On Tue, Aug 24, 2021 at 9:05 PM Kefeng Wang [off-list ref] wrote:

On 2021/8/25 4:08, Saravana Kannan wrote:

quoted

On Tue, Aug 24, 2021 at 1:05 PM Rob Herring [off-list ref] wrote:

quoted

+Saravana

Saravana mentioned to me there may be some issues with this one...


On Mon, Aug 16, 2021 at 2:43 AM Kefeng Wang [off-list ref] wrote:

quoted

of_amba_device_create() uses irq_of_parse_and_map() to translate
a DT interrupt specification into a Linux virtual interrupt number.

But it doesn't properly handle the case where the interrupt controller
is not yet available, eg, when pl011 interrupt is connected to MBIGEN
interrupt controller, because the mbigen initialization is too late,
which will lead to no IRQ due to no IRQ domain found, log is shown below,
   "irq: no irq domain found for uart0 !"

use of_irq_get() to return -EPROBE_DEFER as above, and in the function
amba_device_try_add()/amba_device_add(), it will properly handle in such
case, also return 0 in other fail cases to be consistent as before.

Cc: Russell King <linux@armlinux.org.uk>
Cc: Rob Herring <robh+dt@kernel.org>
Cc: Frank Rowand <redacted>
Reported-by: Ruizhe Lin <redacted>
Signed-off-by: Kefeng Wang <redacted>
---
  drivers/amba/bus.c    | 27 +++++++++++++++++++++++++++
  drivers/of/platform.c |  6 +-----
  2 files changed, 28 insertions(+), 5 deletions(-)

diff --git a/drivers/amba/bus.c b/drivers/amba/bus.c
index 36f2f42c8014..720aa6cdd402 100644
--- a/drivers/amba/bus.c
+++ b/drivers/amba/bus.c

@@ -19,6 +19,7 @@
  #include <linux/clk/clk-conf.h>
  #include <linux/platform_device.h>
  #include <linux/reset.h>
+#include <linux/of_irq.h>

  #include <asm/irq.h>

@@ -371,12 +372,38 @@ static void amba_device_release(struct device *dev)
         kfree(d);
  }

+static int of_amba_device_decode_irq(struct amba_device *dev)
+{
+       struct device_node *node = dev->dev.of_node;
+       int i, irq = 0;
+
+       if (IS_ENABLED(CONFIG_OF_IRQ) && node) {
+               /* Decode the IRQs and address ranges */
+               for (i = 0; i < AMBA_NR_IRQS; i++) {
+                       irq = of_irq_get(node, i);
+                       if (irq < 0) {
+                               if (irq == -EPROBE_DEFER)
+                                       return irq;
+                               irq = 0;
+                       }
+
+                       dev->irq[i] = irq;
+               }
+       }
+
+       return 0;
+}
+
  static int amba_device_try_add(struct amba_device *dev, struct resource *parent)
  {
         u32 size;
         void __iomem *tmp;
         int i, ret;

+       ret = of_amba_device_decode_irq(dev);
+       if (ret)
+               goto err_out;
+

Similar to other resources the AMBA bus "gets" for the device, I think
this should be moved into amba_probe() and not here. There's no reason
to delay the addition of the device (and loading its module) because
the IRQ isn't ready yet.

The following code in the amba_device_try_add() will be called, it uses irq[0]
and irq[1], so I put of_amba_device_decode_irq() into amba_device_try_add().

470         if (dev->irq[0])
471                 ret = device_create_file(&dev->dev, &dev_attr_irq0);
472         if (ret == 0 && dev->irq[1])
473                 ret = device_create_file(&dev->dev, &dev_attr_irq1);
474         if (ret == 0)
475                 return ret;

of_amba_device_decode_irq() in amba_device_try_add() won't lead to issue,
only delay the device add, right?

But delaying the device add is the issue. For example, adding a device
could trigger the loading of the corresponding module using uevents.
But now this change would delay that step. That can have other
unintended consequences -- slowing down boot, what if the driver was
working fine without the IRQ, etc.

If make it into amba_probe(), the above code should be moved too, could we
make a new patch to move both of them, or don't move them?

I'd say move them both. If Russell hasn't already picked this up, then
I'd say redo your Patch 3/3.

Btw, I've been working on [1] cleaning up the one-off deferred probe
solution that we have for amba devices. That causes a bunch of other
headaches. Your patch 3/3 takes us further in the wrong direction by
adding more reasons for delaying the addition of the device.

-Saravana

[1] - https://lore.kernel.org/lkml/CAGETcx8b228nDUho3cX9AAQ-pXOfZTMv8cj2vhdx9yc_pk8q+A@mail.gmail.com/ (local)