1376213 - Crash in [@ mozalloc_abort | abort | webrender::gpu_cache::Texture::push_data ]

Status: RESOLVED FIXED

(Core :: Graphics: WebRender, defect, P3)

Description

[Darkspirit]

User Agent: Mozilla/5.0 (X11; Linux x86_64; rv:56.0) Gecko/20100101 Firefox/56.0
Build ID: 20170625100343

Steps to reproduce:

bp-d46e081e-5997-425b-bfa0-f308a0170625
webrender+webrendest enabled, no gpu-process today on Debian Testing x64 (Radeon RX480).

Comment 1

[Kartikaya Gupta (email:[e-mail])]

Crash reason: "Can't allocate > MAX_VERTEX_TEXTURE_WIDTH per resource!"

Comment 2

[Darkspirit]

Nightly 57 x64 20170813183258 @ Debian Testing (webrender+webrendest+stylo)

STR for me:
1. open view-source:https://tobias-kluge.de/css/bootstrap.css
2. get a browser crash

> Can't allocate > MAX_VERTEX_TEXTURE_WIDTH per resource!
bp-dba6e69c-4ecc-4e1f-9972-932620170814
bp-ebd69828-67ed-4d94-8acf-413c00170814
bp-18244dbe-4ae5-411d-9438-6aaca0170814

Comment 3

[Darkspirit]

(In reply to Jan Andre Ikenmeyer [:darkspirit] from comment #2)
> STR for me:
> 1. open view-source:https://tobias-kluge.de/css/bootstrap.css
> 2. get a browser crash

Also on Windows.
Nightly 57 x64 20170815100349 @ Windows 10
bp-4ccbf717-ead5-4df5-8b2b-29cdd0170816
bp-b36c8b95-b890-41a4-bda4-fd2900170816
> Crash Reason: EXCEPTION_ILLEGAL_INSTRUCTION
but with the same crash stack as on linux.

Comment 4

[Darkspirit]

Nightly 57 x64 20170817100132 @ Debian Testing (layers force accel+webrender+webrendest+blob-images)
1. view-source:https://us.battle.net//d3/en/blog/20976068
2. mark some text
3. grab the scrollbar and move down fast.
4. browser crash

bp-911f6bc4-3db4-40a7-947b-cceec0170817	18.08.17	01:26
[@ mozalloc_abort | abort | webrender::gpu_cache::Texture::push_data ] = this bug

bp-1c0b8253-2d14-48a4-8409-6fb070170817	18.08.17	01:26
[@ mozalloc_abort | abort | webrender::gpu_cache::Texture::push_data ] = this bug

bp-67f4e8a2-5ba3-4a47-8ef5-118d00170817	18.08.17	01:25
[@ @0x405b63 ] = not reported, but seems the same like bug 1377113
> Can't allocate > MAX_VERTEX_TEXTURE_WIDTH per resource!

Comment 5

[Jerry Shih[:jerry] (UTC+8) (inactive)]

The gecko send a big size of glyph array to WR, and WR can't handle this big number of glyph in one batch. Then we have an assert in WR.

There are two problem here.
a) gecko should not send all glyph data into WR. In this case, gecko send all glyph data to WR even though some glyph is not seen.

b) if a page really contain so many glyph on the screen, WR should split this glyph array into several batches.

Comment 6

[Jerry Shih[:jerry] (UTC+8) (inactive)]

This PR[1] is to address the b) in comment 5.

[1]
https://github.com/servo/webrender/pull/1597

Comment 7

[Jerry Shih[:jerry] (UTC+8) (inactive)]

There is no crash with [1] when I follow the STR in comment 2.

[1]
https://github.com/servo/webrender/pull/1597

Comment 8

[Jerry Shih[:jerry] (UTC+8) (inactive)]

Wait for https://github.com/servo/webrender/pull/1597 merged in gecko.

Comment 9

[Darkspirit]

Nightly 57 x64 20170824100243 @ Debian Testing (KDE/Xorg/Radeon RX480)
webrender + webrendest + layers-free + blob-image + layers force acceleration

(In reply to Jan Andre Ikenmeyer [:darkspirit] from comment #4)
> 1. view-source:https://us.battle.net//d3/en/blog/20976068
browser crash just by opening that url
bp-11c10d93-c34a-4faa-8924-5d0dd0170825 [@ @0x405f75 ]
> Can't allocate > MAX_VERTEX_TEXTURE_WIDTH per resource!


(In reply to Jan Andre Ikenmeyer [:darkspirit] from comment #2)
> 1. open view-source:https://tobias-kluge.de/css/bootstrap.css
> 2. get a browser crash
bp-1a279e27-5190-453d-bea8-96f6b0170825 [@ mozalloc_abort | abort | webrender::gpu_cache::Texture::push_data ]
> Can't allocate > MAX_VERTEX_TEXTURE_WIDTH per resource!


(In reply to Jerry Shih[:jerry] (UTC+8) from comment #8)
> Wait for https://github.com/servo/webrender/pull/1597 merged in gecko.
I'm looking forward to get the next Nightly and will re-test then.

Comment 10

[Darkspirit]

STR comment 2: macOS bp-ebd735dd-3758-46cf-ad54-d9cfe0170825

Comment 11

[Jerry Shih[:jerry] (UTC+8) (inactive)]

The https://github.com/servo/webrender/pull/1597 is in gecko now, and I don't see the crash for the STR in comment 2.

Hi Jan,
Could you try with the nightly again?

Comment 12

[Darkspirit]

(In reply to Jerry Shih[:jerry] (UTC+8) from comment #11)
> The https://github.com/servo/webrender/pull/1597 is in gecko now, and I
> don't see the crash for the STR in comment 2.
> 
> Hi Jan,
> Could you try with the nightly again?

Verified fixed in Nightly 57 x64 20170829100404 @ Debian Testing.
STRs in comment 2 and comment 4 do not work anymore.
I'll verify Windows and macOS later.