> -----Original Message----- > From: Sergei Shtylyov [mailto:sergei.shtylyov@xxxxxxxxxxxxxxxxxx] > Sent: Tuesday, November 6, 2018 11:34 PM > To: Guo, Chaohong <chaohong.guo@xxxxxxxxx>; linux-ide@xxxxxxxxxxxxxxx > Cc: b.zolnierkie@xxxxxxxxxxx; axboe@xxxxxxxxx; vireshk@xxxxxxxxxx > Subject: Re: [PATCH v3 1/4] Allow system to allocate IRQ 0 to ATA devices > > On 11/06/2018 05:42 AM, Chaohong guo wrote: > > > Interrupt vector 0 can be used on some platform. > > It exists even on x86. However, it's never passed to request_irq() there IIRC, > just to setup_irq()... [Guo, Chaohong] Yes. normally on x86 platform, the vector 0 won't be allocated to device. It is a system vector for handling the exception of dividing zero. After my changing, the AHCI driver can work with vector 0 in non-root jailhouse cell just because jailhouse will remap it via VT-D interrupt remapping. How do you think about it ? Do we fix it in somewhere else to prevent kernel in non-root cell from allocating vector less than 0x20 ? -minskey > > > In libata, the routine > > ata_host_activate() considers irq=0 as invalid. > > As a matter of fact, Linus told everybody to consider IRQ0 invalid, hence > this code perhaps... > > > As a result, when running > > linux in non-root cell of Jailhouse, if we allocate just one PCI ATA > > device to the guest, the device will get an IRQ of value 0. Although > > IRQ0 is perfectly legal, ATA device will fail to start up. > > I've just rechecked the PCI specs and they say that on x86 IRQ0 is valid, so > your software is OK. > > > Signed-off-by: Chaohong guo <chaohong.guo@xxxxxxxxx> > > --- > > drivers/ata/libata-core.c | 2 +- > > 1 file changed, 1 insertion(+), 1 deletion(-) > > Like Bart said, this patch alone would break some drivers... > > [...] > > MBR, Sergei
