Re: Progress on the Alpha distribution at debian-ports
On 22/02/12 00:46, Phil Carmody wrote: > --- On Tue, 2/21/12, Michael Cree wrote: >> On 13/02/2012, at 10:56 AM, Bob Tracy wrote: [2] Provided that you do not use pulseaudio. If pulseaudio is running with a newer kernel then iceweasel will crash and will be unuseable. >>> >>> This is that #$%@! pulseaudio mutex bug we can't seem to get anyone to >>> fix. There *is* a documented workaround: see Debian bug #649641. >>> Unfortunately, this *does* require building the pulseaudio package from >>> source. >> >> I've found the bug --- it's in the kernel. If you are >> building your own kernel then try the attached patch. >> It should fix the pulseaudio problem (and probably some >> others as well). The patch is needed for any kernels >> since 2.6.39 and has not been sent upstream yet. > > diff --git a/arch/alpha/include/asm/futex.h b/arch/alpha/include/asm/futex.h > index e8a761a..f939794 100644 > --- a/arch/alpha/include/asm/futex.h > +++ b/arch/alpha/include/asm/futex.h > @@ -108,7 +108,7 @@ futex_atomic_cmpxchg_inatomic(u32 *uval, u32 __user > *uaddr, > " lda $31,3b-2b(%0)\n" > " .previous\n" > : "+r"(ret), "=&r"(prev), "=&r"(cmp) > - : "r"(uaddr), "r"((long)oldval), "r"(newval) > + : "r"(uaddr), "r"((long)(int)oldval), "r"(newval) > : "memory"); > > *uval = prev; > > > So this fixes an issue introduced here: > """ > commit 8d7718aa082aaf30a0b4989e1f04858952f941bc > Author: Michel Lespinasse > Date: Thu Mar 10 18:50:58 2011 -0800 > > futex: Sanitize futex ops argument types > > Change futex_atomic_op_inuser and futex_atomic_cmpxchg_inatomic > prototypes to use u32 types for the futex as this is the data type the > futex core code uses all over the place. > """ > > and the fix is in cases where we are expecting "sign" (a set bit 31) > extension, > but are incorrectly getting zero extension? Yes, that's the problem. On Alpha there are only 64-bit compare CPU instructions and the extant assembly code for futex_atomic_cmpxchg_inatomic() uses the ldl_l instruction, which sign extends, to load one of the arguments of the comparison. The other argument, provided by the wrapping C-code, thus must be signed extended to match. > If so - have the other 64-bit architectures been double-checked to see > if they are also victims of this > changed-14-architectures-probably-tested-on-only-1 patch? Not that I know of. The test I used to bisect this was to run the glibc test suite in its nptl directory, in particular, the tst-robustX and tst-robustpiX tests. Most of them fail with the bad Alpha kernel. > Anyway if the author of the above alpha fix is reading, and my > interpretation above is correct, he may have at least a > Reviewed-by, or even an > Acked-by: Phil Carmody While I bisected it, and thought it must be a sign extension issue somewhere, it was actually Richard Henderson who suggested the fix on the linux-alpha email list. I'm not sure if he is intending to submit a proper patch with commit message, or is leaving it up to the likes of me. I will follow it up now that I am satisfied the fix is good (it fixes the glibc test suite failures, and pulseaudio is now [mostly] working.) Cheers Michael. -- To UNSUBSCRIBE, email to debian-alpha-requ...@lists.debian.org with a subject of "unsubscribe". Trouble? Contact listmas...@lists.debian.org Archive: http://lists.debian.org/4f483f60.2020...@orcon.net.nz
Re: Progress on the Alpha distribution at debian-ports
--- On Tue, 2/21/12, Michael Cree wrote: > On 13/02/2012, at 10:56 AM, Bob Tracy wrote: > >> [2] Provided that you do not use pulseaudio. If pulseaudio is running > >> with a newer kernel then iceweasel will crash and will be unuseable. > > > > This is that #$%@! pulseaudio mutex bug we can't seem to get anyone to > > fix. There *is* a documented workaround: see Debian bug #649641. > > Unfortunately, this *does* require building the pulseaudio package from > > source. > > I've found the bug --- it's in the kernel. If you are > building your own kernel then try the attached patch. > It should fix the pulseaudio problem (and probably some > others as well). The patch is needed for any kernels > since 2.6.39 and has not been sent upstream yet. diff --git a/arch/alpha/include/asm/futex.h b/arch/alpha/include/asm/futex.h index e8a761a..f939794 100644 --- a/arch/alpha/include/asm/futex.h +++ b/arch/alpha/include/asm/futex.h @@ -108,7 +108,7 @@ futex_atomic_cmpxchg_inatomic(u32 *uval, u32 __user *uaddr, " lda $31,3b-2b(%0)\n" " .previous\n" : "+r"(ret), "=&r"(prev), "=&r"(cmp) - : "r"(uaddr), "r"((long)oldval), "r"(newval) + : "r"(uaddr), "r"((long)(int)oldval), "r"(newval) : "memory"); *uval = prev; So this fixes an issue introduced here: """ commit 8d7718aa082aaf30a0b4989e1f04858952f941bc Author: Michel Lespinasse Date: Thu Mar 10 18:50:58 2011 -0800 futex: Sanitize futex ops argument types Change futex_atomic_op_inuser and futex_atomic_cmpxchg_inatomic prototypes to use u32 types for the futex as this is the data type the futex core code uses all over the place. """ and the fix is in cases where we are expecting "sign" (a set bit 31) extension, but are incorrectly getting zero extension? If so - have the other 64-bit architectures been double-checked to see if they are also victims of this changed-14-architectures-probably-tested-on-only-1 patch? It looks like we were the only arch using a (long) cast, so hopefully we're alone. However, there may be implicit casts in the compiler's asm itself that are hiding the issue. Anyway if the author of the above alpha fix is reading, and my interpretation above is correct, he may have at least a Reviewed-by, or even an Acked-by: Phil Carmody Phil -- To UNSUBSCRIBE, email to debian-alpha-requ...@lists.debian.org with a subject of "unsubscribe". Trouble? Contact listmas...@lists.debian.org Archive: http://lists.debian.org/1329824817.67943.yahoomailclas...@web171309.mail.ir2.yahoo.com
Re: Progress on the Alpha distribution at debian-ports
On Tue, Feb 21, 2012 at 12:19:59PM +1300, Michael Cree wrote: > On 13/02/2012, at 10:56 AM, Bob Tracy wrote: > >This is that #$%@! pulseaudio mutex bug we can't seem to get anyone to > >fix. There *is* a documented workaround: see Debian bug #649641. > > I've found the bug --- it's in the kernel. If you are building your > own kernel then try the attached patch. It should fix the pulseaudio > problem (and probably some others as well). The patch is needed for > any kernels since 2.6.39 and has not been sent upstream yet. Saw that go flying by in the linux-kernel and linux-alpha mail lists. I'll give it a try in the next day or so. A recent 3.X-rcY didn't work on my x86 box, so I've got additional reasons for a rebuild. --Bob -- To UNSUBSCRIBE, email to debian-alpha-requ...@lists.debian.org with a subject of "unsubscribe". Trouble? Contact listmas...@lists.debian.org Archive: http://lists.debian.org/20120221031115.ga9...@gherkin.frus.com
Re: Progress on the Alpha distribution at debian-ports
On 13/02/2012, at 10:56 AM, Bob Tracy wrote: [2] Provided that you do not use pulseaudio. If pulseaudio is running with a newer kernel then iceweasel will crash and will be unuseable. This is that #$%@! pulseaudio mutex bug we can't seem to get anyone to fix. There *is* a documented workaround: see Debian bug #649641. Unfortunately, this *does* require building the pulseaudio package from source. I've found the bug --- it's in the kernel. If you are building your own kernel then try the attached patch. It should fix the pulseaudio problem (and probably some others as well). The patch is needed for any kernels since 2.6.39 and has not been sent upstream yet. Cheers Michael. alpha-fix-futex.patch Description: Binary data
Re: Progress on the Alpha distribution at debian-ports
On 19/02/12 19:43, Bob Tracy wrote: > Have you tried KDE4 yet? I still haven't tried it with compositing > enabled, but it works well for me otherwise. Can't remember the source > package name I was waiting on (to be rebuilt by the new compiler with > the working optimizer), but I *think* it was qt4-x11 or some such. No, I don't have KDE4 installed, but I guess it is about time I gave it a go. qt4-xll version 4:4.8.0-1 was built with gcc-4.6 on the 3rd February this year, thus is *not* afflicted by the nasty gcc-4.4 optimisation bug. Nevertheless I did find a less severe optimisation bug in gcc-4.6 which can lead to segmentation violations [1], that was fixed in the last week or two (gcc-4.6 version 4.6.2-13 IIRC). So some packages built between the switch to gcc-4.6 (mid Decemeber last year) and about the 11th February could be afflicted by that gcc-4.6 optimisation bug [2]. BTW, I have found the kernel commit that causes the pulseaudio/iceweasel crash [3] but I am still to work exactly what is wrong with the commit and to propose a fix. Cheers Michael. [1] PR middle-end/51994. http://gcc.gnu.org/bugzilla/show_bug.cgi?id=51994 [2] Two that I knew of (because their test suite picked up the problem) were libarchive and git. We have working versions of those in the repository now. [3] And also causes test suite failures in glibc, and the java compiler (javac) to lock up on SMP systems. And probably is responsible for other problems. -- To UNSUBSCRIBE, email to debian-alpha-requ...@lists.debian.org with a subject of "unsubscribe". Trouble? Contact listmas...@lists.debian.org Archive: http://lists.debian.org/4f41f153.6020...@orcon.net.nz
Re: Progress on the Alpha distribution at debian-ports
On Sun, Feb 19, 2012 at 11:45:53AM +1300, Michael Cree wrote: > > So how is Gnome3 for you? > > > > It always starts up in fallback mode for me. I do have radeon KMS > > runnning, but mesa is using the software DRI renderer. That might be > > why it starts in fallback mode. > > Ahh, that's because I am not a member of the video group. Fixed. Now I > get the gallium DRI driver. glxgears works nicely and less than 15% cpu > usage for 60fps when I use the xfce4 desktop. > > But Gnome is not so happy: all I get is a black screen with the cursor > when I log in. After a while it switches to screensave mode, brings up > the password entry prompt when I move the mouse, and I can unlock and > come back to the big black empty nothiningness. This new Gnome > experience is somewhat underwhelming. I'll screw-up my courage and give it a try soon. Had some furnace repairs to accomplish this weekend: installed a new controller board today. Have you tried KDE4 yet? I still haven't tried it with compositing enabled, but it works well for me otherwise. Can't remember the source package name I was waiting on (to be rebuilt by the new compiler with the working optimizer), but I *think* it was qt4-x11 or some such. --Bob -- To UNSUBSCRIBE, email to debian-alpha-requ...@lists.debian.org with a subject of "unsubscribe". Trouble? Contact listmas...@lists.debian.org Archive: http://lists.debian.org/20120219064345.gb30...@gherkin.frus.com
Re: Progress on the Alpha distribution at debian-ports
On 18/02/12 16:01, Michael Cree wrote: > On 13/02/12 10:56, Bob Tracy wrote: >> On Sun, Feb 12, 2012 at 02:37:44PM +1300, Michael Cree wrote: >> I seized the occasion of this announcement to try again and figure out a >> way past my Gnome 3 upgrade issues, > [...] >> When I saw that "evolution", "gnome-core", >> and other essentials would finally be upgraded instead of removed, I knew >> I had the solution :-). > > So how is Gnome3 for you? > > It always starts up in fallback mode for me. I do have radeon KMS > runnning, but mesa is using the software DRI renderer. That might be > why it starts in fallback mode. Ahh, that's because I am not a member of the video group. Fixed. Now I get the gallium DRI driver. glxgears works nicely and less than 15% cpu usage for 60fps when I use the xfce4 desktop. But Gnome is not so happy: all I get is a black screen with the cursor when I log in. After a while it switches to screensave mode, brings up the password entry prompt when I move the mouse, and I can unlock and come back to the big black empty nothiningness. This new Gnome experience is somewhat underwhelming. Cheers Michael. -- To UNSUBSCRIBE, email to debian-alpha-requ...@lists.debian.org with a subject of "unsubscribe". Trouble? Contact listmas...@lists.debian.org Archive: http://lists.debian.org/4f402a21.8040...@orcon.net.nz
Re: Progress on the Alpha distribution at debian-ports
On 13/02/12 10:56, Bob Tracy wrote: > On Sun, Feb 12, 2012 at 02:37:44PM +1300, Michael Cree wrote: >> You may have noticed that we have the majority of the unstable >> distribution (was over 95% a week ago) built at Debian-Ports. It is >> reasonably up-to-date with many problems in the toolchain fixed, the >> kernel up to date, Gnome 3 built and installable [1], and other software >> like iceweasel built and working [2]. I hope some of you have it >> installed and are trying it out! > > I seized the occasion of this announcement to try again and figure out a > way past my Gnome 3 upgrade issues, [...] > When I saw that "evolution", "gnome-core", > and other essentials would finally be upgraded instead of removed, I knew > I had the solution :-). So how is Gnome3 for you? It always starts up in fallback mode for me. I do have radeon KMS runnning, but mesa is using the software DRI renderer. That might be why it starts in fallback mode. My experience is that Gnome generally works but with normal usage it is not long before I hit a segmentation violation somewhere. I have had Gnome totally bomb out and I am back at the login window. It wouldn't surprise me if some (most?) of these crashes are due to either: 1) Packages that were built with the buggy gcc-4.4 and should be rebuilt with gcc-4.6 (but which ones are they). 2) The pulseaudio crash. >> I have put quite a bit of work into this over the last six months, but >> is a work rate that I will not be able to keep up past April this year. > > Left to speculation is what kind of work rate you might find sustainable > past April :-). I am off overseas for a couple of weeks to attend my brother's wedding, and on my return work will be extremely busy for the next couple of months so I probably won't have many spare evenings. >> Quite a number of people in this forum were horrified at the >> discontinuation of Debian Linux on Alpha and spoke up for a rear-guard >> action to get another release of the distribution. > > The reasons for discontinuation of Debian Linux on Alpha were, and > presumably remain, valid. You, Bill, and I leapt into the breach, and > in my case at least, it was an honest attempt to see exactly what kind > of effort was going to be required to keep the Debian distro reasonably > up-to-date. Indeed. And it has become clear that there is a lot of work to get the Alpha port up to speed. And there is still quite a bit of work to do. But we do have the toolchain in much better shape. The nasty optimisation bugs in gcc are no longer plaguing us, the kernel is up to date (but see comments below), and we have got more fixes into libc. Thus we have the core technologies in place for Wheezy --- it is now time to get the higher level/desktop application software going. (The annoying GPREL16 relocation errors bug in the linker remains but we can work around it.) > I'll try to make the sales pitch here... Frankly, it has been a lot of > fun for me to run an Alpha system on the so-called bleeding edge. And I too am having quite a bit of fun doing this. > Haven't heard from traditional community participants like Ivan > Kokshaysky in many months. I don't think I've seen him on the linux-alpha email list some time, but, in general, I have found that he does pop up when prodded. The Alpha port of the kernel is only (relatively) up to date because people like myself and Matt Turner have stepped up to maintain it. We fix and attend to the easier problems, and do our best to prod Ivan or Richard (the old maintainers) into action on the difficult problems. >> [2] Provided that you do not use pulseaudio. If pulseaudio is running >> with a newer kernel then iceweasel will crash and will be unuseable. > > This is that #$%@! pulseaudio mutex bug we can't seem to get anyone to > fix. There *is* a documented workaround: see Debian bug #649641. > Unfortunately, this *does* require building the pulseaudio package from > source. I have started work on this one. I am fairly convinced that it is _not_ a pulseaudio bug, but a kernel bug. And if that is indeed the case then I think it is better that the pulseaudio maintainers do not apply the workaround. Sweeping it under the carpet could allow us to ignore a much deeper and much more signficant problem. Cheers Michael. -- To UNSUBSCRIBE, email to debian-alpha-requ...@lists.debian.org with a subject of "unsubscribe". Trouble? Contact listmas...@lists.debian.org Archive: http://lists.debian.org/4f3f1481.2000...@orcon.net.nz
old Alpha workstation Re: Progress on the Alpha distribution at debian-ports
Hi there, does someone have one or two old Alpha workstations - that he/she doesn't use - to give away? I do numerical model of complex systems. If someone has such a workstation, I will of course pay for the shipment (I live in Rome Italy) and a modest contribution for the wkst, too. If it doesn't have Linux but WIN NT4.0 alpha, that's OK too (I'll install Debian for alpha). Thanks in advance for who wants to help personal numerical computation! (I work both in astrophysics and global change). And, with the occasion, my colleagues and I thank very warmly all the friends who help implement Debian for Alpha processor systems! Dr. Alberto Di Fazio senior scientist, National Institute for Astrophysics/Astronomical Observatory of Rome member of the CNR/IGBP Italian National Commission on Global Change Italian Focal Point of the IGBP/AIMES Core Project (ex-GAIM)(Analysis, Integration, and Modelling of the Earth System) president Global Dynamics Institute permanently accredited to the COP under the UNFCCC as observer scientist > > Da:Bob Tracy >A: Michael Cree >Cc: debian-alpha@lists.debian.org >Inviato: Domenica 12 Febbraio 2012 23:41 >Oggetto: Re: Progress on the Alpha distribution at debian-ports > >On Sun, Feb 12, 2012 at 03:56:41PM -0600, Bob Tracy wrote: >> Got a new bug I need to file against the "radvd" package: the "radvd" >> process ID stored in /var/run/radvd/pid is the predecessor of two child >> processes that get spawned (and detach: ppid == 1). Thus, init level >> changes don't work correctly as far as locating running radvd processes, >> among other anomalies. I run the "gw6c" IPv6 client, which spawns >> "radvd" as part of its startup routine. When the IPv6 tunnel collapses >> and "gw6c" reopens it, two more running instances of "radvd" appear >> because of the broken pidfile <--> running instance(s) relationship. > >The latest "radvd" at least partially fixes the issue: the pidfile now >contains the last-spawned-but-detached child, so a running instance of >"radvd" can now be detected once again. It remains to be seen whether >the other detached "radvd" process will die when its younger sibling >goes away. If so, the fix is valid. > >--Bob > > >-- >To UNSUBSCRIBE, email to debian-alpha-requ...@lists.debian.org >with a subject of "unsubscribe". Trouble? Contact listmas...@lists.debian.org >Archive: http://lists.debian.org/20120212224136.ga18...@gherkin.frus.com > > > >
Re: Progress on the Alpha distribution at debian-ports
On Sun, Feb 12, 2012 at 03:56:41PM -0600, Bob Tracy wrote: > Got a new bug I need to file against the "radvd" package: the "radvd" > process ID stored in /var/run/radvd/pid is the predecessor of two child > processes that get spawned (and detach: ppid == 1). Thus, init level > changes don't work correctly as far as locating running radvd processes, > among other anomalies. I run the "gw6c" IPv6 client, which spawns > "radvd" as part of its startup routine. When the IPv6 tunnel collapses > and "gw6c" reopens it, two more running instances of "radvd" appear > because of the broken pidfile <--> running instance(s) relationship. The latest "radvd" at least partially fixes the issue: the pidfile now contains the last-spawned-but-detached child, so a running instance of "radvd" can now be detected once again. It remains to be seen whether the other detached "radvd" process will die when its younger sibling goes away. If so, the fix is valid. --Bob -- To UNSUBSCRIBE, email to debian-alpha-requ...@lists.debian.org with a subject of "unsubscribe". Trouble? Contact listmas...@lists.debian.org Archive: http://lists.debian.org/20120212224136.ga18...@gherkin.frus.com
Re: Progress on the Alpha distribution at debian-ports
On Sun, Feb 12, 2012 at 02:37:44PM +1300, Michael Cree wrote: > You may have noticed that we have the majority of the unstable > distribution (was over 95% a week ago) built at Debian-Ports. It is > reasonably up-to-date with many problems in the toolchain fixed, the > kernel up to date, Gnome 3 built and installable [1], and other software > like iceweasel built and working [2]. I hope some of you have it > installed and are trying it out! I seized the occasion of this announcement to try again and figure out a way past my Gnome 3 upgrade issues, because "apt-get" was stubbornly refusing to upgrade certain packages without gutting a functional system. Most of the problem was centered around libgnome-desktop-3-2 and friends, which I *finally* resolved by specifying a few key packages with it on an "apt-get install" command. When I saw that "evolution", "gnome-core", and other essentials would finally be upgraded instead of removed, I knew I had the solution :-). > I have put quite a bit of work into this over the last six months, but > is a work rate that I will not be able to keep up past April this year. Left to speculation is what kind of work rate you might find sustainable past April :-). I anticipate being able to take up *some* of the slack this summer, modulo not having access to a suitable buildd platform to make debugging more feasible. > Quite a number of people in this forum were horrified at the > discontinuation of Debian Linux on Alpha and spoke up for a rear-guard > action to get another release of the distribution. The reasons for discontinuation of Debian Linux on Alpha were, and presumably remain, valid. You, Bill, and I leapt into the breach, and in my case at least, it was an honest attempt to see exactly what kind of effort was going to be required to keep the Debian distro reasonably up-to-date. > But really it has > come down to three of us who have been working on this: myself, Bill > (running buildds) and Bob (testing and writing bug reports). It would > be really good if others could help with the work. I'll try to make the sales pitch here... Frankly, it has been a lot of fun for me to run an Alpha system on the so-called bleeding edge. It was *made* fun by knowing there were a few key individuals who cared enough to investigate why things were broken and come up with workarounds and/or fixes. It was made *possible* by Michael and Bill being willing to set up and maintain a buildd infrastructure to provide new packages in a timely manner: if you're still reading this list, you're familiar with my reports on the resources (including time) required to build some of the larger packages. Honestly, I would have given up long ago if forced to rely on my own system to build everything from source. As far as coding skills required, not really so much... The simple truth is, a dogged determination to figure out why something doesn't work properly counts for more in this environment than one might expect. Specific example: I might not be able to fix a broken optimizer, but I can sure recognize the effects of one :-), and bring it to the appropriate maintainer's attention. Finally, the "truth in advertising" disclaimer: when my Alpha finally dies in such manner that it cannot be resurrected, I'll likely not get another one unless I can do so inexpensively. It's only my fondness for antiques in general, and Alpha systems in particular, that has kept me active in this community. > There are number of criteria for architecture recertification one of > which is having Debian Developers actively working on the port. We have > none so we have failed before we even start. Haven't heard from traditional community participants like Ivan Kokshaysky in many months. While there are other people out there who know Alpha assembly language pretty well, Ivan was (hopefully still "is") one of the "go-to" experts I worked with back in 2006 or so (the strncpy.S fix for Alpha). > But if people would like I could approach the Debian-Ports FTP master > about whether an unofficial "testing-like" CD could be cut from the > Alpha unstable repo at the time of the Wheezy release. It would not > have any security support but would provide a reasonably up-to-date (as > far as Debian goes) and comprehensive build of packages for the Alpha. Would *love* to see this, because restoring my current system from scratch is too unpleasant to contemplate presently. > That brings us to another issue, that of an installer. The current > debian-installer fails to build on Alpha and needs some serious work if > such a CD is to be of any use. I am not planning to work on the Debian > installer --- there are plenty of other things to do. I will work on > aboot to bring it some new features but that is it. Someone else will > have to step up and do the installer. In spite of running sid, my Alpha is a major component of my infrastructure. Even with a known good installer, I wouldn't willingly trash a
