\
> +// [159:128] Untyped 2D Surface Pitch: 0x3f \n\
> +// (Pitch minus 1 (in bytes) of the 2D surface, it represents 64) \n\
> +(W) mov (1) r30.4<1>:ud 0x3f:ud \n\
> +// [231:224] Block Width: 0xf (15) \n\
> +// (Specifies the width minus 1 (in number of data elements) for this \n\
> +// rectangular region, it represents 16) \n\
> +// Block width (encoded_value + 1) must be a multiple of DW (4 bytes). \n\
> +// [239:232] Block Height: 0 \n\
> +// (Specifies the height minus 1 (in number of data elements) for \n\
> +// this rectangular region, it represents 1) \n\
> +// [243:240] Array Length: 0 \n\
> +// (Specifies Array Length minus 1 for Load2DBlockArray messages, \n\
> +// must be zero for 2D Block Store messages, it represents 1) \n\
> +(W) mov (1) r30.7<1>:ud 0xf:ud \n\
> +// \n\
> +// dest data payload format is selected by Data Size. \n\
> +// Block Height x Block Width x Data size / GRF Register size \n\
> +// => 1 x 16 x 32bit / 512bit = 1 \n\
> +// data payload format size is 1 GRF Register. \n\
> +// \n\
> +// send.ugm Untyped 2D Block Array Load \n\
> +// Format: send.ugm (1) dst src0 src1 ExtMsg MsgDesc \n\
> +// Execution Mask restriction: SIMT1 \n\
> +// \n\
> +// Extended Message Descriptor (Dataport Extended Descriptor Imm 2D Block) \n\
> +// bspec: 67780 \n\
> +// 0x0 => \n\
> +// [32:22] Global Y_offset: 0 \n\
> +// [21:12] Global X_offset: 0 \n\
> +// \n\
> +// Message Descriptor \n\
> +// bspec: 63972 \n\
> +// 0x2128403 => \n\
> +// [30:29] Address Type: 0 (FLAT) \n\
> +// [28:25] Src0 Length: 1 \n\
> +// [24:20] Dest Length: 1 \n\
> +// [19:16] Cache : 2 (L1UC_L3UC) 10 \n\
> +// [15] Transpose Block: 1 \n\
> +// [11:9] Data Size: 2 (D32) 10 \n\
> +// [7] VNNI Transform: 0 \n\
> +// [5:0] Load Operation: 3 (Load 2D Block) 11 \n\
> +(W) send.ugm (1) r31 r30 null 0x0 0x2128403 \n\
> +#endif \n\
> + ", lower_32_bits(addr), upper_32_bits(addr));
> +}
> \ No newline at end of file
> diff --git a/lib/gpgpu_shader.h b/lib/gpgpu_shader.h
> index 355b128b5..318550c52 100644
> --- a/lib/gpgpu_shader.h
> +++ b/lib/gpgpu_shader.h
> @@ -87,6 +87,7 @@ void gpgpu_shader__write_offset(struct gpgpu_shader *shdr, uint64_t ppgtt_offset
> uint32_t value);
> void gpgpu_shader__write_on_exception(struct gpgpu_shader *shdr, uint32_t dw, uint32_t x_offset,
> uint32_t y_offset, uint32_t mask, uint32_t value);
> +void gpgpu_shader__read_page_fault(struct gpgpu_shader *shdr, uint64_t ppgtt_addr);
> void gpgpu_shader__label(struct gpgpu_shader *shdr, int label_id);
> void gpgpu_shader__jump(struct gpgpu_shader *shdr, int label_id);
> void gpgpu_shader__jump_neq(struct gpgpu_shader *shdr, int label_id,
> diff --git a/lib/iga64_generated_codes.c b/lib/iga64_generated_codes.c
> index b23613ac4..53a705358 100644
> --- a/lib/iga64_generated_codes.c
> +++ b/lib/iga64_generated_codes.c
> @@ -3,7 +3,7 @@
>
> #include "gpgpu_shader.h"
>
> -#define MD5_SUM_IGA64_ASMS 4fcde43dedb9d3212f1d85b5b180b0c1
> +#define MD5_SUM_IGA64_ASMS 01290b5ecda7a6e765463558d6f59952
>
> struct iga64_template const iga64_code_gpgpu_fill[] = {
> { .gen_ver = 2000, .size = 44, .code = (const uint32_t []) {
> @@ -79,6 +79,25 @@ struct iga64_template const iga64_code_gpgpu_fill[] = {
> }}
> };
>
> +struct iga64_template const iga64_code_read_page_fault[] = {
> + { .gen_ver = 2000, .size = 44, .code = (const uint32_t []) {
> + 0x800c0061, 0x1e054330, 0x00000000, 0x00000000,
> + 0x80000061, 0x1e054220, 0x00000000, 0xc0ded000,
> + 0x80000061, 0x1e154220, 0x00000000, 0xc0ded001,
> + 0x80000061, 0x1e254220, 0x00000000, 0x0000003f,
> + 0x80000061, 0x1e354220, 0x00000000, 0x00000000,
> + 0x80000061, 0x1e454220, 0x00000000, 0x0000003f,
> + 0x80000061, 0x1e754220, 0x00000000, 0x0000000f,
> + 0x80032031, 0x1f0c0000, 0xf8061e0c, 0x00a00000,
> + 0x80000001, 0x00010000, 0x20000000, 0x00000000,
> + 0x80000001, 0x00010000, 0x30000000, 0x00000000,
> + 0x80000901, 0x00010000, 0x00000000, 0x00000000,
> + }},
> + { .gen_ver = 0, .size = 0, .code = (const uint32_t []) {
> +
> + }}
> +};
> +
> struct iga64_template const iga64_code_end_system_routine_step_if_eq[] = {
> { .gen_ver = 2000, .size = 44, .code = (const uint32_t []) {
> 0x80000966, 0x80018220, 0x02008000, 0x00008000,
next prev parent reply other threads:[~2024-11-19 11:39 UTC|newest]
Thread overview: 24+ messages / expand[flat|nested] mbox.gz Atom feed top
2024-11-15 14:11 [PATCH i-g-t 0/4] tests/intel/xe_eudebug_online: Introduce read/write pagefault tests Gwan-gyeong Mun
2024-11-15 14:11 ` [PATCH i-g-t 1/4] lib/gppgu_shader: Add write to ppgtt offset Gwan-gyeong Mun
2024-11-18 13:00 ` Hajda, Andrzej
2024-11-21 12:01 ` Gwan-gyeong Mun
2024-11-15 14:11 ` [PATCH i-g-t 2/4] lib/gpgpu_shader: Add causing a read pagefault from the eu thread Gwan-gyeong Mun
2024-11-18 13:08 ` Hajda, Andrzej
2024-11-21 12:02 ` Gwan-gyeong Mun
2024-11-19 11:38 ` Manszewski, Christoph [this message]
2024-11-21 12:07 ` Gwan-gyeong Mun
2024-11-15 14:11 ` [PATCH i-g-t 3/4] eudebug: Add eudebug pagefault event declarations Gwan-gyeong Mun
2024-11-18 16:52 ` Hajda, Andrzej
2024-11-19 8:50 ` Manszewski, Christoph
2024-11-19 12:26 ` Manszewski, Christoph
2024-11-15 14:11 ` [PATCH i-g-t 4/4] tests/intel/xe_eudebug_online: Add read/write pagefault online tests Gwan-gyeong Mun
2024-11-19 8:10 ` Hajda, Andrzej
2024-11-21 12:06 ` Gwan-gyeong Mun
2024-11-19 15:58 ` Hajda, Andrzej
2024-11-21 12:11 ` Gwan-gyeong Mun
2024-11-19 16:49 ` Manszewski, Christoph
2024-11-21 12:15 ` Gwan-gyeong Mun
2024-11-15 14:42 ` ✓ CI.xeBAT: success for tests/intel/xe_eudebug_online: Introduce read/write pagefault tests Patchwork
2024-11-15 14:48 ` ✓ Fi.CI.BAT: " Patchwork
2024-11-15 21:24 ` ✗ Fi.CI.IGT: failure " Patchwork
2024-11-15 23:25 ` ✗ CI.xeFULL: " Patchwork
Reply instructions:
You may reply publicly to this message via plain-text email
using any one of the following methods:
* Save the following mbox file, import it into your mail client,
and reply-to-all from there: mbox
Avoid top-posting and favor interleaved quoting:
https://en.wikipedia.org/wiki/Posting_style#Interleaved_style
* Reply using the --to, --cc, and --in-reply-to
switches of git-send-email(1):
git send-email \
--in-reply-to=9b4d1bd7-9918-48aa-85cf-3ad6c1cb90ac@intel.com \
--to=christoph.manszewski@intel.com \
--cc=dominik.grzegorzek@intel.com \
--cc=gwan-gyeong.mun@intel.com \
--cc=igt-dev@lists.freedesktop.org \
--cc=jonathan.cavitt@intel.com \
--cc=mika.kuoppala@intel.com \
/path/to/YOUR_REPLY
https://kernel.org/pub/software/scm/git/docs/git-send-email.html
* If your mail client supports setting the In-Reply-To header
via mailto: links, try the mailto: link
Be sure your reply has a Subject: header at the top and a blank line
before the message body.
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox