| Message ID | 20170828142307.30061-1-l.stach@pengutronix.de |
|---|---|
| Headers | show |
| Series | DWC host without MSI controller | expand |
On Mon, Aug 28, 2017 at 7:23 AM, Lucas Stach <l.stach@pengutronix.de> wrote: > Hi all, > > this small series tries to fix/workaround a serious design flaw of the DWC PCIe > host controller: it is unable to work with both legacy and MSI IRQs enabled at > the same time. As soon as the first MSI is enabled in the DWC MSI controller, > the host stops forwarding legacy IRQs. > > If the MSI controller is present, MSIs will be used for the PCIe port services > IRQs, leaving endpoint devices which don't support MSIs unable to raise IRQs. > It is only safe to enable the MSI controller if it is validated that all PCIe > devices and drivers in the system support working MSIs. As most devices > support falling back to using legacy PCIe IRQs if MSI support is missing it is > much safer to disable the MSI by default and only enable it on validated > systems. > > Feedback welcome. > > Regards, > Lucas > > Lucas Stach (3): > PCI: designware: only register MSI controller when MSI irq line is > valid > PCI: imx6: allow MSI irq to be absent > ARM: dts: imx6qdl: remove MSI irq line > > .../devicetree/bindings/pci/fsl,imx6q-pcie.txt | 8 ++++---- > arch/arm/boot/dts/imx6qdl.dtsi | 2 -- > drivers/pci/dwc/pci-imx6.c | 23 +++++++++++----------- > drivers/pci/dwc/pcie-designware-host.c | 4 ++-- > 4 files changed, 17 insertions(+), 20 deletions(-) > Lucas, Thank you for following up on this! I tested it with and without the third patch that removes msi from the dt thus with both an ath9k 802.11 device which only supports legacy interrupts and a Marvell sky2 device which supports msi as well as legacy interrupts and it works as expected: - without msi enabled in dts (default): both sky2 and ath9k work - with msi enabled in dt: sky2 works ath9k does not Tested-by: Tim Harvey <tharvey@gateworks.com> Regards, Tim
Hi Bjorn, On Mon, Aug 28, 2017 at 1:59 PM, Tim Harvey <tharvey@gateworks.com> wrote: > On Mon, Aug 28, 2017 at 7:23 AM, Lucas Stach <l.stach@pengutronix.de> wrote: >> Hi all, >> >> this small series tries to fix/workaround a serious design flaw of the DWC PCIe >> host controller: it is unable to work with both legacy and MSI IRQs enabled at >> the same time. As soon as the first MSI is enabled in the DWC MSI controller, >> the host stops forwarding legacy IRQs. >> >> If the MSI controller is present, MSIs will be used for the PCIe port services >> IRQs, leaving endpoint devices which don't support MSIs unable to raise IRQs. >> It is only safe to enable the MSI controller if it is validated that all PCIe >> devices and drivers in the system support working MSIs. As most devices >> support falling back to using legacy PCIe IRQs if MSI support is missing it is >> much safer to disable the MSI by default and only enable it on validated >> systems. >> >> Feedback welcome. >> >> Regards, >> Lucas >> >> Lucas Stach (3): >> PCI: designware: only register MSI controller when MSI irq line is >> valid >> PCI: imx6: allow MSI irq to be absent >> ARM: dts: imx6qdl: remove MSI irq line >> >> .../devicetree/bindings/pci/fsl,imx6q-pcie.txt | 8 ++++---- >> arch/arm/boot/dts/imx6qdl.dtsi | 2 -- >> drivers/pci/dwc/pci-imx6.c | 23 +++++++++++----------- >> drivers/pci/dwc/pcie-designware-host.c | 4 ++-- >> 4 files changed, 17 insertions(+), 20 deletions(-) >> > > Lucas, > > Thank you for following up on this! > > I tested it with and without the third patch that removes msi from the > dt thus with both an ath9k 802.11 device which only supports legacy > interrupts and a Marvell sky2 device which supports msi as well as > legacy interrupts and it works as expected: > - without msi enabled in dts (default): both sky2 and ath9k work > - with msi enabled in dt: sky2 works ath9k does not > > Tested-by: Tim Harvey <tharvey@gateworks.com> Any comments about this series?
Hi Fabio, Am Montag, den 09.10.2017, 09:14 -0300 schrieb Fabio Estevam: > Hi Bjorn, > > On Mon, Aug 28, 2017 at 1:59 PM, Tim Harvey <tharvey@gateworks.com> > wrote: > > On Mon, Aug 28, 2017 at 7:23 AM, Lucas Stach <l.stach@pengutronix.d > > e> wrote: > > > Hi all, > > > > > > this small series tries to fix/workaround a serious design flaw > > > of the DWC PCIe > > > host controller: it is unable to work with both legacy and MSI > > > IRQs enabled at > > > the same time. As soon as the first MSI is enabled in the DWC MSI > > > controller, > > > the host stops forwarding legacy IRQs. > > > > > > If the MSI controller is present, MSIs will be used for the PCIe > > > port services > > > IRQs, leaving endpoint devices which don't support MSIs unable to > > > raise IRQs. > > > It is only safe to enable the MSI controller if it is validated > > > that all PCIe > > > devices and drivers in the system support working MSIs. As most > > > devices > > > support falling back to using legacy PCIe IRQs if MSI support is > > > missing it is > > > much safer to disable the MSI by default and only enable it on > > > validated > > > systems. > > > > > > Feedback welcome. > > > > > > Regards, > > > Lucas > > > > > > Lucas Stach (3): > > > PCI: designware: only register MSI controller when MSI irq line > > > is > > > valid > > > PCI: imx6: allow MSI irq to be absent > > > ARM: dts: imx6qdl: remove MSI irq line > > > > > > .../devicetree/bindings/pci/fsl,imx6q-pcie.txt | 8 ++++---- > > > arch/arm/boot/dts/imx6qdl.dtsi | 2 -- > > > drivers/pci/dwc/pci-imx6.c | 23 > > > +++++++++++----------- > > > drivers/pci/dwc/pcie-designware-host.c | 4 ++-- > > > 4 files changed, 17 insertions(+), 20 deletions(-) > > > > > > > Lucas, > > > > Thank you for following up on this! > > > > I tested it with and without the third patch that removes msi from > > the > > dt thus with both an ath9k 802.11 device which only supports legacy > > interrupts and a Marvell sky2 device which supports msi as well as > > legacy interrupts and it works as expected: > > - without msi enabled in dts (default): both sky2 and ath9k work > > - with msi enabled in dt: sky2 works ath9k does not > > > > Tested-by: Tim Harvey <tharvey@gateworks.com> > > Any comments about this series? Bjorn already commented on the series and I'm trying free up a slot to address the feedback. Regards, Lucas
Hi all, this small series tries to fix/workaround a serious design flaw of the DWC PCIe host controller: it is unable to work with both legacy and MSI IRQs enabled at the same time. As soon as the first MSI is enabled in the DWC MSI controller, the host stops forwarding legacy IRQs. If the MSI controller is present, MSIs will be used for the PCIe port services IRQs, leaving endpoint devices which don't support MSIs unable to raise IRQs. It is only safe to enable the MSI controller if it is validated that all PCIe devices and drivers in the system support working MSIs. As most devices support falling back to using legacy PCIe IRQs if MSI support is missing it is much safer to disable the MSI by default and only enable it on validated systems. Feedback welcome. Regards, Lucas Lucas Stach (3): PCI: designware: only register MSI controller when MSI irq line is valid PCI: imx6: allow MSI irq to be absent ARM: dts: imx6qdl: remove MSI irq line .../devicetree/bindings/pci/fsl,imx6q-pcie.txt | 8 ++++---- arch/arm/boot/dts/imx6qdl.dtsi | 2 -- drivers/pci/dwc/pci-imx6.c | 23 +++++++++++----------- drivers/pci/dwc/pcie-designware-host.c | 4 ++-- 4 files changed, 17 insertions(+), 20 deletions(-)