|
On Monday, 6 June 2016, 18:33:00 EDT, jim burns wrote:
> On Mon June 6 2016, 07:14:26 PM, Julien Grall wrote:
> >
> >
> > Have you tried to revert this patch and see if it works?
> >
> >
> >
> > Regards,
> >
> >
> >
> > --
> > Julien Grall
>
> Thanx for the interest. Unfortunately, as stated, this is an old computer,
> and takes 1 1/2 hrs to compile the kernel at runlevel 3. Re-compiles would
> not be my first choice for the use of my time!
So I actually did try to compile the kernel, and after 1 1/2 hrs, my disk ran out of space! I've since installed Fedora 24 on a faster computer, with lots of space. I recompiled Fedora's 4.5.6 src.rpm with Stefano's patch removed (methodology discussed below), and the long and short of it that the 4.5.6 kernel w/o Stefano's patch does not boot, with exactly the same problem - it hangs at the dracut.initqueue target, then drops to an emergency shell because the hard disk controller (IDE) wasn't initialized.
Methodology: I unpacked the src.rpm, unpacked the kernel, and tried to edit
./arch/x86/pci/xen.c, and realized that it hadn't been patched yet. This became an education in patch procedures.
The patches that end up in the ~/rpmbuild/SOURCES when you unpack the src.rpm are Fedora patches, not kernel patches. However, one of those files is the kernel patch file, patch-4.5.6.xz. I uncompressed this, removed Stefano's legacy irqs patch, recompressed it, overwrote the original patch-4.5.6.xz in the SOURCES dir, ran 'rpmbuild -bb etc.', and 1 1/2 hrs later, installed the rpms on the older, slower computer, and rebooted, with the results as above.
> I noticed that 4.4.12 also has Stefano's patch, and I installed it today,
> and lo and behold, it works (where 4.4.9 didn't). So - preponderance of
> the evidence? I'm waiting for Fedora to supply rpms for 4.6.1 or higher to
> cinch it (since 4.6.1 also has Stefano's patch).
>
> BTW, I've noticed that xen MSI init tends to cluster at the beginning of
> dmesg, instead of being spread out as was the previous norm. Is this a
> side effect of Stefano's patch?:
>
> [ 0.000000] Using NULL legacy PIC
And yes, w/o Stefano's patch, the xen MSI pirq inits don't cluster right after the above stmt. | 
