Re: [PATCH v4 13/18] PCI: dwc: Let glue drivers hide the Root Port MSI capabilities

[Shawn Lin <shawn.lin@rock-chips.com>]

Hi Koichiro

在 2026/10/15 星期四 11:30, Koichiro Den 写道:
> Commit f5cd8a929c82 ("PCI: dwc: Remove MSI/MSIX capability for Root Port
> if iMSI-RX is used as MSI controller") hides the Root Port MSI/MSI-X
> capabilities when iMSI-RX is the MSI controller, as it cannot receive
> the Root Port's own MSIs.
> 
> Some platforms cannot deliver them with an external MSI controller
> either. The DesignWare core signals Root Port interrupts (AER, PME,
> bandwidth, hotplug) as sideband signals on its SII interface and only
> turns them into MSI writes with the optional SII2MSI feature. Otherwise
> it is up to the vendor wrapper. The controller on R-Car Gen4 SoCs
> (DWC 5.20a) appears to lack SII2MSI, and its APP block merely latches
> the signals into status registers wired to platform IRQs, so an MSI
> targeted at the GIC ITS is never sent.
> 
> Add a no_rp_msi flag so glue drivers can request the same INTx fallback
> regardless of the MSI controller in use. It is the counterpart of
> 'keep_rp_msi_en' from commit 3a4e8302e72f ("PCI: imx6: Keep Root Port
> MSI capability with iMSI-RX to work around hardware bug"), which keeps
> precedence: that one is needed for endpoint MSIs to work at all, whereas
> hiding the capability only moves the Root Port services to INTx. The
> removal stays in dw_pcie_setup_rc(), which also runs after a Root Port
> reset.
> 

With it, dw_pcie_rp will carry three flags with overlapping and
priority-dependent semantics:

   use_imsi_rx     - the MSI controller is iMSI-RX
   keep_rp_msi_en  - do NOT remove the RP MSI caps (i.MX6 workaround)
   no_rp_msi       - the platform never gets the RP's own MSIs

and the resulting condition:

   if ((pp->use_imsi_rx || pp->no_rp_msi) && !pp->keep_rp_msi_en)

is getting hard to reason about. Adding no_rp_msi for a single platform
whose controller is not even iMSI-RX makes the combination worse.

Since only R-Car Gen4 needs this, and it has its own host init, would it
be fine for the glue driver to call dw_pcie_remove_capability()
directly? I'd like to avoid growing the shared struct if a local
solution is possible.

> Signed-off-by: Koichiro Den <den@valinux.co.jp>
> ---
> Changes since v3:
>    - No changes.
> 
>   drivers/pci/controller/dwc/pcie-designware-host.c | 6 ++++--
>   drivers/pci/controller/dwc/pcie-designware.h      | 1 +
>   2 files changed, 5 insertions(+), 2 deletions(-)
> 
> diff --git a/drivers/pci/controller/dwc/pcie-designware-host.c b/drivers/pci/controller/dwc/pcie-designware-host.c
> index 774e1936d999..32e179603cf4 100644
> --- a/drivers/pci/controller/dwc/pcie-designware-host.c
> +++ b/drivers/pci/controller/dwc/pcie-designware-host.c
> @@ -1191,9 +1191,11 @@ int dw_pcie_setup_rc(struct dw_pcie_rp *pp)
>   	 * The iMSI-RX module does not support receiving MSI or MSI-X generated
>   	 * by the Root Port. If iMSI-RX is used as the MSI controller, remove
>   	 * the MSI and MSI-X capabilities of the Root Port to allow the drivers
> -	 * to fall back to INTx instead.
> +	 * to fall back to INTx instead. Glue drivers set no_rp_msi when the
> +	 * platform never delivers the Root Port's own MSIs, whatever the MSI
> +	 * controller in use.
>   	 */
> -	if (pp->use_imsi_rx && !pp->keep_rp_msi_en) {
> +	if ((pp->use_imsi_rx || pp->no_rp_msi) && !pp->keep_rp_msi_en) {
>   		dw_pcie_remove_capability(pci, PCI_CAP_ID_MSI);
>   		dw_pcie_remove_capability(pci, PCI_CAP_ID_MSIX);
>   	}
> diff --git a/drivers/pci/controller/dwc/pcie-designware.h b/drivers/pci/controller/dwc/pcie-designware.h
> index 2ce61709b58c..d94116402b74 100644
> --- a/drivers/pci/controller/dwc/pcie-designware.h
> +++ b/drivers/pci/controller/dwc/pcie-designware.h
> @@ -445,6 +445,7 @@ struct dw_pcie_host_ops {
>   struct dw_pcie_rp {
>   	bool			use_imsi_rx:1;
>   	bool			keep_rp_msi_en:1;
> +	bool			no_rp_msi:1;
>   	bool			cfg0_io_shared:1;
>   	u64			cfg0_base;
>   	void __iomem		*va_cfg0_base;