| Message ID | 1470891318-23046-6-git-send-email-gwshan@linux.vnet.ibm.com |
|---|---|
| State | Accepted |
| Headers | show |
On 11/08/16 14:55, Gavin Shan wrote: > The ECRC generation and check can't be enabled on Broadcom's NIC > (14e4:168a) when it seats behind PMC PCIe switch downstream port > (11f8:8546). Otherwise, the NIC's config space can not be accessed Could you specify the NIC's model name as well as its vendor/device ID? Andrew
On Thu, Aug 11, 2016 at 03:07:47PM +1000, Andrew Donnellan wrote: >On 11/08/16 14:55, Gavin Shan wrote: >>The ECRC generation and check can't be enabled on Broadcom's NIC >>(14e4:168a) when it seats behind PMC PCIe switch downstream port >>(11f8:8546). Otherwise, the NIC's config space can not be accessed > >Could you specify the NIC's model name as well as its vendor/device ID? > vendor/device ID are already included in the commit log: 14e4:168a. I will amend with a descriptive string if Stewart asks for a update. Otherwise, it's fine not to provide that as vendor/device IDs are (almost) good enough. Thanks, Gavin
Gavin Shan <gwshan@linux.vnet.ibm.com> writes: > The ECRC generation and check can't be enabled on Broadcom's NIC > (14e4:168a) when it seats behind PMC PCIe switch downstream port > (11f8:8546). Otherwise, the NIC's config space can not be accessed > and returns 0xFF's on read because of EEH error even after the error > is cleared. The issue is reported from Firestone. > > This disables ECRC generation and check on Broadcom's NIC when it > seats behind PMC PCIe switch downstream port. With this applied, > the NIC can be detected successfully. > > Reported-by: Li Meng <shlimeng@cn.ibm.com> > Signed-off-by: Gavin Shan <gwshan@linux.vnet.ibm.com> Should this also go to stable?
On Tue, Oct 11, 2016 at 03:09:53PM +1100, Stewart Smith wrote: >Gavin Shan <gwshan@linux.vnet.ibm.com> writes: >> The ECRC generation and check can't be enabled on Broadcom's NIC >> (14e4:168a) when it seats behind PMC PCIe switch downstream port >> (11f8:8546). Otherwise, the NIC's config space can not be accessed >> and returns 0xFF's on read because of EEH error even after the error >> is cleared. The issue is reported from Firestone. >> >> This disables ECRC generation and check on Broadcom's NIC when it >> seats behind PMC PCIe switch downstream port. With this applied, >> the NIC can be detected successfully. >> >> Reported-by: Li Meng <shlimeng@cn.ibm.com> >> Signed-off-by: Gavin Shan <gwshan@linux.vnet.ibm.com> > >Should this also go to stable? > I don't think it needn't go stable necessarily as the PMC switch isn't a intergarted one. So the issue was found on customized board, not a standard (or official) one. Thanks, Gavin
diff --git a/hw/phb3.c b/hw/phb3.c index 20af9a7..3fce81d 100644 --- a/hw/phb3.c +++ b/hw/phb3.c @@ -405,6 +405,24 @@ static void phb3_switch_port_init(struct phb *phb, pci_cfg_write32(phb, bdfn, aercap + PCIECAP_AER_CAPCTL, val32); } +static inline bool phb3_endpoint_report_ecrc(struct pci_device *pd) +{ + if (!pd || !pd->parent) + return true; + + /* No ECRC generation and check on Broadcom ethernet adapter + * when it seats behind a PMC's PCIe switch downstream port. + * Otherwise, the Broadcom ethernet adapter's config space + * can't be accessed because of frozen PE error even after + * the frozen PE error is cleared. + */ + if (pd->vdid == 0x168a14e4 || + pd->parent->vdid == 0x854611f8) + return false; + + return true; +} + static void phb3_endpoint_init(struct phb *phb, struct pci_device *dev, int ecap, int aercap) @@ -442,8 +460,12 @@ static void phb3_endpoint_init(struct phb *phb, /* Enable ECRC generation and check */ pci_cfg_read32(phb, bdfn, aercap + PCIECAP_AER_CAPCTL, &val32); - val32 |= (PCIECAP_AER_CAPCTL_ECRCG_EN | - PCIECAP_AER_CAPCTL_ECRCC_EN); + if (phb3_endpoint_report_ecrc(dev)) + val32 |= (PCIECAP_AER_CAPCTL_ECRCG_EN | + PCIECAP_AER_CAPCTL_ECRCC_EN); + else + val32 &= ~(PCIECAP_AER_CAPCTL_ECRCG_EN | + PCIECAP_AER_CAPCTL_ECRCC_EN); pci_cfg_write32(phb, bdfn, aercap + PCIECAP_AER_CAPCTL, val32); }
The ECRC generation and check can't be enabled on Broadcom's NIC (14e4:168a) when it seats behind PMC PCIe switch downstream port (11f8:8546). Otherwise, the NIC's config space can not be accessed and returns 0xFF's on read because of EEH error even after the error is cleared. The issue is reported from Firestone. This disables ECRC generation and check on Broadcom's NIC when it seats behind PMC PCIe switch downstream port. With this applied, the NIC can be detected successfully. Reported-by: Li Meng <shlimeng@cn.ibm.com> Signed-off-by: Gavin Shan <gwshan@linux.vnet.ibm.com> --- hw/phb3.c | 26 ++++++++++++++++++++++++-- 1 file changed, 24 insertions(+), 2 deletions(-)