Re: [PATCH] scsi: ufs: qcom: fix platform_msi_domain_free_irqs() reference

[Can Guo]

Hi Mani,

On 1/30/2023 3:45 PM, Manivannan Sadhasivam wrote:
> On Mon, Jan 30, 2023 at 12:03:45PM +0800, Can Guo wrote:
> > Hi Mani and Arnd,
> >
> > On 1/27/2023 2:02 PM, Manivannan Sadhasivam wrote:
> > > On Thu, Jan 26, 2023 at 10:17:31PM +0100, Arnd Bergmann wrote:
> > > > From: Arnd Bergmann <arnd@xxxxxxxx>
> > > >
> > > > The newly added MSI support is mostly hidden inside of an #ifdef,
> > > > except for one line that now causes a build failure when MSI
> > > > is disabled:
> > > >
> > > > drivers/ufs/host/ufs-qcom.c: In function 'ufs_qcom_remove':
> > > > drivers/ufs/host/ufs-qcom.c:1698:9: error: implicit declaration of function 'platform_msi_domain_free_irqs' [-Werror=i]
> > > >    1698 |         platform_msi_domain_free_irqs(hba->dev);
> > > >         |         ^~~~~~~~~~~~~~~~~~~~~~~~~~~~~
> > > >
> > > > Above that, the symbol that guards the other call was recently
> > > > removed, so that is all dead code at the moment.
> > > >
> > > > Remove the incorrect #ifdef and instead of a Kconfig dependency
> > > > to only allow building the driver when CONFIG_GENERIC_MSI_IRQ
> > > > is enabled. This symbol is always present when PCI_MSI
> > > > or ARM_GIC_V3_ITS are enabled, both of which should be present
> > > > on kernels that can run on Qualcomm SoCs.
> > > >
> > > > The 'select RESET_CONTROLLER' in combination with this dependency
> > > > unfortunately causes a dependency loop and this is a user-visible
> > > > symbol, so it's better to change both to 'depends on'.
> > > >
> > > > Fixes: 519b6274a777 ("scsi: ufs: qcom: Add MCQ ESI config vendor specific ops")
> > > > Fixes: 13e7accb81d6 ("genirq: Get rid of GENERIC_MSI_IRQ_DOMAIN")
> > > > Signed-off-by: Arnd Bergmann <arnd@xxxxxxxx>
> > > Reviewed-by: Manivannan Sadhasivam <mani@xxxxxxxxxx>
> > >
> > > > ---
> > > > Not sure if this is the best solution, both the GENERIC_MSI_IRQ
> > > > dependencies and the RESET_CONTROLLER dependencies are a bit
> > > > inconsistent already. Feel free to pick another approach that
> > > > addresses both of the bugs I found.
> > > I think your proposed solution works best at the moment.
> > >
> > > Thanks,
> > > Mani
> > >
> > > > ---
> > > >    drivers/ufs/host/Kconfig    | 3 ++-
> > > >    drivers/ufs/host/ufs-qcom.c | 8 --------
> > > >    2 files changed, 2 insertions(+), 9 deletions(-)
> > > >
> > > > diff --git a/drivers/ufs/host/Kconfig b/drivers/ufs/host/Kconfig
> > > > index 139064e70a34..663881437921 100644
> > > > --- a/drivers/ufs/host/Kconfig
> > > > +++ b/drivers/ufs/host/Kconfig
> > > > @@ -57,8 +57,9 @@ config SCSI_UFS_DWC_TC_PLATFORM
> > > >    config SCSI_UFS_QCOM
> > > >    	tristate "QCOM specific hooks to UFS controller platform driver"
> > > >    	depends on SCSI_UFSHCD_PLATFORM && ARCH_QCOM
> > > > +	depends on GENERIC_MSI_IRQ
> > > > +	depends on RESET_CONTROLLER
> > > >    	select QCOM_SCM if SCSI_UFS_CRYPTO
> > > > -	select RESET_CONTROLLER
> > > >    	help
> > > >    	  This selects the QCOM specific additions to UFSHCD platform driver.
> > > >    	  UFS host on QCOM needs some vendor specific configuration before
> > > > diff --git a/drivers/ufs/host/ufs-qcom.c b/drivers/ufs/host/ufs-qcom.c
> > > > index 681da3ea7154..eb66b5f6cf19 100644
> > > > --- a/drivers/ufs/host/ufs-qcom.c
> > > > +++ b/drivers/ufs/host/ufs-qcom.c
> > > > @@ -1538,7 +1538,6 @@ static int ufs_qcom_get_outstanding_cqs(struct ufs_hba *hba,
> > > >    	return 0;
> > > >    }
> > > > -#ifdef CONFIG_GENERIC_MSI_IRQ_DOMAIN
> > > >    static void ufs_qcom_write_msi_msg(struct msi_desc *desc, struct msi_msg *msg)
> > > >    {
> > > >    	struct device *dev = msi_desc_to_dev(desc);
> > > > @@ -1626,13 +1625,6 @@ static int ufs_qcom_config_esi(struct ufs_hba *hba)
> > > >    	return ret;
> > > >    }
> > > > -#else
> > > > -static int ufs_qcom_config_esi(struct ufs_hba *hba)
> > > > -{
> > > > -	return -EOPNOTSUPP;
> > > > -}
> > > > -#endif
> > > > -
> > > >    /*
> > > >     * struct ufs_hba_qcom_vops - UFS QCOM specific variant operations
> > > >     *
> > > > --
> > > > 2.39.0
> > Thank you for pointing to 13e7accb81d6 ("genirq: Get rid of
> > GENERIC_MSI_IRQ_DOMAIN"),
> >
> > which I was not aware of. I am thinking about fixing it just like how
> >
> > 13e7accb81d6 ("genirq: Get rid of GENERIC_MSI_IRQ_DOMAIN") is fixing
> > drivers/dma/qcom/hidma.c -
> >
> > In ufs-qcom.c, use CONFIG_GENERIC_MSI_IRQ instead of
> > CONFIG_GENERIC_MSI_IRQ_DOMAIN,
> >
> > and meanwhile add #ifdef check before calling
> > platform_msi_domain_free_irqs().
> >
> > Please let me know your idea.
> No. Let's get rid of the ifdef's please... I like Arnd's patch as it models the
> dependency nicely in Kconfig. And most (all?) of the qcom platforms require
> GENERIC_MSI_IRQ and RESET_CONTROLLER one way or the other anyway.
>
> Thanks,
> Mani


Make sense, I agree.


Thanks,

Can Guo.

> > Thanks.
> > Regards,
> > Can Guo