[PATCH Resend] PCI: Limit islolated function probing on bus 0 for LoongArch

[PATCH Resend] PCI: Limit islolated function probing on bus 0 for LoongArch

[Huacai Chen]

We found some discrete AMD graphics devices hide funtion 0 and the whole
is not supposed to be probed.

Since our original purpose is to allow integrated devices (on bus 0) to
be probed without function 0, we can limit the islolated function probing
only on bus 0.

Cc: stable@vger.kernel.org
Fixes: a02fd05661d73a8 ("PCI: Extend isolated function probing to LoongArch")
Signed-off-by: Huacai Chen <chenhuacai@loongson.cn>
---
Resend to correct the subject line.

 drivers/pci/probe.c        | 2 +-
 include/linux/hypervisor.h | 8 +++++---
 2 files changed, 6 insertions(+), 4 deletions(-)

diff --git a/drivers/pci/probe.c b/drivers/pci/probe.c
index c83e75a0ec12..da6a2aef823a 100644
--- a/drivers/pci/probe.c
+++ b/drivers/pci/probe.c
@@ -2883,7 +2883,7 @@ int pci_scan_slot(struct pci_bus *bus, int devfn)
 			 * a hypervisor that passes through individual PCI
 			 * functions.
 			 */
-			if (!hypervisor_isolated_pci_functions())
+			if (!hypervisor_isolated_pci_functions(bus->number))
 				break;
 		}
 		fn = next_fn(bus, dev, fn);
diff --git a/include/linux/hypervisor.h b/include/linux/hypervisor.h
index be5417303ecf..30ece04a16d9 100644
--- a/include/linux/hypervisor.h
+++ b/include/linux/hypervisor.h
@@ -32,13 +32,15 @@ static inline bool jailhouse_paravirt(void)
 
 #endif /* !CONFIG_X86 */
 
-static inline bool hypervisor_isolated_pci_functions(void)
+static inline bool hypervisor_isolated_pci_functions(int bus)
 {
 	if (IS_ENABLED(CONFIG_S390))
 		return true;
 
-	if (IS_ENABLED(CONFIG_LOONGARCH))
-		return true;
+	if (IS_ENABLED(CONFIG_LOONGARCH)) {
+		if (bus == 0)
+			return true;
+	}
 
 	return jailhouse_paravirt();
 }
-- 
2.47.3

Re: [PATCH Resend] PCI: Limit islolated function probing on bus 0 for LoongArch

[Bjorn Helgaas]

On Tue, Oct 14, 2025 at 03:41:00PM +0800, Huacai Chen wrote:
> We found some discrete AMD graphics devices hide funtion 0 and the whole
> is not supposed to be probed.
> 
> Since our original purpose is to allow integrated devices (on bus 0) to
> be probed without function 0, we can limit the islolated function probing
> only on bus 0.

s/islolated/isolated/ (multiple)
s/funtion/function/

I suppose this fixes some problem where:

  - An AMD GPU is on some bus other than 00
  - The GPU has no function 0
  - Without function 0, we normally don't probe other functions
  - a02fd05661d7 means we *do* probe other functions on LoongArch
  - Therefore we find some non-0 function we're not supposed to find

If that's the case, what bad thing happens?  Is there some dmesg hint
we can include in the commit log?

I suppose this means such devices are potentially broken for s390 and
jailhouse as well?

> Cc: stable@vger.kernel.org
> Fixes: a02fd05661d73a8 ("PCI: Extend isolated function probing to LoongArch")

a02fd05661d7 (12-char SHA1 is conventional)

> -static inline bool hypervisor_isolated_pci_functions(void)
> +static inline bool hypervisor_isolated_pci_functions(int bus)
>  {
>  	if (IS_ENABLED(CONFIG_S390))
>  		return true;
>  
> -	if (IS_ENABLED(CONFIG_LOONGARCH))
> -		return true;
> +	if (IS_ENABLED(CONFIG_LOONGARCH)) {
> +		if (bus == 0)

I don't really like this embedded assumption that the root bus is bus
00.  That's not necessarily the case; we have many host bridges that
lead to a bus other than 00.

> +			return true;
> +	}
>  
>  	return jailhouse_paravirt();
>  }
> -- 
> 2.47.3
>