Re: [PATCH][kvmtool] virtio/pci: Size the MSI-X bar according to the number of MSI-X

[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

 



Hi Andre,

On Tue, 31 Aug 2021 12:10:35 +0100,
Andre Przywara <andre.przywara@xxxxxxx> wrote:
> 
> On Fri, 27 Aug 2021 12:54:05 +0100
> Marc Zyngier <maz@xxxxxxxxxx> wrote:
> 
> Hi Marc,
> 
> > Since 45d3b59e8c45 ("kvm tools: Increase amount of possible interrupts
> > per PCI device"), the number of MSI-S has gone from 4 to 33.
> > 
> > However, the corresponding storage hasn't been upgraded, and writing
> > to the MSI-X table is a pretty risky business. Now that the Linux
> > kernel writes to *all* MSI-X entries before doing anything else
> > with the device, kvmtool dies a horrible death.
> > 
> > Fix it by properly defining the size of the MSI-X bar, and make
> > Linux great again.
> > 
> > This includes some fixes the PBA region decoding, as well as minor
> > cleanups to make this code a bit more maintainable.
> > 
> > Signed-off-by: Marc Zyngier <maz@xxxxxxxxxx>
> 
> Many thanks for fixing this, it looks good to me now. Just some
> questions below:
> 
> > ---
> >  virtio/pci.c | 42 ++++++++++++++++++++++++++++++------------
> >  1 file changed, 30 insertions(+), 12 deletions(-)
> > 
> > diff --git a/virtio/pci.c b/virtio/pci.c
> > index eb91f512..41085291 100644
> > --- a/virtio/pci.c
> > +++ b/virtio/pci.c
> > @@ -7,6 +7,7 @@
> >  #include "kvm/irq.h"
> >  #include "kvm/virtio.h"
> >  #include "kvm/ioeventfd.h"
> > +#include "kvm/util.h"
> >  
> >  #include <sys/ioctl.h>
> >  #include <linux/virtio_pci.h>
> > @@ -14,6 +15,13 @@
> >  #include <assert.h>
> >  #include <string.h>
> >  
> > +#define ALIGN_UP(x, s)		ALIGN((x) + (s) - 1, (s))
> > +#define VIRTIO_NR_MSIX		(VIRTIO_PCI_MAX_VQ + VIRTIO_PCI_MAX_CONFIG)
> > +#define VIRTIO_MSIX_TABLE_SIZE	(VIRTIO_NR_MSIX * 16)
> > +#define VIRTIO_MSIX_PBA_SIZE	(ALIGN_UP(VIRTIO_MSIX_TABLE_SIZE, 64) / 8)
> > +#define VIRTIO_MSIX_BAR_SIZE	(1UL << fls_long(VIRTIO_MSIX_TABLE_SIZE + \
> > +						 VIRTIO_MSIX_PBA_SIZE))
> > +
> >  static u16 virtio_pci__port_addr(struct virtio_pci *vpci)
> >  {
> >  	return pci__bar_address(&vpci->pci_hdr, 0);
> > @@ -333,18 +341,27 @@ static void virtio_pci__msix_mmio_callback(struct kvm_cpu *vcpu,
> >  	struct virtio_pci *vpci = vdev->virtio;
> >  	struct msix_table *table;
> >  	u32 msix_io_addr = virtio_pci__msix_io_addr(vpci);
> > +	u32 pba_offset;
> >  	int vecnum;
> >  	size_t offset;
> >  
> > -	if (addr > msix_io_addr + PCI_IO_SIZE) {
> 
> Ouch, the missing "=" looks like another long standing bug you fixed, I
> wonder how this ever worked before? Looking deeper it looks like the
> whole PBA code was quite broken (allowing writes, for instance, and
> mixing with the code for the MSIX table)?

I don't think it ever worked. And to be fair, no known guest ever
reads from it either. It just that as I was reworking it, some of the
pitfalls became obvious.

> 
> > +	BUILD_BUG_ON(VIRTIO_NR_MSIX > (sizeof(vpci->msix_pba) * 8));
> > +
> > +	pba_offset = vpci->pci_hdr.msix.pba_offset & ~PCI_MSIX_TABLE_BIR;
> 
> Any particular reason you read back the offset from the MSIX capability
> instead of just using VIRTIO_MSIX_TABLE_SIZE here? Is that to avoid
> accidentally diverging in the future, by having just one place of
> definition?

Exactly. My first version of this patch actually failed to update the
offset advertised to the guest, so I decided to just have a single
location for this. At least, we won't have to touch this code again if
we change the number of MSI-X.

> 
> > +	if (addr >= msix_io_addr + pba_offset) {
> > +		/* Read access to PBA */
> >  		if (is_write)
> >  			return;
> > -		table  = (struct msix_table *)&vpci->msix_pba;
> > -		offset = addr - (msix_io_addr + PCI_IO_SIZE);
> > -	} else {
> > -		table  = vpci->msix_table;
> > -		offset = addr - msix_io_addr;
> > +		offset = addr - (msix_io_addr + pba_offset);
> > +		if ((offset + len) > sizeof (vpci->msix_pba))
> > +			return;
> > +		memcpy(data, (void *)&vpci->msix_pba + offset, len);
> 
> Should this be a char* cast, since pointer arithmetic on void* is
> somewhat frowned upon (aka "forbidden in the C standard, but allowed as
> a GCC extension")?

I am trying to be consistent. A quick grep shows at least 19
occurrences of pointer arithmetic with '(void *)', and none with
'(char *)'. Happy for someone to go and repaint this, but I don't
think this should be the purpose of this patch.

Thanks,

	M.

-- 
Without deviation from the norm, progress is not possible.
_______________________________________________
kvmarm mailing list
kvmarm@xxxxxxxxxxxxxxxxxxxxx
https://lists.cs.columbia.edu/mailman/listinfo/kvmarm



[Index of Archives]     [Linux KVM]     [Spice Development]     [Libvirt]     [Libvirt Users]     [Linux USB Devel]     [Linux Audio Users]     [Yosemite News]     [Linux Kernel]     [Linux SCSI]

  Powered by Linux