[ Upstream commit c13704f568 ]
Previously, the kernel sometimes assigned more MMIO or MMIO_PREF space than
desired. For example, if the user requested 128M of space with
"pci=realloc,hpmemsize=128M", we sometimes assigned 256M:
pci 0000:06:01.0: BAR 14: assigned [mem 0x90100000-0xa00fffff] = 256M
pci 0000:06:04.0: BAR 14: assigned [mem 0xa0200000-0xb01fffff] = 256M
With this patch applied:
pci 0000:06:01.0: BAR 14: assigned [mem 0x90100000-0x980fffff] = 128M
pci 0000:06:04.0: BAR 14: assigned [mem 0x98200000-0xa01fffff] = 128M
This happened when in the first pass, the MMIO_PREF succeeded but the MMIO
failed. In the next pass, because MMIO_PREF was already assigned, the
attempt to assign MMIO_PREF returned an error code instead of success
(nothing more to do, already allocated). Hence, the size which was actually
allocated, but thought to have failed, was placed in the MMIO window.
The bug resulted in the MMIO_PREF being added to the MMIO window, which
meant doubling if MMIO_PREF size = MMIO size. With a large MMIO_PREF, the
MMIO window would likely fail to be assigned altogether due to lack of
32-bit address space.
Change find_free_bus_resource() to do the following:
- Return first unassigned resource of the correct type.
- If there is none, return first assigned resource of the correct type.
- If none of the above, return NULL.
Returning an assigned resource of the correct type allows the caller to
distinguish between already assigned and no resource of the correct type.
Add checks in pbus_size_io() and pbus_size_mem() to return success if
resource returned from find_free_bus_resource() is already allocated.
This avoids pbus_size_io() and pbus_size_mem() returning error code to
__pci_bus_size_bridges() when a resource has been successfully assigned in
a previous pass. This fixes the existing behaviour where space for a
resource could be reserved multiple times in different parent bridge
windows.
Link: https://lore.kernel.org/lkml/20190531171216.20532-2-logang@deltatee.com/T/#u
Link: https://bugzilla.kernel.org/show_bug.cgi?id=203243
Link: https://lore.kernel.org/r/PS2P216MB075563AA6AD242AA666EDC6A80760@PS2P216MB0755.KORP216.PROD.OUTLOOK.COM
Reported-by: Kit Chow <kchow@gigaio.com>
Reported-by: Nicholas Johnson <nicholas.johnson-opensource@outlook.com.au>
Signed-off-by: Nicholas Johnson <nicholas.johnson-opensource@outlook.com.au>
Signed-off-by: Bjorn Helgaas <bhelgaas@google.com>
Reviewed-by: Mika Westerberg <mika.westerberg@linux.intel.com>
Reviewed-by: Logan Gunthorpe <logang@deltatee.com>
Signed-off-by: Sasha Levin <sashal@kernel.org>
[ Upstream commit 35ff867b76 ]
When sriov_numvfs is being updated, we call the driver->sriov_configure()
function, which may enable VFs and call probe functions, which may make new
devices visible. This all happens before before sriov_numvfs_store()
updates sriov->num_VFs, so previously, concurrent sysfs reads of
sriov_numvfs returned stale values.
Serialize the sysfs read vs the write so the read returns the correct
num_VFs value.
[bhelgaas: hold device_lock instead of checking mutex_is_locked()]
Link: https://bugzilla.kernel.org/show_bug.cgi?id=202991
Link: https://lore.kernel.org/r/20190911072736.32091-1-pierre.cregut@orange.com
Signed-off-by: Pierre Crégut <pierre.cregut@orange.com>
Signed-off-by: Bjorn Helgaas <bhelgaas@google.com>
Signed-off-by: Sasha Levin <sashal@kernel.org>
[ Upstream commit 8a94644b44 ]
kobject_init_and_add() takes a reference even when it fails. If it returns
an error, kobject_put() must be called to clean up the memory associated
with the object.
When kobject_init_and_add() fails, call kobject_put() instead of kfree().
b8eb718348 ("net-sysfs: Fix reference count leak in
rx|netdev_queue_add_kobject") fixed a similar problem.
Link: https://lore.kernel.org/r/20200528021322.1984-1-wu000273@umn.edu
Signed-off-by: Qiushi Wu <wu000273@umn.edu>
Signed-off-by: Bjorn Helgaas <bhelgaas@google.com>
Signed-off-by: Sasha Levin <sashal@kernel.org>
commit de3c4bf648 upstream.
Add tx term offset support to pcie qcom driver need in some revision of
the ipq806x SoC. Ipq8064 needs tx term offset set to 7.
Link: https://lore.kernel.org/r/20200615210608.21469-9-ansuelsmth@gmail.com
Fixes: 82a823833f ("PCI: qcom: Add Qualcomm PCIe controller driver")
Signed-off-by: Sham Muthayyan <smuthayy@codeaurora.org>
Signed-off-by: Ansuel Smith <ansuelsmth@gmail.com>
Signed-off-by: Lorenzo Pieralisi <lorenzo.pieralisi@arm.com>
Acked-by: Stanimir Varbanov <svarbanov@mm-sol.com>
Cc: stable@vger.kernel.org # v4.5+
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
commit 5149901e9e upstream.
Set some specific value for Tx De-Emphasis, Tx Swing and Rx equalization
needed on some ipq8064 based device (Netgear R7800 for example). Without
this the system locks on kernel load.
Link: https://lore.kernel.org/r/20200615210608.21469-8-ansuelsmth@gmail.com
Fixes: 82a823833f ("PCI: qcom: Add Qualcomm PCIe controller driver")
Signed-off-by: Ansuel Smith <ansuelsmth@gmail.com>
Signed-off-by: Lorenzo Pieralisi <lorenzo.pieralisi@arm.com>
Reviewed-by: Rob Herring <robh@kernel.org>
Acked-by: Stanimir Varbanov <svarbanov@mm-sol.com>
Cc: stable@vger.kernel.org # v4.5+
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
commit 2194bc7c39 upstream.
device_attach() returning failure indicates a driver error while trying to
probe the device. In such a scenario, the PCI device should still be added
in the system and be visible to the user.
When device_attach() fails, merely warn about it and keep the PCI device in
the system.
This partially reverts ab1a187bba ("PCI: Check device_attach() return
value always").
Link: https://lore.kernel.org/r/20200706233240.3245512-1-rajatja@google.com
Signed-off-by: Rajat Jain <rajatja@google.com>
Signed-off-by: Bjorn Helgaas <bhelgaas@google.com>
Reviewed-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
Cc: stable@vger.kernel.org # v4.6+
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
commit 45beb31d3a upstream.
We are seeing AMD Radeon Pro W5700 doesn't work when IOMMU is enabled:
iommu ivhd0: AMD-Vi: Event logged [IOTLB_INV_TIMEOUT device=63:00.0 address=0x42b5b01a0]
iommu ivhd0: AMD-Vi: Event logged [IOTLB_INV_TIMEOUT device=63:00.0 address=0x42b5b01c0]
The error also makes graphics driver fail to probe the device.
It appears to be the same issue as commit 5e89cd303e ("PCI: Mark AMD
Navi14 GPU rev 0xc5 ATS as broken") addresses, and indeed the same ATS
quirk can workaround the issue.
See-also: 5e89cd303e ("PCI: Mark AMD Navi14 GPU rev 0xc5 ATS as broken")
See-also: d28ca864c4 ("PCI: Mark AMD Stoney Radeon R7 GPU ATS as broken")
See-also: 9b44b0b09d ("PCI: Mark AMD Stoney GPU ATS as broken")
Bugzilla: https://bugzilla.kernel.org/show_bug.cgi?id=208725
Link: https://lore.kernel.org/r/20200728104554.28927-1-kai.heng.feng@canonical.com
Signed-off-by: Kai-Heng Feng <kai.heng.feng@canonical.com>
Signed-off-by: Bjorn Helgaas <bhelgaas@google.com>
Acked-by: Alex Deucher <alexander.deucher@amd.com>
Cc: stable@vger.kernel.org
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
commit dae68d7fd4 upstream.
If context is not NULL in acpiphp_grab_context(), but the
is_going_away flag is set for the device's parent, the reference
counter of the context needs to be decremented before returning
NULL or the context will never be freed, so make that happen.
Fixes: edf5bf34d4 ("ACPI / dock: Use callback pointers from devices' ACPI hotplug contexts")
Reported-by: Vasily Averin <vvs@virtuozzo.com>
Cc: 3.15+ <stable@vger.kernel.org> # 3.15+
Signed-off-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
commit ec0160891e upstream.
Commit 711419e504 ("irqdomain: Add the missing assignment of
domain->fwnode for named fwnode") unintentionally caused a dangling pointer
page fault issue on firmware nodes that were freed after IRQ domain
allocation. Commit e3beca48a4 fixed that dangling pointer issue by only
freeing the firmware node after an IRQ domain allocation failure. That fix
no longer frees the firmware node immediately, but leaves the firmware node
allocated after the domain is removed.
The firmware node must be kept around through irq_domain_remove, but should be
freed it afterwards.
Add the missing free operations after domain removal where where appropriate.
Fixes: e3beca48a4 ("irqdomain/treewide: Keep firmware node unconditionally allocated")
Signed-off-by: Jon Derrick <jonathan.derrick@intel.com>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Reviewed-by: Andy Shevchenko <andriy.shevchenko@linux.intel.com>
Acked-by: Bjorn Helgaas <bhelgaas@google.com> # drivers/pci
Cc: stable@vger.kernel.org
Link: https://lkml.kernel.org/r/1595363169-7157-1-git-send-email-jonathan.derrick@intel.com
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
[ Upstream commit 090688fa4e ]
The acpi_get_table() should be coupled with acpi_put_table() if the mapped
table is not used at runtime to release the table mapping.
In pci_quirk_amd_sb_acs(), IVRS table is just used for checking AMD IOMMU
is supported, not used at runtime, so put the table after using it.
Fixes: 15b100dfd1 ("PCI: Claim ACS support for AMD southbridge devices")
Link: https://lore.kernel.org/r/1595411068-15440-1-git-send-email-guohanjun@huawei.com
Signed-off-by: Hanjun Guo <guohanjun@huawei.com>
Signed-off-by: Bjorn Helgaas <bhelgaas@google.com>
Signed-off-by: Sasha Levin <sashal@kernel.org>
[ Upstream commit e3bca37d15 ]
Commit 1b79c52844 ("PCI: cadence: Add host driver for Cadence PCIe
controller") in order to update Vendor ID, directly wrote to
PCI_VENDOR_ID register. However PCI_VENDOR_ID in root port configuration
space is read-only register and writing to it will have no effect.
Use local management register to configure Vendor ID and Subsystem Vendor
ID.
Link: https://lore.kernel.org/r/20200722110317.4744-10-kishon@ti.com
Fixes: 1b79c52844 ("PCI: cadence: Add host driver for Cadence PCIe controller")
Signed-off-by: Kishon Vijay Abraham I <kishon@ti.com>
Signed-off-by: Lorenzo Pieralisi <lorenzo.pieralisi@arm.com>
Reviewed-by: Rob Herring <robh@kernel.org>
Signed-off-by: Sasha Levin <sashal@kernel.org>
[ Upstream commit 2a7e32d054 ]
The pci_cfg_wait queue is used to prevent user-space config accesses to
devices while they are recovering from reset.
Previously we used these operations on pci_cfg_wait:
__add_wait_queue(&pci_cfg_wait, ...)
__remove_wait_queue(&pci_cfg_wait, ...)
wake_up_all(&pci_cfg_wait)
The wake_up acquires the wait queue lock, but the add and remove do not.
Originally these were all protected by the pci_lock, but cdcb33f982
("PCI: Avoid possible deadlock on pci_lock and p->pi_lock"), moved
wake_up_all() outside pci_lock, so it could race with add/remove
operations, which caused occasional kernel panics, e.g., during vfio-pci
hotplug/unplug testing:
Unable to handle kernel read from unreadable memory at virtual address ffff802dac469000
Resolve this by using wait_event() instead of __add_wait_queue() and
__remove_wait_queue(). The wait queue lock is held by both wait_event()
and wake_up_all(), so it provides mutual exclusion.
Fixes: cdcb33f982 ("PCI: Avoid possible deadlock on pci_lock and p->pi_lock")
Link: https://lore.kernel.org/linux-pci/79827f2f-9b43-4411-1376-b9063b67aee3@huawei.com/T/#u
Based-on: https://lore.kernel.org/linux-pci/20191210031527.40136-1-zhengxiang9@huawei.com/
Based-on-patch-by: Xiang Zheng <zhengxiang9@huawei.com>
Signed-off-by: Bjorn Helgaas <bhelgaas@google.com>
Tested-by: Xiang Zheng <zhengxiang9@huawei.com>
Cc: Heyi Guo <guoheyi@huawei.com>
Cc: Biaoxiang Ye <yebiaoxiang@huawei.com>
Signed-off-by: Sasha Levin <sashal@kernel.org>
commit e7b856dfce upstream.
As reported in https://bugzilla.kernel.org/206217 , raw_violation_fixup
is causing more harm than good in some common use-cases.
This patch is a partial revert of commit:
191cd6fb5d ("PCI: tegra: Add SW fixup for RAW violations")
and fixes the following regression since then.
* Description:
When both the NIC and MMC are used one can see the following message:
NETDEV WATCHDOG: enp1s0 (r8169): transmit queue 0 timed out
and
pcieport 0000:00:02.0: AER: Uncorrected (Non-Fatal) error received: 0000:01:00.0
r8169 0000:01:00.0: AER: PCIe Bus Error: severity=Uncorrected (Non-Fatal), type=Transaction Layer, (Requester ID)
r8169 0000:01:00.0: AER: device [10ec:8168] error status/mask=00004000/00400000
r8169 0000:01:00.0: AER: [14] CmpltTO (First)
r8169 0000:01:00.0: AER: can't recover (no error_detected callback)
pcieport 0000:00:02.0: AER: device recovery failed
After that, the ethernet NIC is not functional anymore even after
reloading the r8169 module. After a reboot, this is reproducible by
copying a large file over the NIC to the MMC.
For some reason this is not reproducible when files are copied to a tmpfs.
* Little background on the fixup, by Manikanta Maddireddy:
"In the internal testing with dGPU on Tegra124, CmplTO is reported by
dGPU. This happened because FIFO queue in AFI(AXI to PCIe) module
get full by upstream posted writes. Back to back upstream writes
interleaved with infrequent reads, triggers RAW violation and CmpltTO.
This is fixed by reducing the posted write credits and by changing
updateFC timer frequency. These settings are fixed after stress test.
In the current case, RTL NIC is also reporting CmplTO. These settings
seems to be aggravating the issue instead of fixing it."
Link: https://lore.kernel.org/r/20200718100710.15398-1-kwizart@gmail.com
Fixes: 191cd6fb5d ("PCI: tegra: Add SW fixup for RAW violations")
Signed-off-by: Nicolas Chauvet <kwizart@gmail.com>
Signed-off-by: Lorenzo Pieralisi <lorenzo.pieralisi@arm.com>
Reviewed-by: Manikanta Maddireddy <mmaddireddy@nvidia.com>
Cc: stable@vger.kernel.org
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
commit b361663c5a upstream.
Recently ASPM handling was changed to allow ASPM on PCIe-to-PCI/PCI-X
bridges. Unfortunately the ASMedia ASM1083/1085 PCIe to PCI bridge device
doesn't seem to function properly with ASPM enabled. On an Asus PRIME
H270-PRO motherboard, it causes errors like these:
pcieport 0000:00:1c.0: AER: PCIe Bus Error: severity=Corrected, type=Data Link Layer, (Transmitter ID)
pcieport 0000:00:1c.0: AER: device [8086:a292] error status/mask=00003000/00002000
pcieport 0000:00:1c.0: AER: [12] Timeout
pcieport 0000:00:1c.0: AER: Corrected error received: 0000:00:1c.0
pcieport 0000:00:1c.0: AER: can't find device of ID00e0
In addition to flooding the kernel log, this also causes the machine to
wake up immediately after suspend is initiated.
The device advertises ASPM L0s and L1 support in the Link Capabilities
register, but the ASMedia web page for ASM1083 [1] claims "No PCIe ASPM
support".
Windows 10 (build 2004) enables L0s, but it also logs correctable PCIe
errors.
Add a quirk to disable ASPM for this device.
[1] https://www.asmedia.com.tw/eng/e_show_products.php?cate_index=169&item=114
[bhelgaas: commit log]
Fixes: 66ff14e59e ("PCI/ASPM: Allow ASPM on links to PCIe-to-PCI/PCI-X Bridges")
Bugzilla: https://bugzilla.kernel.org/show_bug.cgi?id=208667
Link: https://lore.kernel.org/r/20200722021803.17958-1-hancockrwd@gmail.com
Signed-off-by: Robert Hancock <hancockrwd@gmail.com>
Signed-off-by: Bjorn Helgaas <bhelgaas@google.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
[ Upstream commit e3beca48a4 ]
Quite some non OF/ACPI users of irqdomains allocate firmware nodes of type
IRQCHIP_FWNODE_NAMED or IRQCHIP_FWNODE_NAMED_ID and free them right after
creating the irqdomain. The only purpose of these FW nodes is to convey
name information. When this was introduced the core code did not store the
pointer to the node in the irqdomain. A recent change stored the firmware
node pointer in irqdomain for other reasons and missed to notice that the
usage sites which do the alloc_fwnode/create_domain/free_fwnode sequence
are broken by this. Storing a dangling pointer is dangerous itself, but in
case that the domain is destroyed later on this leads to a double free.
Remove the freeing of the firmware node after creating the irqdomain from
all affected call sites to cure this.
Fixes: 711419e504 ("irqdomain: Add the missing assignment of domain->fwnode for named fwnode")
Reported-by: Andy Shevchenko <andriy.shevchenko@linux.intel.com>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Acked-by: Bjorn Helgaas <bhelgaas@google.com>
Acked-by: Marc Zyngier <maz@kernel.org>
Cc: stable@vger.kernel.org
Link: https://lkml.kernel.org/r/873661qakd.fsf@nanos.tec.linutronix.de
Signed-off-by: Sasha Levin <sashal@kernel.org>
commit c3aaf08670 upstream.
26ad34d510 ("PCI / ACPI: Whitelist D3 for more PCIe hotplug ports") added
the struct pci_platform_pm_ops.bridge_d3() function pointer and
platform_pci_bridge_d3() to use it.
The .bridge_d3() op is implemented by acpi_pci_platform_pm, but not by
mid_pci_platform_pm. We don't expect platform_pci_bridge_d3() to be called
on Intel MID platforms, but nothing in the code itself would prevent that.
Check the .bridge_d3() pointer for NULL before calling it.
Fixes: 26ad34d510 ("PCI / ACPI: Whitelist D3 for more PCIe hotplug ports")
Signed-off-by: Bjorn Helgaas <bhelgaas@google.com>
Reviewed-by: Mika Westerberg <mika.westerberg@linux.intel.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
[ Upstream commit 87dccf0932 ]
The vim3l board does not work with a standard PCIe switch (ASM1184e),
spitting all kind of errors - hinting at HW misconfiguration (no link,
port enumeration issues, etc).
According to the the Synopsys DWC PCIe Reference Manual, in the section
dedicated to the PLCR register, bit 7 is described (FAST_LINK_MODE) as:
"Sets all internal timers to fast mode for simulation purposes."
it is sound to set this bit from a simulation perspective, but on actual
silicon, which expects timers to have a nominal value, it is not.
Make sure the FAST_LINK_MODE bit is cleared when configuring the RC
to solve this problem.
Link: https://lore.kernel.org/r/20200429164230.309922-1-maz@kernel.org
Fixes: 9c0ef6d34f ("PCI: amlogic: Add the Amlogic Meson PCIe controller driver")
Signed-off-by: Marc Zyngier <maz@kernel.org>
[lorenzo.pieralisi@arm.com: commit log]
Signed-off-by: Lorenzo Pieralisi <lorenzo.pieralisi@arm.com>
Reviewed-by: Neil Armstrong <narmstrong@baylibre.com>
Acked-by: Rob Herring <robh@kernel.org>
Signed-off-by: Sasha Levin <sashal@kernel.org>
[ Upstream commit 0414b93e78 ]
On a system that uses the internal DWC MSI widget, I get this
warning from debugfs when CONFIG_GENERIC_IRQ_DEBUGFS is selected:
debugfs: File ':soc:pcie@fc000000' in directory 'domains' already present!
This is due to the fact that the DWC MSI code tries to register two
IRQ domains for the same firmware node, without telling the low
level code how to distinguish them (by setting a bus token). This
further confuses debugfs which tries to create corresponding
files for each domain.
Fix it by tagging the inner domain as DOMAIN_BUS_NEXUS, which is
the closest thing we have as to "generic MSI".
Link: https://lore.kernel.org/r/20200501113921.366597-1-maz@kernel.org
Signed-off-by: Marc Zyngier <maz@kernel.org>
Signed-off-by: Lorenzo Pieralisi <lorenzo.pieralisi@arm.com>
Acked-by: Jingoo Han <jingoohan1@gmail.com>
Signed-off-by: Sasha Levin <sashal@kernel.org>
[ Upstream commit 7b38fd9760 ]
Except for Endpoints, we enable PTM at enumeration-time. Previously we did
not account for the fact that Switch Downstream Ports are not permitted to
have a PTM capability; their PTM behavior is controlled by the Upstream
Port (PCIe r5.0, sec 7.9.16). Since Downstream Ports don't have a PTM
capability, we did not mark them as "ptm_enabled", which meant that
pci_enable_ptm() on an Endpoint failed because there was no PTM path to it.
Mark Downstream Ports as "ptm_enabled" if their Upstream Port has PTM
enabled.
Fixes: eec097d431 ("PCI: Add pci_enable_ptm() for drivers to enable PTM on endpoints")
Reported-by: Aditya Paluri <Venkata.AdityaPaluri@synopsys.com>
Signed-off-by: Bjorn Helgaas <bhelgaas@google.com>
Signed-off-by: Sasha Levin <sashal@kernel.org>
[ Upstream commit ec411e02b7 ]
Kai-Heng Feng reported that it takes a long time (> 1 s) to resume
Thunderbolt-connected devices from both runtime suspend and system sleep
(s2idle).
This was because some Downstream Ports that support > 5 GT/s do not also
support Data Link Layer Link Active reporting. Per PCIe r5.0 sec 6.6.1:
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).
Sec 7.5.3.6 requires such Ports to support DLL Link Active reporting, but
at least the Intel JHL6240 Thunderbolt 3 Bridge [8086:15c0] and the Intel
JHL7540 Thunderbolt 3 Bridge [8086:15ea] do not.
Previously we tried to wait for Link training to complete, but since there
was no DLL Link Active reporting, all we could do was wait the worst-case
1000 ms, then another 100 ms.
Instead of using the supported speeds to determine whether to wait for Link
training, check whether the port supports DLL Link Active reporting. The
Ports in question do not, so we'll wait only the 100 ms required for Ports
that support Link speeds <= 5 GT/s.
This of course assumes these Ports always train the Link within 100 ms even
if they are operating at > 5 GT/s, which is not required by the spec.
[bhelgaas: commit log, comment]
Link: https://bugzilla.kernel.org/show_bug.cgi?id=206837
Link: https://lore.kernel.org/r/20200514133043.27429-1-mika.westerberg@linux.intel.com
Reported-by: Kai-Heng Feng <kai.heng.feng@canonical.com>
Tested-by: Kai-Heng Feng <kai.heng.feng@canonical.com>
Signed-off-by: Mika Westerberg <mika.westerberg@linux.intel.com>
Signed-off-by: Bjorn Helgaas <bhelgaas@google.com>
Signed-off-by: Sasha Levin <sashal@kernel.org>
[ Upstream commit 1b54ae8327 ]
If device_register() has an error, we should bail out of
pci_register_host_bridge() rather than continuing on.
Fixes: 37d6a0a6f4 ("PCI: Add pci_register_host_bridge() interface")
Link: https://lore.kernel.org/r/20200513223859.11295-1-robh@kernel.org
Signed-off-by: Rob Herring <robh@kernel.org>
Signed-off-by: Bjorn Helgaas <bhelgaas@google.com>
Reviewed-by: Lorenzo Pieralisi <lorenzo.pieralisi@arm.com>
Reviewed-by: Arnd Bergmann <arnd@arndb.de>
Signed-off-by: Sasha Levin <sashal@kernel.org>
[ Upstream commit 66ff14e59e ]
7d715a6c1a ("PCI: add PCI Express ASPM support") added the ability for
Linux to enable ASPM, but for some undocumented reason, it didn't enable
ASPM on links where the downstream component is a PCIe-to-PCI/PCI-X Bridge.
Remove this exclusion so we can enable ASPM on these links.
The Dell OptiPlex 7080 mentioned in the bugzilla has a TI XIO2001
PCIe-to-PCI Bridge. Enabling ASPM on the link leading to it allows the
Intel SoC to enter deeper Package C-states, which is a significant power
savings.
[bhelgaas: commit log]
Bugzilla: https://bugzilla.kernel.org/show_bug.cgi?id=207571
Link: https://lore.kernel.org/r/20200505173423.26968-1-kai.heng.feng@canonical.com
Signed-off-by: Kai-Heng Feng <kai.heng.feng@canonical.com>
Signed-off-by: Bjorn Helgaas <bhelgaas@google.com>
Reviewed-by: Mika Westerberg <mika.westerberg@linux.intel.com>
Signed-off-by: Sasha Levin <sashal@kernel.org>
[ Upstream commit 2b9f217433 ]
The outbound windows (PCIEPAUR(x), PCIEPALR(x)) describe a mapping between
a CPU address (which is determined by the window number 'x') and a
programmed PCI address - Thus allowing the controller to translate CPU
accesses into PCI accesses.
However the existing code incorrectly writes the CPU address - lets fix
this by writing the PCI address instead.
For memory transactions, existing DT users describe a 1:1 identity mapping
and thus this change should have no effect. However the same isn't true for
I/O.
Link: https://lore.kernel.org/r/20191004132941.6660-1-andrew.murray@arm.com
Fixes: c25da47788 ("PCI: rcar: Add Renesas R-Car PCIe driver")
Tested-by: Marek Vasut <marek.vasut+renesas@gmail.com>
Signed-off-by: Andrew Murray <andrew.murray@arm.com>
Signed-off-by: Lorenzo Pieralisi <lorenzo.pieralisi@arm.com>
Reviewed-by: Marek Vasut <marek.vasut+renesas@gmail.com>
Signed-off-by: Sasha Levin <sashal@kernel.org>
[ Upstream commit bca718988b ]
If we fails somewhere in 'v3_pci_probe()', we need to free 'host'.
Use the managed version of 'pci_alloc_host_bridge()' to do that easily.
The use of managed resources is already widely used in this driver.
Link: https://lore.kernel.org/r/20200418081637.1585-1-christophe.jaillet@wanadoo.fr
Fixes: 68a15eb7bd ("PCI: v3-semi: Add V3 Semiconductor PCI host driver")
Signed-off-by: Christophe JAILLET <christophe.jaillet@wanadoo.fr>
[lorenzo.pieralisi@arm.com: commit log]
Signed-off-by: Lorenzo Pieralisi <lorenzo.pieralisi@arm.com>
Acked-by: Linus Walleij <linus.walleij@linaro.org>
Signed-off-by: Sasha Levin <sashal@kernel.org>
[ Upstream commit 3e5095eebe ]
Versions of VMD with the Host Physical Address shadow register use this
register to calculate the bus address offset needed to do guest
passthrough of the domain. This register shadows the Host Physical
Address registers including the resource type bits. After calculating
the offset, the extra resource type bits lead to the VMD resources being
over-provisioned at the front and under-provisioned at the back.
Example:
pci 10000:80:02.0: reg 0x10: [mem 0xf801fffc-0xf803fffb 64bit]
Expected:
pci 10000:80:02.0: reg 0x10: [mem 0xf8020000-0xf803ffff 64bit]
If other devices are mapped in the over-provisioned front, it could lead
to resource conflict issues with VMD or those devices.
Link: https://lore.kernel.org/r/20200528030240.16024-3-jonathan.derrick@intel.com
Fixes: a1a3017013 ("PCI: vmd: Fix shadow offsets to reflect spec changes")
Signed-off-by: Jon Derrick <jonathan.derrick@intel.com>
Signed-off-by: Lorenzo Pieralisi <lorenzo.pieralisi@arm.com>
Signed-off-by: Sasha Levin <sashal@kernel.org>
[ Upstream commit c88d191817 ]
This patch fixes two bit conflicts in the pci-bridge-emul driver:
1. Bit 3 of Device Status (19 of Device Control) is marked as both
Write-1-to-Clear and Read-Only. It should be Write-1-to-Clear.
The Read-Only and Reserved bitmasks are shifted by 1 bit due to this
error.
2. Bit 12 of Slot Control is marked as both Read-Write and Reserved.
It should be Read-Write.
Link: https://lore.kernel.org/r/20200511162117.6674-2-jonathan.derrick@intel.com
Signed-off-by: Jon Derrick <jonathan.derrick@intel.com>
Signed-off-by: Lorenzo Pieralisi <lorenzo.pieralisi@arm.com>
Acked-by: Rob Herring <robh@kernel.org>
Signed-off-by: Sasha Levin <sashal@kernel.org>
[ Upstream commit 90c6cb4a35 ]
Trying to change Link Status register does not have any effect as this
is a read-only register. Trying to overwrite bits for Negotiated Link
Width does not make sense.
In future proper change of link width can be done via Lane Count Select
bits in PCIe Control 0 register.
Trying to unconditionally enable ASPM L0s via ASPM Control bits in Link
Control register is wrong. There should be at least some detection if
endpoint supports L0s as isn't mandatory.
Moreover ASPM Control bits in Link Control register are controlled by
pcie/aspm.c code which sets it according to system ASPM settings,
immediately after aardvark driver probes. So setting these bits by
aardvark driver has no long running effect.
Remove code which touches ASPM L0s bits from this driver and let
kernel's ASPM implementation to set ASPM state properly.
Some users are reporting issues that this code is problematic for some
Intel wifi cards and removing it fixes them, see e.g.:
https://bugzilla.kernel.org/show_bug.cgi?id=196339
If problems with Intel wifi cards occur even after this commit, then
pcie/aspm.c code could be modified / hooked to not enable ASPM L0s state
for affected problematic cards.
Link: https://lore.kernel.org/r/20200430080625.26070-3-pali@kernel.org
Tested-by: Tomasz Maciej Nowak <tmn505@gmail.com>
Signed-off-by: Pali Rohár <pali@kernel.org>
Signed-off-by: Lorenzo Pieralisi <lorenzo.pieralisi@arm.com>
Acked-by: Rob Herring <robh@kernel.org>
Acked-by: Thomas Petazzoni <thomas.petazzoni@bootlin.com>
Signed-off-by: Sasha Levin <sashal@kernel.org>
[ Upstream commit d09ddd8190 ]
When resizing a BAR, pci_reassign_bridge_resources() is invoked to bring
the bridge windows of parent bridges in line with the new BAR assignment.
This assumes the device whose BAR is being resized lives on a subordinate
bus, but this is not necessarily the case. A device may live on the root
bus, in which case dev->bus->self is NULL, and passing a NULL pci_dev
pointer to pci_reassign_bridge_resources() will cause it to crash.
So let's make the call to pci_reassign_bridge_resources() conditional on
whether dev->bus->self is non-NULL in the first place.
Fixes: 8bb705e3e7 ("PCI: Add pci_resize_resource() for resizing BARs")
Link: https://lore.kernel.org/r/20200421162256.26887-1-ardb@kernel.org
Signed-off-by: Ard Biesheuvel <ardb@kernel.org>
Signed-off-by: Bjorn Helgaas <bhelgaas@google.com>
Reviewed-by: Christian König <christian.koenig@amd.com>
Signed-off-by: Sasha Levin <sashal@kernel.org>
commit aa0ce96d72 upstream.
Root Complex Integrated Endpoints (RCiEPs) do not have an upstream bridge,
so pci_configure_mps() previously ignored them, which may result in reduced
performance.
Instead, program the Max_Payload_Size of RCiEPs to the maximum supported
value (unless it is limited for the PCIE_BUS_PEER2PEER case). This also
affects the subsequent programming of Max_Read_Request_Size because Linux
programs MRRS based on the MPS value.
Fixes: 9dae3a9729 ("PCI: Move MPS configuration check to pci_configure_device()")
Link: https://lore.kernel.org/r/1585343775-4019-1-git-send-email-ashok.raj@intel.com
Tested-by: Dave Jiang <dave.jiang@intel.com>
Signed-off-by: Ashok Raj <ashok.raj@intel.com>
Signed-off-by: Bjorn Helgaas <bhelgaas@google.com>
Cc: stable@vger.kernel.org
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
[ Upstream commit ec11e5c213 ]
This patch adds support for this VMD device which supports the bus
restriction mode.
Signed-off-by: Jon Derrick <jonathan.derrick@intel.com>
Signed-off-by: Lorenzo Pieralisi <lorenzo.pieralisi@arm.com>
Signed-off-by: Sasha Levin <sashal@kernel.org>
[ Upstream commit 3247bd10a4 ]
All Intel platforms guarantee that all root complex implementations must
send transactions up to IOMMU for address translations. Hence for Intel
RCiEP devices, we can assume some ACS-type isolation even without an ACS
capability.
From the Intel VT-d spec, r3.1, sec 3.16 ("Root-Complex Peer to Peer
Considerations"):
When DMA remapping is enabled, peer-to-peer requests through the
Root-Complex must be handled as follows:
- The input address in the request is translated (through first-level,
second-level or nested translation) to a host physical address (HPA).
The address decoding for peer addresses must be done only on the
translated HPA. Hardware implementations are free to further limit
peer-to-peer accesses to specific host physical address regions (or
to completely disallow peer-forwarding of translated requests).
- Since address translation changes the contents (address field) of
the PCI Express Transaction Layer Packet (TLP), for PCI Express
peer-to-peer requests with ECRC, the Root-Complex hardware must use
the new ECRC (re-computed with the translated address) if it
decides to forward the TLP as a peer request.
- Root-ports, and multi-function root-complex integrated endpoints, may
support additional peer-to-peer control features by supporting PCI
Express Access Control Services (ACS) capability. Refer to ACS
capability in PCI Express specifications for details.
Since Linux didn't give special treatment to allow this exception, certain
RCiEP MFD devices were grouped in a single IOMMU group. This doesn't permit
a single device to be assigned to a guest for instance.
In one vendor system: Device 14.x were grouped in a single IOMMU group.
/sys/kernel/iommu_groups/5/devices/0000:00:14.0
/sys/kernel/iommu_groups/5/devices/0000:00:14.2
/sys/kernel/iommu_groups/5/devices/0000:00:14.3
After this patch:
/sys/kernel/iommu_groups/5/devices/0000:00:14.0
/sys/kernel/iommu_groups/5/devices/0000:00:14.2
/sys/kernel/iommu_groups/6/devices/0000:00:14.3 <<< new group
14.0 and 14.2 are integrated devices, but legacy end points, whereas 14.3
was a PCIe-compliant RCiEP.
00:14.3 Network controller: Intel Corporation Device 9df0 (rev 30)
Capabilities: [40] Express (v2) Root Complex Integrated Endpoint, MSI 00
This permits assigning this device to a guest VM.
[bhelgaas: drop "Fixes" tag since this doesn't fix a bug in that commit]
Link: https://lore.kernel.org/r/1590699462-7131-1-git-send-email-ashok.raj@intel.com
Tested-by: Darrel Goeddel <dgoeddel@forcepoint.com>
Signed-off-by: Ashok Raj <ashok.raj@intel.com>
Signed-off-by: Bjorn Helgaas <bhelgaas@google.com>
Reviewed-by: Alex Williamson <alex.williamson@redhat.com>
Cc: stable@vger.kernel.org
Cc: Lu Baolu <baolu.lu@linux.intel.com>
Cc: Mark Scott <mscott@forcepoint.com>,
Cc: Romil Sharma <rsharma@forcepoint.com>
Signed-off-by: Sasha Levin <sashal@kernel.org>
[ Upstream commit 5727043c73 ]
The AMD Starship USB 3.0 host controller advertises Function Level Reset
support, but it apparently doesn't work. Add a quirk to prevent use of FLR
on this device.
Without this quirk, when attempting to assign (pass through) an AMD
Starship USB 3.0 host controller to a guest OS, the system becomes
increasingly unresponsive over the course of several minutes, eventually
requiring a hard reset. Shortly after attempting to start the guest, I see
these messages:
vfio-pci 0000:05:00.3: not ready 1023ms after FLR; waiting
vfio-pci 0000:05:00.3: not ready 2047ms after FLR; waiting
vfio-pci 0000:05:00.3: not ready 4095ms after FLR; waiting
vfio-pci 0000:05:00.3: not ready 8191ms after FLR; waiting
And then eventually:
vfio-pci 0000:05:00.3: not ready 65535ms after FLR; giving up
INFO: NMI handler (perf_event_nmi_handler) took too long to run: 0.000 msecs
perf: interrupt took too long (642744 > 2500), lowering kernel.perf_event_max_sample_rate to 1000
INFO: NMI handler (perf_event_nmi_handler) took too long to run: 82.270 msecs
INFO: NMI handler (perf_event_nmi_handler) took too long to run: 680.608 msecs
INFO: NMI handler (perf_event_nmi_handler) took too long to run: 100.952 msecs
...
watchdog: BUG: soft lockup - CPU#3 stuck for 22s! [qemu-system-x86:7487]
Tested on a Micro-Star International Co., Ltd. MS-7C59/Creator TRX40
motherboard with an AMD Ryzen Threadripper 3970X.
Link: https://lore.kernel.org/r/20200524003529.598434ff@f31-4.lan
Signed-off-by: Kevin Buettner <kevinb@redhat.com>
Signed-off-by: Bjorn Helgaas <bhelgaas@google.com>
Signed-off-by: Sasha Levin <sashal@kernel.org>
[ Upstream commit 0d14f06cd6 ]
The AMD Matisse HD Audio & USB 3.0 devices advertise Function Level Reset
support, but hang when an FLR is triggered.
To reproduce the problem, attach the device to a VM, then detach and try to
attach again.
Rename the existing quirk_intel_no_flr(), which was not Intel-specific, to
quirk_no_flr(), and apply it to prevent the use of FLR on these AMD
devices.
Link: https://lore.kernel.org/r/CAAri2DpkcuQZYbT6XsALhx2e6vRqPHwtbjHYeiH7MNp4zmt1RA@mail.gmail.com
Signed-off-by: Marcos Scriven <marcos@scriven.org>
Signed-off-by: Bjorn Helgaas <bhelgaas@google.com>
Signed-off-by: Sasha Levin <sashal@kernel.org>
[ Upstream commit 68f5fc4ea9 ]
Both Pericom OHCI and EHCI devices advertise PME# support from all power
states:
06:00.0 USB controller [0c03]: Pericom Semiconductor PI7C9X442SL USB OHCI Controller [12d8:400e] (rev 01) (prog-if 10 [OHCI])
Subsystem: Pericom Semiconductor PI7C9X442SL USB OHCI Controller [12d8:400e]
Capabilities: [80] Power Management version 3
Flags: PMEClk- DSI- D1+ D2+ AuxCurrent=375mA PME(D0+,D1+,D2+,D3hot+,D3cold+)
06:00.2 USB controller [0c03]: Pericom Semiconductor PI7C9X442SL USB EHCI Controller [12d8:400f] (rev 01) (prog-if 20 [EHCI])
Subsystem: Pericom Semiconductor PI7C9X442SL USB EHCI Controller [12d8:400f]
Capabilities: [80] Power Management version 3
Flags: PMEClk- DSI- D1+ D2+ AuxCurrent=375mA PME(D0+,D1+,D2+,D3hot+,D3cold+)
But testing shows that it's unreliable: there is a 20% chance PME# won't be
asserted when a USB device is plugged.
Remove PME support for both devices to make USB plugging work reliably.
Bugzilla: https://bugzilla.kernel.org/show_bug.cgi?id=205981
Link: https://lore.kernel.org/r/20200508065343.32751-2-kai.heng.feng@canonical.com
Signed-off-by: Kai-Heng Feng <kai.heng.feng@canonical.com>
Signed-off-by: Bjorn Helgaas <bhelgaas@google.com>
Cc: stable@vger.kernel.org
Signed-off-by: Sasha Levin <sashal@kernel.org>
[ Upstream commit b6caa1d8c8 ]
Don't disable MEM/IO decoding when a device have both non_compliant_bars
and mmio_always_on.
That would allow us quirk devices with junk in BARs but can't disable
their decoding.
Signed-off-by: Jiaxun Yang <jiaxun.yang@flygoat.com>
Acked-by: Bjorn Helgaas <helgaas@kernel.org>
Signed-off-by: Thomas Bogendoerfer <tsbogend@alpha.franken.de>
Signed-off-by: Sasha Levin <sashal@kernel.org>
[ Upstream commit f044baaff1 ]
The caller of pcie_wait_for_link_delay() specifies the time to wait after
the link becomes active. When the downstream port doesn't support link
active reporting, obviously we can't tell when the link becomes active, so
we waited the worst-case time (1000 ms) plus 100 ms, ignoring the delay
from the caller.
Instead, wait for 1000 ms + the delay from the caller.
Fixes: 4827d63891 ("PCI/PM: Add pcie_wait_for_link_delay()")
Signed-off-by: Bjorn Helgaas <bhelgaas@google.com>
Signed-off-by: Sasha Levin <sashal@kernel.org>
commit 0a8f41023e upstream.
Some Google Apex Edge TPU devices have a class code of 0
(PCI_CLASS_NOT_DEFINED). This prevents the PCI core from assigning
resources for the Apex BARs because __dev_sort_resources() ignores
classless devices, host bridges, and IOAPICs.
On x86, firmware typically assigns those resources, so this was not a
problem. But on some architectures, firmware does *not* assign BARs, and
since the PCI core didn't do it either, the Apex device didn't work
correctly:
apex 0000:01:00.0: can't enable device: BAR 0 [mem 0x00000000-0x00003fff 64bit pref] not claimed
apex 0000:01:00.0: error enabling PCI device
f390d08d8b ("staging: gasket: apex: fixup undefined PCI class") added a
quirk to fix the class code, but it was in the apex driver, and if the
driver was built as a module, it was too late to help.
Move the quirk to the PCI core, where it will always run early enough that
the PCI core will assign resources if necessary.
Link: https://lore.kernel.org/r/CAEzXK1r0Er039iERnc2KJ4jn7ySNUOG9H=Ha8TD8XroVqiZjgg@mail.gmail.com
Fixes: f390d08d8b ("staging: gasket: apex: fixup undefined PCI class")
Reported-by: Luís Mendes <luis.p.mendes@gmail.com>
Debugged-by: Luís Mendes <luis.p.mendes@gmail.com>
Tested-by: Luis Mendes <luis.p.mendes@gmail.com>
Signed-off-by: Bjorn Helgaas <bhelgaas@google.com>
Cc: Todd Poynor <toddpoynor@google.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
commit 299bd044a6 upstream.
Many Zhaoxin Root Ports and Switch Downstream Ports do provide ACS-like
capability but have no ACS Capability Structure. Peer-to-Peer transactions
could be blocked between these ports, so add quirk so devices behind them
could be assigned to different IOMMU group.
Link: https://lore.kernel.org/r/20200327091148.5190-4-RaymondPang-oc@zhaoxin.com
Signed-off-by: Raymond Pang <RaymondPang-oc@zhaoxin.com>
Signed-off-by: Bjorn Helgaas <bhelgaas@google.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
commit 7cf2cba43f upstream.
Most of the ACS quirks have a similar pattern of:
acs_flags &= ~( <controls provided by this device> );
return acs_flags ? 0 : 1;
Pull this out into a helper function to simplify the quirks slightly. The
helper function is also a convenient place for comments about what the list
of ACS controls means. No functional change intended.
Signed-off-by: Bjorn Helgaas <bhelgaas@google.com>
Reviewed-by: Logan Gunthorpe <logang@deltatee.com>
Reviewed-by: Alex Williamson <alex.williamson@redhat.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
commit c8de8ed2dc upstream.
The ACS quirks differ in needless ways, which makes them look more
different than they really are.
Reorder the ACS flags in order of definitions in the spec:
PCI_ACS_SV Source Validation
PCI_ACS_TB Translation Blocking
PCI_ACS_RR P2P Request Redirect
PCI_ACS_CR P2P Completion Redirect
PCI_ACS_UF Upstream Forwarding
PCI_ACS_EC P2P Egress Control
PCI_ACS_DT Direct Translated P2P
(PCIe r5.0, sec 7.7.8.2) and use similar code structure in all. No
functional change intended.
Signed-off-by: Bjorn Helgaas <bhelgaas@google.com>
Reviewed-by: Logan Gunthorpe <logang@deltatee.com>
Reviewed-by: Alex Williamson <alex.williamson@redhat.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
commit 0325837c51 upstream.
Some Zhaoxin endpoints are implemented as multi-function devices without an
ACS capability, but they actually don't support peer-to-peer transactions.
Add ACS quirks to declare DMA isolation.
Link: https://lore.kernel.org/r/20200327091148.5190-3-RaymondPang-oc@zhaoxin.com
Signed-off-by: Raymond Pang <RaymondPang-oc@zhaoxin.com>
Signed-off-by: Bjorn Helgaas <bhelgaas@google.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
commit 2880325bda upstream.
The ASMedia USB XHCI Controller claims to support generating PME# while
in D0:
01:00.0 USB controller: ASMedia Technology Inc. Device 2142 (prog-if 30 [XHCI])
Subsystem: SUNIX Co., Ltd. Device 312b
Capabilities: [78] Power Management version 3
Flags: PMEClk- DSI- D1- D2- AuxCurrent=55mA PME(D0+,D1-,D2-,D3hot-,D3cold-)
Status: D0 NoSoftRst+ PME-Enable+ DSel=0 DScale=0 PME-
However PME# only gets asserted when plugging USB 2.0 or USB 1.1 devices,
but not for USB 3.0 devices.
Remove PCI_PM_CAP_PME_D0 to avoid using PME under D0.
Bugzilla: https://bugzilla.kernel.org/show_bug.cgi?id=205919
Link: https://lore.kernel.org/r/20191219192006.16270-1-kai.heng.feng@canonical.com
Signed-off-by: Kai-Heng Feng <kai.heng.feng@canonical.com>
Signed-off-by: Bjorn Helgaas <bhelgaas@google.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
[ Upstream commit ad9001f2f4 ]
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 5.0 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 5.0 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):
0000:00:1b.0: wait for 100 ms after DLLLA is set before access to 0000:01:00.0
0000:02:00.0: wait for 100 ms after DLLLA is set before access to 0000:03:00.0
0000:02:02.0: wait for 100 ms after DLLLA is set before access to 0000:37:00.0
I've instrumented the kernel with some additional logging so we can see the
actual delays performed:
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: waiting for D3hot delay of 10 ms
pcieport 0000:02:01.0: waiting for D3hot delay of 10 ms
pcieport 0000:02:04.0: waiting for D3hot delay of 10 ms
For the switch upstream port (01:00.0 reachable through 00:1b.0 root port)
we wait for 100 ms but not taking into account the DLLLA requirement. We
then wait 10 ms 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 it turns
out to be 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.
On this particular 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). Below is an example how it looks like when this happens:
pcieport 0000:83:04.0: pciehp: Slot(4): Card not present
pcieport 0000:87:04.0: PME# disabled
pcieport 0000:83:04.0: pciehp: pciehp_unconfigure_device: domain🚌dev = 0000:86:00
pcieport 0000:86:00.0: Refused to change power state, currently in D3
pcieport 0000:86:00.0: restoring config space at offset 0x3c (was 0xffffffff, writing 0x201ff)
pcieport 0000:86:00.0: restoring config space at offset 0x38 (was 0xffffffff, writing 0x0)
...
There is also one reported case (see the bugzilla link below) where the
missing delay causes xHCI on a Titan Ridge controller fail to runtime
resume when USB-C dock is plugged. This does not involve pciehp but instead
the PCI core fails to runtime resume the xHCI device:
pcieport 0000:04:02.0: restoring config space at offset 0xc (was 0x10000, writing 0x10020)
pcieport 0000:04:02.0: restoring config space at offset 0x4 (was 0x100000, writing 0x100406)
xhci_hcd 0000:39:00.0: Refused to change power state, currently in D3
xhci_hcd 0000:39:00.0: restoring config space at offset 0x3c (was 0xffffffff, writing 0x1ff)
xhci_hcd 0000:39:00.0: restoring config space at offset 0x38 (was 0xffffffff, writing 0x0)
...
Add a new function pci_bridge_wait_for_secondary_bus() that is called on
PCI core resume and runtime resume paths accordingly if the bridge entered
D3cold (and thus went through reset).
This is second attempt to add the missing delays. The previous solution in
c2bf1fc212 ("PCI: Add missing link delays required by the PCIe spec") was
reverted because of two issues it caused:
1. One system become unresponsive after S3 resume due to PME service
spinning in pcie_pme_work_fn(). The root port in question reports that
the xHCI sent PME but the xHCI device itself does not have PME status
set. The PME status bit is never cleared in the root port resulting
the indefinite loop in pcie_pme_work_fn().
2. Slows down resume if the root/downstream port does not support Data
Link Layer Active Reporting because pcie_wait_for_link_delay() waits
1100 ms in that case.
This version should avoid the above issues because we restrict the delay to
happen only if the port went into D3cold.
Link: https://lore.kernel.org/linux-pci/SL2P216MB01878BBCD75F21D882AEEA2880C60@SL2P216MB0187.KORP216.PROD.OUTLOOK.COM/
Link: https://bugzilla.kernel.org/show_bug.cgi?id=203885
Link: https://lore.kernel.org/r/20191112091617.70282-3-mika.westerberg@linux.intel.com
Reported-by: Kai-Heng Feng <kai.heng.feng@canonical.com>
Tested-by: Kai-Heng Feng <kai.heng.feng@canonical.com>
Signed-off-by: Mika Westerberg <mika.westerberg@linux.intel.com>
Signed-off-by: Bjorn Helgaas <bhelgaas@google.com>
Signed-off-by: Sasha Levin <sashal@kernel.org>
[ Upstream commit 35efea32b2 ]
Previously Clock PM could not be re-enabled after being disabled by
pci_disable_link_state() because clkpm_capable was reset. Change this by
adding a clkpm_disable field similar to aspm_disable.
Link: https://lore.kernel.org/r/4e8a66db-7d53-4a66-c26c-f0037ffaa705@gmail.com
Signed-off-by: Heiner Kallweit <hkallweit1@gmail.com>
Signed-off-by: Bjorn Helgaas <bhelgaas@google.com>
Signed-off-by: Sasha Levin <sashal@kernel.org>
[ Upstream commit 87d0f2a553 ]
This addresses deadlocks in these common cases in hierarchies containing
two switches:
- All involved ports are runtime suspended and they are unplugged. This
can happen easily if the drivers involved automatically enable runtime
PM (xHCI for example does that).
- System is suspended (e.g., closing the lid on a laptop) with a dock +
something else connected, and the dock is unplugged while suspended.
These cases lead to the following deadlock:
INFO: task irq/126-pciehp:198 blocked for more than 120 seconds.
irq/126-pciehp D 0 198 2 0x80000000
Call Trace:
schedule+0x2c/0x80
schedule_timeout+0x246/0x350
wait_for_completion+0xb7/0x140
kthread_stop+0x49/0x110
free_irq+0x32/0x70
pcie_shutdown_notification+0x2f/0x50
pciehp_remove+0x27/0x50
pcie_port_remove_service+0x36/0x50
device_release_driver+0x12/0x20
bus_remove_device+0xec/0x160
device_del+0x13b/0x350
device_unregister+0x1a/0x60
remove_iter+0x1e/0x30
device_for_each_child+0x56/0x90
pcie_port_device_remove+0x22/0x40
pcie_portdrv_remove+0x20/0x60
pci_device_remove+0x3e/0xc0
device_release_driver_internal+0x18c/0x250
device_release_driver+0x12/0x20
pci_stop_bus_device+0x6f/0x90
pci_stop_bus_device+0x31/0x90
pci_stop_and_remove_bus_device+0x12/0x20
pciehp_unconfigure_device+0x88/0x140
pciehp_disable_slot+0x6a/0x110
pciehp_handle_presence_or_link_change+0x263/0x400
pciehp_ist+0x1c9/0x1d0
irq_thread_fn+0x24/0x60
irq_thread+0xeb/0x190
kthread+0x120/0x140
INFO: task irq/190-pciehp:2288 blocked for more than 120 seconds.
irq/190-pciehp D 0 2288 2 0x80000000
Call Trace:
__schedule+0x2a2/0x880
schedule+0x2c/0x80
schedule_preempt_disabled+0xe/0x10
mutex_lock+0x2c/0x30
pci_lock_rescan_remove+0x15/0x20
pciehp_unconfigure_device+0x4d/0x140
pciehp_disable_slot+0x6a/0x110
pciehp_handle_presence_or_link_change+0x263/0x400
pciehp_ist+0x1c9/0x1d0
irq_thread_fn+0x24/0x60
irq_thread+0xeb/0x190
kthread+0x120/0x140
What happens here is that the whole hierarchy is runtime resumed and the
parent PCIe downstream port, which got the hot-remove event, starts
removing devices below it, taking pci_lock_rescan_remove() lock. When the
child PCIe port is runtime resumed it calls pciehp_check_presence() which
ends up calling pciehp_card_present() and pciehp_check_link_active(). Both
of these use pcie_capability_read_word(), which notices that the underlying
device is already gone and returns PCIBIOS_DEVICE_NOT_FOUND with the
capability value set to 0. When pciehp gets this value it thinks that its
child device is also hot-removed and schedules its IRQ thread to handle the
event.
The deadlock happens when the child's IRQ thread runs and tries to acquire
pci_lock_rescan_remove() which is already taken by the parent and the
parent waits for the child's IRQ thread to finish.
Prevent this from happening by checking the return value of
pcie_capability_read_word() and if it is PCIBIOS_DEVICE_NOT_FOUND stop
performing any hot-removal activities.
[bhelgaas: add common scenarios to commit log]
Link: https://lore.kernel.org/r/20191029170022.57528-2-mika.westerberg@linux.intel.com
Tested-by: Kai-Heng Feng <kai.heng.feng@canonical.com>
Signed-off-by: Mika Westerberg <mika.westerberg@linux.intel.com>
Signed-off-by: Bjorn Helgaas <bhelgaas@google.com>
Signed-off-by: Sasha Levin <sashal@kernel.org>
Nicholas Johnson