Signed-off-by: Rafał Miłecki <[email protected]>
---
I got reply from Bjorn, that we should *not* depend on PCI_EXP_LNKCAP:
http://marc.info/?l=linux-pci&m=140457671930896&w=2
So the change since RFC is adding a list of PCIe 1.0 and 2.0 devices.
Use PCI_EXP_LNKCAP only as a fallback, in case someone tries sth like:
echo "14e4 4360" > /sys/bus/pci/drivers/bcma-pci-bridge/new_id
---
drivers/bcma/host_pci.c | 32 ++++++++++++++++++++++++++++++--
include/linux/bcma/bcma.h | 1 +
include/linux/bcma/bcma_regs.h | 2 ++
3 files changed, 33 insertions(+), 2 deletions(-)
diff --git a/drivers/bcma/host_pci.c b/drivers/bcma/host_pci.c
index e3333053..df20a50 100644
--- a/drivers/bcma/host_pci.c
+++ b/drivers/bcma/host_pci.c
@@ -13,10 +13,12 @@
static void bcma_host_pci_switch_core(struct bcma_device *core)
{
+ int win2 = core->bus->host_is_pcie2 ?
+ BCMA_PCIE2_BAR0_WIN2 : BCMA_PCI_BAR0_WIN2;
+
pci_write_config_dword(core->bus->host_pci, BCMA_PCI_BAR0_WIN,
core->addr);
- pci_write_config_dword(core->bus->host_pci, BCMA_PCI_BAR0_WIN2,
- core->wrap);
+ pci_write_config_dword(core->bus->host_pci, win2, core->wrap);
core->bus->mapped_core = core;
bcma_debug(core->bus, "Switched to core: 0x%X\n", core->id.id);
}
@@ -194,6 +196,32 @@ static int bcma_host_pci_probe(struct pci_dev *dev,
goto err_pci_release_regions;
}
+ /* PCIe Gen 2 has some regs different */
+ switch (dev->device) {
+ case 0x0576:
+ case 0x4313:
+ case 43224:
+ case 0x4331:
+ case 0x4353:
+ case 0x4357:
+ case 0x4358:
+ case 0x4359:
+ case 0x4365:
+ case 0x4727:
+ bus->host_is_pcie2 = false;
+ break;
+ case 0x4360:
+ case 0x43a0:
+ case 0x43b1:
+ bus->host_is_pcie2 = true;
+ break;
+ default:
+ /* Fallback, e.g. if someone uses bcma-pci-bridge/new_id */
+ pcie_capability_read_dword(dev, PCI_EXP_LNKCAP, &val);
+ bus->host_is_pcie2 = !!(val & PCI_EXP_LNKCAP_SLS_5_0GB);
+ break;
+ }
+
/* Map MMIO */
err = -ENOMEM;
bus->mmio = pci_iomap(dev, 0, ~0UL);
diff --git a/include/linux/bcma/bcma.h b/include/linux/bcma/bcma.h
index 452286a..9f98e2f 100644
--- a/include/linux/bcma/bcma.h
+++ b/include/linux/bcma/bcma.h
@@ -315,6 +315,7 @@ struct bcma_bus {
const struct bcma_host_ops *ops;
enum bcma_hosttype hosttype;
+ bool host_is_pcie2; /* Valid for BCMA_HOSTTYPE_PCI only */
union {
/* Pointer to the PCI bus (only for BCMA_HOSTTYPE_PCI) */
struct pci_dev *host_pci;
diff --git a/include/linux/bcma/bcma_regs.h b/include/linux/bcma/bcma_regs.h
index 917dcd7..d02f8b5 100644
--- a/include/linux/bcma/bcma_regs.h
+++ b/include/linux/bcma/bcma_regs.h
@@ -59,6 +59,8 @@
#define BCMA_PCI_GPIO_XTAL 0x40 /* PCI config space GPIO 14 for Xtal powerup */
#define BCMA_PCI_GPIO_PLL 0x80 /* PCI config space GPIO 15 for PLL powerdown */
+#define BCMA_PCIE2_BAR0_WIN2 0x70
+
/* SiliconBackplane Address Map.
* All regions may not exist on all chips.
*/
--
1.8.4.5
On 6 July 2014 14:19, Arend van Spriel <[email protected]> wrote:
> On 06-07-14 13:11, Rafał Miłecki wrote:
>> Signed-off-by: Rafał Miłecki <[email protected]>
>> ---
>> I got reply from Bjorn, that we should *not* depend on PCI_EXP_LNKCAP:
>> http://marc.info/?l=linux-pci&m=140457671930896&w=2
>>
>> So the change since RFC is adding a list of PCIe 1.0 and 2.0 devices.
>> Use PCI_EXP_LNKCAP only as a fallback, in case someone tries sth like:
>> echo "14e4 4360" > /sys/bus/pci/drivers/bcma-pci-bridge/new_id
>
> Do you have to know this before or after enumerating the cores? They
> have a different core id, right?
Sure, having a list of cores would allow me to determine PCIe revision.
Unfortunately right now we have only this single bcma_bus_register
call that does both: scanning and initialization. I'd need to first
scan cores, then determine PCIe revision and finally initialize cores.
I wonder if I could try to use bcma_bus_scan_early that was developer
for SoC needs...
On Mon, Jul 07, 2014 at 09:37:10PM +0200, Rafał Miłecki wrote:
> On 7 July 2014 11:38, Rafał Miłecki <[email protected]> wrote:
> > On 6 July 2014 14:19, Arend van Spriel <[email protected]> wrote:
> >> On 06-07-14 13:11, Rafał Miłecki wrote:
> >>> Signed-off-by: Rafał Miłecki <[email protected]>
> >>> ---
> >>> I got reply from Bjorn, that we should *not* depend on PCI_EXP_LNKCAP:
> >>> http://marc.info/?l=linux-pci&m=140457671930896&w=2
> >>>
> >>> So the change since RFC is adding a list of PCIe 1.0 and 2.0 devices.
> >>> Use PCI_EXP_LNKCAP only as a fallback, in case someone tries sth like:
> >>> echo "14e4 4360" > /sys/bus/pci/drivers/bcma-pci-bridge/new_id
> >>
> >> Do you have to know this before or after enumerating the cores? They
> >> have a different core id, right?
> >
> > Sure, having a list of cores would allow me to determine PCIe revision.
> >
> > Unfortunately right now we have only this single bcma_bus_register
> > call that does both: scanning and initialization. I'd need to first
> > scan cores, then determine PCIe revision and finally initialize cores.
> >
> > I wonder if I could try to use bcma_bus_scan_early that was developer
> > for SoC needs...
>
> John: please kindly drop this patch for now. All other bcma/b43
> patches are unaffected.
OK
--
John W. Linville Someday the world will need a hero, and you
[email protected] might be all we have. Be ready.
On 06-07-14 13:11, Rafał Miłecki wrote:
> Signed-off-by: Rafał Miłecki <[email protected]>
> ---
> I got reply from Bjorn, that we should *not* depend on PCI_EXP_LNKCAP:
> http://marc.info/?l=linux-pci&m=140457671930896&w=2
>
> So the change since RFC is adding a list of PCIe 1.0 and 2.0 devices.
> Use PCI_EXP_LNKCAP only as a fallback, in case someone tries sth like:
> echo "14e4 4360" > /sys/bus/pci/drivers/bcma-pci-bridge/new_id
Do you have to know this before or after enumerating the cores? They
have a different core id, right?
Regards,
Arend
> ---
> drivers/bcma/host_pci.c | 32 ++++++++++++++++++++++++++++++--
> include/linux/bcma/bcma.h | 1 +
> include/linux/bcma/bcma_regs.h | 2 ++
> 3 files changed, 33 insertions(+), 2 deletions(-)
>
> diff --git a/drivers/bcma/host_pci.c b/drivers/bcma/host_pci.c
> index e3333053..df20a50 100644
> --- a/drivers/bcma/host_pci.c
> +++ b/drivers/bcma/host_pci.c
> @@ -13,10 +13,12 @@
>
> static void bcma_host_pci_switch_core(struct bcma_device *core)
> {
> + int win2 = core->bus->host_is_pcie2 ?
> + BCMA_PCIE2_BAR0_WIN2 : BCMA_PCI_BAR0_WIN2;
> +
> pci_write_config_dword(core->bus->host_pci, BCMA_PCI_BAR0_WIN,
> core->addr);
> - pci_write_config_dword(core->bus->host_pci, BCMA_PCI_BAR0_WIN2,
> - core->wrap);
> + pci_write_config_dword(core->bus->host_pci, win2, core->wrap);
> core->bus->mapped_core = core;
> bcma_debug(core->bus, "Switched to core: 0x%X\n", core->id.id);
> }
> @@ -194,6 +196,32 @@ static int bcma_host_pci_probe(struct pci_dev *dev,
> goto err_pci_release_regions;
> }
>
> + /* PCIe Gen 2 has some regs different */
> + switch (dev->device) {
> + case 0x0576:
> + case 0x4313:
> + case 43224:
> + case 0x4331:
> + case 0x4353:
> + case 0x4357:
> + case 0x4358:
> + case 0x4359:
> + case 0x4365:
> + case 0x4727:
> + bus->host_is_pcie2 = false;
> + break;
> + case 0x4360:
> + case 0x43a0:
> + case 0x43b1:
> + bus->host_is_pcie2 = true;
> + break;
> + default:
> + /* Fallback, e.g. if someone uses bcma-pci-bridge/new_id */
> + pcie_capability_read_dword(dev, PCI_EXP_LNKCAP, &val);
> + bus->host_is_pcie2 = !!(val & PCI_EXP_LNKCAP_SLS_5_0GB);
> + break;
> + }
> +
> /* Map MMIO */
> err = -ENOMEM;
> bus->mmio = pci_iomap(dev, 0, ~0UL);
> diff --git a/include/linux/bcma/bcma.h b/include/linux/bcma/bcma.h
> index 452286a..9f98e2f 100644
> --- a/include/linux/bcma/bcma.h
> +++ b/include/linux/bcma/bcma.h
> @@ -315,6 +315,7 @@ struct bcma_bus {
> const struct bcma_host_ops *ops;
>
> enum bcma_hosttype hosttype;
> + bool host_is_pcie2; /* Valid for BCMA_HOSTTYPE_PCI only */
> union {
> /* Pointer to the PCI bus (only for BCMA_HOSTTYPE_PCI) */
> struct pci_dev *host_pci;
> diff --git a/include/linux/bcma/bcma_regs.h b/include/linux/bcma/bcma_regs.h
> index 917dcd7..d02f8b5 100644
> --- a/include/linux/bcma/bcma_regs.h
> +++ b/include/linux/bcma/bcma_regs.h
> @@ -59,6 +59,8 @@
> #define BCMA_PCI_GPIO_XTAL 0x40 /* PCI config space GPIO 14 for Xtal powerup */
> #define BCMA_PCI_GPIO_PLL 0x80 /* PCI config space GPIO 15 for PLL powerdown */
>
> +#define BCMA_PCIE2_BAR0_WIN2 0x70
> +
> /* SiliconBackplane Address Map.
> * All regions may not exist on all chips.
> */
>
On 7 July 2014 11:38, Rafał Miłecki <[email protected]> wrote:
> On 6 July 2014 14:19, Arend van Spriel <[email protected]> wrote:
>> On 06-07-14 13:11, Rafał Miłecki wrote:
>>> Signed-off-by: Rafał Miłecki <[email protected]>
>>> ---
>>> I got reply from Bjorn, that we should *not* depend on PCI_EXP_LNKCAP:
>>> http://marc.info/?l=linux-pci&m=140457671930896&w=2
>>>
>>> So the change since RFC is adding a list of PCIe 1.0 and 2.0 devices.
>>> Use PCI_EXP_LNKCAP only as a fallback, in case someone tries sth like:
>>> echo "14e4 4360" > /sys/bus/pci/drivers/bcma-pci-bridge/new_id
>>
>> Do you have to know this before or after enumerating the cores? They
>> have a different core id, right?
>
> Sure, having a list of cores would allow me to determine PCIe revision.
>
> Unfortunately right now we have only this single bcma_bus_register
> call that does both: scanning and initialization. I'd need to first
> scan cores, then determine PCIe revision and finally initialize cores.
>
> I wonder if I could try to use bcma_bus_scan_early that was developer
> for SoC needs...
John: please kindly drop this patch for now. All other bcma/b43
patches are unaffected.
--
Rafał