That's remind me of something, With the solution you offered in your first reply:
*>Build both instrumented and >non-instrumented versions of libhwui.so, place first in >/system/lib/asan and the second in /system/lib. Add >LD_LIBRARY_PATH=/system/lib/* *>asan to asan_wrapper. This way anything >started with asan_wrapper will prefer instrumented versions of >libraries in /system/lib/asan, and zygote will use the normal version >in /system/lib.*a little confused ,that when we use asan_device_setup script for java apps Asan using, this *LD_LIBRARY_PATH *setting will not helpful for the mem cost problem, because we don't use asanwapper in this situation. And Of cource, I've read the asan_device_setup script, we changed the app_process to a script and use it to exec the real app_process, I think maybe we can add *LD_LIBRARY_PATH *to the app_process script we added, but still not helpful at all,all forked process will use instrumented versions of libraries in /system/lib/asan, and consume all my memory. Am I right? In other way said, we must use asanwapper instead of asan_device_setup script for java apps, we must setprop.Right? 在 2014年7月23日星期三UTC+8下午7时23分38秒，Evgeniy Stepanov写道： > > Well, setprop approach is deprecated. With the new asan_device_setup > script, everything should work automatically, because asan runtime is > already loaded into zygote, and only activated when an instrumented > application is started. > > Modifying init.rc is really not that hard. There is an idea that I > always wanted to implement for ASan applications to re-exec themselves > if they are run with asanwrapper (i.e. without ASan runtime library in > LD_PRELOAD list). In that case ASan would add itself to LD_PRELOAD and > call exec() on itself. It would require changing ASan code. I would > not have time for this in the coming months. > > > On Thu, Jul 17, 2014 at 6:34 PM, ji wang <mason....@gmail.com > <javascript:>> wrote: > > Hi, Evgeniy > > Thanks for your reply. And yes, the OOM issue happed with instrumented > > libhwui.so. I understood why this memory consume so huge now. > > And with the solution u offred, not all process using the instrumented > > libhwui.so, just the process u want to check using asan, this is useful > for > > Problem Two, but we still have Problem One:we must preload > > libasan_preload.so to the process first, the app process way to achive > this > > is "setprop wrap.com.android.appname "logwrapper asanwrapper"", the > native > > procecess way to achive this is modify the init.rc file. and besides > the > > property way for app will not work after reboot, must set it again. Not > easy > > to use. Does we have any solutions for this? Thank u very much. > > > > 在 2014年7月16日星期三UTC+8下午6时48分41秒，Evgeniy Stepanov写道： > >> > >> Does this out-of-memory issue happen with instrumented libhwui.so? > >> This has an easy explanation. ASan activates all of its > >> memory-consuming logic when the first instrumented library is loaded > >> into the process. If such library is loaded into zygote pre-fork, all > >> java apps are running with ASan enabled, and consume all your memory. > >> > >> There is something you can try. Build both instrumented and > >> non-instrumented versions of libhwui.so, place first in > >> /system/lib/asan and the second in /system/lib. Add > >> LD_LIBRARY_PATH=/system/lib/asan to asan_wrapper. This way anything > >> started with asan_wrapper will prefer instrumented versions of > >> libraries in /system/lib/asan, and zygote will use the normal version > >> in /system/lib. > >> > >> > >> On Wed, Jul 16, 2014 at 2:00 PM, ji wang <mason....@gmail.com> wrote: > >> > Hi, > >> > We now want use LLVM Asan to check some android libs like libhwui.so, > >> > but > >> > found out is not very easy to use. Is there any way to solve this? Or > I > >> > use > >> > it the wrong way? > >> > The problem is we must preload libasan_preload.so before those > checking > >> > lib > >> > initlized or used in some processes. For example, when we want check > >> > libhwui.so,we must modify init.rc to use asanwrapper to load > mediaserver > >> > process in order to preload libasan_preload.so , (and the same to > >> > app_process,of cource the asan_device_setup.sh already done the > >> > app_process > >> > part,) otherwise without this step android device will got boot error > >> > failure.This will ask us have to know how many processes use this > >> > checking > >> > lib and what those processes are exactly! wow.. > >> > Besides, there is one more problem, when we use LLVM asan, after use > >> > asan_device_setup.sh to install asan to our device, we didnot add any > >> > libs > >> > or binary for check, but still got a huge memory cost after the > device > >> > bootup, and for 1G ram android devices, we got boot failure, becauce > of > >> > the > >> > LMK kill launcher and some other app process for no memory lost! > >> > > >> > > >> > -- > >> > You received this message because you are subscribed to the Google > >> > Groups > >> > "address-sanitizer" group. > >> > To unsubscribe from this group and stop receiving emails from it, > send > >> > an > >> > email to address-saniti...@googlegroups.com. > >> > For more options, visit https://groups.google.com/d/optout. > > > > -- > > You received this message because you are subscribed to the Google > Groups > > "address-sanitizer" group. > > To unsubscribe from this group and stop receiving emails from it, send > an > > email to address-saniti...@googlegroups.com <javascript:>. > > For more options, visit https://groups.google.com/d/optout. > -- You received this message because you are subscribed to the Google Groups "address-sanitizer" group. To unsubscribe from this group and stop receiving emails from it, send an email to address-sanitizer+unsubscr...@googlegroups.com. For more options, visit https://groups.google.com/d/optout.
