Bug 234 - X fail to start on x86-64 target
Summary: X fail to start on x86-64 target
Status: VERIFIED FIXED
Alias: None
Product: Meta-yocto
Classification: Build System, Metadata & Runtime
Component: meta-yocto (show other bugs)
Version: 0.9
Hardware: x86 x86_64
: High major
Target Milestone: 0.9 M3
Assignee: Edwin Zhai
QA Contact:
URL:
Whiteboard:
Depends on:
Blocks:
 
Reported: 2010-08-25 22:25 UTC by Edwin Zhai
Modified: 2010-12-21 06:07 UTC (History)
5 users (show)

See Also:
OS type for building Yocto: ---
Type of Regression: ---
Verified: Program Management
Documentation change: ---


Attachments

Note You need to log in before you can comment on or make changes to this bug.
Description Edwin Zhai 2010-08-25 22:25:59 UTC
x86-64 target fail to start Xwindow, while x86 target is okay. Xorg + matchbox-terminal can start, but matchbox desktop fail. There are seg fault from v86d and some demsg from uvesafb. Josh pulled some upstream fixes into v86d and uvesafb, but the root cause is not found yet.
Comment 1 Edwin Zhai 2010-08-27 20:55:00 UTC
Finally found the v86d and uvesafb issue has nothing to do with this bug. After rmmod uvesafb, everything remains same.

Matchbox environment start fail on x86-64 target, while ok on x86 target. Root cause is libmatchbox use "0"(int) as termination indicator when calling XftFontOpen, which in turn called FcPatternVapBuild(in fontconfig). It try to get the "0" as char* and fetch wrong value, as int and char* has different size on x86-64. 

See following source code:
==== libmatchbox ======
  font->font = XftFontOpen (font->dpy, DefaultScreen(font->dpy),
                            XFT_FAMILY, XftTypeString , font->family,
                            XFT_SIZE, XftTypeDouble   , (double)font->pt_size,
                            XFT_WEIGHT, XftTypeInteger, weight,
                            XFT_SLANT, XftTypeInteger , slant,
                            0);

==== fontconfig ====
#define FcPatternVapBuild(result, orig, va)                         \
{                                                                   \
    FcPattern   *__p__ = (orig);                                    \
    const char  *__o__;                                             \
    FcValue     __v__;                                              \
                                                                    \
    for (;;)                                                        \
    {                                                               \
> >        __o__ = va_arg (va, const char *);                       \
> >        if (!__o__)                                              \
> >            break;                                               \
        __v__.type = va_arg (va, FcType);                           \
Comment 2 Edwin Zhai 2010-08-29 23:51:12 UTC
More tests with dongxiao showed that this issue only occur when passing many parameter, which leads passing parameter through stack rather than register. Register can handle the extension well, so you can put 32b value and get 64b same value.
Comment 3 Edwin Zhai 2010-09-02 17:40:22 UTC
Fixed by 0d7bcdf9fd1bfaebfb5f0b99606e85a8e0229bba
Comment 4 Edwin Zhai 2010-09-08 17:33:58 UTC
Verified on 2010-09-04 nightly build with commit 18ea4d2a30cf1cd5e5499bd22d193d8732c2798c

The X window can start successfully.