kernel.git/drivers/pci/pcie, branch linux-5.1.y

PCI: Add missing link delays required by the PCIe spec

2019-07-26T07:12:37Z

[ Upstream commit c2bf1fc212f7e6f25ace1af8f0b3ac061ea48ba5 ] Currently Linux does not follow PCIe spec regarding the required delays after reset. A concrete example is a Thunderbolt add-in-card that consists of a PCIe switch and two PCIe endpoints: +-1b.0-[01-6b]----00.0-[02-6b]--+-00.0-[03]----00.0 TBT controller +-01.0-[04-36]-- DS hotplug port +-02.0-[37]----00.0 xHCI controller \-04.0-[38-6b]-- DS hotplug port The root port (1b.0) and the PCIe switch downstream ports are all PCIe gen3 so they support 8GT/s link speeds. We wait for the PCIe hierarchy to enter D3cold (runtime): pcieport 0000:00:1b.0: power state changed by ACPI to D3cold When it wakes up from D3cold, according to the PCIe 4.0 section 5.8 the PCIe switch is put to reset and its power is re-applied. This means that we must follow the rules in PCIe 4.0 section 6.6.1. For the PCIe gen3 ports we are dealing with here, the following applies: With a Downstream Port that supports Link speeds greater than 5.0 GT/s, software must wait a minimum of 100 ms after Link training completes before sending a Configuration Request to the device immediately below that Port. Software can determine when Link training completes by polling the Data Link Layer Link Active bit or by setting up an associated interrupt (see Section 6.7.3.3). Translating this into the above topology we would need to do this (DLLLA stands for Data Link Layer Link Active): pcieport 0000:00:1b.0: wait for 100ms after DLLLA is set before access to 0000:01:00.0 pcieport 0000:02:00.0: wait for 100ms after DLLLA is set before access to 0000:03:00.0 pcieport 0000:02:02.0: wait for 100ms after DLLLA is set before access to 0000:37:00.0 I've instrumented the kernel with additional logging so we can see the actual delays the kernel performs: pcieport 0000:00:1b.0: power state changed by ACPI to D0 pcieport 0000:00:1b.0: waiting for D3cold delay of 100 ms pcieport 0000:00:1b.0: waking up bus pcieport 0000:00:1b.0: waiting for D3hot delay of 10 ms pcieport 0000:00:1b.0: restoring config space at offset 0x2c (was 0x60, writing 0x60) ... pcieport 0000:00:1b.0: PME# disabled pcieport 0000:01:00.0: restoring config space at offset 0x3c (was 0x1ff, writing 0x201ff) ... pcieport 0000:01:00.0: PME# disabled pcieport 0000:02:00.0: restoring config space at offset 0x3c (was 0x1ff, writing 0x201ff) ... pcieport 0000:02:00.0: PME# disabled pcieport 0000:02:01.0: restoring config space at offset 0x3c (was 0x1ff, writing 0x201ff) ... pcieport 0000:02:01.0: restoring config space at offset 0x4 (was 0x100000, writing 0x100407) pcieport 0000:02:01.0: PME# disabled pcieport 0000:02:02.0: restoring config space at offset 0x3c (was 0x1ff, writing 0x201ff) ... pcieport 0000:02:02.0: PME# disabled pcieport 0000:02:04.0: restoring config space at offset 0x3c (was 0x1ff, writing 0x201ff) ... pcieport 0000:02:04.0: PME# disabled pcieport 0000:02:01.0: PME# enabled pcieport 0000:02:01.0: waiting for D3hot delay of 10 ms pcieport 0000:02:04.0: PME# enabled pcieport 0000:02:04.0: waiting for D3hot delay of 10 ms thunderbolt 0000:03:00.0: restoring config space at offset 0x14 (was 0x0, writing 0x8a040000) ... thunderbolt 0000:03:00.0: PME# disabled xhci_hcd 0000:37:00.0: restoring config space at offset 0x10 (was 0x0, writing 0x73f00000) ... xhci_hcd 0000:37:00.0: PME# disabled For the switch upstream port (01:00.0) we wait for 100ms but not taking into account the DLLLA requirement. We then wait 10ms for D3hot -> D0 transition of the root port and the two downstream hotplug ports. This means that we deviate from what the spec requires. Performing the same check for system sleep (s2idle) transitions we can see following when resuming from s2idle: pcieport 0000:00:1b.0: power state changed by ACPI to D0 pcieport 0000:00:1b.0: restoring config space at offset 0x2c (was 0x60, writing 0x60) ... pcieport 0000:01:00.0: restoring config space at offset 0x3c (was 0x1ff, writing 0x201ff) ... pcieport 0000:02:02.0: restoring config space at offset 0x3c (was 0x1ff, writing 0x201ff) pcieport 0000:02:02.0: restoring config space at offset 0x2c (was 0x0, writing 0x0) pcieport 0000:02:01.0: restoring config space at offset 0x3c (was 0x1ff, writing 0x201ff) pcieport 0000:02:04.0: restoring config space at offset 0x3c (was 0x1ff, writing 0x201ff) pcieport 0000:02:02.0: restoring config space at offset 0x28 (was 0x0, writing 0x0) pcieport 0000:02:00.0: restoring config space at offset 0x3c (was 0x1ff, writing 0x201ff) pcieport 0000:02:02.0: restoring config space at offset 0x24 (was 0x10001, writing 0x1fff1) pcieport 0000:02:01.0: restoring config space at offset 0x2c (was 0x0, writing 0x60) pcieport 0000:02:02.0: restoring config space at offset 0x20 (was 0x0, writing 0x73f073f0) pcieport 0000:02:04.0: restoring config space at offset 0x2c (was 0x0, writing 0x60) pcieport 0000:02:01.0: restoring config space at offset 0x28 (was 0x0, writing 0x60) pcieport 0000:02:00.0: restoring config space at offset 0x2c (was 0x0, writing 0x0) pcieport 0000:02:02.0: restoring config space at offset 0x1c (was 0x101, writing 0x1f1) pcieport 0000:02:04.0: restoring config space at offset 0x28 (was 0x0, writing 0x60) pcieport 0000:02:01.0: restoring config space at offset 0x24 (was 0x10001, writing 0x1ff10001) pcieport 0000:02:00.0: restoring config space at offset 0x28 (was 0x0, writing 0x0) pcieport 0000:02:02.0: restoring config space at offset 0x18 (was 0x0, writing 0x373702) pcieport 0000:02:04.0: restoring config space at offset 0x24 (was 0x10001, writing 0x49f12001) pcieport 0000:02:01.0: restoring config space at offset 0x20 (was 0x0, writing 0x73e05c00) pcieport 0000:02:00.0: restoring config space at offset 0x24 (was 0x10001, writing 0x1fff1) pcieport 0000:02:04.0: restoring config space at offset 0x20 (was 0x0, writing 0x89f07400) pcieport 0000:02:01.0: restoring config space at offset 0x1c (was 0x101, writing 0x5151) pcieport 0000:02:00.0: restoring config space at offset 0x20 (was 0x0, writing 0x8a008a00) pcieport 0000:02:02.0: restoring config space at offset 0xc (was 0x10000, writing 0x10020) pcieport 0000:02:04.0: restoring config space at offset 0x1c (was 0x101, writing 0x6161) pcieport 0000:02:01.0: restoring config space at offset 0x18 (was 0x0, writing 0x360402) pcieport 0000:02:00.0: restoring config space at offset 0x1c (was 0x101, writing 0x1f1) pcieport 0000:02:04.0: restoring config space at offset 0x18 (was 0x0, writing 0x6b3802) pcieport 0000:02:02.0: restoring config space at offset 0x4 (was 0x100000, writing 0x100407) pcieport 0000:02:00.0: restoring config space at offset 0x18 (was 0x0, writing 0x30302) pcieport 0000:02:01.0: restoring config space at offset 0xc (was 0x10000, writing 0x10020) pcieport 0000:02:04.0: restoring config space at offset 0xc (was 0x10000, writing 0x10020) pcieport 0000:02:00.0: restoring config space at offset 0xc (was 0x10000, writing 0x10020) pcieport 0000:02:01.0: restoring config space at offset 0x4 (was 0x100000, writing 0x100407) pcieport 0000:02:04.0: restoring config space at offset 0x4 (was 0x100000, writing 0x100407) pcieport 0000:02:00.0: restoring config space at offset 0x4 (was 0x100000, writing 0x100407) xhci_hcd 0000:37:00.0: restoring config space at offset 0x10 (was 0x0, writing 0x73f00000) ... thunderbolt 0000:03:00.0: restoring config space at offset 0x14 (was 0x0, writing 0x8a040000) This is even worse. None of the mandatory delays are performed. If this would be S3 instead of s2idle then according to PCI FW spec 3.2 section 4.6.8. there is a specific _DSM that allows the OS to skip the delays but this platform does not provide the _DSM and does not go to S3 anyway so no firmware is involved that could already handle these delays. In this particular Intel Coffee Lake platform these delays are not actually needed because there is an additional delay as part of the ACPI power resource that is used to turn on power to the hierarchy but since that additional delay is not required by any of standards (PCIe, ACPI) it is not present in the Intel Ice Lake, for example where missing the mandatory delays causes pciehp to start tearing down the stack too early (links are not yet trained). For this reason, change the PCIe portdrv PM resume hooks so that they perform the mandatory delays before the downstream component gets resumed. We perform the delays before port services are resumed because otherwise pciehp might find that the link is not up (even if it is just training) and tears-down the hierarchy. Signed-off-by: Mika Westerberg Signed-off-by: Rafael J. Wysocki Signed-off-by: Sasha Levin

PCI: Work around Pericom PCIe-to-PCI bridge Retrain Link erratum

2019-05-25T16:16:44Z

commit 4ec73791a64bab25cabf16a6067ee478692e506d upstream. Due to an erratum in some Pericom PCIe-to-PCI bridges in reverse mode (conventional PCI on primary side, PCIe on downstream side), the Retrain Link bit needs to be cleared manually to allow the link training to complete successfully. If it is not cleared manually, the link training is continuously restarted and no devices below the PCI-to-PCIe bridge can be accessed. That means drivers for devices below the bridge will be loaded but won't work and may even crash because the driver is only reading 0xffff. See the Pericom Errata Sheet PI7C9X111SLB_errata_rev1.2_102711.pdf for details. Devices known as affected so far are: PI7C9X110, PI7C9X111SL, PI7C9X130. Add a new flag, clear_retrain_link, in struct pci_dev. Quirks for affected devices set this bit. Note that pcie_retrain_link() lives in aspm.c because that's currently the only place we use it, but this erratum is not specific to ASPM, and we may retrain links for other reasons in the future. Signed-off-by: Stefan Mätje [bhelgaas: apply regardless of CONFIG_PCIEASPM] Signed-off-by: Bjorn Helgaas Reviewed-by: Andy Shevchenko CC: stable@vger.kernel.org Signed-off-by: Greg Kroah-Hartman

PCI: Factor out pcie_retrain_link() function

2019-05-25T16:16:44Z

commit 86fa6a344209d9414ea962b1f1ac6ade9dd7563a upstream. Factor out pcie_retrain_link() to use for Pericom Retrain Link quirk. No functional change intended. Signed-off-by: Stefan Mätje Signed-off-by: Bjorn Helgaas Reviewed-by: Andy Shevchenko CC: stable@vger.kernel.org Signed-off-by: Greg Kroah-Hartman

PCI/LINK: Add Kconfig option (default off)

2019-05-02T13:34:32Z

e8303bb7a75c ("PCI/LINK: Report degraded links via link bandwidth notification") added dmesg logging whenever a link changes speed or width to a state that is considered degraded. Unfortunately, it cannot differentiate signal integrity-related link changes from those intentionally initiated by an endpoint driver, including drivers that may live in userspace or VMs when making use of vfio-pci. Some GPU drivers actively manage the link state to save power, which generates a stream of messages like this: vfio-pci 0000:07:00.0: 32.000 Gb/s available PCIe bandwidth, limited by 2.5 GT/s x16 link at 0000:00:02.0 (capable of 64.000 Gb/s with 5 GT/s x16 link) Since we can't distinguish the intentional changes from the signal integrity issues, leave the reporting turned off by default. Add a Kconfig option to turn it on if desired. Fixes: e8303bb7a75c ("PCI/LINK: Report degraded links via link bandwidth notification") Link: https://lore.kernel.org/linux-pci/20190501142942.26972-1-keith.busch@intel.com Signed-off-by: Keith Busch Signed-off-by: Bjorn Helgaas

PCI/portdrv: Use shared MSI/MSI-X vector for Bandwidth Management

2019-05-01T20:34:01Z

The Interrupt Message Number in the PCIe Capabilities register (PCIe r4.0, sec 7.5.3.2) indicates which MSI/MSI-X vector is shared by interrupts related to the PCIe Capability, including Link Bandwidth Management and Link Autonomous Bandwidth Interrupts (Link Control, 7.5.3.7), Command Completed and Hot-Plug Interrupts (Slot Control, 7.5.3.10), and the PME Interrupt (Root Control, 7.5.3.12). pcie_message_numbers() checked whether we want to enable PME or Hot-Plug interrupts but neglected to check for Link Bandwidth Management, so if we only wanted the Bandwidth Management interrupts, it decided we didn't need any vectors at all. Then pcie_port_enable_irq_vec() tried to reallocate zero vectors, which failed, resulting in fallback to INTx. On some systems, e.g., an X79-based workstation, that INTx seems broken or not handled correctly, so we got spurious IRQ16 interrupts for Bandwidth Management events. Change pcie_message_numbers() so that if we want Link Bandwidth Management interrupts, we use the shared MSI/MSI-X vector from the PCIe Capabilities register. Fixes: e8303bb7a75c ("PCI/LINK: Report degraded links via link bandwidth notification") Link: https://lore.kernel.org/lkml/155597243666.19387.1205950870601742062.stgit@gimli.home Signed-off-by: Alex Williamson [bhelgaas: changelog] Signed-off-by: Bjorn Helgaas

PCI/LINK: Deduplicate bandwidth reports for multi-function devices

2019-03-25T22:59:07Z

If a multi-function device's bandwidth is already limited when it is enumerated, a message is logged only for function 0. By contrast, when downtraining occurs after enumeration, a message is logged for all functions. That's because the former uses pcie_report_downtraining(), whereas the latter uses __pcie_print_link_status() (which doesn't filter functions != 0). I am seeing this happen on a MacBookPro9,1 with a GPU (function 0) and an integrated HDA controller (function 1). Avoid this incongruence by calling pcie_report_downtraining() in both cases. Signed-off-by: Lukas Wunner Signed-off-by: Bjorn Helgaas Reviewed-by: Alexandru Gagniuc

PCI/LINK: Clear bandwidth notification interrupt before enabling it

2019-03-25T22:59:06Z

When booting a MacBookPro9,1, duplicate link downtraining messages are logged for the devices directly attached to the two CPU-internal Root Ports of the Core i7 3615QM: Once on device enumeration and once on enablement of the bandwidth notification interrupt on the Root Ports. Duplicate messages do not occur with Root Ports on the PCH and Downstream Ports on the Thunderbolt controller: Only a single message is logged for these, namely on device enumeration. The reason for the duplicate messages is a stale interrupt in the Link Status register of the 3615QM's internal Root Ports. Avoid by clearing the interrupt before enabling it. An alternative approach would be to clear the interrupt already on device enumeration or to report link downtraining only if the speed has changed. That way, link downtraining occurring between device enumeration and enablement of the bandwidth notification interrupt could be caught. However clearing stale interrupts before enabling them is a standard operating procedure for any driver and keeping the two steps in one place makes the code easier to follow. Signed-off-by: Lukas Wunner Signed-off-by: Bjorn Helgaas Reviewed-by: Alexandru Gagniuc

PCI/LINK: Supply IRQ handler so level-triggered IRQs are acked

2019-03-25T22:58:50Z

A threaded IRQ with a NULL handler does not work with level-triggered interrupts. request_threaded_irq() will return an error: genirq: Threaded irq requested with handler=NULL and !ONESHOT for irq 16 pcie_bw_notification: probe of 0000:00:1b.0:pcie010 failed with error -22 For level interrupts we need to silence the interrupt before exiting the IRQ handler, so just clear the PCI_EXP_LNKSTA_LBMS bit there. Fixes: e8303bb7a75c ("PCI/LINK: Report degraded links via link bandwidth notification") Reported-by: Linus Torvalds Reported-by: Borislav Petkov Signed-off-by: Alexandru Gagniuc Signed-off-by: Bjorn Helgaas

Merge branch 'pci/pm'

2019-03-06T21:30:15Z

- Blacklist Gigabyte X299 Root Port power management to fix Thunderbolt hotplug (Mika Westerberg) - Revert runtime PM suspend/resume callbacks that broke PME on network cable plug (Mika Westerberg) - Disable Data Link State Changed interrupts to prevent wakeup immediately after suspend (Mika Westerberg) * pci/pm: PCI/PME: Fix possible use-after-free on remove PCI/PME: Fix hotplug/sysfs remove deadlock in pcie_pme_remove() PCI: pciehp: Disable Data Link Layer State Changed event on suspend Revert "PCI/PME: Implement runtime PM callbacks" PCI: Blacklist power management of Gigabyte X299 DESIGNARE EX PCIe ports

Merge branch 'pci/portdrv'

2019-03-06T21:30:14Z

- Allow portdrv to claim subtractive decode Ports so PCIe services will work for them (Honghui Zhang) - Report PCIe links that become degraded at run-time (Alexandru Gagniuc) * pci/portdrv: PCI/LINK: Report degraded links via link bandwidth notification PCI/portdrv: Support PCIe services on subtractive decode bridges PCI/portdrv: Use conventional Device ID table formatting