On Fri, Mar 08, 2024 at 11:11:52AM +0530, Manivannan Sadhasivam wrote:
> On Thu, Mar 07, 2024 at 10:43:19PM +0100, Niklas Cassel wrote:
> > On Mon, Mar 04, 2024 at 02:52:20PM +0530, Manivannan Sadhasivam wrote:
> > > The PCIe link can go to LINK_DOWN state in one of the following scenarios:
> > >
> > > 1. Fundamental (PERST#)/hot/warm reset
> > > 2. Link transition from L2/L3 to L0
> > >
> > > In those cases, LINK_DOWN causes some non-sticky DWC registers to loose the
> > > state (like REBAR, PTM_CAP etc...). So the drivers need to reinitialize
> > > them to function properly once the link comes back again.
> > >
> > > This is not a problem for drivers supporting PERST# IRQ, since they can
> > > reinitialize the registers in the PERST# IRQ callback. But for the drivers
> > > not supporting PERST#, there is no way they can reinitialize the registers
> > > other than relying on LINK_DOWN IRQ received when the link goes down. So
> > > let's add a DWC generic API dw_pcie_ep_linkdown() that reinitializes the
> > > non-sticky registers and also notifies the EPF drivers about link going
> > > down.
> > >
> > > This API can also be used by the drivers supporting PERST# to handle the
> > > scenario (2) mentioned above.
> > >
> > > Signed-off-by: Manivannan Sadhasivam <manivannan.sadhasivam@xxxxxxxxxx>
> > > ---
> > > drivers/pci/controller/dwc/pcie-designware-ep.c | 111 ++++++++++++++----------
> > > drivers/pci/controller/dwc/pcie-designware.h | 5 ++
> > > 2 files changed, 72 insertions(+), 44 deletions(-)
> > >
> > > diff --git a/drivers/pci/controller/dwc/pcie-designware-ep.c b/drivers/pci/controller/dwc/pcie-designware-ep.c
> > > index 278bdc9b2269..fed4c2936c78 100644
> > > --- a/drivers/pci/controller/dwc/pcie-designware-ep.c
> > > +++ b/drivers/pci/controller/dwc/pcie-designware-ep.c
> > > @@ -14,14 +14,6 @@
> > > #include <linux/pci-epc.h>
> > > #include <linux/pci-epf.h>
> > >
> > > -void dw_pcie_ep_linkup(struct dw_pcie_ep *ep)
> > > -{
> > > - struct pci_epc *epc = ep->epc;
> > > -
> > > - pci_epc_linkup(epc);
> > > -}
> > > -EXPORT_SYMBOL_GPL(dw_pcie_ep_linkup);
> > > -
> > > void dw_pcie_ep_init_notify(struct dw_pcie_ep *ep)
> > > {
> > > struct pci_epc *epc = ep->epc;
> > > @@ -603,19 +595,56 @@ static unsigned int dw_pcie_ep_find_ext_capability(struct dw_pcie *pci, int cap)
> > > return 0;
> > > }
> > >
> > > +static void dw_pcie_ep_init_non_sticky_registers(struct dw_pcie *pci)
> > > +{
> > > + unsigned int offset, ptm_cap_base;
> > > + unsigned int nbars;
> > > + u32 reg, i;
> > > +
> > > + offset = dw_pcie_ep_find_ext_capability(pci, PCI_EXT_CAP_ID_REBAR);
> > > + ptm_cap_base = dw_pcie_ep_find_ext_capability(pci, PCI_EXT_CAP_ID_PTM);
> > > +
> > > + dw_pcie_dbi_ro_wr_en(pci);
> > > +
> > > + if (offset) {
> > > + reg = dw_pcie_readl_dbi(pci, offset + PCI_REBAR_CTRL);
> > > + nbars = (reg & PCI_REBAR_CTRL_NBAR_MASK) >>
> > > + PCI_REBAR_CTRL_NBAR_SHIFT;
> > > +
> > > + for (i = 0; i < nbars; i++, offset += PCI_REBAR_CTRL)
> > > + dw_pcie_writel_dbi(pci, offset + PCI_REBAR_CAP, 0x0);
> >
> > If you look at PCI_REBAR_CAP, you will see that it is sticky,
> > but you have to actually read the databook to see that:
> >
> > "The RESBAR_CTRL_REG_BAR_SIZE field is automatically updated
> > when you write to RESBAR_CAP_REG_0_REG through the DBI."
> >
> > So the reason why we need to write this register, even though
> > it is sticky, is to update the RESBAR_CTRL_REG_BAR_SIZE register,
> > which is not sticky :)
> >
> > (Perhaps we should add that as a comment?)
> >
>
> Yeah, makes sense.
Note that I add a (unrelated) comment related to REBAR_CAP in this patch:
https://lore.kernel.org/linux-pci/20240307111520.3303774-1-cassel@xxxxxxxxxx/T/#u
But once we move/add code to dw_pcie_ep_init_non_sticky_registers(), I think
that it might be a good "rule" to have a small comment for each write in
dw_pcie_ep_init_non_sticky_registers() which explains why the code should be
in dw_pcie_ep_init_non_sticky_registers() instead of dw_pcie_ep_init_registers(),
even if it just a small:
/* Field PCI_XXX_YYY.ZZZ is non-sticky */
writel_dbi(pci, offset + PCI_XXX_YYY, 0);
Kind regards,
Niklas
