platform/kernel/linux-starfive.git
4 years agoiommu/vt-d: Remove real DMA lookup in find_domain
Jon Derrick [Wed, 27 May 2020 16:56:17 +0000 (10:56 -0600)]
iommu/vt-d: Remove real DMA lookup in find_domain

By removing the real DMA indirection in find_domain(), we can allow
sub-devices of a real DMA device to have their own valid
device_domain_info. The dmar lookup and context entry removal paths have
been fixed to account for sub-devices.

Fixes: 2b0140c69637 ("iommu/vt-d: Use pci_real_dma_dev() for mapping")
Signed-off-by: Jon Derrick <jonathan.derrick@intel.com>
Acked-by: Lu Baolu <baolu.lu@linux.intel.com>
Link: https://lore.kernel.org/r/20200527165617.297470-4-jonathan.derrick@intel.com
Bugzilla: https://bugzilla.kernel.org/show_bug.cgi?id=207575
Signed-off-by: Joerg Roedel <jroedel@suse.de>
4 years agoiommu/vt-d: Allocate domain info for real DMA sub-devices
Jon Derrick [Wed, 27 May 2020 16:56:16 +0000 (10:56 -0600)]
iommu/vt-d: Allocate domain info for real DMA sub-devices

Sub-devices of a real DMA device might exist on a separate segment than
the real DMA device and its IOMMU. These devices should still have a
valid device_domain_info, but the current dma alias model won't
allocate info for the subdevice.

This patch adds a segment member to struct device_domain_info and uses
the sub-device's BDF so that these sub-devices won't alias to other
devices.

Fixes: 2b0140c69637e ("iommu/vt-d: Use pci_real_dma_dev() for mapping")
Cc: stable@vger.kernel.org # v5.6+
Signed-off-by: Jon Derrick <jonathan.derrick@intel.com>
Acked-by: Lu Baolu <baolu.lu@linux.intel.com>
Link: https://lore.kernel.org/r/20200527165617.297470-3-jonathan.derrick@intel.com
Signed-off-by: Joerg Roedel <jroedel@suse.de>
4 years agoiommu/vt-d: Only clear real DMA device's context entries
Jon Derrick [Wed, 27 May 2020 16:56:15 +0000 (10:56 -0600)]
iommu/vt-d: Only clear real DMA device's context entries

Domain context mapping can encounter issues with sub-devices of a real
DMA device. A sub-device cannot have a valid context entry due to it
potentially aliasing another device's 16-bit ID. It's expected that
sub-devices of the real DMA device uses the real DMA device's requester
when context mapping.

This is an issue when a sub-device is removed where the context entry is
cleared for all aliases. Other sub-devices are still valid, resulting in
those sub-devices being stranded without valid context entries.

The correct approach is to use the real DMA device when programming the
context entries. The insertion path is correct because device_to_iommu()
will return the bus and devfn of the real DMA device. The removal path
needs to only operate on the real DMA device, otherwise the entire
context entry would be cleared for all sub-devices of the real DMA
device.

This patch also adds a helper to determine if a struct device is a
sub-device of a real DMA device.

Fixes: 2b0140c69637e ("iommu/vt-d: Use pci_real_dma_dev() for mapping")
Cc: stable@vger.kernel.org # v5.6+
Signed-off-by: Jon Derrick <jonathan.derrick@intel.com>
Acked-by: Lu Baolu <baolu.lu@linux.intel.com>
Link: https://lore.kernel.org/r/20200527165617.297470-2-jonathan.derrick@intel.com
Signed-off-by: Joerg Roedel <jroedel@suse.de>
4 years agoiommu/vt-d: fix a GCC warning
Qian Cai [Thu, 21 May 2020 21:50:30 +0000 (17:50 -0400)]
iommu/vt-d: fix a GCC warning

The commit 6ee1b77ba3ac ("iommu/vt-d: Add svm/sva invalidate function")
introduced a GCC warning,

drivers/iommu/intel-iommu.c:5330:1: warning: 'static' is not at beginning of
declaration [-Wold-style-declaration]
 const static int
 ^~~~~

Fixes: 6ee1b77ba3ac0 ("iommu/vt-d: Add svm/sva invalidate function")
Signed-off-by: Qian Cai <cai@lca.pw>
Acked-by: Lu Baolu <baolu.lu@linux.intel.com>
Link: https://lore.kernel.org/r/20200521215030.16938-1-cai@lca.pw
Signed-off-by: Joerg Roedel <jroedel@suse.de>
4 years agoiommu/vt-d: Fix pointer cast warnings on 32 bit
Lu Baolu [Tue, 19 May 2020 01:34:23 +0000 (09:34 +0800)]
iommu/vt-d: Fix pointer cast warnings on 32 bit

Pointers should be casted to unsigned long to avoid "cast from pointer
to integer of different size" warnings.

drivers/iommu/intel-pasid.c:818:6: warning:
    cast from pointer to integer of different size [-Wpointer-to-int-cast]
drivers/iommu/intel-pasid.c:821:9: warning:
    cast from pointer to integer of different size [-Wpointer-to-int-cast]
drivers/iommu/intel-pasid.c:824:23: warning:
    cast from pointer to integer of different size [-Wpointer-to-int-cast]
drivers/iommu/intel-svm.c:343:45: warning:
    cast to pointer from integer of different size [-Wint-to-pointer-cast]

Fixes: b0d1f8741b81 ("iommu/vt-d: Add nested translation helper function")
Fixes: 56722a4398a3 ("iommu/vt-d: Add bind guest PASID support")
Signed-off-by: Lu Baolu <baolu.lu@linux.intel.com>
Link: https://lore.kernel.org/r/20200519013423.11971-1-baolu.lu@linux.intel.com
Signed-off-by: Joerg Roedel <jroedel@suse.de>
4 years agoiommu/vt-d: Remove IOVA handling code from the non-dma_ops path
Tom Murphy [Sat, 16 May 2020 06:21:01 +0000 (14:21 +0800)]
iommu/vt-d: Remove IOVA handling code from the non-dma_ops path

There's no need for the non-dma_ops path to keep track of IOVAs. The
whole point of the non-dma_ops path is that it allows the IOVAs to be
handled separately. The IOVA handling code removed in this patch is
pointless.

Signed-off-by: Tom Murphy <murphyt7@tcd.ie>
Signed-off-by: Lu Baolu <baolu.lu@linux.intel.com>
Link: https://lore.kernel.org/r/20200516062101.29541-19-baolu.lu@linux.intel.com
Signed-off-by: Joerg Roedel <jroedel@suse.de>
4 years agoiommu/vt-d: Remove duplicated check in intel_svm_bind_mm()
Lu Baolu [Sat, 16 May 2020 06:21:00 +0000 (14:21 +0800)]
iommu/vt-d: Remove duplicated check in intel_svm_bind_mm()

The info and info->pasid_support have already been checked in previous
intel_iommu_enable_pasid() call. No need to check again.

Signed-off-by: Lu Baolu <baolu.lu@linux.intel.com>
Link: https://lore.kernel.org/r/20200516062101.29541-18-baolu.lu@linux.intel.com
Signed-off-by: Joerg Roedel <jroedel@suse.de>
4 years agoiommu/vt-d: Remove redundant IOTLB flush
Lu Baolu [Sat, 16 May 2020 06:20:59 +0000 (14:20 +0800)]
iommu/vt-d: Remove redundant IOTLB flush

IOTLB flush already included in the PASID tear down and the page request
drain process. There is no need to flush again.

Signed-off-by: Jacob Pan <jacob.jun.pan@linux.intel.com>
Signed-off-by: Lu Baolu <baolu.lu@linux.intel.com>
Reviewed-by: Kevin Tian <kevin.tian@intel.com>
Link: https://lore.kernel.org/r/20200516062101.29541-17-baolu.lu@linux.intel.com
Signed-off-by: Joerg Roedel <jroedel@suse.de>
4 years agoiommu/vt-d: Add page request draining support
Lu Baolu [Sat, 16 May 2020 06:20:58 +0000 (14:20 +0800)]
iommu/vt-d: Add page request draining support

When a PASID is stopped or terminated, there can be pending PRQs
(requests that haven't received responses) in remapping hardware.
This adds the interface to drain page requests and call it when a
PASID is terminated.

Signed-off-by: Jacob Pan <jacob.jun.pan@linux.intel.com>
Signed-off-by: Liu Yi L <yi.l.liu@intel.com>
Signed-off-by: Lu Baolu <baolu.lu@linux.intel.com>
Link: https://lore.kernel.org/r/20200516062101.29541-16-baolu.lu@linux.intel.com
Signed-off-by: Joerg Roedel <jroedel@suse.de>
4 years agoiommu/vt-d: Disable non-recoverable fault processing before unbind
Lu Baolu [Sat, 16 May 2020 06:20:57 +0000 (14:20 +0800)]
iommu/vt-d: Disable non-recoverable fault processing before unbind

When a PASID is used for SVA by the device, it's possible that the PASID
entry is cleared before the device flushes all ongoing DMA requests. The
IOMMU should tolerate and ignore the non-recoverable faults caused by the
untranslated requests from this device.

For example, when an exception happens, the process terminates before the
device driver stops DMA and call IOMMU driver to unbind PASID. The flow
of process exist is as follows:

do_exit() {
     exit_mm() {
             mm_put();
             exit_mmap() {
                     intel_invalidate_range() //mmu notifier
                     tlb_finish_mmu()
                     mmu_notifier_release(mm) {
                             intel_iommu_release() {
[2]                                  intel_iommu_teardown_pasid();
                                     intel_iommu_flush_tlbs();
                             }
                     }
                     unmap_vmas();
                     free_pgtables();
             };
     }
     exit_files(tsk) {
             close_files() {
                     dsa_close();
[1]                  dsa_stop_dma();
                     intel_svm_unbind_pasid();
             }
     }
}

Care must be taken on VT-d to avoid unrecoverable faults between the time
window of [1] and [2]. [Process exist flow was contributed by Jacob Pan.]

Intel VT-d provides such function through the FPD bit of the PASID entry.
This sets FPD bit when PASID entry is changing from present to nonpresent
in the mm notifier and will clear it when the pasid is unbound.

Signed-off-by: Lu Baolu <baolu.lu@linux.intel.com>
Reviewed-by: Jacob Pan <jacob.jun.pan@linux.intel.com>
Link: https://lore.kernel.org/r/20200516062101.29541-15-baolu.lu@linux.intel.com
Signed-off-by: Joerg Roedel <jroedel@suse.de>
4 years agoiommu/vt-d: debugfs: Add support to show inv queue internals
Lu Baolu [Sat, 16 May 2020 06:20:56 +0000 (14:20 +0800)]
iommu/vt-d: debugfs: Add support to show inv queue internals

Export invalidation queue internals of each iommu device through the
debugfs.

Example of such dump on a Skylake machine:

$ sudo cat /sys/kernel/debug/iommu/intel/invalidation_queue
Invalidation queue on IOMMU: dmar1
 Base: 0x1672c9000      Head: 80        Tail: 80
Index           qw0                     qw1                     status
    0   0000000000000004        0000000000000000        0000000000000000
    1   0000000200000025        00000001672be804        0000000000000000
    2   0000000000000011        0000000000000000        0000000000000000
    3   0000000200000025        00000001672be80c        0000000000000000
    4   00000000000000d2        0000000000000000        0000000000000000
    5   0000000200000025        00000001672be814        0000000000000000
    6   0000000000000014        0000000000000000        0000000000000000
    7   0000000200000025        00000001672be81c        0000000000000000
    8   0000000000000014        0000000000000000        0000000000000000
    9   0000000200000025        00000001672be824        0000000000000000

Signed-off-by: Lu Baolu <baolu.lu@linux.intel.com>
Reviewed-by: Kevin Tian <kevin.tian@intel.com>
Link: https://lore.kernel.org/r/20200516062101.29541-14-baolu.lu@linux.intel.com
Signed-off-by: Joerg Roedel <jroedel@suse.de>
4 years agoiommu/vt-d: Multiple descriptors per qi_submit_sync()
Lu Baolu [Sat, 16 May 2020 06:20:55 +0000 (14:20 +0800)]
iommu/vt-d: Multiple descriptors per qi_submit_sync()

Current qi_submit_sync() only supports single invalidation descriptor
per submission and appends wait descriptor after each submission to
poll the hardware completion. This extends the qi_submit_sync() helper
to support multiple descriptors, and add an option so that the caller
could specify the Page-request Drain (PD) bit in the wait descriptor.

Signed-off-by: Jacob Pan <jacob.jun.pan@linux.intel.com>
Signed-off-by: Lu Baolu <baolu.lu@linux.intel.com>
Reviewed-by: Kevin Tian <kevin.tian@intel.com>
Link: https://lore.kernel.org/r/20200516062101.29541-13-baolu.lu@linux.intel.com
Signed-off-by: Joerg Roedel <jroedel@suse.de>
4 years agoiommu/vt-d: Replace intel SVM APIs with generic SVA APIs
Jacob Pan [Sat, 16 May 2020 06:20:54 +0000 (14:20 +0800)]
iommu/vt-d: Replace intel SVM APIs with generic SVA APIs

This patch is an initial step to replace Intel SVM code with the
following IOMMU SVA ops:
intel_svm_bind_mm() => iommu_sva_bind_device()
intel_svm_unbind_mm() => iommu_sva_unbind_device()
intel_svm_is_pasid_valid() => iommu_sva_get_pasid()

The features below will continue to work but are not included in this patch
in that they are handled mostly within the IOMMU subsystem.
- IO page fault
- mmu notifier

Consolidation of the above will come after merging generic IOMMU sva
code[1]. There should not be any changes needed for SVA users such as
accelerator device drivers during this time.

[1] http://jpbrucker.net/sva/

Signed-off-by: Jacob Pan <jacob.jun.pan@linux.intel.com>
Signed-off-by: Lu Baolu <baolu.lu@linux.intel.com>
Link: https://lore.kernel.org/r/20200516062101.29541-12-baolu.lu@linux.intel.com
Signed-off-by: Joerg Roedel <jroedel@suse.de>
4 years agoiommu/vt-d: Report SVA feature with generic flag
Jacob Pan [Sat, 16 May 2020 06:20:53 +0000 (14:20 +0800)]
iommu/vt-d: Report SVA feature with generic flag

Query Shared Virtual Address/Memory capability is a generic feature.
SVA feature check is the required first step before calling
iommu_sva_bind_device().

VT-d checks SVA feature enabling at per IOMMU level during this step,
SVA bind device will check and enable PCI ATS, PRS, and PASID capabilities
at device level.

This patch reports Intel SVM as SVA feature such that generic code
(e.g. Uacce [1]) can use it.

[1] https://lkml.org/lkml/2020/1/15/604

Signed-off-by: Jacob Pan <jacob.jun.pan@linux.intel.com>
Signed-off-by: Lu Baolu <baolu.lu@linux.intel.com>
Link: https://lore.kernel.org/r/20200516062101.29541-11-baolu.lu@linux.intel.com
Signed-off-by: Joerg Roedel <jroedel@suse.de>
4 years agoiommu/vt-d: Add get_domain_info() helper
Lu Baolu [Sat, 16 May 2020 06:20:52 +0000 (14:20 +0800)]
iommu/vt-d: Add get_domain_info() helper

Add a get_domain_info() helper to retrieve the valid per-device
iommu private data.

Signed-off-by: Lu Baolu <baolu.lu@linux.intel.com>
Link: https://lore.kernel.org/r/20200516062101.29541-10-baolu.lu@linux.intel.com
Signed-off-by: Joerg Roedel <jroedel@suse.de>
4 years agoiommu/vt-d: Add custom allocator for IOASID
Jacob Pan [Sat, 16 May 2020 06:20:51 +0000 (14:20 +0800)]
iommu/vt-d: Add custom allocator for IOASID

When VT-d driver runs in the guest, PASID allocation must be
performed via virtual command interface. This patch registers a
custom IOASID allocator which takes precedence over the default
XArray based allocator. The resulting IOASID allocation will always
come from the host. This ensures that PASID namespace is system-
wide.

Virtual command registers are used in the guest only, to prevent
vmexit cost, we cache the capability and store it during initialization.

Signed-off-by: Liu, Yi L <yi.l.liu@intel.com>
Signed-off-by: Jacob Pan <jacob.jun.pan@linux.intel.com>
Signed-off-by: Lu Baolu <baolu.lu@linux.intel.com>
Reviewed-by: Eric Auger <eric.auger@redhat.com>
Link: https://lore.kernel.org/r/20200516062101.29541-9-baolu.lu@linux.intel.com
Signed-off-by: Joerg Roedel <jroedel@suse.de>
4 years agoiommu/vt-d: Enlightened PASID allocation
Lu Baolu [Sat, 16 May 2020 06:20:50 +0000 (14:20 +0800)]
iommu/vt-d: Enlightened PASID allocation

Enabling IOMMU in a guest requires communication with the host
driver for certain aspects. Use of PASID ID to enable Shared Virtual
Addressing (SVA) requires managing PASID's in the host. VT-d 3.0 spec
provides a Virtual Command Register (VCMD) to facilitate this.
Writes to this register in the guest are trapped by vIOMMU which
proxies the call to the host driver.

This virtual command interface consists of a capability register,
a virtual command register, and a virtual response register. Refer
to section 10.4.42, 10.4.43, 10.4.44 for more information.

This patch adds the enlightened PASID allocation/free interfaces
via the virtual command interface.

Signed-off-by: Liu Yi L <yi.l.liu@intel.com>
Signed-off-by: Lu Baolu <baolu.lu@linux.intel.com>
Signed-off-by: Jacob Pan <jacob.jun.pan@linux.intel.com>
Reviewed-by: Eric Auger <eric.auger@redhat.com>
Reviewed-by: Kevin Tian <kevin.tian@intel.com>
Link: https://lore.kernel.org/r/20200516062101.29541-8-baolu.lu@linux.intel.com
Signed-off-by: Joerg Roedel <jroedel@suse.de>
4 years agoiommu/vt-d: Add svm/sva invalidate function
Jacob Pan [Sat, 16 May 2020 06:20:49 +0000 (14:20 +0800)]
iommu/vt-d: Add svm/sva invalidate function

When Shared Virtual Address (SVA) is enabled for a guest OS via
vIOMMU, we need to provide invalidation support at IOMMU API and driver
level. This patch adds Intel VT-d specific function to implement
iommu passdown invalidate API for shared virtual address.

The use case is for supporting caching structure invalidation
of assigned SVM capable devices. Emulated IOMMU exposes queue
invalidation capability and passes down all descriptors from the guest
to the physical IOMMU.

The assumption is that guest to host device ID mapping should be
resolved prior to calling IOMMU driver. Based on the device handle,
host IOMMU driver can replace certain fields before submit to the
invalidation queue.

Signed-off-by: Liu Yi L <yi.l.liu@intel.com>
Signed-off-by: Jacob Pan <jacob.jun.pan@linux.intel.com>
Signed-off-by: Lu Baolu <baolu.lu@linux.intel.com>
Reviewed-by: Eric Auger <eric.auger@redhat.com>
Link: https://lore.kernel.org/r/20200516062101.29541-7-baolu.lu@linux.intel.com
Signed-off-by: Joerg Roedel <jroedel@suse.de>
4 years agoiommu/vt-d: Support flushing more translation cache types
Jacob Pan [Sat, 16 May 2020 06:20:48 +0000 (14:20 +0800)]
iommu/vt-d: Support flushing more translation cache types

When Shared Virtual Memory is exposed to a guest via vIOMMU, scalable
IOTLB invalidation may be passed down from outside IOMMU subsystems.
This patch adds invalidation functions that can be used for additional
translation cache types.

Signed-off-by: Jacob Pan <jacob.jun.pan@linux.intel.com>
Signed-off-by: Lu Baolu <baolu.lu@linux.intel.com>
Reviewed-by: Eric Auger <eric.auger@redhat.com>
Link: https://lore.kernel.org/r/20200516062101.29541-6-baolu.lu@linux.intel.com
Signed-off-by: Joerg Roedel <jroedel@suse.de>
4 years agoiommu/vt-d: Add bind guest PASID support
Jacob Pan [Sat, 16 May 2020 06:20:47 +0000 (14:20 +0800)]
iommu/vt-d: Add bind guest PASID support

When supporting guest SVA with emulated IOMMU, the guest PASID
table is shadowed in VMM. Updates to guest vIOMMU PASID table
will result in PASID cache flush which will be passed down to
the host as bind guest PASID calls.

For the SL page tables, it will be harvested from device's
default domain (request w/o PASID), or aux domain in case of
mediated device.

    .-------------.  .---------------------------.
    |   vIOMMU    |  | Guest process CR3, FL only|
    |             |  '---------------------------'
    .----------------/
    | PASID Entry |--- PASID cache flush -
    '-------------'                       |
    |             |                       V
    |             |                CR3 in GPA
    '-------------'
Guest
------| Shadow |--------------------------|--------
      v        v                          v
Host
    .-------------.  .----------------------.
    |   pIOMMU    |  | Bind FL for GVA-GPA  |
    |             |  '----------------------'
    .----------------/  |
    | PASID Entry |     V (Nested xlate)
    '----------------\.------------------------------.
    |             |   |SL for GPA-HPA, default domain|
    |             |   '------------------------------'
    '-------------'
Where:
 - FL = First level/stage one page tables
 - SL = Second level/stage two page tables

Signed-off-by: Jacob Pan <jacob.jun.pan@linux.intel.com>
Signed-off-by: Liu Yi L <yi.l.liu@intel.com>
Signed-off-by: Lu Baolu <baolu.lu@linux.intel.com>
Link: https://lore.kernel.org/r/20200516062101.29541-5-baolu.lu@linux.intel.com
Signed-off-by: Joerg Roedel <jroedel@suse.de>
4 years agoiommu/vt-d: Add nested translation helper function
Jacob Pan [Sat, 16 May 2020 06:20:46 +0000 (14:20 +0800)]
iommu/vt-d: Add nested translation helper function

Nested translation mode is supported in VT-d 3.0 Spec.CH 3.8.
With PASID granular translation type set to 0x11b, translation
result from the first level(FL) also subject to a second level(SL)
page table translation. This mode is used for SVA virtualization,
where FL performs guest virtual to guest physical translation and
SL performs guest physical to host physical translation.

This patch adds a helper function for setting up nested translation
where second level comes from a domain and first level comes from
a guest PGD.

Signed-off-by: Jacob Pan <jacob.jun.pan@linux.intel.com>
Signed-off-by: Liu Yi L <yi.l.liu@intel.com>
Signed-off-by: Lu Baolu <baolu.lu@linux.intel.com>
Reviewed-by: Eric Auger <eric.auger@redhat.com>
Link: https://lore.kernel.org/r/20200516062101.29541-4-baolu.lu@linux.intel.com
Signed-off-by: Joerg Roedel <jroedel@suse.de>
4 years agoiommu/vt-d: Use a helper function to skip agaw for SL
Jacob Pan [Sat, 16 May 2020 06:20:45 +0000 (14:20 +0800)]
iommu/vt-d: Use a helper function to skip agaw for SL

An Intel iommu domain uses 5-level page table by default. If the iommu
that the domain tries to attach supports less page levels, the top level
page tables should be skipped. Add a helper to do this so that it could
be used in other places.

Signed-off-by: Jacob Pan <jacob.jun.pan@linux.intel.com>
Signed-off-by: Lu Baolu <baolu.lu@linux.intel.com>
Reviewed-by: Eric Auger <eric.auger@redhat.com>
Reviewed-by: Kevin Tian <kevin.tian@intel.com>
Link: https://lore.kernel.org/r/20200516062101.29541-3-baolu.lu@linux.intel.com
Signed-off-by: Joerg Roedel <jroedel@suse.de>
4 years agoiommu/vt-d: Move domain helper to header
Jacob Pan [Sat, 16 May 2020 06:20:44 +0000 (14:20 +0800)]
iommu/vt-d: Move domain helper to header

Move domain helper to header to be used by SVA code.

Signed-off-by: Jacob Pan <jacob.jun.pan@linux.intel.com>
Signed-off-by: Lu Baolu <baolu.lu@linux.intel.com>
Reviewed-by: Eric Auger <eric.auger@redhat.com>
Reviewed-by: Kevin Tian <kevin.tian@intel.com>
Link: https://lore.kernel.org/r/20200516062101.29541-2-baolu.lu@linux.intel.com
Signed-off-by: Joerg Roedel <jroedel@suse.de>
4 years agoiommu: Remove functions that support private domain
Sai Praneeth Prakhya [Wed, 13 May 2020 22:47:21 +0000 (15:47 -0700)]
iommu: Remove functions that support private domain

After moving iommu_group setup to iommu core code [1][2] and removing
private domain support in vt-d [3], there are no users for functions such
as iommu_request_dm_for_dev(), iommu_request_dma_domain_for_dev() and
request_default_domain_for_dev(). So, remove these functions.

[1] commit dce8d6964ebd ("iommu/amd: Convert to probe/release_device()
    call-backs")
[2] commit e5d1841f18b2 ("iommu/vt-d: Convert to probe/release_device()
    call-backs")
[3] commit 327d5b2fee91 ("iommu/vt-d: Allow 32bit devices to uses DMA
    domain")

Signed-off-by: Sai Praneeth Prakhya <sai.praneeth.prakhya@intel.com>
Cc: Joerg Roedel <joro@8bytes.org>
Cc: Lu Baolu <baolu.lu@linux.intel.com>
Link: https://lore.kernel.org/r/20200513224721.20504-1-sai.praneeth.prakhya@intel.com
Signed-off-by: Joerg Roedel <jroedel@suse.de>
4 years agoiommu/vt-d: Unify format of the printed messages
Andy Shevchenko [Thu, 7 May 2020 16:18:02 +0000 (19:18 +0300)]
iommu/vt-d: Unify format of the printed messages

Unify format of the printed messages, i.e. replace printk(LEVEL ... )
with pr_level(...).

Signed-off-by: Andy Shevchenko <andriy.shevchenko@linux.intel.com>
Reviewed-by: Lu Baolu <baolu.lu@linux.intel.com>
Link: https://lore.kernel.org/r/20200507161804.13275-1-andriy.shevchenko@linux.intel.com
Signed-off-by: Joerg Roedel <jroedel@suse.de>
4 years agoiommu/vt-d: Apply per-device dma_ops
Lu Baolu [Wed, 6 May 2020 01:59:47 +0000 (09:59 +0800)]
iommu/vt-d: Apply per-device dma_ops

Current Intel IOMMU driver sets the system level dma_ops. This causes
each dma API to go through the IOMMU driver even the devices are using
identity mapped domains. This sets per-device dma_ops only if a device
is using a DMA domain. Otherwise, use the default system level dma_ops
for direct dma.

Signed-off-by: Lu Baolu <baolu.lu@linux.intel.com>
Tested-by: Daniel Drake <drake@endlessm.com>
Reviewed-by: Jon Derrick <jonathan.derrick@intel.com>
Reviewed-by: Jerry Snitselaar <jsnitsel@redhat.com>
Link: https://lore.kernel.org/r/20200506015947.28662-4-baolu.lu@linux.intel.com
Signed-off-by: Joerg Roedel <jroedel@suse.de>
4 years agoiommu/vt-d: Allow PCI sub-hierarchy to use DMA domain
Lu Baolu [Wed, 6 May 2020 01:59:46 +0000 (09:59 +0800)]
iommu/vt-d: Allow PCI sub-hierarchy to use DMA domain

Before commit fa954e6831789 ("iommu/vt-d: Delegate the dma domain
to upper layer"), Intel IOMMU started off with all devices in the
identity domain, and took them out later if it found they couldn't
access all of memory. This required devices behind a PCI bridge to
use a DMA domain at the beginning because all PCI devices behind
the bridge use the same source-id in their transactions and the
domain couldn't be changed at run-time.

Intel IOMMU driver is now aligned with the default domain framework,
there's no need to keep this requirement anymore.

Signed-off-by: Lu Baolu <baolu.lu@linux.intel.com>
Tested-by: Daniel Drake <drake@endlessm.com>
Reviewed-by: Jon Derrick <jonathan.derrick@intel.com>
Reviewed-by: Jerry Snitselaar <jsnitsel@redhat.com>
Link: https://lore.kernel.org/r/20200506015947.28662-3-baolu.lu@linux.intel.com
Signed-off-by: Joerg Roedel <jroedel@suse.de>
4 years agoiommu/vt-d: Allow 32bit devices to uses DMA domain
Lu Baolu [Wed, 6 May 2020 01:59:45 +0000 (09:59 +0800)]
iommu/vt-d: Allow 32bit devices to uses DMA domain

Currently, if a 32bit device initially uses an identity domain, Intel
IOMMU driver will convert it forcibly to a DMA one if its address
capability is not enough for the whole system memory. The motivation was
to overcome the overhead caused by possible bounced buffer.

Unfortunately, this improvement has led to many problems. For example,
some 32bit devices are required to use an identity domain, forcing them
to use DMA domain will cause the device not to work anymore. On the
other hand, the VMD sub-devices share a domain but each sub-device might
have different address capability. Forcing a VMD sub-device to use DMA
domain blindly will impact the operation of other sub-devices without
any notification. Further more, PCI aliased devices (PCI bridge and all
devices beneath it, VMD devices and various devices quirked with
pci_add_dma_alias()) must use the same domain. Forcing one device to
switch to DMA domain during runtime will cause in-fligh DMAs for other
devices to abort or target to other memory which might cause undefind
system behavior.

With the last private domain usage in iommu_need_mapping() removed, all
private domain helpers are also cleaned in this patch. Otherwise, the
compiler will complain that some functions are defined but not used.

Signed-off-by: Lu Baolu <baolu.lu@linux.intel.com>
Tested-by: Daniel Drake <drake@endlessm.com>
Reviewed-by: Jon Derrick <jonathan.derrick@intel.com>
Reviewed-by: Jerry Snitselaar <jsnitsel@redhat.com>
Cc: Daniel Drake <drake@endlessm.com>
Cc: Derrick Jonathan <jonathan.derrick@intel.com>
Cc: Jerry Snitselaar <jsnitsel@redhat.com>
Link: https://lore.kernel.org/r/20200506015947.28662-2-baolu.lu@linux.intel.com
Signed-off-by: Joerg Roedel <jroedel@suse.de>
4 years agoMerge tag 'v5.7-rc4' into core
Joerg Roedel [Wed, 13 May 2020 10:01:33 +0000 (12:01 +0200)]
Merge tag 'v5.7-rc4' into core

Linux 5.7-rc4

4 years agoiommu/iova: Unify format of the printed messages
Andy Shevchenko [Thu, 7 May 2020 16:18:03 +0000 (19:18 +0300)]
iommu/iova: Unify format of the printed messages

Unify format of the printed messages, i.e. replace printk(LEVEL ... )
with pr_level(...).

Signed-off-by: Andy Shevchenko <andriy.shevchenko@linux.intel.com>
Link: https://lore.kernel.org/r/20200507161804.13275-2-andriy.shevchenko@linux.intel.com
Signed-off-by: Joerg Roedel <jroedel@suse.de>
4 years agoiommu/renesas: Fix unused-function warning
Arnd Bergmann [Fri, 8 May 2020 22:02:16 +0000 (00:02 +0200)]
iommu/renesas: Fix unused-function warning

gcc warns because the only reference to ipmmu_find_group
is inside of an #ifdef:

drivers/iommu/ipmmu-vmsa.c:878:28: error: 'ipmmu_find_group' defined but not used [-Werror=unused-function]

Change the #ifdef to an equivalent IS_ENABLED().

Fixes: 6580c8a78424 ("iommu/renesas: Convert to probe/release_device() call-backs")
Signed-off-by: Arnd Bergmann <arnd@arndb.de>
Reviewed-by: Simon Horman <horms@verge.net.au>
Link: https://lore.kernel.org/r/20200508220224.688985-1-arnd@arndb.de
Signed-off-by: Joerg Roedel <jroedel@suse.de>
4 years agoiommu: Do not probe devices on IOMMU-less busses
Thierry Reding [Mon, 11 May 2020 16:10:00 +0000 (18:10 +0200)]
iommu: Do not probe devices on IOMMU-less busses

The host1x bus implemented on Tegra SoCs is primarily an abstraction to
create logical device from multiple platform devices. Since the devices
in such a setup are typically hierarchical, DMA setup still needs to be
done so that DMA masks can be properly inherited, but we don't actually
want to attach the host1x logical devices to any IOMMU. The platform
devices that make up the logical device are responsible for memory bus
transactions, so it is them that will need to be attached to the IOMMU.

Add a check to __iommu_probe_device() that aborts IOMMU setup early for
busses that don't have the IOMMU operations pointer set since they will
cause a crash otherwise.

Signed-off-by: Thierry Reding <treding@nvidia.com>
Link: https://lore.kernel.org/r/20200511161000.3853342-1-thierry.reding@gmail.com
Signed-off-by: Joerg Roedel <jroedel@suse.de>
4 years agoiommu/amd: Fix variable "iommu" set but not used
Qian Cai [Sat, 9 May 2020 01:56:45 +0000 (21:56 -0400)]
iommu/amd: Fix variable "iommu" set but not used

The commit dce8d6964ebd ("iommu/amd: Convert to probe/release_device()
call-backs") introduced an unused variable,

drivers/iommu/amd_iommu.c: In function 'amd_iommu_uninit_device':
drivers/iommu/amd_iommu.c:422:20: warning: variable 'iommu' set but not
used [-Wunused-but-set-variable]
  struct amd_iommu *iommu;
                    ^~~~~

Signed-off-by: Qian Cai <cai@lca.pw>
Link: https://lore.kernel.org/r/20200509015645.3236-1-cai@lca.pw
Fixes: dce8d6964ebd ("iommu/amd: Convert to probe/release_device() call-backs")
Signed-off-by: Joerg Roedel <jroedel@suse.de>
4 years agoiommu: Unexport iommu_group_get_for_dev()
Joerg Roedel [Wed, 29 Apr 2020 13:37:12 +0000 (15:37 +0200)]
iommu: Unexport iommu_group_get_for_dev()

The function is now only used in IOMMU core code and shouldn't be used
outside of it anyway, so remove the export for it.

Signed-off-by: Joerg Roedel <jroedel@suse.de>
Tested-by: Marek Szyprowski <m.szyprowski@samsung.com>
Acked-by: Marek Szyprowski <m.szyprowski@samsung.com>
Link: https://lore.kernel.org/r/20200429133712.31431-35-joro@8bytes.org
Signed-off-by: Joerg Roedel <jroedel@suse.de>
4 years agoiommu: Move more initialization to __iommu_probe_device()
Joerg Roedel [Wed, 29 Apr 2020 13:37:11 +0000 (15:37 +0200)]
iommu: Move more initialization to __iommu_probe_device()

Move the calls to dev_iommu_get() and try_module_get() into
__iommu_probe_device(), so that the callers don't have to do it on
their own.

Signed-off-by: Joerg Roedel <jroedel@suse.de>
Tested-by: Marek Szyprowski <m.szyprowski@samsung.com>
Acked-by: Marek Szyprowski <m.szyprowski@samsung.com>
Link: https://lore.kernel.org/r/20200429133712.31431-34-joro@8bytes.org
Signed-off-by: Joerg Roedel <jroedel@suse.de>
4 years agoiommu: Remove add_device()/remove_device() code-paths
Joerg Roedel [Wed, 29 Apr 2020 13:37:10 +0000 (15:37 +0200)]
iommu: Remove add_device()/remove_device() code-paths

All drivers are converted to use the probe/release_device()
call-backs, so the add_device/remove_device() pointers are unused and
the code using them can be removed.

Signed-off-by: Joerg Roedel <jroedel@suse.de>
Tested-by: Marek Szyprowski <m.szyprowski@samsung.com>
Acked-by: Marek Szyprowski <m.szyprowski@samsung.com>
Link: https://lore.kernel.org/r/20200429133712.31431-33-joro@8bytes.org
Signed-off-by: Joerg Roedel <jroedel@suse.de>
4 years agoiommu/exynos: Convert to probe/release_device() call-backs
Joerg Roedel [Wed, 29 Apr 2020 13:37:09 +0000 (15:37 +0200)]
iommu/exynos: Convert to probe/release_device() call-backs

Convert the Exynos IOMMU driver to use the probe_device() and
release_device() call-backs of iommu_ops, so that the iommu core code
does the group and sysfs setup.

Signed-off-by: Joerg Roedel <jroedel@suse.de>
Tested-by: Marek Szyprowski <m.szyprowski@samsung.com>
Acked-by: Marek Szyprowski <m.szyprowski@samsung.com>
Link: https://lore.kernel.org/r/20200429133712.31431-32-joro@8bytes.org
Signed-off-by: Joerg Roedel <jroedel@suse.de>
4 years agoiommu/exynos: Use first SYSMMU in controllers list for IOMMU core
Joerg Roedel [Wed, 29 Apr 2020 13:37:08 +0000 (15:37 +0200)]
iommu/exynos: Use first SYSMMU in controllers list for IOMMU core

On Exynos platforms there can be more than one SYSMMU (IOMMU) for one
DMA master device. Since the IOMMU core code expects only one hardware
IOMMU, use the first SYSMMU in the list.

Signed-off-by: Joerg Roedel <jroedel@suse.de>
Tested-by: Marek Szyprowski <m.szyprowski@samsung.com>
Acked-by: Marek Szyprowski <m.szyprowski@samsung.com>
Link: https://lore.kernel.org/r/20200429133712.31431-31-joro@8bytes.org
Signed-off-by: Joerg Roedel <jroedel@suse.de>
4 years agoiommu/omap: Convert to probe/release_device() call-backs
Joerg Roedel [Wed, 29 Apr 2020 13:37:07 +0000 (15:37 +0200)]
iommu/omap: Convert to probe/release_device() call-backs

Convert the OMAP IOMMU driver to use the probe_device() and
release_device() call-backs of iommu_ops, so that the iommu core code
does the group and sysfs setup.

Signed-off-by: Joerg Roedel <jroedel@suse.de>
Link: https://lore.kernel.org/r/20200429133712.31431-30-joro@8bytes.org
Signed-off-by: Joerg Roedel <jroedel@suse.de>
4 years agoiommu/omap: Remove orphan_dev tracking
Joerg Roedel [Wed, 29 Apr 2020 13:37:06 +0000 (15:37 +0200)]
iommu/omap: Remove orphan_dev tracking

Remove the tracking of device which could not be probed because
their IOMMU is not probed yet. Replace it with a call to
bus_iommu_probe() when a new IOMMU is probed.

Signed-off-by: Joerg Roedel <jroedel@suse.de>
Link: https://lore.kernel.org/r/20200429133712.31431-29-joro@8bytes.org
Signed-off-by: Joerg Roedel <jroedel@suse.de>
4 years agoiommu/renesas: Convert to probe/release_device() call-backs
Joerg Roedel [Wed, 29 Apr 2020 13:37:05 +0000 (15:37 +0200)]
iommu/renesas: Convert to probe/release_device() call-backs

Convert the Renesas IOMMU driver to use the probe_device() and
release_device() call-backs of iommu_ops, so that the iommu core code
does the group and sysfs setup.

Signed-off-by: Joerg Roedel <jroedel@suse.de>
Link: https://lore.kernel.org/r/20200429133712.31431-28-joro@8bytes.org
Signed-off-by: Joerg Roedel <jroedel@suse.de>
4 years agoiommu/tegra: Convert to probe/release_device() call-backs
Joerg Roedel [Wed, 29 Apr 2020 13:37:04 +0000 (15:37 +0200)]
iommu/tegra: Convert to probe/release_device() call-backs

Convert the Tegra IOMMU drivers to use the probe_device() and
release_device() call-backs of iommu_ops, so that the iommu core code
does the group and sysfs setup.

Signed-off-by: Joerg Roedel <jroedel@suse.de>
Link: https://lore.kernel.org/r/20200429133712.31431-27-joro@8bytes.org
Signed-off-by: Joerg Roedel <jroedel@suse.de>
4 years agoiommu/rockchip: Convert to probe/release_device() call-backs
Joerg Roedel [Wed, 29 Apr 2020 13:37:03 +0000 (15:37 +0200)]
iommu/rockchip: Convert to probe/release_device() call-backs

Convert the Rockchip IOMMU driver to use the probe_device() and
release_device() call-backs of iommu_ops, so that the iommu core code
does the group and sysfs setup.

Signed-off-by: Joerg Roedel <jroedel@suse.de>
Link: https://lore.kernel.org/r/20200429133712.31431-26-joro@8bytes.org
Signed-off-by: Joerg Roedel <jroedel@suse.de>
4 years agoiommu/qcom: Convert to probe/release_device() call-backs
Joerg Roedel [Wed, 29 Apr 2020 13:37:02 +0000 (15:37 +0200)]
iommu/qcom: Convert to probe/release_device() call-backs

Convert the QCOM IOMMU driver to use the probe_device() and
release_device() call-backs of iommu_ops, so that the iommu core code
does the group and sysfs setup.

Signed-off-by: Joerg Roedel <jroedel@suse.de>
Link: https://lore.kernel.org/r/20200429133712.31431-25-joro@8bytes.org
Signed-off-by: Joerg Roedel <jroedel@suse.de>
4 years agoiommu/mediatek-v1 Convert to probe/release_device() call-backs
Joerg Roedel [Wed, 29 Apr 2020 13:37:01 +0000 (15:37 +0200)]
iommu/mediatek-v1 Convert to probe/release_device() call-backs

Convert the Mediatek-v1 IOMMU driver to use the probe_device() and
release_device() call-backs of iommu_ops, so that the iommu core code
does the group and sysfs setup.

Signed-off-by: Joerg Roedel <jroedel@suse.de>
Link: https://lore.kernel.org/r/20200429133712.31431-24-joro@8bytes.org
Signed-off-by: Joerg Roedel <jroedel@suse.de>
4 years agoiommu/mediatek: Convert to probe/release_device() call-backs
Joerg Roedel [Wed, 29 Apr 2020 13:37:00 +0000 (15:37 +0200)]
iommu/mediatek: Convert to probe/release_device() call-backs

Convert the Mediatek IOMMU driver to use the probe_device() and
release_device() call-backs of iommu_ops, so that the iommu core code
does the group and sysfs setup.

Signed-off-by: Joerg Roedel <jroedel@suse.de>
Link: https://lore.kernel.org/r/20200429133712.31431-23-joro@8bytes.org
Signed-off-by: Joerg Roedel <jroedel@suse.de>
4 years agoiommu/msm: Convert to probe/release_device() call-backs
Joerg Roedel [Wed, 29 Apr 2020 13:36:59 +0000 (15:36 +0200)]
iommu/msm: Convert to probe/release_device() call-backs

Convert the MSM IOMMU driver to use the probe_device() and
release_device() call-backs of iommu_ops, so that the iommu core code
does the group and sysfs setup.

Signed-off-by: Joerg Roedel <jroedel@suse.de>
Link: https://lore.kernel.org/r/20200429133712.31431-22-joro@8bytes.org
Signed-off-by: Joerg Roedel <jroedel@suse.de>
4 years agoiommu/virtio: Convert to probe/release_device() call-backs
Joerg Roedel [Wed, 29 Apr 2020 13:36:58 +0000 (15:36 +0200)]
iommu/virtio: Convert to probe/release_device() call-backs

Convert the VirtIO IOMMU driver to use the probe_device() and
release_device() call-backs of iommu_ops, so that the iommu core code
does the group and sysfs setup.

Signed-off-by: Joerg Roedel <jroedel@suse.de>
Link: https://lore.kernel.org/r/20200429133712.31431-21-joro@8bytes.org
Signed-off-by: Joerg Roedel <jroedel@suse.de>
4 years agoiommu/s390: Convert to probe/release_device() call-backs
Joerg Roedel [Wed, 29 Apr 2020 13:36:57 +0000 (15:36 +0200)]
iommu/s390: Convert to probe/release_device() call-backs

Convert the S390 IOMMU driver to use the probe_device() and
release_device() call-backs of iommu_ops, so that the iommu core code
does the group and sysfs setup.

Signed-off-by: Joerg Roedel <jroedel@suse.de>
Link: https://lore.kernel.org/r/20200429133712.31431-20-joro@8bytes.org
Signed-off-by: Joerg Roedel <jroedel@suse.de>
4 years agoiommu/pamu: Convert to probe/release_device() call-backs
Joerg Roedel [Wed, 29 Apr 2020 13:36:56 +0000 (15:36 +0200)]
iommu/pamu: Convert to probe/release_device() call-backs

Convert the PAMU IOMMU driver to use the probe_device() and
release_device() call-backs of iommu_ops, so that the iommu core code
does the group and sysfs setup.

Signed-off-by: Joerg Roedel <jroedel@suse.de>
Link: https://lore.kernel.org/r/20200429133712.31431-19-joro@8bytes.org
Signed-off-by: Joerg Roedel <jroedel@suse.de>
4 years agoiommu/arm-smmu: Convert to probe/release_device() call-backs
Joerg Roedel [Wed, 29 Apr 2020 13:36:55 +0000 (15:36 +0200)]
iommu/arm-smmu: Convert to probe/release_device() call-backs

Convert the arm-smmu and arm-smmu-v3 drivers to use the probe_device() and
release_device() call-backs of iommu_ops, so that the iommu core code does the
group and sysfs setup.

Signed-off-by: Joerg Roedel <jroedel@suse.de>
Link: https://lore.kernel.org/r/20200429133712.31431-18-joro@8bytes.org
Signed-off-by: Joerg Roedel <jroedel@suse.de>
4 years agoiommu/vt-d: Convert to probe/release_device() call-backs
Joerg Roedel [Wed, 29 Apr 2020 13:36:54 +0000 (15:36 +0200)]
iommu/vt-d: Convert to probe/release_device() call-backs

Convert the Intel IOMMU driver to use the probe_device() and
release_device() call-backs of iommu_ops, so that the iommu core code
does the group and sysfs setup.

Signed-off-by: Joerg Roedel <jroedel@suse.de>
Reviewed-by: Lu Baolu <baolu.lu@linux.intel.com>
Link: https://lore.kernel.org/r/20200429133712.31431-17-joro@8bytes.org
Signed-off-by: Joerg Roedel <jroedel@suse.de>
4 years agoiommu/amd: Convert to probe/release_device() call-backs
Joerg Roedel [Wed, 29 Apr 2020 13:36:53 +0000 (15:36 +0200)]
iommu/amd: Convert to probe/release_device() call-backs

Convert the AMD IOMMU Driver to use the probe_device() and
release_device() call-backs of iommu_ops, so that the iommu core code
does the group and sysfs setup.

Signed-off-by: Joerg Roedel <jroedel@suse.de>
Link: https://lore.kernel.org/r/20200429133712.31431-16-joro@8bytes.org
Signed-off-by: Joerg Roedel <jroedel@suse.de>
4 years agoiommu/amd: Remove dev_data->passthrough
Joerg Roedel [Wed, 29 Apr 2020 13:36:52 +0000 (15:36 +0200)]
iommu/amd: Remove dev_data->passthrough

Make use of generic IOMMU infrastructure to gather the same information
carried in dev_data->passthrough and remove the struct member.

Signed-off-by: Joerg Roedel <jroedel@suse.de>
Link: https://lore.kernel.org/r/20200429133712.31431-15-joro@8bytes.org
Signed-off-by: Joerg Roedel <jroedel@suse.de>
4 years agoiommu: Export bus_iommu_probe() and make is safe for re-probing
Joerg Roedel [Wed, 29 Apr 2020 13:36:51 +0000 (15:36 +0200)]
iommu: Export bus_iommu_probe() and make is safe for re-probing

Add a check to the bus_iommu_probe() call-path to make sure it ignores
devices which have already been successfully probed. Then export the
bus_iommu_probe() function so it can be used by IOMMU drivers.

Signed-off-by: Joerg Roedel <jroedel@suse.de>
Tested-by: Marek Szyprowski <m.szyprowski@samsung.com>
Acked-by: Marek Szyprowski <m.szyprowski@samsung.com>
Link: https://lore.kernel.org/r/20200429133712.31431-14-joro@8bytes.org
Signed-off-by: Joerg Roedel <jroedel@suse.de>
4 years agoiommu: Move iommu_group_create_direct_mappings() out of iommu_group_add_device()
Joerg Roedel [Wed, 29 Apr 2020 13:36:50 +0000 (15:36 +0200)]
iommu: Move iommu_group_create_direct_mappings() out of iommu_group_add_device()

After the previous changes the iommu group may not have a default
domain when iommu_group_add_device() is called. With no default domain
iommu_group_create_direct_mappings() will do nothing and no direct
mappings will be created.

Rename iommu_group_create_direct_mappings() to
iommu_create_device_direct_mappings() to better reflect that the
function creates direct mappings only for one device and not for all
devices in the group. Then move the call to the places where a default
domain actually exists.

Signed-off-by: Joerg Roedel <jroedel@suse.de>
Tested-by: Marek Szyprowski <m.szyprowski@samsung.com>
Acked-by: Marek Szyprowski <m.szyprowski@samsung.com>
Link: https://lore.kernel.org/r/20200429133712.31431-13-joro@8bytes.org
Signed-off-by: Joerg Roedel <jroedel@suse.de>
4 years agoiommu: Split off default domain allocation from group assignment
Joerg Roedel [Wed, 29 Apr 2020 13:36:49 +0000 (15:36 +0200)]
iommu: Split off default domain allocation from group assignment

When a bus is initialized with iommu-ops, all devices on the bus are
scanned and iommu-groups are allocated for them, and each groups will
also get a default domain allocated.

Until now this happened as soon as the group was created and the first
device added to it. When other devices with different default domain
requirements were added to the group later on, the default domain was
re-allocated, if possible.

This resulted in some back and forth and unnecessary allocations, so
change the flow to defer default domain allocation until all devices
have been added to their respective IOMMU groups.

The default domains are allocated for newly allocated groups after
each device on the bus is handled and was probed by the IOMMU driver.

Signed-off-by: Joerg Roedel <jroedel@suse.de>
Tested-by: Marek Szyprowski <m.szyprowski@samsung.com>
Acked-by: Marek Szyprowski <m.szyprowski@samsung.com>
Link: https://lore.kernel.org/r/20200429133712.31431-12-joro@8bytes.org
Signed-off-by: Joerg Roedel <jroedel@suse.de>
4 years agoiommu: Move new probe_device path to separate function
Joerg Roedel [Wed, 29 Apr 2020 13:36:48 +0000 (15:36 +0200)]
iommu: Move new probe_device path to separate function

This makes it easier to remove to old code-path when all drivers are
converted. As a side effect that it also fixes the error cleanup
path.

Signed-off-by: Joerg Roedel <jroedel@suse.de>
Tested-by: Marek Szyprowski <m.szyprowski@samsung.com>
Acked-by: Marek Szyprowski <m.szyprowski@samsung.com>
Link: https://lore.kernel.org/r/20200429133712.31431-11-joro@8bytes.org
Signed-off-by: Joerg Roedel <jroedel@suse.de>
4 years agoiommu: Keep a list of allocated groups in __iommu_probe_device()
Joerg Roedel [Wed, 29 Apr 2020 13:36:47 +0000 (15:36 +0200)]
iommu: Keep a list of allocated groups in __iommu_probe_device()

This is needed to defer default_domain allocation for new IOMMU groups
until all devices have been added to the group.

Signed-off-by: Joerg Roedel <jroedel@suse.de>
Tested-by: Marek Szyprowski <m.szyprowski@samsung.com>
Acked-by: Marek Szyprowski <m.szyprowski@samsung.com>
Link: https://lore.kernel.org/r/20200429133712.31431-10-joro@8bytes.org
Signed-off-by: Joerg Roedel <jroedel@suse.de>
4 years agoiommu: Move default domain allocation to iommu_probe_device()
Joerg Roedel [Wed, 29 Apr 2020 13:36:46 +0000 (15:36 +0200)]
iommu: Move default domain allocation to iommu_probe_device()

Well, not really. The call to iommu_alloc_default_domain() in
iommu_group_get_for_dev() has to stay around as long as there are
IOMMU drivers using the add/remove_device() call-backs instead of
probe/release_device().

Those drivers expect that iommu_group_get_for_dev() returns the device
attached to a group and the group set up with a default domain (and
the device attached to the groups current domain).

But when all drivers are converted this compatability mess can be
removed.

Signed-off-by: Joerg Roedel <jroedel@suse.de>
Tested-by: Marek Szyprowski <m.szyprowski@samsung.com>
Acked-by: Marek Szyprowski <m.szyprowski@samsung.com>
Link: https://lore.kernel.org/r/20200429133712.31431-9-joro@8bytes.org
Signed-off-by: Joerg Roedel <jroedel@suse.de>
4 years agoiommu: Add probe_device() and release_device() call-backs
Joerg Roedel [Wed, 29 Apr 2020 13:36:45 +0000 (15:36 +0200)]
iommu: Add probe_device() and release_device() call-backs

Add call-backs to 'struct iommu_ops' as an alternative to the
add_device() and remove_device() call-backs, which will be removed when
all drivers are converted.

The new call-backs will not setup IOMMU groups and domains anymore,
so also add a probe_finalize() call-back where the IOMMU driver can do
per-device setup work which require the device to be set up with a
group and a domain.

Signed-off-by: Joerg Roedel <jroedel@suse.de>
Tested-by: Marek Szyprowski <m.szyprowski@samsung.com>
Acked-by: Marek Szyprowski <m.szyprowski@samsung.com>
Link: https://lore.kernel.org/r/20200429133712.31431-8-joro@8bytes.org
Signed-off-by: Joerg Roedel <jroedel@suse.de>
4 years agoiommu/amd: Return -ENODEV in add_device when device is not handled by IOMMU
Joerg Roedel [Wed, 29 Apr 2020 13:36:44 +0000 (15:36 +0200)]
iommu/amd: Return -ENODEV in add_device when device is not handled by IOMMU

When check_device() fails on the device, it is not handled by the
IOMMU and amd_iommu_add_device() needs to return -ENODEV.

Signed-off-by: Joerg Roedel <jroedel@suse.de>
Link: https://lore.kernel.org/r/20200429133712.31431-7-joro@8bytes.org
Signed-off-by: Joerg Roedel <jroedel@suse.de>
4 years agoiommu/amd: Remove dma_mask check from check_device()
Joerg Roedel [Wed, 29 Apr 2020 13:36:43 +0000 (15:36 +0200)]
iommu/amd: Remove dma_mask check from check_device()

The check was only needed for the DMA-API implementation in the AMD
IOMMU driver, which no longer exists.

Signed-off-by: Joerg Roedel <jroedel@suse.de>
Link: https://lore.kernel.org/r/20200429133712.31431-6-joro@8bytes.org
Signed-off-by: Joerg Roedel <jroedel@suse.de>
4 years agoiommu/vt-d: Wire up iommu_ops->def_domain_type
Joerg Roedel [Wed, 29 Apr 2020 13:36:42 +0000 (15:36 +0200)]
iommu/vt-d: Wire up iommu_ops->def_domain_type

The Intel VT-d driver already has a matching function to determine the
default domain type for a device. Wire it up in intel_iommu_ops.

Signed-off-by: Joerg Roedel <jroedel@suse.de>
Reviewed-by: Lu Baolu <baolu.lu@linux.intel.com>
Link: https://lore.kernel.org/r/20200429133712.31431-5-joro@8bytes.org
Signed-off-by: Joerg Roedel <jroedel@suse.de>
4 years agoiommu/amd: Implement iommu_ops->def_domain_type call-back
Joerg Roedel [Wed, 29 Apr 2020 13:36:41 +0000 (15:36 +0200)]
iommu/amd: Implement iommu_ops->def_domain_type call-back

Implement the new def_domain_type call-back for the AMD IOMMU driver.

Signed-off-by: Joerg Roedel <jroedel@suse.de>
Link: https://lore.kernel.org/r/20200429133712.31431-4-joro@8bytes.org
Signed-off-by: Joerg Roedel <jroedel@suse.de>
4 years agoiommu: Add def_domain_type() callback in iommu_ops
Sai Praneeth Prakhya [Wed, 29 Apr 2020 13:36:40 +0000 (15:36 +0200)]
iommu: Add def_domain_type() callback in iommu_ops

Some devices are reqired to use a specific type (identity or dma)
of default domain when they are used with a vendor iommu. When the
system level default domain type is different from it, the vendor
iommu driver has to request a new default domain with
iommu_request_dma_domain_for_dev() and iommu_request_dm_for_dev()
in the add_dev() callback. Unfortunately, these two helpers only
work when the group hasn't been assigned to any other devices,
hence, some vendor iommu driver has to use a private domain if
it fails to request a new default one.

This adds def_domain_type() callback in the iommu_ops, so that
any special requirement of default domain for a device could be
aware by the iommu generic layer.

Signed-off-by: Sai Praneeth Prakhya <sai.praneeth.prakhya@intel.com>
Signed-off-by: Lu Baolu <baolu.lu@linux.intel.com>
[ jroedel@suse.de: Added iommu_get_def_domain_type() function and use
                   it to allocate the default domain ]
Co-developed-by: Joerg Roedel <jroedel@suse.de>
Signed-off-by: Joerg Roedel <jroedel@suse.de>
Tested-by: Marek Szyprowski <m.szyprowski@samsung.com>
Acked-by: Marek Szyprowski <m.szyprowski@samsung.com>
Link: https://lore.kernel.org/r/20200429133712.31431-3-joro@8bytes.org
Signed-off-by: Joerg Roedel <jroedel@suse.de>
4 years agoiommu: Move default domain allocation to separate function
Joerg Roedel [Wed, 29 Apr 2020 13:36:39 +0000 (15:36 +0200)]
iommu: Move default domain allocation to separate function

Move the code out of iommu_group_get_for_dev() into a separate
function.

Signed-off-by: Joerg Roedel <jroedel@suse.de>
Tested-by: Marek Szyprowski <m.szyprowski@samsung.com>
Acked-by: Marek Szyprowski <m.szyprowski@samsung.com>
Link: https://lore.kernel.org/r/20200429133712.31431-2-joro@8bytes.org
Signed-off-by: Joerg Roedel <jroedel@suse.de>
4 years agoLinux 5.7-rc4
Linus Torvalds [Sun, 3 May 2020 21:56:04 +0000 (14:56 -0700)]
Linux 5.7-rc4

4 years agoMerge tag 'for-5.7-rc3-tag' of git://git.kernel.org/pub/scm/linux/kernel/git/kdave...
Linus Torvalds [Sun, 3 May 2020 18:30:08 +0000 (11:30 -0700)]
Merge tag 'for-5.7-rc3-tag' of git://git./linux/kernel/git/kdave/linux

Pull more btrfs fixes from David Sterba:
 "A few more stability fixes, minor build warning fixes and git url
  fixup:

   - fix partial loss of prealloc extent past i_size after fsync

   - fix potential deadlock due to wrong transaction handle passing via
     journal_info

   - fix gcc 4.8 struct intialization warning

   - update git URL in MAINTAINERS entry"

* tag 'for-5.7-rc3-tag' of git://git.kernel.org/pub/scm/linux/kernel/git/kdave/linux:
  MAINTAINERS: btrfs: fix git repo URL
  btrfs: fix gcc-4.8 build warning for struct initializer
  btrfs: transaction: Avoid deadlock due to bad initialization timing of fs_info::journal_info
  btrfs: fix partial loss of prealloc extent past i_size after fsync

4 years agoMerge tag 'iommu-fixes-v5.7-rc3' of git://git.kernel.org/pub/scm/linux/kernel/git...
Linus Torvalds [Sun, 3 May 2020 18:04:57 +0000 (11:04 -0700)]
Merge tag 'iommu-fixes-v5.7-rc3' of git://git./linux/kernel/git/joro/iommu

Pull IOMMU fixes from Joerg Roedel:

 - Fix a memory leak when dev_iommu gets freed and a sub-pointer does
   not

 - Build dependency fixes for Mediatek, spapr_tce, and Intel IOMMU
   driver

 - Export iommu_group_get_for_dev() only for GPLed modules

 - Fix AMD IOMMU interrupt remapping when x2apic is enabled

 - Fix error path in the QCOM IOMMU driver probe function

* tag 'iommu-fixes-v5.7-rc3' of git://git.kernel.org/pub/scm/linux/kernel/git/joro/iommu:
  iommu/qcom: Fix local_base status check
  iommu: Properly export iommu_group_get_for_dev()
  iommu/vt-d: Use right Kconfig option name
  iommu/amd: Fix legacy interrupt remapping for x2APIC-enabled system
  iommu: spapr_tce: Disable compile testing to fix build on book3s_32 config
  iommu/mediatek: Fix MTK_IOMMU dependencies
  iommu: Fix the memory leak in dev_iommu_free()

4 years agoMAINTAINERS: btrfs: fix git repo URL
Eric Biggers [Fri, 1 May 2020 23:44:17 +0000 (16:44 -0700)]
MAINTAINERS: btrfs: fix git repo URL

The git repo listed for btrfs hasn't been updated in over a year.
List the current one instead.

Signed-off-by: Eric Biggers <ebiggers@google.com>
Reviewed-by: David Sterba <dsterba@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
4 years agoMerge tag 'pm-5.7-rc4' of git://git.kernel.org/pub/scm/linux/kernel/git/rafael/linux-pm
Linus Torvalds [Sat, 2 May 2020 20:45:30 +0000 (13:45 -0700)]
Merge tag 'pm-5.7-rc4' of git://git./linux/kernel/git/rafael/linux-pm

Pull power management fixes from Rafael Wysocki:

 - prevent the intel_pstate driver from printing excessive diagnostic
   messages in some cases (Chris Wilson)

 - make the hibernation restore kernel freeze kernel threads as well as
   user space tasks (Dexuan Cui)

 - fix the ACPI device PM disagnostic messages to include the correct
   power state name (Kai-Heng Feng).

* tag 'pm-5.7-rc4' of git://git.kernel.org/pub/scm/linux/kernel/git/rafael/linux-pm:
  PM: ACPI: Output correct message on target power state
  PM: hibernate: Freeze kernel threads in software_resume()
  cpufreq: intel_pstate: Only mention the BIOS disabling turbo mode once

4 years agoMerge branches 'pm-cpufreq' and 'pm-sleep'
Rafael J. Wysocki [Sat, 2 May 2020 19:39:17 +0000 (21:39 +0200)]
Merge branches 'pm-cpufreq' and 'pm-sleep'

* pm-cpufreq:
  cpufreq: intel_pstate: Only mention the BIOS disabling turbo mode once

* pm-sleep:
  PM: hibernate: Freeze kernel threads in software_resume()

4 years agoMerge tag 'iomap-5.7-fixes-1' of git://git.kernel.org/pub/scm/fs/xfs/xfs-linux
Linus Torvalds [Sat, 2 May 2020 18:31:12 +0000 (11:31 -0700)]
Merge tag 'iomap-5.7-fixes-1' of git://git./fs/xfs/xfs-linux

Pull iomap fix from Darrick Wong:
 "Hoist the check for an unrepresentable FIBMAP return value into
  ioctl_fibmap.

  The internal kernel function can handle 64-bit values (and is needed
  to fix a regression on ext4 + jbd2). It is only the userspace ioctl
  that is so old that it cannot deal"

* tag 'iomap-5.7-fixes-1' of git://git.kernel.org/pub/scm/fs/xfs/xfs-linux:
  fibmap: Warn and return an error in case of block > INT_MAX

4 years agoMerge tag 'nfs-for-5.7-4' of git://git.linux-nfs.org/projects/trondmy/linux-nfs
Linus Torvalds [Sat, 2 May 2020 18:24:01 +0000 (11:24 -0700)]
Merge tag 'nfs-for-5.7-4' of git://git.linux-nfs.org/projects/trondmy/linux-nfs

Pull NFS client bugfixes from Trond Myklebust:
 "Highlights include:

  Stable fixes:
   - fix handling of backchannel binding in BIND_CONN_TO_SESSION

  Bugfixes:
   - Fix a credential use-after-free issue in pnfs_roc()
   - Fix potential posix_acl refcnt leak in nfs3_set_acl
   - defer slow parts of rpc_free_client() to a workqueue
   - Fix an Oopsable race in __nfs_list_for_each_server()
   - Fix trace point use-after-free race
   - Regression: the RDMA client no longer responds to server disconnect
     requests
   - Fix return values of xdr_stream_encode_item_{present, absent}
   - _pnfs_return_layout() must always wait for layoutreturn completion

  Cleanups:
   - Remove unreachable error conditions"

* tag 'nfs-for-5.7-4' of git://git.linux-nfs.org/projects/trondmy/linux-nfs:
  NFS: Fix a race in __nfs_list_for_each_server()
  NFSv4.1: fix handling of backchannel binding in BIND_CONN_TO_SESSION
  SUNRPC: defer slow parts of rpc_free_client() to a workqueue.
  NFSv4: Remove unreachable error condition due to rpc_run_task()
  SUNRPC: Remove unreachable error condition
  xprtrdma: Fix use of xdr_stream_encode_item_{present, absent}
  xprtrdma: Fix trace point use-after-free race
  xprtrdma: Restore wake-up-all to rpcrdma_cm_event_handler()
  nfs: Fix potential posix_acl refcnt leak in nfs3_set_acl
  NFS/pnfs: Fix a credential use-after-free issue in pnfs_roc()
  NFS/pnfs: Ensure that _pnfs_return_layout() waits for layoutreturn completion

4 years agoMerge tag 'dmaengine-fix-5.7-rc4' of git://git.infradead.org/users/vkoul/slave-dma
Linus Torvalds [Sat, 2 May 2020 18:16:14 +0000 (11:16 -0700)]
Merge tag 'dmaengine-fix-5.7-rc4' of git://git.infradead.org/users/vkoul/slave-dma

Pull dmaengine fixes from Vinod Koul:
 "Core:
   - Documentation typo fixes
   - fix the channel indexes
   - dmatest: fixes for process hang and iterations

  Drivers:
   - hisilicon: build error fix without PCI_MSI
   - ti-k3: deadlock fix
   - uniphier-xdmac: fix for reg region
   - pch: fix data race
   - tegra: fix clock state"

* tag 'dmaengine-fix-5.7-rc4' of git://git.infradead.org/users/vkoul/slave-dma:
  dmaengine: dmatest: Fix process hang when reading 'wait' parameter
  dmaengine: dmatest: Fix iteration non-stop logic
  dmaengine: tegra-apb: Ensure that clock is enabled during of DMA synchronization
  dmaengine: fix channel index enumeration
  dmaengine: mmp_tdma: Reset channel error on release
  dmaengine: mmp_tdma: Do not ignore slave config validation errors
  dmaengine: pch_dma.c: Avoid data race between probe and irq handler
  dt-bindings: dma: uniphier-xdmac: switch to single reg region
  include/linux/dmaengine: Typos fixes in API documentation
  dmaengine: xilinx_dma: Add missing check for empty list
  dmaengine: ti: k3-psil: fix deadlock on error path
  dmaengine: hisilicon: Fix build error without PCI_MSI

4 years agoMerge tag 'vfio-v5.7-rc4' of git://github.com/awilliam/linux-vfio
Linus Torvalds [Sat, 2 May 2020 00:19:15 +0000 (17:19 -0700)]
Merge tag 'vfio-v5.7-rc4' of git://github.com/awilliam/linux-vfio

Pull VFIO fixes from Alex Williamson:

 - copy_*_user validity check for new vfio_dma_rw interface (Yan Zhao)

 - Fix a potential math overflow (Yan Zhao)

 - Use follow_pfn() for calculating PFNMAPs (Sean Christopherson)

* tag 'vfio-v5.7-rc4' of git://github.com/awilliam/linux-vfio:
  vfio/type1: Fix VA->PA translation for PFNMAP VMAs in vaddr_get_pfn()
  vfio: avoid possible overflow in vfio_iommu_type1_pin_pages
  vfio: checking of validity of user vaddr in vfio_dma_rw

4 years agoMerge tag 'arm64-fixes' of git://git.kernel.org/pub/scm/linux/kernel/git/arm64/linux
Linus Torvalds [Sat, 2 May 2020 00:09:31 +0000 (17:09 -0700)]
Merge tag 'arm64-fixes' of git://git./linux/kernel/git/arm64/linux

Pull arm64 fix from Catalin Marinas:
 "Add -fasynchronous-unwind-tables to the vDSO CFLAGS"

* tag 'arm64-fixes' of git://git.kernel.org/pub/scm/linux/kernel/git/arm64/linux:
  arm64: vdso: Add -fasynchronous-unwind-tables to cflags

4 years agoMerge tag 'io_uring-5.7-2020-05-01' of git://git.kernel.dk/linux-block
Linus Torvalds [Sat, 2 May 2020 00:03:06 +0000 (17:03 -0700)]
Merge tag 'io_uring-5.7-2020-05-01' of git://git.kernel.dk/linux-block

Pull io_uring fixes from Jens Axboe:

 - Fix for statx not grabbing the file table, making AT_EMPTY_PATH fail

 - Cover a few cases where async poll can handle retry, eliminating the
   need for an async thread

 - fallback request busy/free fix (Bijan)

 - syzbot reported SQPOLL thread exit fix for non-preempt (Xiaoguang)

 - Fix extra put of req for sync_file_range (Pavel)

 - Always punt splice async. We'll improve this for 5.8, but wanted to
   eliminate the inode mutex lock from the non-blocking path for 5.7
   (Pavel)

* tag 'io_uring-5.7-2020-05-01' of git://git.kernel.dk/linux-block:
  io_uring: punt splice async because of inode mutex
  io_uring: check non-sync defer_list carefully
  io_uring: fix extra put in sync_file_range()
  io_uring: use cond_resched() in io_ring_ctx_wait_and_kill()
  io_uring: use proper references for fallback_req locking
  io_uring: only force async punt if poll based retry can't handle it
  io_uring: enable poll retry for any file with ->read_iter / ->write_iter
  io_uring: statx must grab the file table for valid fd

4 years agoMerge tag 'block-5.7-2020-05-01' of git://git.kernel.dk/linux-block
Linus Torvalds [Fri, 1 May 2020 18:13:36 +0000 (11:13 -0700)]
Merge tag 'block-5.7-2020-05-01' of git://git.kernel.dk/linux-block

Pull block fixes from Jens Axboe:
 "A few fixes for this release:

   - NVMe pull request from Christoph, with a single fix for a double
     free in the namespace error handling.

   - Kill the bd_openers check in blk_drop_partitions(), fixing a
     regression in this merge window (Christoph)"

* tag 'block-5.7-2020-05-01' of git://git.kernel.dk/linux-block:
  block: remove the bd_openers checks in blk_drop_partitions
  nvme: prevent double free in nvme_alloc_ns() error handling

4 years agoMerge branch 'i2c/for-current-fixed' of git://git.kernel.org/pub/scm/linux/kernel...
Linus Torvalds [Fri, 1 May 2020 18:10:09 +0000 (11:10 -0700)]
Merge branch 'i2c/for-current-fixed' of git://git./linux/kernel/git/wsa/linux

Pull i2c fixes from Wolfram Sang:
 "Three driver bugfixes, and two reverts because the original patches
  revealed underlying problems which the Tegra guys are now working on"

* 'i2c/for-current-fixed' of git://git.kernel.org/pub/scm/linux/kernel/git/wsa/linux:
  i2c: aspeed: Avoid i2c interrupt status clear race condition.
  i2c: amd-mp2-pci: Fix Oops in amd_mp2_pci_init() error handling
  Revert "i2c: tegra: Better handle case where CPU0 is busy for a long time"
  Revert "i2c: tegra: Synchronize DMA before termination"
  i2c: iproc: generate stop event for slave writes

4 years agoMerge tag 'sound-5.7-rc4' of git://git.kernel.org/pub/scm/linux/kernel/git/tiwai...
Linus Torvalds [Fri, 1 May 2020 18:05:28 +0000 (11:05 -0700)]
Merge tag 'sound-5.7-rc4' of git://git./linux/kernel/git/tiwai/sound

Pull sound fixes from Takashi Iwai:
 "Just a collection of small fixes around this time:

   - One more try for fixing PCM OSS regression

   - HD-audio: a new quirk for Lenovo, the improved driver blacklisting,
     a lock fix in the minor error path, and a fix for the possible race
     at monitor notifiaction

   - USB-audio: a quirk ID fix, a fix for POD HD500 workaround"

* tag 'sound-5.7-rc4' of git://git.kernel.org/pub/scm/linux/kernel/git/tiwai/sound:
  ALSA: usb-audio: Correct a typo of NuPrime DAC-10 USB ID
  ALSA: opti9xx: shut up gcc-10 range warning
  ALSA: hda/hdmi: fix without unlocked before return
  ALSA: hda/hdmi: fix race in monitor detection during probe
  ALSA: hda/realtek - Two front mics on a Lenovo ThinkCenter
  ALSA: line6: Fix POD HD500 audio playback
  ALSA: pcm: oss: Place the plugin buffer overflow checks correctly (for 5.7)
  ALSA: pcm: oss: Place the plugin buffer overflow checks correctly
  ALSA: hda: Match both PCI ID and SSID for driver blacklist

4 years agoMerge tag 'drm-fixes-2020-05-01' of git://anongit.freedesktop.org/drm/drm
Linus Torvalds [Fri, 1 May 2020 18:01:51 +0000 (11:01 -0700)]
Merge tag 'drm-fixes-2020-05-01' of git://anongit.freedesktop.org/drm/drm

Pull drm fixes from Dave Airlie:
 "Regular scheduled fixes for graphics. Nothing to extreme bunch of
  amdgpu fixes, i915 and qxl fixes, along with some misc ones.

  All seems to be progressing normally.

  core:
   - EDID off by one DTD fix
   - DP mst write return code fix

  dma-buf:
   - fix SET_NAME ioctl uapi
   - doc fixes

  amdgpu:
   - Fix a green screen on resume issue
   - PM fixes for SR-IOV SDMA fix for navi
   - Renoir display fixes
   - Cursor and pageflip stuttering fixes
   - Misc additional display fixes
   - (uapi) Add additional DCC tiling flags for navi1x

  i915:
   - Fix selftest refcnt leak (Xiyu)
   - Fix gem vma lock (Chris)
   - Fix gt's i915_request.timeline acquire by checking if cacheline is
     valid (Chris)
   - Fix IRQ postinistall fault masks (Matt)

  qxl:
   - use after gree fix
   - fix lost kunmap
   - release leak fix

  virtio:
   - context destruction fix"

* tag 'drm-fixes-2020-05-01' of git://anongit.freedesktop.org/drm/drm: (26 commits)
  dma-buf: fix documentation build warnings
  drm/qxl: qxl_release use after free
  drm/qxl: lost qxl_bo_kunmap_atomic_page in qxl_image_init_helper()
  drm/i915: Use proper fault mask in interrupt postinstall too
  drm/amd/display: Use cursor locking to prevent flip delays
  drm/amd/display: Update downspread percent to match spreadsheet for DCN2.1
  drm/amd/display: Defer cursor update around VUPDATE for all ASIC
  drm/amd/display: fix rn soc bb update
  drm/amd/display: check if REFCLK_CNTL register is present
  drm/amdgpu: bump version for invalidate L2 before SDMA IBs
  drm/amdgpu: invalidate L2 before SDMA IBs (v2)
  drm/amdgpu: add tiling flags from Mesa
  drm/amd/powerplay: avoid using pm_en before it is initialized revised
  Revert "drm/amd/powerplay: avoid using pm_en before it is initialized"
  drm/qxl: qxl_release leak in qxl_hw_surface_alloc()
  drm/qxl: qxl_release leak in qxl_draw_dirty_fb()
  drm/virtio: only destroy created contexts
  drm/dp_mst: Fix drm_dp_send_dpcd_write() return code
  drm/i915/gt: Check cacheline is valid before acquiring
  drm/i915/gem: Hold obj->vma.lock over for_each_ggtt_vma()
  ...

4 years agoMerge tag 'scsi-fixes' of git://git.kernel.org/pub/scm/linux/kernel/git/jejb/scsi
Linus Torvalds [Fri, 1 May 2020 18:00:07 +0000 (11:00 -0700)]
Merge tag 'scsi-fixes' of git://git./linux/kernel/git/jejb/scsi

Pull SCSI fixes from James Bottomley:
 "Four minor fixes: three in drivers and one in the core.

  The core one allows an additional state change that fixes a regression
  introduced by an update to the aacraid driver in the previous merge
  window"

* tag 'scsi-fixes' of git://git.kernel.org/pub/scm/linux/kernel/git/jejb/scsi:
  scsi: target/iblock: fix WRITE SAME zeroing
  scsi: qla2xxx: check UNLOADING before posting async work
  scsi: qla2xxx: set UNLOADING before waiting for session deletion
  scsi: core: Allow the state change from SDEV_QUIESCE to SDEV_BLOCK

4 years agoio_uring: punt splice async because of inode mutex
Pavel Begunkov [Fri, 1 May 2020 14:09:38 +0000 (17:09 +0300)]
io_uring: punt splice async because of inode mutex

Nonblocking do_splice() still may wait for some time on an inode mutex.
Let's play safe and always punt it async.

Reported-by: Jens Axboe <axboe@kernel.dk>
Signed-off-by: Pavel Begunkov <asml.silence@gmail.com>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
4 years agoio_uring: check non-sync defer_list carefully
Pavel Begunkov [Fri, 1 May 2020 14:09:37 +0000 (17:09 +0300)]
io_uring: check non-sync defer_list carefully

io_req_defer() do double-checked locking. Use proper helpers for that,
i.e. list_empty_careful().

Signed-off-by: Pavel Begunkov <asml.silence@gmail.com>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
4 years agoio_uring: fix extra put in sync_file_range()
Pavel Begunkov [Fri, 1 May 2020 14:09:36 +0000 (17:09 +0300)]
io_uring: fix extra put in sync_file_range()

[   40.179474] refcount_t: underflow; use-after-free.
[   40.179499] WARNING: CPU: 6 PID: 1848 at lib/refcount.c:28 refcount_warn_saturate+0xae/0xf0
...
[   40.179612] RIP: 0010:refcount_warn_saturate+0xae/0xf0
[   40.179617] Code: 28 44 0a 01 01 e8 d7 01 c2 ff 0f 0b 5d c3 80 3d 15 44 0a 01 00 75 91 48 c7 c7 b8 f5 75 be c6 05 05 44 0a 01 01 e8 b7 01 c2 ff <0f> 0b 5d c3 80 3d f3 43 0a 01 00 0f 85 6d ff ff ff 48 c7 c7 10 f6
[   40.179619] RSP: 0018:ffffb252423ebe18 EFLAGS: 00010286
[   40.179623] RAX: 0000000000000000 RBX: ffff98d65e929400 RCX: 0000000000000000
[   40.179625] RDX: 0000000000000001 RSI: 0000000000000086 RDI: 00000000ffffffff
[   40.179627] RBP: ffffb252423ebe18 R08: 0000000000000001 R09: 000000000000055d
[   40.179629] R10: 0000000000000c8c R11: 0000000000000001 R12: 0000000000000000
[   40.179631] R13: ffff98d68c434400 R14: ffff98d6a9cbaa20 R15: ffff98d6a609ccb8
[   40.179634] FS:  0000000000000000(0000) GS:ffff98d6af580000(0000) knlGS:0000000000000000
[   40.179636] CS:  0010 DS: 0000 ES: 0000 CR0: 0000000080050033
[   40.179638] CR2: 00000000033e3194 CR3: 000000006480a003 CR4: 00000000003606e0
[   40.179641] Call Trace:
[   40.179652]  io_put_req+0x36/0x40
[   40.179657]  io_free_work+0x15/0x20
[   40.179661]  io_worker_handle_work+0x2f5/0x480
[   40.179667]  io_wqe_worker+0x2a9/0x360
[   40.179674]  ? _raw_spin_unlock_irqrestore+0x24/0x40
[   40.179681]  kthread+0x12c/0x170
[   40.179685]  ? io_worker_handle_work+0x480/0x480
[   40.179690]  ? kthread_park+0x90/0x90
[   40.179695]  ret_from_fork+0x35/0x40
[   40.179702] ---[ end trace 85027405f00110aa ]---

Opcode handler must never put submission ref, but that's what
io_sync_file_range_finish() do. use io_steal_work() there.

Signed-off-by: Pavel Begunkov <asml.silence@gmail.com>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
4 years agoiommu/qcom: Fix local_base status check
Tang Bin [Sat, 18 Apr 2020 13:47:03 +0000 (21:47 +0800)]
iommu/qcom: Fix local_base status check

The function qcom_iommu_device_probe() does not perform sufficient
error checking after executing devm_ioremap_resource(), which can
result in crashes if a critical error path is encountered.

Fixes: 0ae349a0f33f ("iommu/qcom: Add qcom_iommu")
Signed-off-by: Tang Bin <tangbin@cmss.chinamobile.com>
Reviewed-by: Bjorn Andersson <bjorn.andersson@linaro.org>
Link: https://lore.kernel.org/r/20200418134703.1760-1-tangbin@cmss.chinamobile.com
Signed-off-by: Joerg Roedel <jroedel@suse.de>
4 years agoiommu: Properly export iommu_group_get_for_dev()
Greg Kroah-Hartman [Thu, 30 Apr 2020 12:01:20 +0000 (14:01 +0200)]
iommu: Properly export iommu_group_get_for_dev()

In commit a7ba5c3d008d ("drivers/iommu: Export core IOMMU API symbols to
permit modular drivers") a bunch of iommu symbols were exported, all
with _GPL markings except iommu_group_get_for_dev().  That export should
also be _GPL like the others.

Fixes: a7ba5c3d008d ("drivers/iommu: Export core IOMMU API symbols to permit modular drivers")
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
Acked-by: Will Deacon <will@kernel.org>
Cc: Joerg Roedel <jroedel@suse.de>
Cc: John Garry <john.garry@huawei.com>
Cc: Will Deacon <will@kernel.org>
Link: https://lore.kernel.org/r/20200430120120.2948448-1-gregkh@linuxfoundation.org
Signed-off-by: Joerg Roedel <jroedel@suse.de>
4 years agoiommu/vt-d: Use right Kconfig option name
Lu Baolu [Fri, 1 May 2020 07:24:27 +0000 (15:24 +0800)]
iommu/vt-d: Use right Kconfig option name

The CONFIG_ prefix should be added in the code.

Fixes: 046182525db61 ("iommu/vt-d: Add Kconfig option to enable/disable scalable mode")
Reported-and-tested-by: Kumar, Sanjay K <sanjay.k.kumar@intel.com>
Signed-off-by: Lu Baolu <baolu.lu@linux.intel.com>
Cc: Ashok Raj <ashok.raj@intel.com>
Link: https://lore.kernel.org/r/20200501072427.14265-1-baolu.lu@linux.intel.com
Signed-off-by: Joerg Roedel <jroedel@suse.de>
4 years agoiommu/amd: Fix legacy interrupt remapping for x2APIC-enabled system
Suravee Suthikulpanit [Wed, 22 Apr 2020 13:30:02 +0000 (08:30 -0500)]
iommu/amd: Fix legacy interrupt remapping for x2APIC-enabled system

Currently, system fails to boot because the legacy interrupt remapping
mode does not enable 128-bit IRTE (GA), which is required for x2APIC
support.

Fix by using AMD_IOMMU_GUEST_IR_LEGACY_GA mode when booting with
kernel option amd_iommu_intr=legacy instead. The initialization
logic will check GASup and automatically fallback to using
AMD_IOMMU_GUEST_IR_LEGACY if GA mode is not supported.

Fixes: 3928aa3f5775 ("iommu/amd: Detect and enable guest vAPIC support")
Signed-off-by: Suravee Suthikulpanit <suravee.suthikulpanit@amd.com>
Link: https://lore.kernel.org/r/1587562202-14183-1-git-send-email-suravee.suthikulpanit@amd.com
Signed-off-by: Joerg Roedel <jroedel@suse.de>
4 years agoio_uring: use cond_resched() in io_ring_ctx_wait_and_kill()
Xiaoguang Wang [Fri, 1 May 2020 00:52:56 +0000 (08:52 +0800)]
io_uring: use cond_resched() in io_ring_ctx_wait_and_kill()

While working on to make io_uring sqpoll mode support syscalls that need
struct files_struct, I got cpu soft lockup in io_ring_ctx_wait_and_kill(),

    while (ctx->sqo_thread && !wq_has_sleeper(&ctx->sqo_wait))
        cpu_relax();

above loop never has an chance to exit, it's because preempt isn't enabled
in the kernel, and the context calling io_ring_ctx_wait_and_kill() and
io_sq_thread() run in the same cpu, if io_sq_thread calls a cond_resched()
yield cpu and another context enters above loop, then io_sq_thread() will
always in runqueue and never exit.

Use cond_resched() can fix this issue.

Reported-by: syzbot+66243bb7126c410cefe6@syzkaller.appspotmail.com
Signed-off-by: Xiaoguang Wang <xiaoguang.wang@linux.alibaba.com>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
4 years agoio_uring: use proper references for fallback_req locking
Bijan Mottahedeh [Thu, 30 Apr 2020 00:47:50 +0000 (17:47 -0700)]
io_uring: use proper references for fallback_req locking

Use ctx->fallback_req address for test_and_set_bit_lock() and
clear_bit_unlock().

Signed-off-by: Bijan Mottahedeh <bijan.mottahedeh@oracle.com>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
4 years agoio_uring: only force async punt if poll based retry can't handle it
Jens Axboe [Tue, 28 Apr 2020 19:16:53 +0000 (13:16 -0600)]
io_uring: only force async punt if poll based retry can't handle it

We do blocking retry from our poll handler, if the file supports polled
notifications. Only mark the request as needing an async worker if we
can't poll for it.

Signed-off-by: Jens Axboe <axboe@kernel.dk>
4 years agoio_uring: enable poll retry for any file with ->read_iter / ->write_iter
Jens Axboe [Tue, 28 Apr 2020 19:15:06 +0000 (13:15 -0600)]
io_uring: enable poll retry for any file with ->read_iter / ->write_iter

We can have files like eventfd where it's perfectly fine to do poll
based retry on them, right now io_file_supports_async() doesn't take
that into account.

Pass in data direction and check the f_op instead of just always needing
an async worker.

Signed-off-by: Jens Axboe <axboe@kernel.dk>
4 years agoMerge tag 'amd-drm-fixes-5.7-2020-04-29' of git://people.freedesktop.org/~agd5f/linux...
Dave Airlie [Fri, 1 May 2020 01:19:54 +0000 (11:19 +1000)]
Merge tag 'amd-drm-fixes-5.7-2020-04-29' of git://people.freedesktop.org/~agd5f/linux into drm-fixes

amd-drm-fixes-5.7-2020-04-29:

amdgpu:
- Fix a green screen on resume issue
- PM fixes for SR-IOV
- SDMA fix for navi
- Renoir display fixes
- Cursor and pageflip stuttering fixes
- Misc additional display fixes

UAPI:
- Add additional DCC tiling flags for navi1x
  Used by: https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/4697

Signed-off-by: Dave Airlie <airlied@redhat.com>
From: Alex Deucher <alexdeucher@gmail.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20200429212008.4306-1-alexander.deucher@amd.com
4 years agoMerge tag 'drm-intel-fixes-2020-04-30' of git://anongit.freedesktop.org/drm/drm-intel...
Dave Airlie [Fri, 1 May 2020 00:56:13 +0000 (10:56 +1000)]
Merge tag 'drm-intel-fixes-2020-04-30' of git://anongit.freedesktop.org/drm/drm-intel into drm-fixes

- Fix selftest refcnt leak (Xiyu)
- Fix gem vma lock (Chris)
- Fix gt's i915_request.timeline acquire by checking if cacheline is valid (Chris)
- Fix IRQ postinistall fault masks (Matt)

Signed-off-by: Dave Airlie <airlied@redhat.com>
From: Rodrigo Vivi <rodrigo.vivi@intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20200430140042.GA270140@intel.com
4 years agoMerge tag 'drm-misc-fixes-2020-04-30' of git://anongit.freedesktop.org/drm/drm-misc...
Dave Airlie [Fri, 1 May 2020 00:41:59 +0000 (10:41 +1000)]
Merge tag 'drm-misc-fixes-2020-04-30' of git://anongit.freedesktop.org/drm/drm-misc into drm-fixes

A few resources-related fixes for qxl, some doc build warnings and ioctl
fixes for dma-buf, an off-by-one fix in edid, and a return code fix in
DP-MST

Signed-off-by: Dave Airlie <airlied@redhat.com>
From: Maxime Ripard <maxime@cerno.tech>
Link: https://patchwork.freedesktop.org/patch/msgid/20200430153201.wx6of2b2gsoip7bk@gilmour.lan
4 years agoMerge tag 'for-5.7/dm-fixes-2' of git://git.kernel.org/pub/scm/linux/kernel/git/devic...
Linus Torvalds [Thu, 30 Apr 2020 23:45:08 +0000 (16:45 -0700)]
Merge tag 'for-5.7/dm-fixes-2' of git://git./linux/kernel/git/device-mapper/linux-dm

Pull device mapper fixes from Mike Snitzer:

 - Document DM integrity allow_discard feature that was added during 5.7
   merge window.

 - Fix potential for DM writecache data corruption during DM table
   reloads.

 - Fix DM verity's FEC support's hash block number calculation in
   verity_fec_decode().

 - Fix bio-based DM multipath crash due to use of stale copy of
   MPATHF_QUEUE_IO flag state in __map_bio().

* tag 'for-5.7/dm-fixes-2' of git://git.kernel.org/pub/scm/linux/kernel/git/device-mapper/linux-dm:
  dm multipath: use updated MPATHF_QUEUE_IO on mapping for bio-based mpath
  dm verity fec: fix hash block number in verity_fec_decode
  dm writecache: fix data corruption when reloading the target
  dm integrity: document allow_discard option

4 years agoMerge tag 'selinux-pr-20200430' of git://git.kernel.org/pub/scm/linux/kernel/git...
Linus Torvalds [Thu, 30 Apr 2020 23:35:45 +0000 (16:35 -0700)]
Merge tag 'selinux-pr-20200430' of git://git./linux/kernel/git/pcmoore/selinux

Pull SELinux fixes from Paul Moore:
 "Two more SELinux patches to fix problems in the v5.7-rcX releases.

  Wei Yongjun's patch fixes a return code in an error path, and my patch
  fixes a problem where we were not correctly applying access controls
  to all of the netlink messages in the netlink_send LSM hook"

* tag 'selinux-pr-20200430' of git://git.kernel.org/pub/scm/linux/kernel/git/pcmoore/selinux:
  selinux: properly handle multiple messages in selinux_netlink_send()
  selinux: fix error return code in cond_read_list()