54268 - Broswer crashes randomly without being used while at cnn.com

Status: RESOLVED WORKSFORME

(SeaMonkey :: General, defect, P3)

Description

[Chris]

When www.cnn.com is loaded on mozilla (build 2000080712), it seemingly crashes
at random (without user input).  I've seen it happen while reading a page, but
generally, I if leave mozilla open, it will crash randomly with the average time
to crash around 30min.  It exits with status 0 and give no death messages or
core-dump.  I ran a strace on it with the following results:

mozilla does its standard polling and a few lseeks and reads,
begins to mprotect abunch of 4k blocks of memory,
it calls old_mmap to grab 2MB of memory,
pitches the first half with munmap,
then protects the rest in 4k chunks,
after a few cycles of the mapping, unmapping and protecting, it seg faults and
dies, with exit status 0, no core file and no printed error messages, not even
"seg fault".

Anyway, here is the relevant parts of the strace (... inserted to save space and
only the last of the mapping cycles is given):

gettimeofday({969930974, 689985}, NULL) = 0
write(8, "\f\0\5\0\355bA\6\3\0\6\0\0\0\0\0\0\0\0\0\f\0\5\0\355bA"..., 40) = 40
ioctl(8, FIONREAD, [0])                 = 0
gettimeofday({969930974, 690507}, NULL) = 0
ioctl(8, FIONREAD, [0])                 = 0
poll([{fd=8, events=POLLIN}, {fd=6, events=POLLIN, revents=POLLIN}, {fd=13,
events=POLLIN}, {fd=18, events=POLLIN}], 4, 0) = 1
read(6, "\372", 1)                      = 1
read(6, "\372", 1)                      = 1
lseek(16, 16384, SEEK_SET)              = 16384
read(16, "X\0\202?~?@?<?8?\251=u=q=m=\270;|;x;V;R;N;"..., 16384) = 16384
lseek(16, 1703936, SEEK_SET)            = 1703936
read(16, "\\\0\300?\274?n?j?f?:>6>\250<a<]<Y<\267:\2:\3769\3729"..., 16384) =
16384
...
mprotect(0x428c5000, 4096, PROT_READ|PROT_WRITE) = 0
mprotect(0x428c6000, 4096, PROT_READ|PROT_WRITE) = 0
mprotect(0x428c7000, 4096, PROT_READ|PROT_WRITE) = 0
...
<some old_mmap/unmap/protect cycles>
...
old_mmap(NULL, 2097152, PROT_NONE, MAP_PRIVATE|MAP_ANONYMOUS|MAP_NORESERVE, -1,
0) = 0x43500000
munmap(0x43500000, 1048576)             = 0
munmap(0x43700000, 0)                   = -1 EINVAL (Invalid argument)
mprotect(0x43600000, 32768, PROT_READ|PROT_WRITE) = 0
mprotect(0x43608000, 4096, PROT_READ|PROT_WRITE) = 0
...
mprotect(0x43681000, 4096, PROT_READ|PROT_WRITE) = 0
mprotect(0x43682000, 4096, PROT_READ|PROT_WRITE) = 0
mprotect(0x43683000, 4096, PROT_READ|PROT_WRITE) = 0
--- SIGSEGV (Segmentation fault) ---
--- SIGSEGV (Segmentation fault) ---

Comment 1

[Chris]

This also occurs on www.weather.com.  I thought it may be the refresh tags,
because of the apparent 30min. delay, but, even though they both refresh at 1800
secs, I successfully had mozilla refresh the saved versions of the pages with
short refresh times.

Comment 2

[Doron Rosenberg (IBM)]

could you get a talkback build, by getting the sea version?

Comment 3

[Doron Rosenberg (IBM)]

could you get a talkback build, by getting the sea version?

Comment 4

[Asa Dotzler [:asa]]

the sea.tar.gz builds have talkback if you select complete or custom install.  A
talkback report would be helpful.

Comment 5

[Chris]

I have the talkback build (which has caught other stuff), which makes this even
weirder.

Comment 6

[Doron Rosenberg (IBM)]

*mass spam*
adding crash keyword to all crash bugs 

Comment 7

[buster]

I sent mail to some linux users here at NS to see if anyone can reproduce.

Comment 8

[waqar]

I let it run for more than two hours and I did not crash. Using branch from 
2000/10/17.

Comment 9

[Doron Rosenberg (IBM)]

reporter - are you still seeing this with new builds?

Comment 10

[Chris]

I ran a nightly build overnight on cnn.com with no problems, will try M18.
Maybe it got fixed somewhere else.

Comment 11

[Asa Dotzler [:asa]]

resolving as worksforme.  Please reopen if these problems return.

Comment 12

[Akkana Peck]

I left a branch build (was this seen on trunk or branch?) pointed at this site
overnight, and it was still up the next morning, though it showed a dialog about
not having been able to contact cnn.com from some refresh or other.