lkml.org 
[lkml]   [2024]   [Apr]   [17]   [last100]   RSS Feed
Views: [wrap][no wrap]   [headers]  [forward] 
 
Messages in this thread
Patch in this message
/
From
Date
SubjectRe: [PATCH v2] of: property: Add fw_devlink support for interrupt-map property
On Wed, Apr 17, 2024 at 6:34 AM Saravana Kannan <saravanak@google.com> wrote:
>
> On Tue, Apr 16, 2024 at 7:11 AM Rob Herring <robh@kernel.org> wrote:
> >
> > On Sat, Apr 13, 2024 at 02:49:42PM +0530, Anup Patel wrote:
> > > Some of the PCI controllers (such as generic PCI host controller)
> > > use "interrupt-map" DT property to describe the mapping between
> > > PCI endpoints and PCI interrupt pins. This the only case where
> > > the interrupts are not described in DT.
> > >
> > > Currently, there is no fw_devlink created based on "interrupt-map"
> > > DT property
>
> parse_interrupts() calls of_irq_parse_one() that in turn calls
> of_irq_parse_one() that does the "interrupts-map" handling. In fact
> of_irq_parse_pci() also calls of_irq_parse_one() if the PCI device has
> a DT node. So I don't think any new functionality needs to be added.
> If I'm not mistaken we just need parse_interrupts to not ignore
> interrupts-map? A one line change?
>
> Why do we need all of this code you wrote below?

The of_irq_parse_one() calls of_irq_parse_raw() only if the DT
node has "interrupts" or "interrupts-extended" DT property. This
means for most PCI host controller DT nodes, the of_irq_parse_one()
will not return any interrupts.

Here's an example PCI host DT node from the RISC-V world
(but this also applies to other architectures):

pci@30000000 {
compatible = "pci-host-ecam-generic";
device_type = "pci";
#address-cells = <0x03>;
#size-cells = <0x02>;
#interrupt-cells = <0x01>;
interrupt-map-mask = <0x0 0x00 0x00 0x07>;
interrupt-map = <0x00 0x00 0x00 0x01 &aplic_slevel 28 0x04>,
<0x00 0x00 0x00 0x02 &aplic_slevel 29 0x04>,
<0x00 0x00 0x00 0x03 &aplic_slevel 30 0x04>,
<0x00 0x00 0x00 0x04 &aplic_slevel 31 0x04>;
reg = <0x0 0x30000000 0x0 0x10000000>;
ranges = <0x01000000 0x0 0x0 0x0 0x1fff0000 0x00000000 0x10000>,
<0x02000000 0x0 0x40000000 0x0 0x40000000 0x0 0x40000000>,
<0x03000000 0x100 0x0 0x100 0x0 0x4 0x0>;
interrupt-parent = <&aplic_slevel>;
msi-parent = <&imsic_slevel>;
bus-range = <0x00 0xff>;
linux,pci-domain = <0x00>;
};

In the above example, the "interrupt-map" DT property maps a set
of PCIe endpoints to APLIC interrupt sources but this only applies
if the corresponding PCIe endpoints are actually present. Also,
which entry of "interrupt-map" DT property is used for a PCIe
endpoint also depends on the PCI requester ID (bus-device-func)
assigned to the PCIe endpoint.

Regards,
Anup

>
> -Saravana
>
> > > so interrupt controller is not guaranteed to be probed
> > > before PCI host controller. This affects every platform where both
> > > PCI host controller and interrupt controllers are probed as regular
> > > platform devices.
> > >
> > > This creates fw_devlink between consumers (PCI host controller) and
> > > supplier (interrupt controller) based on "interrupt-map" DT property.
> > >
> > > Signed-off-by: Anup Patel <apatel@ventanamicro.com>
> > > ---
> > > Changes since v1:
> > > - Updated commit description based on Rob's suggestion
> > > - Use of_irq_parse_raw() for parsing interrupt-map DT property
> > > ---
> > > drivers/of/property.c | 58 +++++++++++++++++++++++++++++++++++++++++++
> > > 1 file changed, 58 insertions(+)
> > >
> > > diff --git a/drivers/of/property.c b/drivers/of/property.c
> > > index a6358ee99b74..67be66384dac 100644
> > > --- a/drivers/of/property.c
> > > +++ b/drivers/of/property.c
> > > @@ -1311,6 +1311,63 @@ static struct device_node *parse_interrupts(struct device_node *np,
> > > return of_irq_parse_one(np, index, &sup_args) ? NULL : sup_argsnp;
> > > }
> > >
> > > +static struct device_node *parse_interrupt_map(struct device_node *np,
> > > + const char *prop_name, int index)
> > > +{
> > > + const __be32 *imap, *imap_end, *addr;
> > > + struct of_phandle_args sup_args;
> > > + struct device_node *tn, *ipar;
> > > + u32 addrcells, intcells;
> > > + int i, j, imaplen;
> > > +
> > > + if (!IS_ENABLED(CONFIG_OF_IRQ))
> > > + return NULL;
> > > +
> > > + if (strcmp(prop_name, "interrupt-map"))
> > > + return NULL;
> > > +
> > > + ipar = of_node_get(np);
> > > + do {
> > > + if (!of_property_read_u32(ipar, "#interrupt-cells", &intcells))
> > > + break;
> > > + tn = ipar;
> > > + ipar = of_irq_find_parent(ipar);
> > > + of_node_put(tn);
> > > + } while (ipar);
> >
> > No need for this loop. We've only gotten here if 'interrupt-map' is
> > present in the node and '#interrupt-cells' is required if
> > 'interrupt-map' is present.
> >
> > > + if (!ipar)
> > > + return NULL;
> > > + addrcells = of_bus_n_addr_cells(ipar);
> > > + of_node_put(ipar);
> > > +
> > > + imap = of_get_property(np, "interrupt-map", &imaplen);
> > > + if (!imap || imaplen <= (addrcells + intcells))
> > > + return NULL;
> > > + imap_end = imap + imaplen;
> > > +
> > > + sup_args.np = NULL;
> > > + for (i = 0; i <= index && imap < imap_end; i++) {
> > > + if (sup_args.np) {
> > > + of_node_put(sup_args.np);
> > > + sup_args.np = NULL;
> > > + }
> > > +
> > > + addr = imap;
> > > + imap += addrcells;
> > > +
> > > + sup_args.np = np;
> > > + sup_args.args_count = intcells;
> > > + for (j = 0; j < intcells; j++)
> > > + sup_args.args[j] = be32_to_cpu(imap[j]);
> > > + imap += intcells;
> > > +
> > > + if (of_irq_parse_raw(addr, &sup_args))
> > > + return NULL;
> > > + imap += sup_args.args_count + 1;
> > > + }
> >
> > Doesn't this leak a ref on the last time the function is invoked? For
> > example, if we have 2 entries and index is 2. We'll get index=1, but
> > then exit because imap==imap_end. We need a put on index==1 node.
> >
> > Look at my next branch where I've converted things to use __free()
> > cleanups. I don't see it helping here as-is, but maybe when it is
> > correct.
> >
> > Rob

\
 
 \ /
  Last update: 2024-04-17 08:34    [W:0.356 / U:0.128 seconds]
©2003-2020 Jasper Spaans|hosted at Digital Ocean and TransIP|Read the blog|Advertise on this site