RE: Boot time +8sec with freedesktop.org changes
Like Andi says, the basic flaw is that the freedesktop standard doesn't include a program to run to update the cache, so let's fix that. I would add a cache and instate the rule that whoever makes changes in these directories needs to run the cache-update program. You may even be able to tell the debian package system to do that for you, whenever a package installs something there. If you give it a generic name (xdg-update-application-cache) and symlink it to a moblin specific implementation (e.g. moblin-update-application-cache) then the desktop distributions can adopt it as well and make it call gnome and KDE specific versions of the same. Cheers, Waldo -- Intel Corporation - Hillsboro, Oregon Ultra Mobility Group - Platform Software Architecture Tel: +1 503 264 6237 - Mobile: +1 503 703-7327 From: Spencer, Bob Sent: Friday, April 18, 2008 2:01 PM To: ubuntu-mobile@lists.ubuntu.com Cc: Li, Horace; Kleen, Andi; Bastian, Waldo; Banginwar, Rajesh Subject: Boot time +8sec with freedesktop.org changes FYI: with the latest freedesktop.org changes, we see +8sec bootup time increase (shortened to +6sec by Horace). Some notes below. Bob From: Kleen, Andi - Checksum is difficult -- either you risk not catching a change in some extreme cases or have to read everything again to compute it. - Timestamp assumes your time never goes backwards and device time never gets reset (in practice likely a dangerous assumption) If you trust the time stamps or add some way to have a always monotonic time stamp (e.g. include the inode generation number) it's still difficult. There were some attempts like trying to implement inheriting mtime that flows up a directory which might work (but still has the time stamp problem). There were various other ideas I heard over time, but none really good. Another approach was to use a fuse file system that then transparently updates caches as files are written, but that also has some issues. The basic flaw is that the freedesktop standard doesn't include a program to run to update the cache. -Andi From: Spencer, Bob +7sec! Let's see if we can maintain freedesktop.org compliance and leverage caching to get our time back <1sec. Ideas: - checksum of file name/sizes in /usr/share/applications, if no change then go to cached version - try to simplify / optimize loading code . Maybe we are reading applications.menu file multiple times? Load it once. - see if a smaller /etc/xdg/menus/applications.menu could improve time - look into the gnome_menus code to see why it is taking so long. In 8sec you could start every app on the MID. Parsing a few text files should be trivial. Bob From: Kleen, Andi Distributions have spent a lot of effort in the past in trying to make the freedesktop icons/menu spec boot fast and the general conclusion was that it is hopeless with the standard without major infrastructure changes (like serious file system/VFS support) That was with spinning hard disks where seeks cost more, but i suspect it is not that different on SSDs. The only sane way would be probably to go to some other format for the standard applications and only support freedesktop for optional addons. -Andi From: Yang, Lei A Horace, Does the gnome-menu interface read the .desktop files every time? Can we add some cache file to store the list for next boot up? We may use some mechanism that just read the .desktop files which later than the cache file. Thanks, Lei Y From: Li, Horace Hi, All, Thanks Fan & Martin for taking lots of effort to narrow down the time consumption issue. This afternoon, I am debugging into mobile-basic-flash source, and the root cause is that changing the way to build up application list increased the booting time almost 7 seconds. The main purpose of changing the way is to get Moblin aligned to freedesktop.org menu specification, which is part of MCT. I added a timer in three mobile-basic-flash versions to record elapsed time to build up applications list in different way. Three versions are a fast one, whi
RE: Fix for -psb X issue
Can you see where (e.g. with a calltrace) the first lock happens and where the second lock happens? Cheers, Waldo Intel Corporation - Platform Software Engineering, UMG - Hillsboro, Oregon -Original Message- From: [EMAIL PROTECTED] [mailto:[EMAIL PROTECTED] On Behalf Of Bryce Harrington Sent: Thursday, January 17, 2008 12:53 PM To: Amit Kucheria Cc: ubuntu-mobile@lists.ubuntu.com Subject: Fix for -psb X issue I've uploaded a fix for the -psb X issue to PPA. [1] Yesterday, Amit and I found that a psbDRILock() routine was being called twice; this routine requests a lock in the kernel, which works the first time, but not the second. So this morning, I disabled the second lock/unlock in -psb[2], and sure enough, X came up and seems to be working as expected now[3]. Bryce 1: https://launchpad.net/~ubuntu-mobile/+archive 2: http://people.ubuntu.com/~bryce/UME/103_psb_debugs.patch 3: http://bryceharrington.org/drupal/ume-psb-fix -- Ubuntu-mobile mailing list Ubuntu-mobile@lists.ubuntu.com Modify settings or unsubscribe at: https://lists.ubuntu.com/mailman/listinfo/ubuntu-mobile -- Ubuntu-mobile mailing list Ubuntu-mobile@lists.ubuntu.com Modify settings or unsubscribe at: https://lists.ubuntu.com/mailman/listinfo/ubuntu-mobile
RE: Backtrace
DRM_IOCTL_LOCK ends up with drm_lock() in drm_lock.c part of drm.ko
Cheers,
Waldo
Intel Corporation - Platform Software Engineering, UMG - Hillsboro,
Oregon
-Original Message-
From: [EMAIL PROTECTED]
[mailto:[EMAIL PROTECTED] On Behalf Of Bryce
Harrington
Sent: Wednesday, January 16, 2008 2:55 PM
To: Gunawardana, Inuka
Cc: ubuntu-mobile@lists.ubuntu.com
Subject: Backtrace
Finally got a readable backtrace:
#0 0xb7fd440e in __kernel_vsyscall ()
#1 0xb7e25bc9 in ioctl () from /lib/libc.so.6
#2 0xb7bc4a1d in drmGetLock (fd=10, context=2, flags=3213662516) at
xf86drm.c:1259
#3 0xb7bb951b in psbDRILock (pScrn=0x820bcd0, flags=0) at psb_dri.c:724
#4 0xb7baf983 in psbScreenInit (scrnIndex=0, pScreen=0x821b2d8, argc=6,
argv=0xbf8c9b74) at psb_driver.c:1342
#5 0x08073ddb in AddScreen (pfnInit=0xb7baf7f0 , argc=6,
argv=0xbf8c9b74) at ../../dix/main.c:769
#6 0x080a6e39 in InitOutput (pScreenInfo=0x81fa640, argc=6,
argv=0xbf8c9b74) at ../../../../hw/xfree86/common/xf86Init.c:850
#7 0x08074590 in main (argc=6, argv=0xbf8c9b74, envp=Cannot access
memory at address 0x40086432
) at ../../dix/main.c:369
This also confirms my explorations from the other side, that the issue
is within psbDRILock.
drmGetLock is defined in libdrm, as this:
int drmGetLock(int fd, drm_context_t context, drmLockFlags flags)
{
drm_lock_t lock;
lock.context = context;
lock.flags = 0;
if (flags & DRM_LOCK_READY) lock.flags |= _DRM_LOCK_READY;
if (flags & DRM_LOCK_QUIESCENT) lock.flags |= _DRM_LOCK_QUIESCENT;
if (flags & DRM_LOCK_FLUSH) lock.flags |= _DRM_LOCK_FLUSH;
if (flags & DRM_LOCK_FLUSH_ALL) lock.flags |= _DRM_LOCK_FLUSH_ALL;
if (flags & DRM_HALT_ALL_QUEUES) lock.flags |= _DRM_HALT_ALL_QUEUES;
if (flags & DRM_HALT_CUR_QUEUES) lock.flags |= _DRM_HALT_CUR_QUEUES;
while (ioctl(fd, DRM_IOCTL_LOCK, &lock))
;
return 0;
}
Bryce
--
Ubuntu-mobile mailing list
Ubuntu-mobile@lists.ubuntu.com
Modify settings or unsubscribe at:
https://lists.ubuntu.com/mailman/listinfo/ubuntu-mobile
--
Ubuntu-mobile mailing list
Ubuntu-mobile@lists.ubuntu.com
Modify settings or unsubscribe at:
https://lists.ubuntu.com/mailman/listinfo/ubuntu-mobile
RE: Failures packaging Psb gfx driver / libdrm
The changes to libdrm are already upstream, but that version hasn't found its way into a release yet. [1] The patch to libexa required by the psb driver has been comitted upstream to libexa [2], but it wasn't part of libexa 2.2. [1] http://gitweb.freedesktop.org/?p=mesa/drm.git;a=commit;h=e9805de5c736f89 9dea5220ab8b521cc69c9eca6 [2] http://gitweb.freedesktop.org/?p=xorg/xserver.git;a=commit;h=56cc24ffb21 f7fd41f9ea9e8f969aa85021b9f53 Cheers, Waldo Intel Corporation - Platform Software Engineering, UMG - Hillsboro, Oregon -Original Message- From: [EMAIL PROTECTED] [mailto:[EMAIL PROTECTED] On Behalf Of Bryce Harrington Sent: Friday, January 04, 2008 4:38 PM To: ubuntu-mobile@lists.ubuntu.com Subject: Failures packaging Psb gfx driver / libdrm I attempted to get the Psb driver updated to the latest from moblin.org, but without success. The way these packages are put together, it is very difficult to port the changes to Ubuntu. The principle problem is libdrm. The changes required for -psb are packages as a patch, however it fails to apply against Ubuntu's libdrm, and is too large and time consuming to manually take apart and apply in pieces. By the size of the libdrm patch, it appears that the objective is to rebase the libdrm to a newer version. Since many of the files being added by this patch are already in Ubuntu's libdrm, this is leading to a lot of conflicts. I think the only solution would be to generate a similar patch, but against ubuntu. Can someone who has access to this newer version of the tree please send me a patch built against ubuntu's libdrm? Some other questions: The libdrm tree on moblin.org has been incrementing the debian package number (e.g. 2.3.0-6 instead of 2.3.0-4psb2). I'm able to work around this, but it would be better to adjust the versioning procedures to use the latter numbering scheme. Rob/Jacob, would this be possible? The changes to libdrm result in it being a forked dependency for -psb, which makes packaging updates of -psb more involved and time consuming. To fix this, the changes to libdrm really should be pushed upstream. Is there a person responsible for doing this? What is the status? The -psb package is forcing an overwrite of xserver's libexa with -psb's local version. This seems like a dangerous and brittle solution to the EXA 2.2 issue. Do you really want this hack ported for the Ubuntu -psb package? Bryce -- Ubuntu-mobile mailing list Ubuntu-mobile@lists.ubuntu.com Modify settings or unsubscribe at: https://lists.ubuntu.com/mailman/listinfo/ubuntu-mobile -- Ubuntu-mobile mailing list Ubuntu-mobile@lists.ubuntu.com Modify settings or unsubscribe at: https://lists.ubuntu.com/mailman/listinfo/ubuntu-mobile
RE: Request change in xorg-crownbeach.conf
Note that a process still needs to authenticate itself through the server after opening the device. Without authentication you can't do much. The entries marked with DRM_AUTH in the drm_ioctls[] table in drm_drv.c all require prior authentication. Cheers, Waldo Intel Corporation - Platform Software Engineering, UMG - Hillsboro, Oregon -Original Message- From: [EMAIL PROTECTED] [mailto:[EMAIL PROTECTED] On Behalf Of Matthew Garrett Sent: Monday, November 19, 2007 3:46 PM To: ubuntu-mobile@lists.ubuntu.com Subject: Re: Request change in xorg-crownbeach.conf On Fri, Nov 16, 2007 at 02:13:52PM -0800, Tobin Davis wrote: > I'd like to request a change to the xorg.conf in moblin and ubuntu-ume to add > the following: > > Section "DRI" > Mode 0666 > EndSection This should be 0660 at most, otherwise it gives arbitrary uids access to the graphics hardware. -- Matthew Garrett | [EMAIL PROTECTED] -- Ubuntu-mobile mailing list Ubuntu-mobile@lists.ubuntu.com Modify settings or unsubscribe at: https://lists.ubuntu.com/mailman/listinfo/ubuntu-mobile -- Ubuntu-mobile mailing list Ubuntu-mobile@lists.ubuntu.com Modify settings or unsubscribe at: https://lists.ubuntu.com/mailman/listinfo/ubuntu-mobile
RE: Internal compiler errors when building on Crownbeach
Seems like you have hit an old time classic: http://www.ibiblio.org/pub/Linux/docs/faqs/GCC-SIG11-FAQ Cheers, Waldo Intel Corporation - Platform Software Engineering, UMG - Hillsboro, Oregon -Original Message- From: [EMAIL PROTECTED] [mailto:[EMAIL PROTECTED] On Behalf Of Levinson, Aaron N Sent: Tuesday, October 02, 2007 6:47 PM To: ubuntu-mobile@lists.ubuntu.com Subject: Internal compiler errors when building on Crownbeach I tried to build the kernel on the Crownbeach system, but the same lpia gcc compiler that works fine to build the kernel on my desktop development system hits internal compiler errors on the Crownbeach. The internal compiler errors aren't consistent as well, and it may get past one file just fine in one attempt, but building this same file in a prior attempt resulted in an internal compiler error. Any ideas? Thanks, Aaron Levinson -- Ubuntu-mobile mailing list Ubuntu-mobile@lists.ubuntu.com Modify settings or unsubscribe at: https://lists.ubuntu.com/mailman/listinfo/ubuntu-mobile -- Ubuntu-mobile mailing list Ubuntu-mobile@lists.ubuntu.com Modify settings or unsubscribe at: https://lists.ubuntu.com/mailman/listinfo/ubuntu-mobile
RE: UME theme and related tools
I agree, I think that is exactly the kind of tool needed for our customers to feel confident about doing their own theming. Cheers, Waldo Intel Corporation - Platform Software Engineering, UMG - Hillsboro, Oregon From: [EMAIL PROTECTED] [mailto:[EMAIL PROTECTED] On Behalf Of kyle.nitzsche Sent: Friday, September 21, 2007 10:48 AM To: Bill Filler Cc: ubuntu-mobile@lists.ubuntu.com Subject: Re: UME theme and related tools I think the theme SDK should provide a single approach/tool set for all themeable images, whether they are gtk, hildon, window manager or icons. (Not just widgets, as it seems to do now.) This would enable third-parties to easily create a whole theme without having to understand those distinctions. Perhaps the good work with the template.png and related tools could be readily expanded to include all themeable images by using multiple png files: * one for all themeable images in the UME base * one for each application Perhaps the home area background is themeable but isn't templatized png, since its just a single big file. This approach would bring all themeable images under control, so that a new comprehensive list of images with their locations doesn't have to be created every time a custom theme needs to be created. It would also enable pulling together custom sets of applications for a distribution and themeing them. When I made a list of all such base customizable UME images a month or so ago, it included gtk, hildon, and window manager called images. Kyle Bill Filler wrote: Thanks for the clarification Ken. On Sep 21, 2007, at 11:07 AM, Kenneth Wimer wrote: On Friday 21 September 2007 16:44:10 Bill Filler wrote: Sorry if I am asking a dumb question here as I haven't been following the specific details, but I want to make sure I understand something. Hildon has recently documented their themeing support (see approriate links on http://live.gnome.org/Hildon). Are we basing the UME themeing on this and making improvements? If so, great and hopefully we will be merging the stuff upstream so that when future versions of Hildon themeing come out we can use them with our improvements. If not, how come? Thanks Yes, we are using the hildon themeing tools and infrastructure. The main thing we are changing in this aspect is the layout file which defines which parts we use and where they are located in the template.png file. Note, however that the themeing discussed here does not include icons. When we talk about hildon themeing we are discussing the style of the widgets/controls themselves, not the icons which appear on/in them. -- Ken -- Ubuntu-mobile mailing list Ubuntu-mobile@lists.ubuntu.com Modify settings or unsubscribe at: https://lists.ubuntu.com/mailman/ listinfo/ubuntu-mobile -- Ubuntu-mobile mailing list Ubuntu-mobile@lists.ubuntu.com Modify settings or unsubscribe at: https://lists.ubuntu.com/mailman/listinfo/ubuntu-mobile
mobile-hw-decode status update
A first alpha of the HW video accelleration driver and infrastructure will be part of the next graphics driver (Alpha2) release which is expected to become available in WW38. For Ubuntu the most interesting component will be the VA API in the form of libva since that will be open source and a first version (0.20) will be made available on freedesktop.org once the fdo sitewranglers process https://bugs.freedesktop.org/show_bug.cgi?id=12305 The VA API has been expanded with additional video-postprocessing features such as subpicture support for subtitles. See http://www.freedesktop.org/wiki/Software/vaapi for the latest additions. At the development side, MPEG2 and MPEG4 support is complete in the Linux Poulsbo driver. Unfortunately we still have an unresolved issue that causes many of the larger video clips to fail under Linux. H264 support will be completed next. Cheers, Waldo Intel Corporation - Platform Software Engineering, UMG - Hillsboro, Oregon -- Ubuntu-mobile mailing list Ubuntu-mobile@lists.ubuntu.com Modify settings or unsubscribe at: https://lists.ubuntu.com/mailman/listinfo/ubuntu-mobile
mobile-hw-decode status update
A first beta of the HW video accelleration driver and infrastructure will be part of the next graphics driver (Alpha2) release which is expected to become available in WW37 or WW38. For Ubuntu the most interesting component will be the VA API in the form of libva since that will be open source and a first version (0.20) will be made available on freedesktop.org shortly. The VA API is still somewhat in flux, in particular we still expect to make some changes wrt video-postprocessing such as subpicture support for subtitles. See http://www.freedesktop.org/wiki/Software/vaapi for the latest additions. Cheers, Waldo Intel Corporation - Platform Software Engineering, UMG - Hillsboro, Oregon -- Ubuntu-mobile mailing list Ubuntu-mobile@lists.ubuntu.com Modify settings or unsubscribe at: https://lists.ubuntu.com/mailman/listinfo/ubuntu-mobile
mobile-hw-decode status update
A first beta of the HW video accelleration driver and infrastructure will be part of the next graphics driver (alpha) release. The schedule of which is still being worked on, the schedule is expected to be finalized by next week. For Ubuntu the most interesting component will be the VA API in the form of libva since that will be open source. The VA API is still somewhat in flux, in particular we still expect to make some changes wrt video-postprocessing such as subpicture support for subtitles. Cheers, Waldo Intel Corporation - Platform Software Engineering, UMG - Hillsboro, Oregon -- Ubuntu-mobile mailing list Ubuntu-mobile@lists.ubuntu.com Modify settings or unsubscribe at: https://lists.ubuntu.com/mailman/listinfo/ubuntu-mobile
RE: mockups
What does the orange rectangle in the marquee indicate? It disrupts the overal slick appearance a bit IMHO. What about giving it more rounded corners? Cheers, Waldo Intel Corporation - Platform Software Engineering, UMG - Hillsboro, Oregon -Original Message- From: [EMAIL PROTECTED] [mailto:[EMAIL PROTECTED] On Behalf Of Kenneth Wimer Sent: Thursday, July 19, 2007 5:13 PM To: ubuntu-mobile@lists.ubuntu.com Subject: Re: mockups It appears that this did not make it through due to size problems... Here are links to the pics: http://sinecera.de/home_0.3.png http://sinecera.de/apps_0.3.png http://sinecera.de/browser_0.3.png http://sinecera.de/im_0.3.png http://sinecera.de/mediaPlay_0.3.png On Thursday 19 July 2007 16:59:51 Kenneth Wimer wrote: > Hi, > > Here is a series of different apps/screens all using the same basic > themeing. The IM app is still very rough as are some of the icons in the > browser. > > Once I finish making it, I will add a wifi icon in the "status bar" at the > top right. > > Ken -- Ubuntu-mobile mailing list Ubuntu-mobile@lists.ubuntu.com Modify settings or unsubscribe at: https://lists.ubuntu.com/mailman/listinfo/ubuntu-mobile -- Ubuntu-mobile mailing list Ubuntu-mobile@lists.ubuntu.com Modify settings or unsubscribe at: https://lists.ubuntu.com/mailman/listinfo/ubuntu-mobile
mobile-hw-decode Blueprint
I updated mobile-hw-decode two weeks ago and moved it's state to "Review" but haven't seen any feedback so far. So please take a look at it as well. Cheers, Waldo Intel Corporation - Platform Software Engineering, UMG - Hillsboro, Oregon -Original Message- From: [EMAIL PROTECTED] [mailto:[EMAIL PROTECTED] On Behalf Of Johnson, Charles F Sent: Wednesday, June 27, 2007 10:38 AM To: ubuntu-mobile@lists.ubuntu.com Subject: mobile-graphics, mobile-usb-client Blueprints Updated I updated the mobile-graphics, mobile-usb-client blueprints. I'm very unclear on what is required to move these specs to the next state. Or even what the progression of states is. Charles Johnson Ultra-Mobility Group Platform Software Engineering Intel Corporation [EMAIL PROTECTED] 503-712-5181 -- Ubuntu-mobile mailing list Ubuntu-mobile@lists.ubuntu.com Modify settings or unsubscribe at: https://lists.ubuntu.com/mailman/listinfo/ubuntu-mobile -- Ubuntu-mobile mailing list Ubuntu-mobile@lists.ubuntu.com Modify settings or unsubscribe at: https://lists.ubuntu.com/mailman/listinfo/ubuntu-mobile
