On Wed, Jan 14, 2026 at 12:54:37PM +0900, Koichiro Den wrote:

quoted

I realized that I missed one case in v7.

I think dw_pcie_ep_clear_ib_maps() should also be called from
dw_pcie_ep_ib_atu_bar() to tear down any existing inbound mappings for the
same BAR before re-programming it in BAR Match Mode.

This matters when updating inbound mappings for a BAR without resetting the
BAR in between. There are four possible transition patterns, and pattern #4
below was overlooked:

  1. BAR Match Mode -> BAR Match Mode
     As the current implementation does, the mapping is simply updated
     (with the same atu index)

  2. BAR Match Mode -> Address Match Mode
     This patch series already ensures the old BAR Match mapping is
     torn down before reprogramming.

  3. Address Match Mode -> Address Match Mode
     Likewise, existing Address Match mappings are cleared first.

  4. Address Match Mode  -> BAR Match Mode
     This case was not handled. The change below adds the missing
     teardown so that stale Address Match mappings do not remain active.

     --- a/drivers/pci/controller/dwc/pcie-designware-ep.c
     +++ b/drivers/pci/controller/dwc/pcie-designware-ep.c
     @@ -148,9 +148,12 @@ static int dw_pcie_ep_ib_atu_bar(struct dw_pcie_ep *ep, u8 func_no, int type,
             u32 free_win;
             struct dw_pcie *pci = to_dw_pcie_from_ep(ep);
     
     -       if (!ep->bar_to_atu[bar])
     +       if (!ep->bar_to_atu[bar]) {
     +               /* Tear down existing mappings before (re)programming. */
     +               dw_pcie_ep_clear_ib_maps(ep, bar);
     +
                     free_win = find_first_zero_bit(ep->ib_window_map,
                                                   pci->num_ib_windows);
     -       else
     +       } else
                     free_win = ep->bar_to_atu[bar] - 1;

If one of the branches has braces, both branches should have braces:
https://www.kernel.org/doc/html/latest/process/coding-style.html#placing-braces-and-spaces

quoted

Unless there are objections, I'll include this fix in v8.

Isn't it easier/cleaner if we call dw_pcie_ep_clear_ib_maps() in
dw_pcie_ep_set_bar(), rather than calling it in both dw_pcie_ep_ib_atu_addr()
and dw_pcie_ep_ib_atu_bar() ?

dw_pcie_ep_set_bar() knows the condition if we are dynamically reprogramming
a BAR or not, and all the four cases are when dynamically reprogramming a BAR.

I.e. instead of adding additional code to dw_pcie_ep_ib_atu_bar(), we do
something like:

diff --git a/drivers/pci/controller/dwc/pcie-designware-ep.c b/drivers/pci/controller/dwc/pcie-designware-ep.c
index b2ea2c2c986f..63ae5471fe13 100644
--- a/drivers/pci/controller/dwc/pcie-designware-ep.c
+++ b/drivers/pci/controller/dwc/pcie-designware-ep.c

@@ -318,9 +318,6 @@ static int dw_pcie_ep_ib_atu_addr(struct dw_pcie_ep *ep, u8 func_no, int type,
                return -EINVAL;
        }
 
-       /* Tear down any existing mappings before (re)programming. */
-       dw_pcie_ep_clear_ib_maps(ep, bar);
-
        for (i = 0; i < epf_bar->num_submap; i++) {
                off = submap[i].offset;
                size = submap[i].size;

@@ -571,6 +568,9 @@ static int dw_pcie_ep_set_bar(struct pci_epc *epc, u8 func_no, u8 vfunc_no,
                    ep->epf_bar[bar]->flags != flags)
                        return -EINVAL;
 
+               if (ep->epf_bar[bar]->num_submap || epf_bar->num_submap)
+                       dw_pcie_ep_clear_ib_maps(ep, bar);
+
                /*
                 * When dynamically changing a BAR, skip writing the BAR reg, as
                 * that would clear the BAR's PCI address assigned by the host.