Pull virtio updates from Michael Tsirkin:
- Some bug fixes
- The new vdpa subsystem with two first drivers
* tag 'for_linus' of git://git.kernel.org/pub/scm/linux/kernel/git/mst/vhost:
virtio-balloon: Revert "virtio-balloon: Switch back to OOM handler for VIRTIO_BALLOON_F_DEFLATE_ON_OOM"
vdpa: move to drivers/vdpa
virtio: Intel IFC VF driver for VDPA
vdpasim: vDPA device simulator
vhost: introduce vDPA-based backend
virtio: introduce a vDPA based transport
vDPA: introduce vDPA bus
vringh: IOTLB support
vhost: factor out IOTLB
vhost: allow per device message handler
vhost: refine vhost and vringh kconfig
virtio-balloon: Switch back to OOM handler for VIRTIO_BALLOON_F_DEFLATE_ON_OOM
virtio-net: Introduce hash report feature
virtio-net: Introduce RSS receive steering feature
virtio-net: Introduce extended RSC feature
tools/virtio: option to build an out of tree module
Commit 71994620bb ("virtio_balloon: replace oom notifier with shrinker")
changed the behavior when deflation happens automatically. Instead of
deflating when called by the OOM handler, the shrinker is used.
However, the balloon is not simply some other slab cache that should be
shrunk when under memory pressure. The shrinker does not have a concept
of priorities yet, so this behavior cannot be configured. Eventually once
that is in place, we might want to switch back after doing proper testing.
There was a report that this results in undesired side effects when
inflating the balloon to shrink the page cache. [1]
"When inflating the balloon against page cache (i.e. no free memory
remains) vmscan.c will both shrink page cache, but also invoke the
shrinkers -- including the balloon's shrinker. So the balloon
driver allocates memory which requires reclaim, vmscan gets this
memory by shrinking the balloon, and then the driver adds the
memory back to the balloon. Basically a busy no-op."
The name "deflate on OOM" makes it pretty clear when deflation should
happen - after other approaches to reclaim memory failed, not while
reclaiming. This allows to minimize the footprint of a guest - memory
will only be taken out of the balloon when really needed.
Keep using the shrinker for VIRTIO_BALLOON_F_FREE_PAGE_HINT, because
this has no such side effects. Always register the shrinker with
VIRTIO_BALLOON_F_FREE_PAGE_HINT now. We are always allowed to reuse free
pages that are still to be processed by the guest. The hypervisor takes
care of identifying and resolving possible races between processing a
hinting request and the guest reusing a page.
In contrast to pre commit 71994620bb ("virtio_balloon: replace oom
notifier with shrinker"), don't add a module parameter to configure the
number of pages to deflate on OOM. Can be re-added if really needed.
Also, pay attention that leak_balloon() returns the number of 4k pages -
convert it properly in virtio_balloon_oom_notify().
Testing done by Tyler for future reference:
Test setup: VM with 16 CPU, 64GB RAM. Running Debian 10. We have a 42
GB file full of random bytes that we continually cat to /dev/null.
This fills the page cache as the file is read. Meanwhile, we trigger
the balloon to inflate, with a target size of 53 GB. This setup causes
the balloon inflation to pressure the page cache as the page cache is
also trying to grow. Afterwards we shrink the balloon back to zero (so
total deflate == total inflate).
Without this patch (kernel 4.19.0-5):
Inflation never reaches the target until we stop the "cat file >
/dev/null" process. Total inflation time was 542 seconds. The longest
period that made no net forward progress was 315 seconds.
Result of "grep balloon /proc/vmstat" after the test:
balloon_inflate 154828377
balloon_deflate 154828377
With this patch (kernel 5.6.0-rc4+):
Total inflation duration was 63 seconds. No deflate-queue activity
occurs when pressuring the page-cache.
Result of "grep balloon /proc/vmstat" after the test:
balloon_inflate 12968539
balloon_deflate 12968539
Conclusion: This patch fixes the issue. In the test it reduced
inflate/deflate activity by 12x, and reduced inflation time by 8.6x.
But more importantly, if we hadn't killed the "cat file > /dev/null"
process then, without the patch, the inflation process would never reach
the target.
[1] https://www.spinics.net/lists/linux-virtualization/msg40863.html
Link: http://lkml.kernel.org/r/20200311135523.18512-2-david@redhat.com
Fixes: 71994620bb ("virtio_balloon: replace oom notifier with shrinker")
Signed-off-by: David Hildenbrand <david@redhat.com>
Reported-by: Tyler Sanderson <tysand@google.com>
Tested-by: Tyler Sanderson <tysand@google.com>
Acked-by: David Rientjes <rientjes@google.com>
Acked-by: Michael S. Tsirkin <mst@redhat.com>
Cc: Wei Wang <wei.w.wang@intel.com>
Cc: Alexander Duyck <alexander.h.duyck@linux.intel.com>
Cc: David Rientjes <rientjes@google.com>
Cc: Nadav Amit <namit@vmware.com>
Cc: Michal Hocko <mhocko@kernel.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
This reverts commit 5a6b4cc5b7.
It has been queued properly in the akpm tree, this version is just
creating conflicts.
Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
We have both vhost and virtio drivers that depend on vdpa.
It's easier to locate it at a top level directory otherwise
we run into issues e.g. if vhost is built-in but virtio
is modular. Let's just move it up a level.
Reported-by: Randy Dunlap <rdunlap@infradead.org>
Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
This patch implements a software vDPA networking device. The datapath
is implemented through vringh and workqueue. The device has an on-chip
IOMMU which translates IOVA to PA. For kernel virtio drivers, vDPA
simulator driver provides dma_ops. For vhost driers, set_map() methods
of vdpa_config_ops is implemented to accept mappings from vhost.
Currently, vDPA device simulator will loopback TX traffic to RX. So
the main use case for the device is vDPA feature testing, prototyping
and development.
Note, there's no management API implemented, a vDPA device will be
registered once the module is probed. We need to handle this in the
future development.
Signed-off-by: Jason Wang <jasowang@redhat.com>
Link: https://lore.kernel.org/r/20200326140125.19794-9-jasowang@redhat.com
Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
This patch introduces a vDPA transport for virtio. This is used to
use kernel virtio driver to drive the vDPA device that is capable
of populating virtqueue directly.
A new virtio-vdpa driver will be registered to the vDPA bus, when a
new virtio-vdpa device is probed, it will register the device with
vdpa based config ops. This means it is a software transport between
vDPA driver and vDPA device. The transport was implemented through
bus_ops of vDPA parent.
Signed-off-by: Jason Wang <jasowang@redhat.com>
Link: https://lore.kernel.org/r/20200326140125.19794-7-jasowang@redhat.com
Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
vDPA device is a device that uses a datapath which complies with the
virtio specifications with vendor specific control path. vDPA devices
can be both physically located on the hardware or emulated by
software. vDPA hardware devices are usually implemented through PCIE
with the following types:
- PF (Physical Function) - A single Physical Function
- VF (Virtual Function) - Device that supports single root I/O
virtualization (SR-IOV). Its Virtual Function (VF) represents a
virtualized instance of the device that can be assigned to different
partitions
- ADI (Assignable Device Interface) and its equivalents - With
technologies such as Intel Scalable IOV, a virtual device (VDEV)
composed by host OS utilizing one or more ADIs. Or its equivalent
like SF (Sub function) from Mellanox.
>From a driver's perspective, depends on how and where the DMA
translation is done, vDPA devices are split into two types:
- Platform specific DMA translation - From the driver's perspective,
the device can be used on a platform where device access to data in
memory is limited and/or translated. An example is a PCIE vDPA whose
DMA request was tagged via a bus (e.g PCIE) specific way. DMA
translation and protection are done at PCIE bus IOMMU level.
- Device specific DMA translation - The device implements DMA
isolation and protection through its own logic. An example is a vDPA
device which uses on-chip IOMMU.
To hide the differences and complexity of the above types for a vDPA
device/IOMMU options and in order to present a generic virtio device
to the upper layer, a device agnostic framework is required.
This patch introduces a software vDPA bus which abstracts the
common attributes of vDPA device, vDPA bus driver and the
communication method (vdpa_config_ops) between the vDPA device
abstraction and the vDPA bus driver. This allows multiple types of
drivers to be used for vDPA device like the virtio_vdpa and vhost_vdpa
driver to operate on the bus and allow vDPA device could be used by
either kernel virtio driver or userspace vhost drivers as:
virtio drivers vhost drivers
| |
[virtio bus] [vhost uAPI]
| |
virtio device vhost device
virtio_vdpa drv vhost_vdpa drv
\ /
[vDPA bus]
|
vDPA device
hardware drv
|
[hardware bus]
|
vDPA hardware
With the abstraction of vDPA bus and vDPA bus operations, the
difference and complexity of the under layer hardware is hidden from
upper layer. The vDPA bus drivers on top can use a unified
vdpa_config_ops to control different types of vDPA device.
Signed-off-by: Jason Wang <jasowang@redhat.com>
Link: https://lore.kernel.org/r/20200326140125.19794-6-jasowang@redhat.com
Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
Commit 71994620bb ("virtio_balloon: replace oom notifier with shrinker")
changed the behavior when deflation happens automatically. Instead of
deflating when called by the OOM handler, the shrinker is used.
However, the balloon is not simply some slab cache that should be
shrunk when under memory pressure. The shrinker does not have a concept of
priorities, so this behavior cannot be configured.
There was a report that this results in undesired side effects when
inflating the balloon to shrink the page cache. [1]
"When inflating the balloon against page cache (i.e. no free memory
remains) vmscan.c will both shrink page cache, but also invoke the
shrinkers -- including the balloon's shrinker. So the balloon
driver allocates memory which requires reclaim, vmscan gets this
memory by shrinking the balloon, and then the driver adds the
memory back to the balloon. Basically a busy no-op."
The name "deflate on OOM" makes it pretty clear when deflation should
happen - after other approaches to reclaim memory failed, not while
reclaiming. This allows to minimize the footprint of a guest - memory
will only be taken out of the balloon when really needed.
Especially, a drop_slab() will result in the whole balloon getting
deflated - undesired. While handling it via the OOM handler might not be
perfect, it keeps existing behavior. If we want a different behavior, then
we need a new feature bit and document it properly (although, there should
be a clear use case and the intended effects should be well described).
Keep using the shrinker for VIRTIO_BALLOON_F_FREE_PAGE_HINT, because
this has no such side effects. Always register the shrinker with
VIRTIO_BALLOON_F_FREE_PAGE_HINT now. We are always allowed to reuse free
pages that are still to be processed by the guest. The hypervisor takes
care of identifying and resolving possible races between processing a
hinting request and the guest reusing a page.
In contrast to pre commit 71994620bb ("virtio_balloon: replace oom
notifier with shrinker"), don't add a moodule parameter to configure the
number of pages to deflate on OOM. Can be re-added if really needed.
Also, pay attention that leak_balloon() returns the number of 4k pages -
convert it properly in virtio_balloon_oom_notify().
Note1: using the OOM handler is frowned upon, but it really is what we
need for this feature.
Note2: without VIRTIO_BALLOON_F_MUST_TELL_HOST (iow, always with QEMU) we
could actually skip sending deflation requests to our hypervisor,
making the OOM path *very* simple. Besically freeing pages and
updating the balloon. If the communication with the host ever
becomes a problem on this call path.
[1] https://www.spinics.net/lists/linux-virtualization/msg40863.html
Test report by Tyler Sanderson:
Test setup: VM with 16 CPU, 64GB RAM. Running Debian 10. We have a 42
GB file full of random bytes that we continually cat to /dev/null.
This fills the page cache as the file is read. Meanwhile we trigger
the balloon to inflate, with a target size of 53 GB. This setup causes
the balloon inflation to pressure the page cache as the page cache is
also trying to grow. Afterwards we shrink the balloon back to zero (so
total deflate = total inflate).
Without patch (kernel 4.19.0-5):
Inflation never reaches the target until we stop the "cat file >
/dev/null" process. Total inflation time was 542 seconds. The longest
period that made no net forward progress was 315 seconds (see attached
graph).
Result of "grep balloon /proc/vmstat" after the test:
balloon_inflate 154828377
balloon_deflate 154828377
With patch (kernel 5.6.0-rc4+):
Total inflation duration was 63 seconds. No deflate-queue activity
occurs when pressuring the page-cache.
Result of "grep balloon /proc/vmstat" after the test:
balloon_inflate 12968539
balloon_deflate 12968539
Conclusion: This patch fixes the issue. In the test it reduced
inflate/deflate activity by 12x, and reduced inflation time by 8.6x.
But more importantly, if we hadn't killed the "grep balloon
/proc/vmstat" process then, without the patch, the inflation process
would never reach the target.
Attached [1] is a png of a graph showing the problematic behavior without
this patch. It shows deflate-queue activity increasing linearly while
balloon size stays constant over the course of more than 8 minutes of
the test.
[1] https://lore.kernel.org/linux-mm/CAJuQAmphPcfew1v_EOgAdSFiprzjiZjmOf3iJDmFX0gD6b9TYQ@mail.gmail.com/2-without_patch.png
Full test report and discussion [2]:
[2] https://lore.kernel.org/r/CAJuQAmphPcfew1v_EOgAdSFiprzjiZjmOf3iJDmFX0gD6b9TYQ@mail.gmail.com
Tested-by: Tyler Sanderson <tysand@google.com>
Reported-by: Tyler Sanderson <tysand@google.com>
Cc: Michael S. Tsirkin <mst@redhat.com>
Cc: Wei Wang <wei.w.wang@intel.com>
Cc: Alexander Duyck <alexander.h.duyck@linux.intel.com>
Cc: David Rientjes <rientjes@google.com>
Cc: Nadav Amit <namit@vmware.com>
Cc: Michal Hocko <mhocko@kernel.org>
Signed-off-by: David Hildenbrand <david@redhat.com>
Link: https://lore.kernel.org/r/20200205163402.42627-4-david@redhat.com
Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
The functions vring_new_virtqueue() and __vring_new_virtqueue() are used
with split rings, and any allocations within these functions are managed
outside of the .we_own_ring flag. The commit cbeedb72b9 ("virtio_ring:
allocate desc state for split ring separately") allocates the desc state
within the __vring_new_virtqueue() but frees it only when the .we_own_ring
flag is set. This leads to a memory leak when freeing such allocated
virtqueues with the vring_del_virtqueue() function.
Fix this by moving the desc_state free code outside the flag and only
for split rings. Issue was discovered during testing with remoteproc
and virtio_rpmsg.
Fixes: cbeedb72b9 ("virtio_ring: allocate desc state for split ring separately")
Signed-off-by: Suman Anna <s-anna@ti.com>
Link: https://lore.kernel.org/r/20200224212643.30672-1-s-anna@ti.com
Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
Acked-by: Jason Wang <jasowang@redhat.com>
Make sure, at build time, that pfn array is big enough to hold a single
page. It happens to be true since the PAGE_SHIFT value at the moment is
20, which is 1M - exactly 256 4K balloon pages.
Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
Reviewed-by: David Hildenbrand <david@redhat.com>
VQs without a name specified are not valid; they are skipped in the
later loop that assigns MSI-X vectors to queues, but the per_vq_vectors
loop above that counts the required number of vectors previously still
counted any queue with a non-NULL callback as needing a vector.
Add a check to the per_vq_vectors loop so that vectors with no name are
not counted to make the two loops consistent. This prevents
over-counting unnecessary vectors (e.g. for features which were not
negotiated with the device).
Cc: stable@vger.kernel.org
Fixes: 86a559787e ("virtio-balloon: VIRTIO_BALLOON_F_FREE_PAGE_HINT")
Reviewed-by: Cornelia Huck <cohuck@redhat.com>
Signed-off-by: Daniel Verkamp <dverkamp@chromium.org>
Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
Reviewed-by: Wang, Wei W <wei.w.wang@intel.com>
Ensure that elements of the callbacks array that correspond to
unavailable features are set to NULL; previously, they would be left
uninitialized.
Since the corresponding names array elements were explicitly set to
NULL, the uninitialized callback pointers would not actually be
dereferenced; however, the uninitialized callbacks elements would still
be read in vp_find_vqs_msix() and used to calculate the number of MSI-X
vectors required.
Cc: stable@vger.kernel.org
Fixes: 86a559787e ("virtio-balloon: VIRTIO_BALLOON_F_FREE_PAGE_HINT")
Reviewed-by: Cornelia Huck <cohuck@redhat.com>
Signed-off-by: Daniel Verkamp <dverkamp@chromium.org>
Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
Use devm_platform_ioremap_resource() to simplify code, which
contains platform_get_resource, devm_request_mem_region and
devm_ioremap.
Signed-off-by: Yangtao Li <tiny.windzz@gmail.com>
Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
We managed to get confused about the shift direction at least once.
Let's switch to division/multiplcation instead. Add a number of pages
macro for this purpose. We still keep the order macro around too since
this is what alloc/free pages want.
Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
Reviewed-by: Wei Wang <wei.w.wang@intel.com>
Reviewed-by: David Hildenbrand <david@redhat.com>
free_page_order is a confusing name. It's not a page order
actually, it's the order of the block of memory we are hinting.
Rename to hint_block_order. Also, rename SIZE to BYTES
to make it clear it's the block size in bytes.
Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
Reviewed-by: Wei Wang <wei.w.wang@intel.com>
Reviewed-by: David Hildenbrand <david@redhat.com>
Instead of multiplying by page order, virtio balloon divided by page
order. The result is that it can return 0 if there are a bit less
than MAX_ORDER - 1 pages in use, and then shrinker scan won't be called.
Cc: stable@vger.kernel.org
Fixes: 71994620bb ("virtio_balloon: replace oom notifier with shrinker")
Signed-off-by: Wei Wang <wei.w.wang@intel.com>
Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
Reviewed-by: David Hildenbrand <david@redhat.com>
virtio_balloon_shrinker_scan should return number of system pages freed,
but because it's calling functions that deal with balloon pages, it gets
confused and sometimes returns the number of balloon pages.
It does not matter practically as the exact number isn't
used, but it seems better to be consistent in case someone
starts using this API.
Further, if we ever tried to iteratively leak pages as
virtio_balloon_shrinker_scan tries to do, we'd run into issues - this is
because freed_pages was accumulating total freed pages, but was also
subtracted on each iteration from pages_to_free, which can result in
either leaking less memory than we were supposed to free, or more if
pages_to_free underruns.
On a system with 4K pages we are lucky that we are never asked to leak
more than 128 pages while we can leak up to 256 at a time,
but it looks like a real issue for systems with page size != 4K.
Fixes: 71994620bb ("virtio_balloon: replace oom notifier with shrinker")
Reported-by: Khazhismel Kumykov <khazhy@google.com>
Reviewed-by: Wei Wang <wei.w.wang@intel.com>
Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
Commit 780bc7903a ("virtio_ring: Support DMA APIs") makes
virtqueue_add() return -EIO when we fail to map our I/O buffers. This is
a very realistic scenario for guests with encrypted memory, as swiotlb
may run out of space, depending on it's size and the I/O load.
The virtio-blk driver interprets -EIO form virtqueue_add() as an IO
error, despite the fact that swiotlb full is in absence of bugs a
recoverable condition.
Let us change the return code to -ENOMEM, and make the block layer
recover form these failures when virtio-blk encounters the condition
described above.
Cc: stable@vger.kernel.org
Fixes: 780bc7903a ("virtio_ring: Support DMA APIs")
Signed-off-by: Halil Pasic <pasic@linux.ibm.com>
Tested-by: Michael Mueller <mimu@linux.ibm.com>
Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
Linus Torvalds
David Hildenbrand
Alexander Duyck
Michael S. Tsirkin
Zhu Lingshan
Jason Wang
Nathan Chancellor
Suman Anna
Daniel Verkamp
Yangtao Li
Wei Wang
Halil Pasic
