It is possible that XP + e10s has higher crash rates than non-e10s, this bug tracks the work to figure out if this is in fact the case, and if so what we need to do to address it. Once those issues are address we'll enable e10s for XP users on release channels. Currently we don't have strong evidence of a crash problem under XP with e10s on Nightly/Aurora, but we don't have a lot of data either. Most XP users are on release channels where e10s is explicitly disabled for those users.) https://crash-stats.mozilla.com/search/?product=Firefox&dom_ipc_enabled=%21__null__&platform_pretty_version=Windows%20XP&version=51.0a1&version=50.0a2&version=50.0a1&_sort=-date&_facets=signature&_columns=date&_columns=signature&_columns=product&_columns=version&_columns=build_id&_columns=platform#facet-signature Historical summary of XP + e10s changes: 1) a rendering issue while scrolling was seen on a lab device (bug 1237769) with gfx acceleration. 2016-01-17 checkin: We disabled acceleration (D3D9) and use basic composition instead when e10s is enabled. - Bug 1237770 - Disable acceleration (D3D9) when e10s is enabled - https://hg.mozilla.org/mozilla-central/rev/269ad7540a30 2) It was found that the above fix caused severe video performance issues on XP due to perf issues with unaccelerated video. (These isues have since been addressed in other work by the gfx team.) - https://bugzilla.mozilla.org/show_bug.cgi?id=1237769#c12 We decided to disable e10s on XP in general. This reverted the previous change, and tied e10s to acceleration prefs. - https://bugzilla.mozilla.org/show_bug.cgi?id=1237769#c18 - https://bugzilla.mozilla.org/show_bug.cgi?id=1237769#c22 2016-04-07 checkin: - https://hg.mozilla.org/mozilla-central/rev/a3f8a4e83cdd At this point, having layers.acceleration.disabled = false or layers.acceleration.force-enabled = true disables e10s on Windows XP. 3) We found that XP + e10s was crashy on beta, so we disable e10s for Windows XP on release channels: - Bug 1275039 - [e10s] Block e10s on Windows XP 2016-05-26 checkin: - https://hg.mozilla.org/mozilla-central/rev/e3977905475d
Based on the data we have on beta for e10s users, XP with unaccelerated gfx is less crashy. https://sql.telemetry.mozilla.org/queries/1022/source#table
(In reply to Jim Mathies [:jimm] from comment #1) > Based on the data we have on beta for e10s users, XP with unaccelerated gfx > is less crashy. Less crashy than what?
XP users, 49 beta cycle, e10s on, no accelerated graphics: Cohort Khours M M+c M+(c-s) C S C Rate S Rate C - S Rate Percent Plugin Gmp control 167.60 12.73 17.27 12.81 760.00 747.00 4.53 4.46 0.08 0.98 69.61 0.00 test 149.52 7.33 34.79 15.34 4,106.00 2,909.00 27.46 19.46 8.01 0.71 63.10 0.00 We have a final regression of about 19.75% here (12.81 vs. 15.34).
The number of users we have here is incredibly small though, so I'm not sure these crash numbers are at all useful. Here's the crash data from 49 for XP users who got into the experiment, implying they didn't have acceleration enabled. https://crash-stats.mozilla.com/search/?platform_pretty_version=%3DWindows%20XP&dom_ipc_enabled=%21__null__&date=%3E2016-08-01&product=Firefox&version=49.0b&_sort=-date&_facets=signature&_facets=e10s_cohort&_columns=date&_columns=signature&_columns=product&_columns=version&_columns=build_id&_columns=platform&_columns=e10s_cohort#facet-signature
btw, XP support will be dropped after Firefox 52 (xp-eol bug 1130266). XP and Vista users will be automatically switched to the ESR 52 channel and receive security patches at least until some time in 2018.
Depends on: xp-eol
Jim, now that we are dropping XP and Vista support in Firefox 52 (bug 1130266), can we resolve all "support e10s on XP" bugs as WONTFIX? Or is there a chance we will enable e10s for XP in Firefox 52 release or ESR?
(In reply to Chris Peterson [:cpeterson] from comment #6) > Jim, now that we are dropping XP and Vista support in Firefox 52 (bug > 1130266), can we resolve all "support e10s on XP" bugs as WONTFIX? Or is > there a chance we will enable e10s for XP in Firefox 52 release or ESR? Not yet, we still haven't made the xp/e10s decision. Waiting on some crash data from 51.
We disabled all tests on XP (bug 1310836). If we enable E10S for 51/52, we will uplift untested code. Moreover, we are already landing some changes that break testing on XP (for example, bug 1310855). Is it really possible to develop to enable E10S on XP?
XP isn't getting e10s on ESR52, so there's nothing to do here.
You need to log in before you can comment on or make changes to this bug.