Bug 2692

Summary: [Eclipse]tcf file system null error
Product: [Yocto Project Subprojects] Eclipse Plugin Reporter: Hongna Xu <hongnax.xu>
Component: eclipse-pluginAssignee: Jessica <jessica.zhang>
Status: VERIFIED WONTFIX QA Contact: Alexandru Georgescu <alexandru.c.georgescu>
Severity: minor    
Priority: Low CC: alexandru.c.georgescu, bogdanx.a.voiculescu, jiajun.xu, lianhao.lu, yp.ep.watcher, yp.watcher
Version: 1.3   
Target Milestone: Future   
Hardware: x86   
OS: Multiple   
Whiteboard:
OS type for building Yocto: --- Type of Regression: ---
Verified: Documentation change: Don't know
Attachments:
Description Flags
deploy to remote target
none
the tool of oprofile connecting to target none

Description Hongna Xu 2012-07-04 08:33:19 UTC 
Created attachment 621 [details]
deploy to remote target

if i want to connect to qemuarm/mips/ppc/x86/x86-64 sato sdk image first time in eclipse, i encounter the file system null error as attachment, but when i try for the second time, it would connect to target.

the image location:http://autobuilder.yoctoproject.org/pub/nightly/20120613-1/machines/qemu/, built from poky/master,commitID 7f017cab0b3e4d5a4fc356eccd8eb8429b0531d0
Comment 1 Hongna Xu 2012-07-04 08:35:18 UTC 
Created attachment 622 [details]
the tool of oprofile connecting to target
Comment 2 Lianhao Lu 2012-07-04 09:22:26 UTC 
To reproduce this issue, we need an abnormal usage scenario.

1. Launch a qemu instance and use the Eclipse to connect to it on a TCF connection. So an underlying tcp connection was established between the host and the qemu target.

2. Directly kill the qemu instance. The underlying tcp connection was still in ESTABLISHED state since there is no FIN tcp segment ever exchanged between the host and the target, and the default system tcp_keepalive_time is set to 20 minutes so the system can not detect it until then.

3. Launch a new qemu instance, and try to redo the operations in step 1 again. The  RSE/TCF subsystem thinks the TCF connection is still in a valid connected status, so it will directly send the data on that connection. However on the target qemu side, there is no such a TCP connection so it would result an error we see in Hongna's screen shot and it would cause the host tcp connection to be invalidated. So if the user tries for the second time, the RSE/TCF subsystem would reconnect and succeed.

If the qemu is correctly shutdown, the RSE/TCF subsystem would get notified of the connection shutdown event and handle it properly.
Comment 3 Hongna Xu 2012-07-04 09:38:16 UTC 
thank you for lianhao's clarification. but i use the same way to test in previous testing, i encounter this issue for most 2-3 times before in all, but now it happens frequently, why? do they change the socket communication mechanism? i don't think so, what expose it?
Comment 4 Hongna Xu 2012-07-19 08:55:55 UTC 
the issue seems gone on 1.3M2rc1 testing.

Tree/Branch: poky/master
Commit id: eb0cb7e8234f5d2e5623406e9660be91cf52f65e

plugin location:http://autobuilder.pokylinux.org/pub/nightly/20120712-1/eclipse-plugin/indigo/
image location: http://autobuilder.pokylinux.org/pub/nightly/20120712-1/machines/qemu/
Comment 5 Jessica 2012-07-26 21:58:59 UTC 
change the status per last comments, maybe reopened if run into it again
Comment 6 Alexandru Georgescu 2012-09-25 07:27:50 UTC 
> change the status per last comments, maybe reopened if run into it again

Hi,
I managed to reproduce the issue Hongna found following Linhao's scenario when trying to run PowerTop.

I have tested installing Eclipse, adt using 1.3_M4 RC3: e5c3513ca8abb21eb5e9b8c19d1b9199b14ae07c
Comment 7 Jessica 2013-11-15 00:07:46 UTC 
We are moving away from TCF connection overall
Comment 8 Alexandru Georgescu 2013-11-15 13:23:14 UTC 
Marking this as verified.
TCF is not present in Kepler and from the above comments it seems that will be dropped.