On Fri, 2011-11-18 at 13:52 +0000, Konrad Rzeszutek Wilk wrote:So we have two candidates:
We are using this combo in swiotlb-xen and as well in the xen pci.+ xen_raw_printk(str);I expect you've just copied this style from elsewhere but I really
+ panic(str);
dislike this duplication of prints. If panic is not useful here we
really ought to address that at the root instead of going around
patching things to print every panic message twice. I thought
earlyprintk was supposed to solve this problem. Perhaps a generic
early_panic_print could be added to the panic code?
We could declere a 'xen_raw_panic' that would do the job?
The problem is that panic() uses the "late" printk mechanism (so
it goes through the buffer that ends up not beign flushed) and the
panic never sees the light.
So lets fix that instead of working around it...
The 'xen_raw_printk' is synchronous..
But I wonder if the panic surfaces if 'earlyprintk=xen' is used?
At which point it might be that the those extra xen_raw_printk
become pointless?
I think panic's do come out with earlyprintk (unless they are truly
super early).
Ian.--