| Message ID | 20200505161318.26200-1-nsaenzjulienne@suse.de |
|---|---|
| Headers | show |
| Series | USB: pci-quirks: Add Raspberry Pi 4 quirk | expand |
On Tue, May 05, 2020 at 06:13:13PM +0200, Nicolas Saenz Julienne wrote: > On the Raspberry Pi 4, after a PCI reset, VL805's firmware may either be > loaded directly from an EEPROM or, if not present, by the SoC's > co-processor, VideoCore. This series adds support for the later. > > Note that there are a set of constraints we have to consider: > - We need to make sure the VideoCore firmware interface is up and > running before running the VL805 firmware load call. > > - There is no way to discern RPi4's VL805 chip from other platforms', > so we need the firmware load to happen *before* running > quirk_usb_handoff_xhci(). Failure to do so results in an unwarranted > 5 second wait while the fixup code polls xHC's non-existing state. > > By Florian's suggestion I've been spending some time exploring the device > link[1] API in order to see if that could save us from explicitly creating > probe dependencies between pcie-brcmstb and firmware/raspberrypi (patch #3). > Technically these dependencies could be inferred from DT. It turns out Saravana > Kannan has been looking at this already. A new boot mechanism, activated with > fw_devlink=on takes care of the device probe ordering on devices with > consumer/supplier relationships. For now this relationship is created based on > the usage of generic DT properties, but has no support for vendor-specifc DT > properties, which we'd be forced to use in order to create a relationship > between our two devices since our setup is highly non generic. There will > probably be at some point support for such properties, and we will then be able > to revisit some of this code. > > All this is based on the work by Tim Gover in RPi's downstream > kernel[2]. > > [1] https://www.kernel.org/doc/html/v4.13/driver-api/device_link.html > [2] https://github.com/raspberrypi/linux/commit/9935b4c7e360b4494b4cb6e3ce797238a1ab78bd > > --- > > Changes since v7: > - Address Stefan's comments > > Changes since v6: > - Make rpi_firmware_init_vl805() more robust > - Rewrite comments and patch descriptions to be more accessible to non RPi > fluent people > - Removed Florian's Reviewed-by in patch #2 as function changed > substantially > - Tested with/witout u-boot > > Changes since v5: > - Fix issues reported by Kbuild test robot > > Changes since v4: > - Addressed Sergei's comments > - Fix potential warning in patch #2 > > Changes since v3: > - Addressed Greg's comments > > There was no v2, my bad. > > Changes since v1: > - Addressed Floarians comments > > Nicolas Saenz Julienne (4): > soc: bcm2835: Add notify xHCI reset property > firmware: raspberrypi: Introduce vl805 init routine > PCI: brcmstb: Wait for Raspberry Pi's firmware when present > USB: pci-quirks: Add Raspberry Pi 4 quirk > > drivers/firmware/Kconfig | 3 +- > drivers/firmware/raspberrypi.c | 61 ++++++++++++++++++++++ > drivers/pci/controller/pcie-brcmstb.c | 17 ++++++ > drivers/usb/host/pci-quirks.c | 16 ++++++ > include/soc/bcm2835/raspberrypi-firmware.h | 9 +++- > 5 files changed, 104 insertions(+), 2 deletions(-) Hi Nicolas, should I queue this series via the PCI tree ? Just let me know, most of the changes are not in the PCI tree, asking in order to minimize/simplify conflicts handling if possible. Lorenzo
On Wed, 2020-05-13 at 12:17 +0100, Lorenzo Pieralisi wrote: > On Tue, May 05, 2020 at 06:13:13PM +0200, Nicolas Saenz Julienne wrote: > > On the Raspberry Pi 4, after a PCI reset, VL805's firmware may either be > > loaded directly from an EEPROM or, if not present, by the SoC's > > co-processor, VideoCore. This series adds support for the later. > > > > Note that there are a set of constraints we have to consider: > > - We need to make sure the VideoCore firmware interface is up and > > running before running the VL805 firmware load call. > > > > - There is no way to discern RPi4's VL805 chip from other platforms', > > so we need the firmware load to happen *before* running > > quirk_usb_handoff_xhci(). Failure to do so results in an unwarranted > > 5 second wait while the fixup code polls xHC's non-existing state. > > > > By Florian's suggestion I've been spending some time exploring the device > > link[1] API in order to see if that could save us from explicitly creating > > probe dependencies between pcie-brcmstb and firmware/raspberrypi (patch #3). > > Technically these dependencies could be inferred from DT. It turns out > > Saravana > > Kannan has been looking at this already. A new boot mechanism, activated > > with > > fw_devlink=on takes care of the device probe ordering on devices with > > consumer/supplier relationships. For now this relationship is created based > > on > > the usage of generic DT properties, but has no support for vendor-specifc DT > > properties, which we'd be forced to use in order to create a relationship > > between our two devices since our setup is highly non generic. There will > > probably be at some point support for such properties, and we will then be > > able > > to revisit some of this code. > > > > All this is based on the work by Tim Gover in RPi's downstream > > kernel[2]. > > > > [1] https://www.kernel.org/doc/html/v4.13/driver-api/device_link.html > > [2] > > https://github.com/raspberrypi/linux/commit/9935b4c7e360b4494b4cb6e3ce797238a1ab78bd > > > > --- > > > > Changes since v7: > > - Address Stefan's comments > > > > Changes since v6: > > - Make rpi_firmware_init_vl805() more robust > > - Rewrite comments and patch descriptions to be more accessible to non RPi > > fluent people > > - Removed Florian's Reviewed-by in patch #2 as function changed > > substantially > > - Tested with/witout u-boot > > > > Changes since v5: > > - Fix issues reported by Kbuild test robot > > > > Changes since v4: > > - Addressed Sergei's comments > > - Fix potential warning in patch #2 > > > > Changes since v3: > > - Addressed Greg's comments > > > > There was no v2, my bad. > > > > Changes since v1: > > - Addressed Floarians comments > > > > Nicolas Saenz Julienne (4): > > soc: bcm2835: Add notify xHCI reset property > > firmware: raspberrypi: Introduce vl805 init routine > > PCI: brcmstb: Wait for Raspberry Pi's firmware when present > > USB: pci-quirks: Add Raspberry Pi 4 quirk > > > > drivers/firmware/Kconfig | 3 +- > > drivers/firmware/raspberrypi.c | 61 ++++++++++++++++++++++ > > drivers/pci/controller/pcie-brcmstb.c | 17 ++++++ > > drivers/usb/host/pci-quirks.c | 16 ++++++ > > include/soc/bcm2835/raspberrypi-firmware.h | 9 +++- > > 5 files changed, 104 insertions(+), 2 deletions(-) > > Hi Nicolas, > > should I queue this series via the PCI tree ? Just let me know, most of > the changes are not in the PCI tree, asking in order to > minimize/simplify conflicts handling if possible. Yes, I agree, it's better if you take the whole thing. Thanks, Nicolas
On Tue, May 05, 2020 at 06:13:13PM +0200, Nicolas Saenz Julienne wrote: > On the Raspberry Pi 4, after a PCI reset, VL805's firmware may either be > loaded directly from an EEPROM or, if not present, by the SoC's > co-processor, VideoCore. This series adds support for the later. > > Note that there are a set of constraints we have to consider: > - We need to make sure the VideoCore firmware interface is up and > running before running the VL805 firmware load call. > > - There is no way to discern RPi4's VL805 chip from other platforms', > so we need the firmware load to happen *before* running > quirk_usb_handoff_xhci(). Failure to do so results in an unwarranted > 5 second wait while the fixup code polls xHC's non-existing state. > > By Florian's suggestion I've been spending some time exploring the device > link[1] API in order to see if that could save us from explicitly creating > probe dependencies between pcie-brcmstb and firmware/raspberrypi (patch #3). > Technically these dependencies could be inferred from DT. It turns out Saravana > Kannan has been looking at this already. A new boot mechanism, activated with > fw_devlink=on takes care of the device probe ordering on devices with > consumer/supplier relationships. For now this relationship is created based on > the usage of generic DT properties, but has no support for vendor-specifc DT > properties, which we'd be forced to use in order to create a relationship > between our two devices since our setup is highly non generic. There will > probably be at some point support for such properties, and we will then be able > to revisit some of this code. > > All this is based on the work by Tim Gover in RPi's downstream > kernel[2]. > > [1] https://www.kernel.org/doc/html/v4.13/driver-api/device_link.html > [2] https://github.com/raspberrypi/linux/commit/9935b4c7e360b4494b4cb6e3ce797238a1ab78bd > > --- > > Changes since v7: > - Address Stefan's comments > > Changes since v6: > - Make rpi_firmware_init_vl805() more robust > - Rewrite comments and patch descriptions to be more accessible to non RPi > fluent people > - Removed Florian's Reviewed-by in patch #2 as function changed > substantially > - Tested with/witout u-boot > > Changes since v5: > - Fix issues reported by Kbuild test robot > > Changes since v4: > - Addressed Sergei's comments > - Fix potential warning in patch #2 > > Changes since v3: > - Addressed Greg's comments > > There was no v2, my bad. > > Changes since v1: > - Addressed Floarians comments > > Nicolas Saenz Julienne (4): > soc: bcm2835: Add notify xHCI reset property > firmware: raspberrypi: Introduce vl805 init routine > PCI: brcmstb: Wait for Raspberry Pi's firmware when present > USB: pci-quirks: Add Raspberry Pi 4 quirk > > drivers/firmware/Kconfig | 3 +- > drivers/firmware/raspberrypi.c | 61 ++++++++++++++++++++++ > drivers/pci/controller/pcie-brcmstb.c | 17 ++++++ > drivers/usb/host/pci-quirks.c | 16 ++++++ > include/soc/bcm2835/raspberrypi-firmware.h | 9 +++- > 5 files changed, 104 insertions(+), 2 deletions(-) Applied to pci/brcmstb, thanks ! Lorenzo