Re: [PATCH] cxl/acpi: Defer probe when ACPI0016 PCI root bridge is not ready

[Richard Cheng]

On Thu, May 14, 2026 at 10:32:38AM +0800, Chen Pei wrote:
> On some platforms (e.g., RISC-V and ARM64) that use the generic
> pci_acpi_scan_root() implementation, cxl_acpi_probe may run before
> acpi_pci_root driver has bound to ACPI0016 (CXL host bridge) devices.
> In this case, acpi_pci_find_root() returns NULL, causing
> to_cxl_host_bridge() to skip the device silently. This results in
> incomplete CXL port enumeration on first boot.
> 
> Fix this by detecting the case where an ACPI0016 device exists but its
> PCI root bridge is not yet ready, and returning -EPROBE_DEFER to trigger
> a deferred probe retry.
> 
> Signed-off-by: Chen Pei <cp0613@xxxxxxxxxxxxxxxxx>
> ---
>  drivers/cxl/acpi.c | 26 ++++++++++++++++++++++++--
>  1 file changed, 24 insertions(+), 2 deletions(-)
>

Hi Chen Pei,

Thanks for the patch.
I have a few questions and suggestions regarding to your changes.

First of all I would like in which scenario did you encounter the bug?
Any specific CONFIG options and the devices ? what's the error log ?

It would be nice if you can attach it for us. 

> diff --git a/drivers/cxl/acpi.c b/drivers/cxl/acpi.c
> index 127537628817..9952d0cff903 100644
> --- a/drivers/cxl/acpi.c
> +++ b/drivers/cxl/acpi.c
> @@ -631,8 +631,21 @@ static int add_host_bridge_dport(struct device *match, void *arg)
>  	struct acpi_pci_root *pci_root;
>  	struct cxl_port *root_port = arg;
>  	struct device *host = root_port->dev.parent;
> -	struct acpi_device *hb = to_cxl_host_bridge(host, match);
> +	struct acpi_device *adev = to_acpi_device(match);
> +	struct acpi_device *hb;
>  
> +	/*
> +	 * If this is an ACPI0016 device but acpi_pci_find_root() hasn't
> +	 * found the PCI root yet (driver not probed), defer the probe
> +	 * to allow acpi_pci_root to bind first.
> +	 */
> +	if (strcmp(acpi_device_hid(adev), "ACPI0016") == 0 &&
> +	    !acpi_pci_find_root(adev->handle)) {
> +		dev_dbg(host, "deferring probe, ACPI0016 PCI root not ready\n");
> +		return -EPROBE_DEFER;
> +	}

What about strncpy() here since we already know we're comparing against "ACPI0016" ?
At the same time, why not just use "acpi_dev_hid_match()" ? it's widely used across
numerous files.

> +
> +	hb = to_cxl_host_bridge(host, match);
>  	if (!hb)
>  		return 0;
>  
> @@ -688,7 +701,8 @@ static int add_host_bridge_uport(struct device *match, void *arg)
>  {
>  	struct cxl_port *root_port = arg;
>  	struct device *host = root_port->dev.parent;
> -	struct acpi_device *hb = to_cxl_host_bridge(host, match);
> +	struct acpi_device *adev = to_acpi_device(match);
> +	struct acpi_device *hb;
>  	struct acpi_pci_root *pci_root;
>  	struct cxl_dport *dport;
>  	struct cxl_port *port;
> @@ -697,6 +711,14 @@ static int add_host_bridge_uport(struct device *match, void *arg)
>  	resource_size_t component_reg_phys;
>  	int rc;
>  
> +	/* Same deferral check as in add_host_bridge_dport() */
> +	if (strcmp(acpi_device_hid(adev), "ACPI0016") == 0 &&
> +	    !acpi_pci_find_root(adev->handle)) {
> +		dev_dbg(host, "deferring probe, ACPI0016 PCI root not ready\n");
> +		return -EPROBE_DEFER;
> +	}
> +
> +	hb = to_cxl_host_bridge(host, match);
>  	if (!hb)
>  		return 0;
>  
> -- 
> 2.50.1
> 
> 

These 2 checks are basically the same, can we put it in a static inline helper or
a macro if possible? something like the following might be better

```
static int cxl_acpi_defer_host_bridge(struct device *host,
				      struct acpi_device *adev)
{
	if (acpi_dev_hid_match(adev, "ACPI0016") &&
	    !acpi_pci_find_root(adev->handle)) {
		dev_dbg(host, "deferring probe, ACPI0016 PCI root not ready\n");
		return -EPROBE_DEFER;
	}
	return 0;
}
```
and use it in your code like

```
int rc = cxl_acpi_defer_host_bridge(host, adev);
if (rc)
	return rc;
```


Last but not least, have you run the kselftests of CXL ? some mock bridges
are platform devices, not ACPI devices, you are using "to_acpi_device(match)", this
is not a safe runtime check when "match" is a platform_device, the code will read the memory
layout wrongly.


Best regards,
Richard Cheng.