On Thu, Mar 7, 2024 at 8:31 PM Björn Töpel <bjorn@xxxxxxxxxx> wrote:
>
> Anup Patel <apatel@xxxxxxxxxxxxxxxx> writes:
>
> > On Wed, Mar 6, 2024 at 9:22 PM Björn Töpel <bjorn@xxxxxxxxxx> wrote:
> >>
> >> Anup Patel <apatel@xxxxxxxxxxxxxxxx> writes:
> >>
> >> > diff --git a/drivers/irqchip/irq-riscv-aplic-msi.c b/drivers/irqchip/irq-riscv-aplic-msi.c
> >> > new file mode 100644
> >> > index 000000000000..b2a25e011bb2
> >> > --- /dev/null
> >> > +++ b/drivers/irqchip/irq-riscv-aplic-msi.c
> >> > +static void aplic_msi_write_msg(struct irq_data *d, struct msi_msg *msg)
> >> > +{
> >> > + unsigned int group_index, hart_index, guest_index, val;
> >> > + struct aplic_priv *priv = irq_data_get_irq_chip_data(d);
> >> > + struct aplic_msicfg *mc = &priv->msicfg;
> >> > + phys_addr_t tppn, tbppn, msg_addr;
> >> > + void __iomem *target;
> >> > +
> >> > + /* For zeroed MSI, simply write zero into the target register */
> >> > + if (!msg->address_hi && !msg->address_lo && !msg->data) {
> >> > + target = priv->regs + APLIC_TARGET_BASE;
> >> > + target += (d->hwirq - 1) * sizeof(u32);
> >> > + writel(0, target);
> >>
> >> Is the fence needed here (writel_relaxed())...
> >
> > The pci_write_msg_msix() (called via pci_msi_domain_write_msg())
> > uses writel() hence taking inspiration from that we use writel() over here
> > as well.
> >
> > If that's wrong then pci_write_msg_msix() must be fixed as well.
>
> Huh? The writel()s in pci_write_msg_msix() are because there's an
> ordering constraint, and code would be broken w/o it. My question was
> "what are the ordering constraints for this piece of code", because it
> looks like this is a single I/O write without any ordering constraints.
Whatever ordering constraints apply to pci_write_msg_msix() also
apply to APLIC MSI-mode because both create the leaf-level IRQ
domain for the client device driver (PCIe or Platform device) whose
parent is IMSIC base domain.
>
> I'm not a fan of sprinkling fences around "to be safe", but I don't want
> to delay the v16 because of it. It can be fixed later, if it's not
> needed.
I don't think there is a clear way of proving that using write_relaxed()
in aplic_msi_write_msg() is safe considering there is a vast variety
of platform drivers who would be clients of the APLIC MSI-mode
domain.
I agree that we should deal with this later.
Regards,
Anup
