[EMAIL PROTECTED] wrote:
In my mail, I have the following from a very useful newsgroup served from vmware themselves:On Wed, 8 Jan 2003, Mike Burger wrote:http://www.vmware.comYou're right. I wasn't specific. I meant to the patch itself... :-)
Subject:
Re: Redhat 8.0 Host, VMware for Linux WS 3.2 seems to work..
From:
Petr Vandrovec <[EMAIL PROTECTED]>
Date:
Fri, 11 Oct 2002 19:25:55 +0200
Newsgroups:
vmware.for-linux.experimental
Yue Shi Lai wrote:
>>
>> Quantum Mechanics wrote:
>>
>
>>> > mharris said it is a performance issue. They change the priority of io
>>> > threads to -10 automatically in the kernel. This is what causes the
>>> > problem with VMware. Reverting the patch will likely cause a performance
>>> > problem relating to anti-aliased fonts on slower machines. On my Athlon XP
>>> > 1700+(1.47ghz) I haven't noticed a difference.
>
>>
>> But still, with two other methods available, which do not change the
>> priority behavior (much), it is not a good idea to simply shutdown the
>> prioritisation and circumvent the problem by this way.
No. Prioritisation RedHat put into kernel is just stupid beyond belief.
"a = nice(0); iopl(3); b = nice(0);" must return same values for a and b.
As iopl() can be done only by root's apps, they can call nice() themselves
if they think that they need higher priority.
>> If you use my libnice.so and activate the debugging mode, you will
>> notice that the problem is related to nice() returning -1 and not simply
>> setting the thread to -10 by default. Returning -1 is caused by changing
>> from the priority -10 by -19 to -20 (minimum nice) and then from -20 by
>> 19. Getting the priority to 0 and then -19 to change to 0 circumvents
>> problem, but this is very crude and rude.
>> Therefore, your description "this is what causes the problem with
>> VMware" is incorrect. Changing this will trigger another prioritisation
>> behavior circumventing the problem, but it is not *the* problem.
New nice() behavior is just pure crap invented by braindead engineer.
You must do:
errno = 0;
ret = nice(xxx);
if (ret == -1 && errno != 0) panic("AIO: NOT IMPLEMENTED... ")
and it is stupid, as other parts of standard say that errno is
undefined when function did not fail. So nice() always setting
errno to 12345 on success is also conforming.
And BTW, I put new vmware update package on platan:
ftp://platan.vc.cvut.cz/pub/vmware/vmware-any-any-update19.tar.gz.
As you can derive from filename, we finally have one update
package for all VMware versions here, beginning with
VMware Express, through VMware 2/3, ending with GSX 1/2.
VMware 3.2 support is tested on latest 2.5.41-bk (the one which
does not provide get_free_page anymore), GSX2 on 2.5.41, and
other systems... well, they worked before and I hope that I
did not break them. You can complain here, if it does not work.
Petr Vandrovec
[EMAIL PROTECTED]
--
redhat-list mailing list
unsubscribe mailto:[EMAIL PROTECTED]?subject=unsubscribe
https://listman.redhat.com/mailman/listinfo/redhat-list
