Re: [PATCH] PCI: Do not use bus number zero from EA capability

From: sundeep subbaraya
Date: Tue Sep 24 2019 - 05:56:58 EST


Hi Andrew,

On Mon, Sep 23, 2019 at 6:05 PM Andrew Murray <andrew.murray@xxxxxxx> wrote:
>
> On Mon, Sep 02, 2019 at 09:00:03PM +0530, sundeep.lkml@xxxxxxxxx wrote:
> > From: Subbaraya Sundeep <sbhatta@xxxxxxxxxxx>
> >
> > As per the spec, "Enhanced Allocation (EA) for Memory
> > and I/O Resources" ECN, approved 23 October 2014,
> > sec 6.9.1.2, fixed bus numbers of a bridge can be zero
>
> s/can/must/
>
> The spec uses the term *must*. "Can" implies that this is optional.
>
Yes will change to must.

> > when no function that uses EA is located behind it.
> > Hence assign bus numbers sequentially when fixed bus
> > numbers are zero.
>
> Perhaps s/sequentially/as per normal/ or similar. As we're not doing
> anything different here.
>
Ok will change

> >
> > Fixes: 2dbce590117981196fe355efc0569bc6f949ae9b
>
> Is it worth describing what actually goes wrong without this patch - and
> when this occurs? I guess it's possible for a bridge to have an EA
> capability, but no devices using EA behind it - and thus in this suitation
> the downstream devices have unnecessary bus number constraints?
>
EA is for functions which are permanently connected to host bridge.
In our case all the on chip devices use EA and bridges which are there for
connecting off chip devices use EA with fixed bus numbers as zero.
Bus numbers for those bridges need to be configured as per
normal enumeration, failing to do so makes those bridges not functional
because secondary and subordinate bus numbers are 0.

> >
> > Signed-off-by: Subbaraya Sundeep <sbhatta@xxxxxxxxxxx>
>
> Does this need to be CC'd to stable?
>
Ok will CC stable from v2

> > ---
> > drivers/pci/probe.c | 25 +++++++++++++------------
> > 1 file changed, 13 insertions(+), 12 deletions(-)
> >
> > diff --git a/drivers/pci/probe.c b/drivers/pci/probe.c
> > index a3c7338..c06ca4c 100644
> > --- a/drivers/pci/probe.c
> > +++ b/drivers/pci/probe.c
> > @@ -1095,27 +1095,28 @@ static unsigned int pci_scan_child_bus_extend(struct pci_bus *bus,
> > * @sub: updated with subordinate bus number from EA
> > *
> > * If @dev is a bridge with EA capability, update @sec and @sub with
> > - * fixed bus numbers from the capability and return true. Otherwise,
> > - * return false.
> > + * fixed bus numbers from the capability. Otherwise @sec and @sub
> > + * will be zeroed.
> > */
> > -static bool pci_ea_fixed_busnrs(struct pci_dev *dev, u8 *sec, u8 *sub)
> > +static void pci_ea_fixed_busnrs(struct pci_dev *dev, u8 *sec, u8 *sub)
> > {
> > int ea, offset;
> > u32 dw;
> >
> > + *sec = *sub = 0;
> > +
> > if (dev->hdr_type != PCI_HEADER_TYPE_BRIDGE)
> > - return false;
> > + return;
> >
> > /* find PCI EA capability in list */
> > ea = pci_find_capability(dev, PCI_CAP_ID_EA);
> > if (!ea)
> > - return false;
> > + return;
> >
> > offset = ea + PCI_EA_FIRST_ENT;
> > pci_read_config_dword(dev, offset, &dw);
> > *sec = dw & PCI_EA_SEC_BUS_MASK;
> > *sub = (dw & PCI_EA_SUB_BUS_MASK) >> PCI_EA_SUB_BUS_SHIFT;
>
> Is there any value in doing any sanity checking here? E.g. sub !=0, sub > sec?
>

I don't think it is needed since we read hardwired values from HW.

> > - return true;
> > }
> >
> > /*
> > @@ -1151,7 +1152,6 @@ static int pci_scan_bridge_extend(struct pci_bus *bus, struct pci_dev *dev,
> > u16 bctl;
> > u8 primary, secondary, subordinate;
> > int broken = 0;
> > - bool fixed_buses;
> > u8 fixed_sec, fixed_sub;
> > int next_busnr;
> >
> > @@ -1254,11 +1254,12 @@ static int pci_scan_bridge_extend(struct pci_bus *bus, struct pci_dev *dev,
> > pci_write_config_word(dev, PCI_STATUS, 0xffff);
> >
> > /* Read bus numbers from EA Capability (if present) */
> > - fixed_buses = pci_ea_fixed_busnrs(dev, &fixed_sec, &fixed_sub);
> > - if (fixed_buses)
> > + pci_ea_fixed_busnrs(dev, &fixed_sec, &fixed_sub);
> > +
> > + next_busnr = max + 1;
> > + /* Use secondary bus number in EA */
> > + if (fixed_sec)
> > next_busnr = fixed_sec;
> > - else
> > - next_busnr = max + 1;
>
> There is a subtle style change here (assigning and then potentially reassigning
> with a new value vs assigning once using both if/else). No idea if this matters
> but I thought I'd point it out in case it wasn't intentional.
>
This is intentional just to avoid else case.

Thanks for review. I will send v2.
Sundeep

> Thanks,
>
> Andrew Murray
>
> >
> > /*
> > * Prevent assigning a bus number that already exists.
> > @@ -1336,7 +1337,7 @@ static int pci_scan_bridge_extend(struct pci_bus *bus, struct pci_dev *dev,
> > * If fixed subordinate bus number exists from EA
> > * capability then use it.
> > */
> > - if (fixed_buses)
> > + if (fixed_sub)
> > max = fixed_sub;
> > pci_bus_update_busn_res_end(child, max);
> > pci_write_config_byte(dev, PCI_SUBORDINATE_BUS, max);
> > --
> > 2.7.4
> >