| Message ID | 1474769434-5756-1-git-send-email-okaya@codeaurora.org (mailing list archive) |
|---|---|
| State | New, archived |
| Delegated to: | Bjorn Helgaas |
| Headers | show |
On 9/24/2016 10:10 PM, Sinan Kaya wrote: > Each PCIe device can issue up to 32 transactions at a time by default. > Each transaction is tracked by a tag number on the bus. 32 outstanding > transactions is not enough for some performance critical applications > especially when a lot of small sized frames are transmitted. > > Extended tags support increases this number to 256. Devices not > supporting extended tags tie-off this field to 0. According to ECN, it > is safe to enable this feature for all PCIe endpoints. > > Signed-off-by: Sinan Kaya <okaya@codeaurora.org> > --- > drivers/pci/probe.c | 7 +++++++ > 1 file changed, 7 insertions(+) > > diff --git a/drivers/pci/probe.c b/drivers/pci/probe.c > index 93f280d..2424f38 100644 > --- a/drivers/pci/probe.c > +++ b/drivers/pci/probe.c > @@ -1505,12 +1505,19 @@ static void program_hpp_type2(struct pci_dev *dev, struct hpp_type2 *hpp) > */ > } > > +static int pci_configure_extended_tags(struct pci_dev *dev) > +{ I should have checked the capability here before trying to enable it. I'll post a follow up patch on this. Is there any other feedback? > + return pcie_capability_set_word(dev, PCI_EXP_DEVCTL, > + PCI_EXP_DEVCTL_EXT_TAG); > +} > + > static void pci_configure_device(struct pci_dev *dev) > { > struct hotplug_params hpp; > int ret; > > pci_configure_mps(dev); > + pci_configure_extended_tags(dev); > > memset(&hpp, 0, sizeof(hpp)); > ret = pci_get_hp_params(dev, &hpp); >
On Thu, Nov 10, 2016 at 01:35:41PM -0500, Sinan Kaya wrote: > On 9/24/2016 10:10 PM, Sinan Kaya wrote: > > Each PCIe device can issue up to 32 transactions at a time by default. > > Each transaction is tracked by a tag number on the bus. 32 outstanding > > transactions is not enough for some performance critical applications > > especially when a lot of small sized frames are transmitted. > > > > Extended tags support increases this number to 256. Devices not > > supporting extended tags tie-off this field to 0. According to ECN, it > > is safe to enable this feature for all PCIe endpoints. > > > > Signed-off-by: Sinan Kaya <okaya@codeaurora.org> > > --- > > drivers/pci/probe.c | 7 +++++++ > > 1 file changed, 7 insertions(+) > > > > diff --git a/drivers/pci/probe.c b/drivers/pci/probe.c > > index 93f280d..2424f38 100644 > > --- a/drivers/pci/probe.c > > +++ b/drivers/pci/probe.c > > @@ -1505,12 +1505,19 @@ static void program_hpp_type2(struct pci_dev *dev, struct hpp_type2 *hpp) > > */ > > } > > > > +static int pci_configure_extended_tags(struct pci_dev *dev) > > +{ > > I should have checked the capability here before trying to enable it. > I'll post a follow up patch on this. > > Is there any other feedback? If this were completely safe to enable for every device that supported it, why would there be an enable bit in Device Control? I don't know anything about extended tags, but it worries me a little when there's a "go-fast" switch and no explanation about when and why we might need to go slow. > > + return pcie_capability_set_word(dev, PCI_EXP_DEVCTL, > > + PCI_EXP_DEVCTL_EXT_TAG); > > +} > > + > > static void pci_configure_device(struct pci_dev *dev) > > { > > struct hotplug_params hpp; > > int ret; > > > > pci_configure_mps(dev); > > + pci_configure_extended_tags(dev); > > > > memset(&hpp, 0, sizeof(hpp)); > > ret = pci_get_hp_params(dev, &hpp); > > > > > -- > Sinan Kaya > Qualcomm Datacenter Technologies, Inc. as an affiliate of Qualcomm Technologies, Inc. > Qualcomm Technologies, Inc. is a member of the Code Aurora Forum, a Linux Foundation Collaborative Project. > -- > To unsubscribe from this list: send the line "unsubscribe linux-pci" in > the body of a message to majordomo@vger.kernel.org > More majordomo info at http://vger.kernel.org/majordomo-info.html -- To unsubscribe from this list: send the line "unsubscribe linux-pci" in the body of a message to majordomo@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html
On 11/11/2016 3:58 PM, Bjorn Helgaas wrote: >> I should have checked the capability here before trying to enable it. >> > I'll post a follow up patch on this. >> > >> > Is there any other feedback? > If this were completely safe to enable for every device that supported > it, why would there be an enable bit in Device Control? reading from the ECN here. https://pcisig.com/sites/default/files/specification_documents/ECN_Extended_Tag_Enable_Default_05Sept2008_final.pdf The initial value is implementation specific and functions are allowed to set it to 1 by default. > > I don't know anything about extended tags, but it worries me a little > when there's a "go-fast" switch and no explanation about when and why > we might need to go slo Based on my observation, extended tags increase the number of reads that can be queued up back to back downstream. Otherwise, the requests will not make progress until 1 tag out of 32 gets available.
Hi Bjorn, On 11/11/2016 3:58 PM, Bjorn Helgaas wrote: >> Is there any other feedback? > If this were completely safe to enable for every device that supported > it, why would there be an enable bit in Device Control? > > I don't know anything about extended tags, but it worries me a little > when there's a "go-fast" switch and no explanation about when and why > we might need to go slow. > I have v2 posted. Do you feel like applying to linux-next to get some testing exposure or do you want to tie it to some DMI so that we enable it only on recent HW? Sinan
On 1/25/2017 12:49 PM, Sinan Kaya wrote: > Hi Bjorn, > > On 11/11/2016 3:58 PM, Bjorn Helgaas wrote: >>> Is there any other feedback? >> If this were completely safe to enable for every device that supported >> it, why would there be an enable bit in Device Control? >> >> I don't know anything about extended tags, but it worries me a little >> when there's a "go-fast" switch and no explanation about when and why >> we might need to go slow. >> I tried to answer your question in the new commit message. All PCIe completers are required to support 8 bit tags. Generation of 8 bit tags is optional. That's why, there is a supported and an enable/disable bit. > > I have v2 posted. Do you feel like applying to linux-next to get some testing > exposure or do you want to tie it to some DMI so that we enable it only on > recent HW? > > Sinan >
diff --git a/drivers/pci/probe.c b/drivers/pci/probe.c index 93f280d..2424f38 100644 --- a/drivers/pci/probe.c +++ b/drivers/pci/probe.c @@ -1505,12 +1505,19 @@ static void program_hpp_type2(struct pci_dev *dev, struct hpp_type2 *hpp) */ } +static int pci_configure_extended_tags(struct pci_dev *dev) +{ + return pcie_capability_set_word(dev, PCI_EXP_DEVCTL, + PCI_EXP_DEVCTL_EXT_TAG); +} + static void pci_configure_device(struct pci_dev *dev) { struct hotplug_params hpp; int ret; pci_configure_mps(dev); + pci_configure_extended_tags(dev); memset(&hpp, 0, sizeof(hpp)); ret = pci_get_hp_params(dev, &hpp);
Each PCIe device can issue up to 32 transactions at a time by default. Each transaction is tracked by a tag number on the bus. 32 outstanding transactions is not enough for some performance critical applications especially when a lot of small sized frames are transmitted. Extended tags support increases this number to 256. Devices not supporting extended tags tie-off this field to 0. According to ECN, it is safe to enable this feature for all PCIe endpoints. Signed-off-by: Sinan Kaya <okaya@codeaurora.org> --- drivers/pci/probe.c | 7 +++++++ 1 file changed, 7 insertions(+)