* pci/resource:
PCI: Work around poweroff & suspend-to-RAM issue on Macbook Pro 11
PCI: Do not disregard parent resources starting at 0x0
Conflicts:
arch/x86/pci/fixup.c
* pci/portdrv:
PCI/portdrv: Allocate MSI/MSI-X vector for Downstream Port Containment
PCI/portdrv: Support multiple interrupts for MSI as well as MSI-X
* pci/pm:
PCI/PM: Avoid using device_may_wakeup() for runtime PM
x86/PCI: Avoid AMD SB7xx EHCI USB wakeup defect
PCI/PM: Restore the status of PCI devices across hibernation
drm/radeon: make MacBook Pro d3_delay quirk more generic
drm/amdgpu: remove unnecessary save/restore of pdev->d3_delay
PCI/PM: Add needs_resume flag to avoid suspend complete optimization
PCI: imx6: Fix config read timeout handling
switchtec: Fix minor bug with partition ID register
switchtec: Use new cdev_device_add() helper function
PCI: endpoint: Make PCI_ENDPOINT depend on HAS_DMA
* pci/enumeration:
PCI: Enable ECRC only if device supports it
PCI: Add sysfs max_link_speed/width, current_link_speed/width, etc
PCI: Test INTx masking during enumeration, not at run-time
pci_target_state() calls device_may_wakeup() which checks whether or not
the device may wake up the system from sleep states, but pci_target_state()
is used for runtime PM too.
Since runtime PM is expected to always enable remote wakeup if possible,
modify pci_target_state() to take additional argument indicating whether or
not it should look for a state from which the device can signal wakeup and
pass either the return value of device_can_wakeup(), or "false" (if the
device itself is not wakeup-capable) to it from the code related to runtime
PM.
While at it, fix the comment in pci_dev_run_wake() which is not about sleep
states.
Signed-off-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
Signed-off-by: Bjorn Helgaas <bhelgaas@google.com>
Reviewed-by: Mika Westerberg <mika.westerberg@linux.intel.com>
On an AMD Carrizo laptop, when EHCI runtime PM is enabled, EHCI ports do
not assert PME# for device plug/unplug events while in D3.
As Alan Stern points out [1], the PME signal is not enabled when controller
is in D3, therefore it's not being woken up when new devices get plugged
in.
Testing shows PME signal works when the EHCI power state is D2.
Clear the PCI_PM_CAP_PME_D3 and PCI_PM_CAP_PME_D3cold bits in
dev->pme_support to indicate the device will not assert PME# from those
states.
[1] http://lkml.kernel.org/r/Pine.LNX.4.44L0.1706121010010.2092-100000@iolanthe.rowland.org
Link: https://bugzilla.kernel.org/show_bug.cgi?id=196091
Link: https://support.amd.com/TechDocs/46837.pdf (Section 23)
Link: https://support.amd.com/TechDocs/42413.pdf (Appendix A2)
Signed-off-by: Kai-Heng Feng <kai.heng.feng@canonical.com>
[bhelgaas: changelog, add parens in quirk]
Signed-off-by: Bjorn Helgaas <bhelgaas@google.com>
Currently we saw a lot of "No irq handler" errors during hibernation, which
caused the system hang finally:
ata4.00: qc timeout (cmd 0xec)
ata4.00: failed to IDENTIFY (I/O error, err_mask=0x4)
ata4.00: revalidation failed (errno=-5)
ata4: SATA link up 6.0 Gbps (SStatus 133 SControl 300)
do_IRQ: 31.151 No irq handler for vector
According to above logs, there is an interrupt triggered and it is
dispatched to CPU31 with a vector number 151, but there is no handler for
it, thus this IRQ will not get acked and will cause an IRQ flood which
kills the system. To be more specific, the 31.151 is an interrupt from the
AHCI host controller.
After some investigation, the reason why this issue is triggered is because
the thaw_noirq() function does not restore the MSI/MSI-X settings across
hibernation.
The scenario is illustrated below:
1. Before hibernation, IRQ 34 is the handler for the AHCI device, which
is bound to CPU31.
2. Hibernation starts, the AHCI device is put into low power state.
3. All the nonboot CPUs are put offline, so IRQ 34 has to be migrated to
the last alive one - CPU0.
4. After the snapshot has been created, all the nonboot CPUs are brought
up again; IRQ 34 remains bound to CPU0.
5. AHCI devices are put into D0.
6. The snapshot is written to the disk.
The issue is triggered in step 6. The AHCI interrupt should be delivered
to CPU0, however it is delivered to the original CPU31 instead, which
causes the "No irq handler" issue.
Ying Huang has provided a clue that, in step 3 it is possible that writing
to the register might not take effect as the PCI devices have been
suspended.
In step 3, the IRQ 34 affinity should be modified from CPU31 to CPU0, but
in fact it is not. In __pci_write_msi_msg(), if the device is already in
low power state, the low level MSI message entry will not be updated but
cached. During the device restore process after a normal suspend/resume,
pci_restore_msi_state() writes the cached MSI back to the hardware.
But this is not the case for hibernation. pci_restore_msi_state() is not
currently called in pci_pm_thaw_noirq(), although pci_save_state() has
saved the necessary PCI cached information in pci_pm_freeze_noirq().
Restore the PCI status for the device during hibernation. Otherwise the
status might be lost across hibernation (for example, settings for MSI,
MSI-X, ATS, ACS, IOV, etc.), which might cause problems during hibernation.
Suggested-by: Ying Huang <ying.huang@intel.com>
Suggested-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
Signed-off-by: Chen Yu <yu.c.chen@intel.com>
[bhelgaas: changelog]
Signed-off-by: Bjorn Helgaas <bhelgaas@google.com>
Reviewed-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
Cc: stable@vger.kernel.org
Cc: Len Brown <len.brown@intel.com>
Cc: Dan Williams <dan.j.williams@intel.com>
Cc: Rui Zhang <rui.zhang@intel.com>
Cc: Ying Huang <ying.huang@intel.com>
The PCI Power Management Spec, r1.2, sec 5.6.1, requires a 10 millisecond
delay when powering on a device, i.e., transitioning from state D3hot to
D0.
Apparently some devices require more time, and d1f9809ed1 ("drm/radeon:
add quirk for d3 delay during switcheroo poweron for apple macbooks") added
an additional delay for the Radeon device in a MacBook Pro. 4807c5a8a0
("drm/radeon: add a PX quirk list") made the affected device more explicit.
Add a generic PCI quirk to increase the d3_delay. This means we will use
the additional delay for *all* wakeups from D3, not just those initiated by
radeon_switcheroo_set_state().
Signed-off-by: Bjorn Helgaas <bhelgaas@google.com>
Reviewed-by: Andreas Boll <andreas.boll.dev@gmail.com>
Acked-by: Alex Deucher <alexander.deucher@amd.com>
CC: Maarten Lankhorst <maarten.lankhorst@canonical.com>
Remove unnecessary save/restore of pdev->d3_delay.
The only assignments to pdev->d3_delay are in radeon_switcheroo_set_state()
and some quirks, none of which should be relevant in the
amdgpu_switcheroo_set_state() path.
Signed-off-by: Bjorn Helgaas <bhelgaas@google.com>
Reviewed-by: Andreas Boll <andreas.boll.dev@gmail.com>
Acked-by: Alex Deucher <alexander.deucher@amd.com>
Neither soft poweroff (transition to ACPI power state S5) nor
suspend-to-RAM (transition to state S3) works on the Macbook Pro 11,4 and
11,5.
The problem is related to the [mem 0x7fa00000-0x7fbfffff] space. When we
use that space, e.g., by assigning it to the 00:1c.0 Root Port, the ACPI
Power Management 1 Control Register (PM1_CNT) at [io 0x1804] doesn't work
anymore.
Linux does a soft poweroff (transition to S5) by writing to PM1_CNT. The
theory about why this doesn't work is:
- The write to PM1_CNT causes an SMI
- The BIOS SMI handler depends on something in
[mem 0x7fa00000-0x7fbfffff]
- When Linux assigns [mem 0x7fa00000-0x7fbfffff] to the 00:1c.0 Port, it
covers up whatever the SMI handler uses, so the SMI handler no longer
works correctly
Reserve the [mem 0x7fa00000-0x7fbfffff] space so we don't assign it to
anything.
This is voodoo programming, since we don't know what the real conflict is,
but we've failed to find the root cause.
Bugzilla: https://bugzilla.kernel.org/show_bug.cgi?id=103211
Tested-by: thejoe@gmail.com
Signed-off-by: Bjorn Helgaas <bhelgaas@google.com>
Cc: stable@vger.kernel.org
Cc: Rafael J. Wysocki <rafael@kernel.org>
Cc: Lukas Wunner <lukas@wunner.de>
Cc: Chen Yu <yu.c.chen@intel.com>
John reported that an Intel QuickAssist crypto accelerator didn't work in a
Dell PowerEdge R730. The problem seems to be that we enabled ECRC when the
device doesn't support it:
85:00.0 Co-processor [0b40]: Intel Corporation DH895XCC Series QAT [8086:0435]
Capabilities: [100 v1] Advanced Error Reporting
AERCap: First Error Pointer: 00, GenCap- CGenEn+ ChkCap- ChkEn+
1302fcf0d0 ("PCI: Configure *all* devices, not just hot-added ones")
exposed the problem because it applies settings from the _HPX method to all
devices, not just hot-added ones. The R730 supplies an _HPX method that
allows the kernel to enable ECRC.
Only enable ECRC if the device advertises support for it.
Link: https://bugs.launchpad.net/ubuntu/+source/linux/+bug/1571798
Fixes: 1302fcf0d0 ("PCI: Configure *all* devices, not just hot-added ones")
Reported-by: John Mazzie <john_mazzie@dell.com>
Signed-off-by: Bjorn Helgaas <bhelgaas@google.com>
Expose PCIe bridges attributes such as secondary bus number, subordinate
bus number, max link speed and link width, current link speed and link
width via sysfs in /sys/bus/pci/devices/...
This information is available via lspci, but that requires root privilege.
Signed-off-by: Wong Vee Khee <vee.khee.wong@ni.com>
Signed-off-by: Hui Chun Ong <hui.chun.ong@ni.com>
[bhelgaas: changelog, return errors early to unindent usual case, return
errors with same style throughout]
Signed-off-by: Bjorn Helgaas <bhelgaas@google.com>
Currently pcie_port_enable_irq_vec() only allocates MSI/MSI-X vectors for
PME, hotplug, and AER.
The Downstream Port Containment feature also supports MSI/MSI-X interrupts,
so allocate a vector for it, too.
Signed-off-by: Liudongdong <liudongdong3@huawei.com>
Signed-off-by: Gabriele Paoloni <gabriele.paoloni@huawei.com>
[bhelgaas: changelog, comment]
Signed-off-by: Bjorn Helgaas <bhelgaas@google.com>
Reviewed-by: Christoph Hellwig <hch@lst.de>
Root Ports can generate several different interrupts using either MSI or
MSI-X, but we only support that for MSI-X. Ports that support MSI but not
MSI-X are currently limited to sharing a single interrupt.
Rename pcie_port_enable_msix() to pcie_port_enable_irq_vec() and extend it
to support multiple interrupts using either MSI-X (preferred) or MSI.
Signed-off-by: Gabriele Paoloni <gabriele.paoloni@huawei.com>
[bhelgaas: changelog, reword comments, simplify PME/hotplug no-MSI logic]
Signed-off-by: Bjorn Helgaas <bhelgaas@google.com>
Reviewed-by: Christoph Hellwig <hch@lst.de>
The test for INTx masking via PCI_COMMAND_INTX_DISABLE performed in
pci_intx_mask_supported() should be done before the device can be used.
This is to avoid writing PCI_COMMAND while the driver owns the device, in
case that has any effect on MSI/MSI-X interrupts.
Move the content of pci_intx_mask_supported() to pci_intx_mask_broken() and
call it from pci_setup_device().
The test result can be queried at any time later using the same
pci_intx_mask_supported() interface as before (though with changed
implementation), so callers (uio, vfio) should be unaffected.
Signed-off-by: Piotr Gregor <piotrgregor@rsyncme.org>
[bhelgaas: changelog, remove quirk check, remove locking, move
dev->broken_intx_masking assignment to caller]
Signed-off-by: Bjorn Helgaas <bhelgaas@google.com>
Reviewed-by: Alex Williamson <alex.williamson@redhat.com>
Acked-by: Michael S. Tsirkin <mst@redhat.com>
No need for such convoluted code, when all we need is to call one function
in one specific case.
Tested-by: Narendra K <Narendra.K@dell.com> # DellEMC PowerEdge 1950, R730XD
Signed-off-by: Jean Delvare <jdelvare@suse.de>
Signed-off-by: Bjorn Helgaas <bhelgaas@google.com>
The function find_smbios_instance_string() does not consider the
PCI domain number. As a result, SMBIOS type 41 device type instance
would be exported to sysfs for all the PCI domains which have a
PCI device with same bus/device/function, though PCI bus/device/func
from a specific PCI domain has SMBIOS type 41 device type instance
defined.
Address the issue by making find_smbios_instance_string() check PCI domain
number as well.
Reported-by: Shai Fultheim <Shai@ScaleMP.com>
Suggested-by: Shai Fultheim <Shai@ScaleMP.com>
Tested-by: Shai Fultheim <Shai@ScaleMP.com>
Signed-off-by: Sujith Pandel <sujithpshankar@gmail.com>
Signed-off-by: Narendra K <Narendra_K@Dell.com>
Signed-off-by: Bjorn Helgaas <bhelgaas@google.com>
Remove the space from "PCI :" to make the message consistent with other PCI
messages.
Signed-off-by: Vincent Legoll <vincent.legoll@gmail.com>
Signed-off-by: Bjorn Helgaas <bhelgaas@google.com>
PCI_STD_RESOURCE_END is (confusingly) the index of the last valid BAR, not
the *number* of BARs. To iterate through all possible BARs, we need to
include PCI_STD_RESOURCE_END.
Fixes: 9fe373f999 ("PCI: Increase IBM ipr SAS Crocodile BARs to at least system page size")
Signed-off-by: Bjorn Helgaas <bhelgaas@google.com>
PCI_STD_RESOURCE_END is (confusingly) the index of the last valid BAR, not
the *number* of BARs. To iterate through all possible BARs, we need to
include PCI_STD_RESOURCE_END.
Fixes: 55d728a40d ("efi/fb: Avoid reconfiguration of BAR that covers the framebuffer")
Signed-off-by: Bjorn Helgaas <bhelgaas@google.com>
Some drivers - like i915 - may not support the system suspend direct
complete optimization due to differences in their runtime and system
suspend sequence. Add a flag that when set resumes the device before
calling the driver's system suspend handlers which effectively disables
the optimization.
Needed by a future patch fixing suspend/resume on i915.
Suggested by Rafael.
Signed-off-by: Imre Deak <imre.deak@intel.com>
Signed-off-by: Bjorn Helgaas <bhelgaas@google.com>
Acked-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
Cc: stable@vger.kernel.org
This driver was OR'ing desired bits from the existing control setting.
That could create an invalid DPC Trigger Enabled configuration if the
platform previously set this to "ERR_FATAL", 01b. The driver currently
wants to set this to ERR_NONFATAL/ERR_FATAL, 10b, and the logical OR of
this gets 11b, which is reserved. Fix that by masking off the fields it is
setting.
Signed-off-by: Keith Busch <keith.busch@intel.com>
Signed-off-by: Bjorn Helgaas <bhelgaas@google.com>
The DPC interupt may be executed on a device that is being removed. Skip
queuing event handling if the status is all 1's, which should be seen only
if the device is not present.
Signed-off-by: Keith Busch <keith.busch@intel.com>
Signed-off-by: Bjorn Helgaas <bhelgaas@google.com>
Commit cc7b0d4955 ("PCI: designware: Update PCI config space remap
function") made PCI configuration requests non-posted, which means we now
get a synchronous abort when the CFG space read to probe for downstream
devices times out.
Synchronous aborts need to be handled differently from the async aborts we
were getting before, in particular the PC needs to be advanced when
resolving the abort. This is mostly a copy of what other PCI drivers do on
ARM to handle those aborts.
[bhelgaas: changelog, "Fixes"]
Fixes: cc7b0d4955 ("PCI: designware: Update PCI config space remap function")
Tested-by: Fabio Estevam <fabio.estevam@nxp.com>
Tested-by: Peter Senna Tschudin <peter.senna@collabora.com>
Signed-off-by: Lucas Stach <l.stach@pengutronix.de>
Signed-off-by: Bjorn Helgaas <bhelgaas@google.com>
Acked-by: Richard Zhu <hongxing.zhu@nxp.com>
When a switch endpoint is configured without NTB, the mmio_ntb registers
will read all zeros. However, in corner case configurations where the
partition ID is not zero and NTB is not enabled, the code will have the
wrong partition ID and this causes the driver to use the wrong set of
drivers. To fix this we simply take the partition ID from the system info
region.
Reported-by: Dingbao Chen <dingbao.chen@microsemi.com>
Signed-off-by: Logan Gunthorpe <logang@deltatee.com>
Signed-off-by: Bjorn Helgaas <bhelgaas@google.com>
Convert from "cdev_add() + device_add()" to cdev_device_add(), and from
"device_del() + cdev_del()" to cdev_device_del().
[bhelgaas: changelog]
Signed-off-by: Logan Gunthorpe <logang@deltatee.com>
Signed-off-by: Bjorn Helgaas <bhelgaas@google.com>
If NO_DMA=y:
drivers/built-in.o: In function `__pci_epc_create':
(.text+0xef4e): undefined reference to `bad_dma_ops'
drivers/built-in.o: In function `pci_epc_add_epf':
(.text+0xf676): undefined reference to `bad_dma_ops'
drivers/built-in.o: In function `pci_epf_alloc_space':
(.text+0xfa32): undefined reference to `bad_dma_ops'
drivers/built-in.o: In function `pci_epf_free_space':
(.text+0xfac4): undefined reference to `bad_dma_ops'
Add a dependency on HAS_DMA to fix this.
Signed-off-by: Geert Uytterhoeven <geert@linux-m68k.org>
Signed-off-by: Bjorn Helgaas <bhelgaas@google.com>
min_vecs is the minimum amount of vectors needed to operate in MSI-X mode
which may just include the vectors that don't need affinity.
Disabling affinity settings causes the qla2xxx driver scsi_add_host() to fail
when blk_mq is enabled as the blk_mq_pci_map_queues() expects affinity masks
on each vector.
Fixes: dfef358bd1 ("PCI/MSI: Don't apply affinity if there aren't enough vectors left")
Signed-off-by: Michael Hernandez <michael.hernandez@cavium.com>
Signed-off-by: Himanshu Madhani <himanshu.madhani@cavium.com>
Signed-off-by: Bjorn Helgaas <bhelgaas@google.com>
Reviewed-by: Christoph Hellwig <hch@lst.de>
Cc: stable@vger.kernel.org # v4.10+
pci_add_resource_offset() is for host bridge windows where the bridge
translates CPU addresses to PCI bus addresses by adding an offset. To my
knowledge, no host bridge translates bus numbers, so this is only useful
for MEM and IO windows. In any event, host->busn_offset is never set to
anything other than zero, so pci_add_resource() is sufficient.
a2e50f53d5 ("MIPS: PCI: Add a hook for IORESOURCE_BUS in
pci_controller/bridge_controller") also added busn_resource itself. This
is currently unused but may be used by future SGI IP27 fixes, so I left it
there.
Tested-by: Joshua Kinard <kumba@gentoo.org> # SGI IP30 and IP27
Signed-off-by: Bjorn Helgaas <bhelgaas@google.com>
Acked-by: Joshua Kinard <kumba@gentoo.org>
Commit f44116ae88 ("PCI: Remove pci_find_parent_resource() use for
allocation") updated the logic that iterates over all bus resources and
compares them to a given resource, in order to decide whether one is the
parent of the latter.
This change inadvertently causes pci_find_parent_resource() to disregard
resources starting at address 0x0, resulting in an error such as the one
below on ARM systems whose I/O window starts at 0x0.
pci_bus 0000:00: root bus resource [mem 0x10000000-0x3efeffff window]
pci_bus 0000:00: root bus resource [io 0x0000-0xffff window]
pci_bus 0000:00: root bus resource [mem 0x8000000000-0xffffffffff window]
pci_bus 0000:00: root bus resource [bus 00-0f]
pci 0000:00:01.0: PCI bridge to [bus 01]
pci 0000:00:02.0: PCI bridge to [bus 02]
pci 0000:00:03.0: PCI bridge to [bus 03]
pci 0000:00:03.0: can't claim BAR 13 [io 0x0000-0x0fff]: no compatible bridge window
pci 0000:03:01.0: can't claim BAR 0 [io 0x0000-0x001f]: no compatible bridge window
While this never happens on x86, it is perfectly legal in general for a PCI
MMIO or IO window to start at address 0x0, and it was supported in the code
before commit f44116ae88.
Drop the test for res->start != 0; resource_contains() already checks
whether [start, end) completely covers the resource, and so it should be
redundant.
Signed-off-by: Ard Biesheuvel <ard.biesheuvel@linaro.org>
Signed-off-by: Bjorn Helgaas <bhelgaas@google.com>
Pull some more input subsystem updates from Dmitry Torokhov:
"An updated xpad driver with a few more recognized device IDs, and a
new psxpad-spi driver, allowing connecting Playstation 1 and 2 joypads
via SPI bus"
* 'for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/dtor/input:
Input: cros_ec_keyb - remove extraneous 'const'
Input: add support for PlayStation 1/2 joypads connected via SPI
Input: xpad - add USB IDs for Mad Catz Brawlstick and Razer Sabertooth
Input: xpad - sync supported devices with xboxdrv
Input: xpad - sort supported devices by USB ID
Pull UML fixes from Richard Weinberger:
"No new stuff, just fixes"
* 'for-linus-4.12-rc1' of git://git.kernel.org/pub/scm/linux/kernel/git/rw/uml:
um: Add missing NR_CPUS include
um: Fix to call read_initrd after init_bootmem
um: Include kbuild.h instead of duplicating its macros
um: Fix PTRACE_POKEUSER on x86_64
um: Set number of CPUs
um: Fix _print_addr()
Merge misc fixes from Andrew Morton:
"15 fixes"
* emailed patches from Andrew Morton <akpm@linux-foundation.org>:
mm, docs: update memory.stat description with workingset* entries
mm: vmscan: scan until it finds eligible pages
mm, thp: copying user pages must schedule on collapse
dax: fix PMD data corruption when fault races with write
dax: fix data corruption when fault races with write
ext4: return to starting transaction in ext4_dax_huge_fault()
mm: fix data corruption due to stale mmap reads
dax: prevent invalidation of mapped DAX entries
Tigran has moved
mm, vmalloc: fix vmalloc users tracking properly
mm/khugepaged: add missed tracepoint for collapse_huge_page_swapin
gcov: support GCC 7.1
mm, vmstat: Remove spurious WARN() during zoneinfo print
time: delete current_fs_time()
hwpoison, memcg: forcibly uncharge LRU pages
Commit 4b4cea91691d ("mm: vmscan: fix IO/refault regression in cache
workingset transition") introduced three new entries in memory stat
file:
- workingset_refault
- workingset_activate
- workingset_nodereclaim
This commit adds a corresponding description to the cgroup v2 docs.
Link: http://lkml.kernel.org/r/1494530293-31236-1-git-send-email-guro@fb.com
Signed-off-by: Roman Gushchin <guro@fb.com>
Cc: Johannes Weiner <hannes@cmpxchg.org>
Cc: Michal Hocko <mhocko@kernel.org>
Cc: Vladimir Davydov <vdavydov.dev@gmail.com>
Cc: Tejun Heo <tj@kernel.org>
Cc: Li Zefan <lizefan@huawei.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
We have encountered need_resched warnings in __collapse_huge_page_copy()
while doing {clear,copy}_user_highpage() over HPAGE_PMD_NR source pages.
mm->mmap_sem is held for write, but the iteration is well bounded.
Reschedule as needed.
Link: http://lkml.kernel.org/r/alpine.DEB.2.10.1705101426380.109808@chino.kir.corp.google.com
Signed-off-by: David Rientjes <rientjes@google.com>
Acked-by: Vlastimil Babka <vbabka@suse.cz>
Cc: "Kirill A. Shutemov" <kirill.shutemov@linux.intel.com>
Cc: Johannes Weiner <hannes@cmpxchg.org>
Cc: Mel Gorman <mgorman@techsingularity.net>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
This is based on a patch from Jan Kara that fixed the equivalent race in
the DAX PTE fault path.
Currently DAX PMD read fault can race with write(2) in the following
way:
CPU1 - write(2) CPU2 - read fault
dax_iomap_pmd_fault()
->iomap_begin() - sees hole
dax_iomap_rw()
iomap_apply()
->iomap_begin - allocates blocks
dax_iomap_actor()
invalidate_inode_pages2_range()
- there's nothing to invalidate
grab_mapping_entry()
- we add huge zero page to the radix tree
and map it to page tables
The result is that hole page is mapped into page tables (and thus zeros
are seen in mmap) while file has data written in that place.
Fix the problem by locking exception entry before mapping blocks for the
fault. That way we are sure invalidate_inode_pages2_range() call for
racing write will either block on entry lock waiting for the fault to
finish (and unmap stale page tables after that) or read fault will see
already allocated blocks by write(2).
Fixes: 9f141d6ef6 ("dax: Call ->iomap_begin without entry lock during dax fault")
Link: http://lkml.kernel.org/r/20170510172700.18991-1-ross.zwisler@linux.intel.com
Signed-off-by: Ross Zwisler <ross.zwisler@linux.intel.com>
Reviewed-by: Jan Kara <jack@suse.cz>
Cc: Dan Williams <dan.j.williams@intel.com>
Cc: <stable@vger.kernel.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
Currently DAX read fault can race with write(2) in the following way:
CPU1 - write(2) CPU2 - read fault
dax_iomap_pte_fault()
->iomap_begin() - sees hole
dax_iomap_rw()
iomap_apply()
->iomap_begin - allocates blocks
dax_iomap_actor()
invalidate_inode_pages2_range()
- there's nothing to invalidate
grab_mapping_entry()
- we add zero page in the radix tree
and map it to page tables
The result is that hole page is mapped into page tables (and thus zeros
are seen in mmap) while file has data written in that place.
Fix the problem by locking exception entry before mapping blocks for the
fault. That way we are sure invalidate_inode_pages2_range() call for
racing write will either block on entry lock waiting for the fault to
finish (and unmap stale page tables after that) or read fault will see
already allocated blocks by write(2).
Fixes: 9f141d6ef6
Link: http://lkml.kernel.org/r/20170510085419.27601-5-jack@suse.cz
Signed-off-by: Jan Kara <jack@suse.cz>
Reviewed-by: Ross Zwisler <ross.zwisler@linux.intel.com>
Cc: Dan Williams <dan.j.williams@intel.com>
Cc: <stable@vger.kernel.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
DAX will return to locking exceptional entry before mapping blocks for a
page fault to fix possible races with concurrent writes. To avoid lock
inversion between exceptional entry lock and transaction start, start
the transaction already in ext4_dax_huge_fault().
Fixes: 9f141d6ef6
Link: http://lkml.kernel.org/r/20170510085419.27601-4-jack@suse.cz
Signed-off-by: Jan Kara <jack@suse.cz>
Cc: Ross Zwisler <ross.zwisler@linux.intel.com>
Cc: Dan Williams <dan.j.williams@intel.com>
Cc: <stable@vger.kernel.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
Currently, we didn't invalidate page tables during invalidate_inode_pages2()
for DAX. That could result in e.g. 2MiB zero page being mapped into
page tables while there were already underlying blocks allocated and
thus data seen through mmap were different from data seen by read(2).
The following sequence reproduces the problem:
- open an mmap over a 2MiB hole
- read from a 2MiB hole, faulting in a 2MiB zero page
- write to the hole with write(3p). The write succeeds but we
incorrectly leave the 2MiB zero page mapping intact.
- via the mmap, read the data that was just written. Since the zero
page mapping is still intact we read back zeroes instead of the new
data.
Fix the problem by unconditionally calling invalidate_inode_pages2_range()
in dax_iomap_actor() for new block allocations and by properly
invalidating page tables in invalidate_inode_pages2_range() for DAX
mappings.
Fixes: c6dcf52c23
Link: http://lkml.kernel.org/r/20170510085419.27601-3-jack@suse.cz
Signed-off-by: Jan Kara <jack@suse.cz>
Signed-off-by: Ross Zwisler <ross.zwisler@linux.intel.com>
Cc: Dan Williams <dan.j.williams@intel.com>
Cc: <stable@vger.kernel.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
Patch series "mm,dax: Fix data corruption due to mmap inconsistency",
v4.
This series fixes data corruption that can happen for DAX mounts when
page faults race with write(2) and as a result page tables get out of
sync with block mappings in the filesystem and thus data seen through
mmap is different from data seen through read(2).
The series passes testing with t_mmap_stale test program from Ross and
also other mmap related tests on DAX filesystem.
This patch (of 4):
dax_invalidate_mapping_entry() currently removes DAX exceptional entries
only if they are clean and unlocked. This is done via:
invalidate_mapping_pages()
invalidate_exceptional_entry()
dax_invalidate_mapping_entry()
However, for page cache pages removed in invalidate_mapping_pages()
there is an additional criteria which is that the page must not be
mapped. This is noted in the comments above invalidate_mapping_pages()
and is checked in invalidate_inode_page().
For DAX entries this means that we can can end up in a situation where a
DAX exceptional entry, either a huge zero page or a regular DAX entry,
could end up mapped but without an associated radix tree entry. This is
inconsistent with the rest of the DAX code and with what happens in the
page cache case.
We aren't able to unmap the DAX exceptional entry because according to
its comments invalidate_mapping_pages() isn't allowed to block, and
unmap_mapping_range() takes a write lock on the mapping->i_mmap_rwsem.
Since we essentially never have unmapped DAX entries to evict from the
radix tree, just remove dax_invalidate_mapping_entry().
Fixes: c6dcf52c23 ("mm: Invalidate DAX radix tree entries only if appropriate")
Link: http://lkml.kernel.org/r/20170510085419.27601-2-jack@suse.cz
Signed-off-by: Ross Zwisler <ross.zwisler@linux.intel.com>
Signed-off-by: Jan Kara <jack@suse.cz>
Reported-by: Jan Kara <jack@suse.cz>
Cc: Dan Williams <dan.j.williams@intel.com>
Cc: <stable@vger.kernel.org> [4.10+]
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
Commit 1f5307b1e0 ("mm, vmalloc: properly track vmalloc users") has
pulled asm/pgtable.h include dependency to linux/vmalloc.h and that
turned out to be a bad idea for some architectures. E.g. m68k fails
with
In file included from arch/m68k/include/asm/pgtable_mm.h:145:0,
from arch/m68k/include/asm/pgtable.h:4,
from include/linux/vmalloc.h:9,
from arch/m68k/kernel/module.c:9:
arch/m68k/include/asm/mcf_pgtable.h: In function 'nocache_page':
>> arch/m68k/include/asm/mcf_pgtable.h:339:43: error: 'init_mm' undeclared (first use in this function)
#define pgd_offset_k(address) pgd_offset(&init_mm, address)
as spotted by kernel build bot. nios2 fails for other reason
In file included from include/asm-generic/io.h:767:0,
from arch/nios2/include/asm/io.h:61,
from include/linux/io.h:25,
from arch/nios2/include/asm/pgtable.h:18,
from include/linux/mm.h:70,
from include/linux/pid_namespace.h:6,
from include/linux/ptrace.h:9,
from arch/nios2/include/uapi/asm/elf.h:23,
from arch/nios2/include/asm/elf.h:22,
from include/linux/elf.h:4,
from include/linux/module.h:15,
from init/main.c:16:
include/linux/vmalloc.h: In function '__vmalloc_node_flags':
include/linux/vmalloc.h:99:40: error: 'PAGE_KERNEL' undeclared (first use in this function); did you mean 'GFP_KERNEL'?
which is due to the newly added #include <asm/pgtable.h>, which on nios2
includes <linux/io.h> and thus <asm/io.h> and <asm-generic/io.h> which
again includes <linux/vmalloc.h>.
Tweaking that around just turns out a bigger headache than necessary.
This patch reverts 1f5307b1e0 and reimplements the original fix in a
different way. __vmalloc_node_flags can stay static inline which will
cover vmalloc* functions. We only have one external user
(kvmalloc_node) and we can export __vmalloc_node_flags_caller and
provide the caller directly. This is much simpler and it doesn't really
need any games with header files.
[akpm@linux-foundation.org: coding-style fixes]
[mhocko@kernel.org: revert old comment]
Link: http://lkml.kernel.org/r/20170509211054.GB16325@dhcp22.suse.cz
Fixes: 1f5307b1e0 ("mm, vmalloc: properly track vmalloc users")
Link: http://lkml.kernel.org/r/20170509153702.GR6481@dhcp22.suse.cz
Signed-off-by: Michal Hocko <mhocko@suse.com>
Cc: Tobias Klauser <tklauser@distanz.ch>
Cc: Geert Uytterhoeven <geert@linux-m68k.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
One return case of `__collapse_huge_page_swapin()` does not invoke
tracepoint while every other return case does. This commit adds a
tracepoint invocation for the case.
Link: http://lkml.kernel.org/r/20170507101813.30187-1-sj38.park@gmail.com
Signed-off-by: SeongJae Park <sj38.park@gmail.com>
Cc: Kirill A. Shutemov <kirill.shutemov@linux.intel.com>
Cc: Johannes Weiner <hannes@cmpxchg.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
Bjorn Helgaas