Re: [REGRESSION][BISECTED] 5.15-rc1: Broken AHCI on NVIDIA ION (MCP79)

[Josef Johansson <josef@oderland.se>]

On 10/12/21 15:33, Josef Johansson wrote:
> On 10/12/21 15:07, Jason Andryuk wrote:
>> On Tue, Oct 12, 2021 at 2:09 AM Josef Johansson <josef@oderland.se> wrote:
>>> On 10/11/21 21:34, Josef Johansson wrote:
>>>> On 10/11/21 20:47, Josef Johansson wrote:
>>>>> More can be read over at freedesktop:
>>>>> https://gitlab.freedesktop.org/drm/amd/-/issues/1715
>> Hi, Josef,
>>
>> If you compare
>> commit fcacdfbef5a1633211ebfac1b669a7739f5b553e "PCI/MSI: Provide a
>> new set of mask and unmask functions"
>> and
>> commit 446a98b19fd6da97a1fb148abb1766ad89c9b767 "PCI/MSI: Use new
>> mask/unmask functions" some of the replacement functions in 446198b1
>> no longer exit early for the pci_msi_ignore_mask flag.
>>
>> Josef, I'd recommend you try adding pci_msi_ignore_mask checks to the
>> new functions in fcacdfbef5a to see if that helps.
>>
>> There was already a pci_msi_ignore_mask fixup in commit
>> 1a519dc7a73c977547d8b5108d98c6e769c89f4b "PCI/MSI: Skip masking MSI-X
>> on Xen PV" though the kernel was crashing in that case.
>>
>> Regards,
>> Jason
> Hi Jason,
>
> Makes sense. I am compiling now, will try it as soon as it's done.
>
> diff --git a/drivers/pci/msi.c b/drivers/pci/msi.c
> index 0099a00af361..620928fd0065 100644
> --- a/drivers/pci/msi.c
> +++ b/drivers/pci/msi.c
> @@ -148,6 +148,9 @@ static noinline void pci_msi_update_mask(struct
> msi_desc *desc, u32 clear, u32 s
>      raw_spinlock_t *lock = &desc->dev->msi_lock;
>      unsigned long flags;
>  
> +    if (pci_msi_ignore_mask)
> +        return;
> +
>      raw_spin_lock_irqsave(lock, flags);
>      desc->msi_mask &= ~clear;
>      desc->msi_mask |= set;
> @@ -179,6 +182,9 @@ static inline void __iomem
> *pci_msix_desc_addr(struct msi_desc *desc)
>   */
>  static void pci_msix_write_vector_ctrl(struct msi_desc *desc, u32 ctrl)
>  {
> +    if (pci_msi_ignore_mask)
> +        return;
> +
>      void __iomem *desc_addr = pci_msix_desc_addr(desc);
>  
>      writel(ctrl, desc_addr + PCI_MSIX_ENTRY_VECTOR_CTRL);
> @@ -186,6 +192,9 @@ static void pci_msix_write_vector_ctrl(struct
> msi_desc *desc, u32 ctrl)
>  
>  static inline void pci_msix_mask(struct msi_desc *desc)
>  {
> +    if (pci_msi_ignore_mask)
> +        return;
> +
>      desc->msix_ctrl |= PCI_MSIX_ENTRY_CTRL_MASKBIT;
>      pci_msix_write_vector_ctrl(desc, desc->msix_ctrl);
>      /* Flush write to device */
> @@ -194,6 +203,9 @@ static inline void pci_msix_mask(struct msi_desc *desc)
>  
>  static inline void pci_msix_unmask(struct msi_desc *desc)
>  {
> +    if (pci_msi_ignore_mask)
> +        return;
> +
>      desc->msix_ctrl &= ~PCI_MSIX_ENTRY_CTRL_MASKBIT;
>      pci_msix_write_vector_ctrl(desc, desc->msix_ctrl);
>  }
>
I love open source. It just works. Was my patch correct btw?

Thanks Jason!

Regards

- Josef