drivers/pci: Fix FIELD_PREP compilation error with gcc-8

[PATCH] drivers/pci: Fix FIELD_PREP compilation error with gcc-8

Posted by Anders Roxell 1 month ago

Commit cbc654d18d37 ("bitops: Add __attribute_const__ to generic
ffs()-family implementations") causes a compilation failure on ARM
footbridge_defconfig with gcc-8:

  FIELD_PREP: value too large for the field

The error occurs in pcie_set_readrq() at:
  v = FIELD_PREP(PCI_EXP_DEVCTL_READRQ, ffs(rq) - 8);

With __attribute_const__, gcc-8 now performs wrong compile-time
validation in FIELD_PREP and cannot guarantee that ffs(rq) - 8 will
always produce values that fit in the 3-bit PCI_EXP_DEVCTL_READRQ field.

Avoid FIELD_PREP entirely by using direct bit manipulation. Replace
FIELD_PREP(PCI_EXP_DEVCTL_READRQ, ffs(rq) - 8) with the equivalent
manual bit operations: ((ffs(rq) - 8) << 12) & PCI_EXP_DEVCTL_READRQ.

This bypasses the compile-time validation while maintaining identical
runtime behavior and functionality.

Fixes: cbc654d18d37 ("bitops: Add __attribute_const__ to generic ffs()-family implementations")
Reported-by: Linux Kernel Functional Testing <lkft@linaro.org>
Closes: https://lore.kernel.org/linux-pci/CA+G9fYuysVr6qT8bjF6f08WLyCJRG7aXAeSd2F7=zTaHHd7L+Q@mail.gmail.com/T/#u
Signed-off-by: Anders Roxell <anders.roxell@linaro.org>
---
 drivers/pci/pci.c | 3 ++-
 1 file changed, 2 insertions(+), 1 deletion(-)

diff --git a/drivers/pci/pci.c b/drivers/pci/pci.c
index e698278229f2..9f9607bd9f51 100644
--- a/drivers/pci/pci.c
+++ b/drivers/pci/pci.c
@@ -5893,7 +5893,8 @@ int pcie_set_readrq(struct pci_dev *dev, int rq)
 			rq = mps;
 	}
 
-	v = FIELD_PREP(PCI_EXP_DEVCTL_READRQ, ffs(rq) - 8);
+	/* Ideally we would used FIELD_PREP() but this is a work around for gcc-8 */
+	v = ((ffs(rq) - 8) << 12) & PCI_EXP_DEVCTL_READRQ;
 
 	if (bridge->no_inc_mrrs) {
 		int max_mrrs = pcie_get_readrq(dev);
-- 
2.50.1

Re: [PATCH] drivers/pci: Fix FIELD_PREP compilation error with gcc-8

Posted by David Laight 3 weeks, 6 days ago

On Thu, 28 Aug 2025 12:12:37 +0200
Anders Roxell <anders.roxell@linaro.org> wrote:

> Commit cbc654d18d37 ("bitops: Add __attribute_const__ to generic
> ffs()-family implementations") causes a compilation failure on ARM
> footbridge_defconfig with gcc-8:
> 
>   FIELD_PREP: value too large for the field
> 
> The error occurs in pcie_set_readrq() at:
>   v = FIELD_PREP(PCI_EXP_DEVCTL_READRQ, ffs(rq) - 8);
> 
> With __attribute_const__, gcc-8 now performs wrong compile-time
> validation in FIELD_PREP and cannot guarantee that ffs(rq) - 8 will
> always produce values that fit in the 3-bit PCI_EXP_DEVCTL_READRQ field.

Which is actually quite correct - in principle pcie_get_mps() could
return a small value.

What is probably happening is that two copies of the FIELD_PREP()
are being generated for ffs(rq) and ffs(mps).
The latter might be (mps ? asm_fun(mps) + 1 : 0) leading to an extra
copy for ffs(0) - which will cause the warning in FIELD_PREP.

An alternate fix you be to move the validation of rq below the
'performance' clamp.

	David


> 
> Avoid FIELD_PREP entirely by using direct bit manipulation. Replace
> FIELD_PREP(PCI_EXP_DEVCTL_READRQ, ffs(rq) - 8) with the equivalent
> manual bit operations: ((ffs(rq) - 8) << 12) & PCI_EXP_DEVCTL_READRQ.
> 
> This bypasses the compile-time validation while maintaining identical
> runtime behavior and functionality.
> 
> Fixes: cbc654d18d37 ("bitops: Add __attribute_const__ to generic ffs()-family implementations")
> Reported-by: Linux Kernel Functional Testing <lkft@linaro.org>
> Closes: https://lore.kernel.org/linux-pci/CA+G9fYuysVr6qT8bjF6f08WLyCJRG7aXAeSd2F7=zTaHHd7L+Q@mail.gmail.com/T/#u
> Signed-off-by: Anders Roxell <anders.roxell@linaro.org>
> ---
>  drivers/pci/pci.c | 3 ++-
>  1 file changed, 2 insertions(+), 1 deletion(-)
> 
> diff --git a/drivers/pci/pci.c b/drivers/pci/pci.c
> index e698278229f2..9f9607bd9f51 100644
> --- a/drivers/pci/pci.c
> +++ b/drivers/pci/pci.c
> @@ -5893,7 +5893,8 @@ int pcie_set_readrq(struct pci_dev *dev, int rq)
>  			rq = mps;
>  	}
>  
> -	v = FIELD_PREP(PCI_EXP_DEVCTL_READRQ, ffs(rq) - 8);
> +	/* Ideally we would used FIELD_PREP() but this is a work around for gcc-8 */
> +	v = ((ffs(rq) - 8) << 12) & PCI_EXP_DEVCTL_READRQ;
>  
>  	if (bridge->no_inc_mrrs) {
>  		int max_mrrs = pcie_get_readrq(dev);

Re: [PATCH] drivers/pci: Fix FIELD_PREP compilation error with gcc-8

Posted by Kees Cook 4 weeks, 1 day ago

On Thu, Aug 28, 2025 at 12:12:37PM +0200, Anders Roxell wrote:
> Commit cbc654d18d37 ("bitops: Add __attribute_const__ to generic
> ffs()-family implementations") causes a compilation failure on ARM
> footbridge_defconfig with gcc-8:
> 
>   FIELD_PREP: value too large for the field
> 
> The error occurs in pcie_set_readrq() at:
>   v = FIELD_PREP(PCI_EXP_DEVCTL_READRQ, ffs(rq) - 8);
> 
> With __attribute_const__, gcc-8 now performs wrong compile-time
> validation in FIELD_PREP and cannot guarantee that ffs(rq) - 8 will
> always produce values that fit in the 3-bit PCI_EXP_DEVCTL_READRQ field.

Thanks for examining this! It seems rather alarming -- why did it
work before?

> Avoid FIELD_PREP entirely by using direct bit manipulation. Replace
> FIELD_PREP(PCI_EXP_DEVCTL_READRQ, ffs(rq) - 8) with the equivalent
> manual bit operations: ((ffs(rq) - 8) << 12) & PCI_EXP_DEVCTL_READRQ.
> 
> This bypasses the compile-time validation while maintaining identical
> runtime behavior and functionality.

Did you dig into why this happened? It seems like a fragile situation,
so I'm worried we'll see more of these pop up.

> Fixes: cbc654d18d37 ("bitops: Add __attribute_const__ to generic ffs()-family implementations")
> Reported-by: Linux Kernel Functional Testing <lkft@linaro.org>
> Closes: https://lore.kernel.org/linux-pci/CA+G9fYuysVr6qT8bjF6f08WLyCJRG7aXAeSd2F7=zTaHHd7L+Q@mail.gmail.com/T/#u
> Signed-off-by: Anders Roxell <anders.roxell@linaro.org>
> ---
>  drivers/pci/pci.c | 3 ++-
>  1 file changed, 2 insertions(+), 1 deletion(-)
> 
> diff --git a/drivers/pci/pci.c b/drivers/pci/pci.c
> index e698278229f2..9f9607bd9f51 100644
> --- a/drivers/pci/pci.c
> +++ b/drivers/pci/pci.c
> @@ -5893,7 +5893,8 @@ int pcie_set_readrq(struct pci_dev *dev, int rq)
>  			rq = mps;
>  	}
>  
> -	v = FIELD_PREP(PCI_EXP_DEVCTL_READRQ, ffs(rq) - 8);
> +	/* Ideally we would used FIELD_PREP() but this is a work around for gcc-8 */
> +	v = ((ffs(rq) - 8) << 12) & PCI_EXP_DEVCTL_READRQ;
>  
>  	if (bridge->no_inc_mrrs) {
>  		int max_mrrs = pcie_get_readrq(dev);

If you're sure this is okay, I'll take it with the series, but I feel
like we should justify it better. :)

-Kees

-- 
Kees Cook

Re: [PATCH] drivers/pci: Fix FIELD_PREP compilation error with gcc-8

Posted by Kees Cook 4 weeks ago

On Wed, Sep 03, 2025 at 09:29:40PM -0700, Kees Cook wrote:
> Thanks for examining this! It seems rather alarming -- why did it
> work before?

I've proposed an alternative:
https://lore.kernel.org/lkml/20250905052836.work.425-kees@kernel.org/

-- 
Kees Cook