From: Simon L. <sim...@in...> - 2012-04-19 12:51:09
|
Hi all, I have found 2 segfaults in enlightenment and have attached various back traces. I am ok with gdb but was unable to create patches as i am not at all familiar with the code. Relevant Info the version is rev 70308 although the first issue has been present for quite some time. The second i only found tonight, neither really stop me from using enlightenment though. I am running openSUSE from a rolling repository and am compiling e17, libs and modules using easy_e17.sh I am using the proprietary NVIDIA driver with 3 screens across 2 graphics cards running with separate x servers. So 3 running. Segfault 1 comes when i try to remove the e weather widget from being a gadget on my primary desktop or try and unload the module. I added the 3 weather widgets and was going to keep the one i liked best which is forcast. Backtraces for this are the 2 files mentioning e weather to me what should be the useful code seems to be optimised out. #2 seems to happen whenever i try to add a gadget to my 2nd or 3rd display (off the secondary graphics card) i tried the pager and 1 or 2 other widgets i was mostly testing the first issue at the time. Add pager and add gadget 3rd display logs. The issue here is there is a NULL deference but i couldn't work out why. I can replicate both issues every time if more information is required just guide me as to what. Cheers Simon -- http://simotek.net/tech |
From: Carsten H. (T. R. <ra...@ra...> - 2012-04-22 02:38:46
|
On Thu, 19 Apr 2012 22:20:43 +0930 Simon Lees <sim...@in...> said: > Hi all, > I have found 2 segfaults in enlightenment and have attached various back > traces. I am ok with gdb but was unable to create patches as i am not at > all familiar with the code. > Relevant Info the version is rev 70308 although the first issue has been > present for quite some time. The second i only found tonight, neither > really stop me from using enlightenment though. > I am running openSUSE from a rolling repository and am compiling e17, > libs and modules using easy_e17.sh I am using the proprietary NVIDIA > driver with 3 screens across 2 graphics cards running with separate x > servers. So 3 running. > Segfault 1 comes when i try to remove the e weather widget from being a > gadget on my primary desktop or try and unload the module. I added the 3 > weather widgets and was going to keep the one i liked best which is > forcast. Backtraces for this are the 2 files mentioning e weather to me > what should be the useful code seems to be optimised out. core issue: sigill. illegal instruction. this didn't even happen inside the assembly bits in evas (which have detection mechanisms to not use ones that your cpu doesn't support - it uses illegal instruction traps to turn it off). as such this is happening during regular c code, so your possible issues are: 1. compiler is broken and producing wrong assembly for the arch it is being asked to. 2. you are passing incorrect optimization flags to the compiler telling it to optimize for an architecture you don't have. 3. you got the binary from someone else and they have optimized for an arch you don't have. 4. hardware problem (memory corruption, bus, cpu, disk corruption). 5. the backtrace is screwed and saying something that isn't reality at all. the issue is in ecore's binary here. the actual machinecode in memory should be read-only so it cannot be written over or modified at runtime. thats how the linker works. > #2 seems to happen whenever i try to add a gadget to my 2nd or 3rd > display (off the secondary graphics card) i tried the pager and 1 or 2 > other widgets i was mostly testing the first issue at the time. Add > pager and add gadget 3rd display logs. The issue here is there is a NULL > deference but i couldn't work out why. > I can replicate both issues every time if more information is required > just guide me as to what. > Cheers > Simon the 2nd one indicates timer is probably an invalid pointer. that's rather odd. you are using multihead not xinerama? i don't think i've tested multihead with e for like ... hmm 8 years. :) in theory it has code to do it, but it's never tested or maintained. -- ------------- Codito, ergo sum - "I code, therefore I am" -------------- The Rasterman (Carsten Haitzler) ra...@ra... |
From: Michael B. <mic...@gm...> - 2012-04-22 02:45:42
|
On Sun, 22 Apr 2012 11:39:49 +0900 Carsten Haitzler (The Rasterman) <ra...@ra...> wrote: > On Thu, 19 Apr 2012 22:20:43 +0930 Simon Lees <sim...@in...> > said: > > > Hi all, > > I have found 2 segfaults in enlightenment and have attached various back > > traces. I am ok with gdb but was unable to create patches as i am not at > > all familiar with the code. > > Relevant Info the version is rev 70308 although the first issue has been > > present for quite some time. The second i only found tonight, neither > > really stop me from using enlightenment though. > > I am running openSUSE from a rolling repository and am compiling e17, > > libs and modules using easy_e17.sh I am using the proprietary NVIDIA > > driver with 3 screens across 2 graphics cards running with separate x > > servers. So 3 running. > > Segfault 1 comes when i try to remove the e weather widget from being a > > gadget on my primary desktop or try and unload the module. I added the 3 > > weather widgets and was going to keep the one i liked best which is > > forcast. Backtraces for this are the 2 files mentioning e weather to me > > what should be the useful code seems to be optimised out. > > core issue: sigill. illegal instruction. this didn't even happen inside the > assembly bits in evas (which have detection mechanisms to not use ones that > your cpu doesn't support - it uses illegal instruction traps to turn it off). > as such this is happening during regular c code, so your possible issues are: > > 1. compiler is broken and producing wrong assembly for the arch it is being > asked to. > 2. you are passing incorrect optimization flags to the compiler telling it to > optimize for an architecture you don't have. > 3. you got the binary from someone else and they have optimized for an arch > you don't have. > 4. hardware problem (memory corruption, bus, cpu, disk corruption). > 5. the backtrace is screwed and saying something that isn't reality at all. > > the issue is in ecore's binary here. the actual machinecode in memory should > be read-only so it cannot be written over or modified at runtime. thats how > the linker works. > > > #2 seems to happen whenever i try to add a gadget to my 2nd or 3rd > > display (off the secondary graphics card) i tried the pager and 1 or 2 > > other widgets i was mostly testing the first issue at the time. Add > > pager and add gadget 3rd display logs. The issue here is there is a NULL > > deference but i couldn't work out why. > > I can replicate both issues every time if more information is required > > just guide me as to what. > > Cheers > > Simon > > the 2nd one indicates timer is probably an invalid pointer. that's rather odd. > you are using multihead not xinerama? i don't think i've tested multihead with > e for like ... hmm 8 years. :) in theory it has code to do it, but it's never > tested or maintained. I've been using true multihead with e for years, it works fine except for (obviously) compositing |