Adds support for virtio-pmem device as an alternative for virtio-blk.
Exposing disk image to guest as virtio-blk device results in both guest
and host independently caching the disk I/O. Using virtio-pmem device
allows to mount disk image as direct access (DAX) in the guest and thus
bypass the guest cache. This will reduce memory foodprint of the VMs.
BUG=None
TEST=cargo test
TEST=Boot patched termina kernel in crosvm; mount virtio-pmem device as
DAX and run xfstests.
Change-Id: I935fc8fc7527f79e5169f07ec7927e4ea4fa6027
Reviewed-on: https://chromium-review.googlesource.com/c/chromiumos/platform/crosvm/+/1605517
Tested-by: kokoro <noreply+kokoro@google.com>
Reviewed-by: Zach Reizner <zachr@chromium.org>
Commit-Queue: Jakub Staroń <jstaron@google.com>
This manifested itself in a couple places that were turning shared
memory buffers into slices for the purposes of passing these slices to
`Read` and `Write` trait methods.
However, this required the removal of the methods that took `Read` and
`Write` instances. This was a convenient interface but impossible to
implement safely because making slices from raw pointers without
enforcing safety guarantees causes undefined behaviour in Rust. It turns
out lots of code in crosvm was using these interfaces indirectly, which
explains why this CL touches so much.
TEST=crosvm run
BUG=chromium:938767
Change-Id: I4ff40c98da6ed08a4a42f4c31f0717f81b1c5863
Reviewed-on: https://chromium-review.googlesource.com/c/chromiumos/platform/crosvm/+/1636685
Reviewed-by: Zach Reizner <zachr@chromium.org>
Tested-by: Zach Reizner <zachr@chromium.org>
Tested-by: kokoro <noreply+kokoro@google.com>
Commit-Queue: Zach Reizner <zachr@chromium.org>
The RequestType::Flush handler correctly uses fsync(), which issues an
fsync to the underlying disk image file. However, the flush timer
(started on write and cancelled if a flush request is executed) was only
calling flush(), which is insufficient when the disk image is a raw file
- it just flushes in-memory buffers and does not issue an fsync.
BUG=None
TEST=Issue writes in crosvm; verify fsync in strace output
Change-Id: I1de8a35615031b5fdf5599dd6b49015d0b245c31
Signed-off-by: Daniel Verkamp <dverkamp@chromium.org>
Reviewed-on: https://chromium-review.googlesource.com/1632876
Tested-by: kokoro <noreply+kokoro@google.com>
Legacy-Commit-Queue: Commit Bot <commit-bot@chromium.org>
Reviewed-by: Dylan Reid <dgreid@chromium.org>
We don't need schema anymore. Will use bool and custom enums.
BUG=None
TEST=local build and run crosvm
Change-Id: I1396916878f2903b17a75f375aee4eec1ced0583
Signed-off-by: Daniel Verkamp <dverkamp@chromium.org>
Reviewed-on: https://chromium-review.googlesource.com/1564780
Tested-by: kokoro <noreply+kokoro@google.com>
Legacy-Commit-Queue: Commit Bot <commit-bot@chromium.org>
Reviewed-by: Zach Reizner <zachr@chromium.org>
These type of requests are not necessarily specific to the virtio-wl,
and other devices (virtio-gpu) may want to use them.
BUG=chromium:924405
TEST=compile
Change-Id: Iad0889da8ab3d23bb2378448fc05e3c840a93d93
Reviewed-on: https://chromium-review.googlesource.com/1626791
Commit-Ready: ChromeOS CL Exonerator Bot <chromiumos-cl-exonerator@appspot.gserviceaccount.com>
Tested-by: kokoro <noreply+kokoro@google.com>
Legacy-Commit-Queue: Commit Bot <commit-bot@chromium.org>
Reviewed-by: Zach Reizner <zachr@chromium.org>
Moves logic of consuming available descriptor heads from AvailIter
to Queue. Adds pop() function to the public interface of the Queue.
This will let us remove the used_desc_heads arrays from various virtio
device implementations because pop() is not borrowing the queue for
the whole loop scope.
BUG=None
TEST=tast run ${IP} vm.CrostiniStartEverything
Change-Id: I8eda48e41b5ec1b2d59f3177435abafb9ad5f3a3
Reviewed-on: https://chromium-review.googlesource.com/1611013
Commit-Ready: Jakub Staroń <jstaron@google.com>
Tested-by: kokoro <noreply+kokoro@google.com>
Legacy-Commit-Queue: Commit Bot <commit-bot@chromium.org>
Reviewed-by: Zach Reizner <zachr@chromium.org>
The IN_BUFFER_LEN variable limits the amount of data that will be read
into a single page sized descriptor. The old value for it left room for
the 16 byte header but reserved no space for VFDs. This happens to work
fine if the size of the read data and VFDs did not exceed the buffer
size, but, in rare circumstance, the maximum amount of data would be
read along with a FD getting received, spilling the descriptor and
causing it to fail to write to it. The guest driver does not handle this
gracefully and usually panics due to corruption.
The new value reserves room for the max number of VFDs so that the
descriptors will not end up with too much data.
BUG=chromium:951576
TEST=while true; do
(/opt/google/cros-containers/bin/wayland_demo&);
pkill -f /opt/google/cros-containers/lib/ld-linux-x86-64.so.2;
done
Change-Id: Ic0c1c10f81a91b5e5cd076e3ded8d3cc0564b614
Reviewed-on: https://chromium-review.googlesource.com/1623558
Commit-Ready: Zach Reizner <zachr@chromium.org>
Tested-by: Zach Reizner <zachr@chromium.org>
Tested-by: kokoro <noreply+kokoro@google.com>
Legacy-Commit-Queue: Commit Bot <commit-bot@chromium.org>
Reviewed-by: Daniel Verkamp <dverkamp@chromium.org>
Reviewed-by: Dylan Reid <dgreid@chromium.org>
Transfer requests have a flag called Interrupt on Short Packet (ISP)
that we have been totally ignoring. Now that the control request
execution phases have been rearranged in previous changes, we can
correctly trigger an event when a transfer fills fewer bytes than
requested.
This fixes firmware update on the Edge TPU USB Accelerator (Coral):
https://coral.withgoogle.com/docs/accelerator/get-started/
BUG=chromium:831850
TEST=Run Edge TPU test model from get-started page
Change-Id: I0d21e408a19c2e2eba562362bfe636c75dbb7160
Signed-off-by: Daniel Verkamp <dverkamp@chromium.org>
Reviewed-on: https://chromium-review.googlesource.com/1593716
Tested-by: kokoro <noreply+kokoro@google.com>
Legacy-Commit-Queue: Commit Bot <commit-bot@chromium.org>
Reviewed-by: Zach Reizner <zachr@chromium.org>
This matches the QEMU CMOS implementation and is used by BIOSes to
determine the valid memory regions to add to the e820 map.
BUG=b:133358982
TEST=Boot u-boot qemu build; observe memory size
Change-Id: I27956bc05738b5dd5b84240d5137cb06846aaab9
Signed-off-by: Daniel Verkamp <dverkamp@chromium.org>
Reviewed-on: https://chromium-review.googlesource.com/1625330
Tested-by: kokoro <noreply+kokoro@google.com>
Legacy-Commit-Queue: Commit Bot <commit-bot@chromium.org>
Reviewed-by: Dylan Reid <dgreid@chromium.org>
This fixes the Linux serial driver loopback test for the 4th serial port
(COM4, aka /dev/ttyS3).
The default value of the modem status register is always returned in the
current crosvm implementation; however, this doesn't match the expected
value in the loopback mode test used by the Linux kernel 8250 serial
driver. The other 3 serial ports (/dev/ttyS[012]) skip this test
because of the UPF_SKIP_TEST flag in STD_COMX_FLAGS, but the 4th port
uses STD_COM4_FLAGS instead, which doesn't have UPF_SKIP_TEST.
The mapping of MCR to MSR bits is defined in the 16550 UART data sheet:
http://www.ti.com/product/pc16550d
BUG=chromium:953983
TEST=`echo test > /dev/ttyS3` in crosvm
Change-Id: I1b52111235a500fef2dee00c5803a2221a25e0c6
Signed-off-by: Daniel Verkamp <dverkamp@chromium.org>
Reviewed-on: https://chromium-review.googlesource.com/1620704
Tested-by: kokoro <noreply+kokoro@google.com>
Legacy-Commit-Queue: Commit Bot <commit-bot@chromium.org>
Reviewed-by: Zach Reizner <zachr@chromium.org>
This change allows an output to be set for each serial device for a
guest machine (stdout, syslog, or sink).
BUG=chromium:953983
TEST=FEATURES=test emerge-sarien crosvm; cd sys_util; cargo test;
./build_test; manual testing on x86_64 and aarch_64
Change-Id: I9e7fcb0b296c0f8a5aa8d54b1a74ae801f6badc8
Reviewed-on: https://chromium-review.googlesource.com/1572813
Commit-Ready: Trent Begin <tbegin@chromium.org>
Tested-by: kokoro <noreply+kokoro@google.com>
Tested-by: Trent Begin <tbegin@chromium.org>
Reviewed-by: Dylan Reid <dgreid@chromium.org>
If a control request has a DataStage, execute the request during the
data stage so that we can get the actual length of the transferred data.
This will allow the ISP (Interrupt on Short Packets) bit to be
implemented correctly; we need to know the actual length during
processing of the DataStage TRB rather than at the StatusStage as we did
previously. (Note that ISP isn't implemented yet in this change; this
just passes the correct length to xhci_transfer.on_transfer_complete,
which doesn't do anything useful with the length yet in the short
transfer case.)
BUG=chromium:831850
TEST=Test adb on Crostini
Change-Id: I340424269aa139b0d9698f44ce9d3a457f3eb899
Signed-off-by: Daniel Verkamp <dverkamp@chromium.org>
Reviewed-on: https://chromium-review.googlesource.com/1593715
Commit-Ready: ChromeOS CL Exonerator Bot <chromiumos-cl-exonerator@appspot.gserviceaccount.com>
Tested-by: kokoro <noreply+kokoro@google.com>
Reviewed-by: Zach Reizner <zachr@chromium.org>
Certain control requests need to be intercepted for special handling
rather than being passed through directly to the device. Clean up the
implementation of these intercepted requests by getting rid of the
confusing intermediate analyze_request_setup function and merging it
with the actual handling of each intercepted request in the new
execute_control_transfer function. This keeps the custom handling all
in one place rather than scattered around the file and removes the need
for the extra HostToDeviceControlRequest enum.
Also add a check in get_standard_request to verify that the request is
indeed a standard control request so that the caller does not need to
check.
BUG=chromium:831850
TEST=Test adb on Crostini
Change-Id: I73b1db76941c39f124cfd0f51f14c15017ba9141
Signed-off-by: Daniel Verkamp <dverkamp@chromium.org>
Reviewed-on: https://chromium-review.googlesource.com/1593714
Commit-Ready: ChromeOS CL Exonerator Bot <chromiumos-cl-exonerator@appspot.gserviceaccount.com>
Tested-by: kokoro <noreply+kokoro@google.com>
Reviewed-by: Zach Reizner <zachr@chromium.org>
The get_type and get_direction helpers return Options, even though
they can only possibly return a non-None value; all bit patterns of the
fields they are interpreting are defined. Drop the Options and return
the enum values directly to simplify callers and remove dead code.
Fix up a typo ("recipienet" -> "recipient") while we're in the
neighborhood.
BUG=chromium:831850
TEST=Test adb in Crostini
Change-Id: Ie26a1ed1c15f5f17b5ae80be78ce5f8ff51fab28
Signed-off-by: Daniel Verkamp <dverkamp@chromium.org>
Reviewed-on: https://chromium-review.googlesource.com/1593713
Commit-Ready: ChromeOS CL Exonerator Bot <chromiumos-cl-exonerator@appspot.gserviceaccount.com>
Tested-by: kokoro <noreply+kokoro@google.com>
Reviewed-by: Zach Reizner <zachr@chromium.org>
The device to host and host to device branches of HostDevice's
handle_control_transfer function are basically identical, differing only
in whether to do a read or write and when to do it. Split this code out
into its own function, execute_control_transfer, to reduce code
duplication and prepare for future changes.
BUG=chromium:831850
TEST=Test adb in crostini
Change-Id: Ic659f18ca97e2c1c71f67f3543096fe69228969a
Signed-off-by: Daniel Verkamp <dverkamp@chromium.org>
Reviewed-on: https://chromium-review.googlesource.com/1593712
Commit-Ready: ChromeOS CL Exonerator Bot <chromiumos-cl-exonerator@appspot.gserviceaccount.com>
Tested-by: kokoro <noreply+kokoro@google.com>
Reviewed-by: Zach Reizner <zachr@chromium.org>
Rather than acquiring the lock twice, lock it once and hold it while
doing the check and potential follow-up call to interrupt(). This looks
like it could be a race to me, but I don't know if it can actually cause
problems in practice. In any case, it's better to only acquire the lock
once.
BUG=chromium:831850
TEST=Test adb in Crostini
Change-Id: Id7aa76e543cd5b858faf128f516e8d63e27cf3e7
Signed-off-by: Daniel Verkamp <dverkamp@chromium.org>
Reviewed-on: https://chromium-review.googlesource.com/1592579
Commit-Ready: ChromeOS CL Exonerator Bot <chromiumos-cl-exonerator@appspot.gserviceaccount.com>
Tested-by: kokoro <noreply+kokoro@google.com>
Reviewed-by: Zach Reizner <zachr@chromium.org>
Set the name of the thread created to run the Xhci controller event loop
so that it can be identified more easily in a debugger.
BUG=None
TEST=Attach to running crosvm with gdb and verify 'info threads' name
Change-Id: Id73a580b35231ec7fa7aec5bd51027d32e483bff
Signed-off-by: Daniel Verkamp <dverkamp@chromium.org>
Reviewed-on: https://chromium-review.googlesource.com/1594192
Commit-Ready: ChromeOS CL Exonerator Bot <chromiumos-cl-exonerator@appspot.gserviceaccount.com>
Tested-by: kokoro <noreply+kokoro@google.com>
Reviewed-by: Zach Reizner <zachr@chromium.org>
The UsbHub::try_detach function, which is only called from the libusb
hotplug callback when devices are removed, calls into
port.get_backend_device(), which acquires the backend_device lock, and
then calls port.detach(), which also tries to acquire the backend_device
lock, while still holding onto the backend_device. This clearly leads to
a deadlock.
Add an extra block in try_detach to keep the backend_device lock scoped
so that it is dropped before calling port.detach().
BUG=chromium:958117
TEST=Hot remove usb device from Crostini; verify it is gone in lsusb
Change-Id: Ibbbf68623390e3d25576f544b815c2a5607934fc
Signed-off-by: Daniel Verkamp <dverkamp@chromium.org>
Reviewed-on: https://chromium-review.googlesource.com/c/chromiumos/platform/crosvm/+/1594158
Tested-by: kokoro <noreply+kokoro@google.com>
Reviewed-by: Dylan Reid <dgreid@chromium.org>
Reviewed-by: Zach Reizner <zachr@chromium.org>
1. Removed for device slot reset and evaluate context. The verification was
unnecessary and may cause some guest kernel operations to fail.
2. The context was updated after dequeue pointer set
3. Reset device when it's attached.
4. Add seccomp rules to allow the above reset.
The verification was copied from another implementation which works for
adb, but does not work with serial devices. The verification is also not
part of the spec, so we removed it here.
BUG=b:131336977
TEST=local build and test
Change-Id: Ifd7994ff5512346d1bab27654e60c97a602da8a6
Signed-off-by: Daniel Verkamp <dverkamp@chromium.org>
Signed-off-by: Zach Reizner <zachr@google.com>
Reviewed-on: https://chromium-review.googlesource.com/1558934
Tested-by: kokoro <noreply+kokoro@google.com>
Originally, crosvm would list details about an attached usb device for a
given port. This change allows getting details about multiple ports at
once. This is intended to simplify command line usage and downstream
consumers like concierge.
TEST=various vmc commands
Chrome UI for handling USB devices
BUG=chromium:831850
Change-Id: I55681a7fea7425c897a22a579dcc15567683ef54
Reviewed-on: https://chromium-review.googlesource.com/1529765
Commit-Ready: Zach Reizner <zachr@chromium.org>
Tested-by: Zach Reizner <zachr@chromium.org>
Tested-by: kokoro <noreply+kokoro@google.com>
Reviewed-by: Zach Reizner <zachr@chromium.org>
Extracts BalloonAdjust from VmRequest into BalloonControlCommand.
BUG=None
TEST=cargo test
TEST=cargo test --package msg_socket
TEST=cargo test --package devices
TEST=cargo test --package vm_control
TEST=tast -verbose run ${IP} vm.CrostiniStartEverything
Change-Id: Ia9f5778c37c8fd4fa560df413134d1b441142f64
Reviewed-on: https://chromium-review.googlesource.com/1565298
Commit-Ready: Jakub Staroń <jstaron@google.com>
Tested-by: kokoro <noreply+kokoro@google.com>
Reviewed-by: Stephen Barber <smbarber@chromium.org>
A few files were missing license blurbs at the top, so update them all
to include them.
BUG=none
TEST=none
Change-Id: Ida101be2e5c255b8cffeb15f5b93f63bfd1b130b
Reviewed-on: https://chromium-review.googlesource.com/1577900
Commit-Ready: Stephen Barber <smbarber@chromium.org>
Tested-by: kokoro <noreply+kokoro@google.com>
Reviewed-by: Dylan Reid <dgreid@chromium.org>
Fix up the Display impl for ExecuteError so that it's clear which
direction data is moving for the Read and Write variants.
BUG=None
TEST=cargo build
Change-Id: Ide4ea5cb453e4d7f6bd2812a1696df96daec511b
Signed-off-by: Daniel Verkamp <dverkamp@chromium.org>
Reviewed-on: https://chromium-review.googlesource.com/1574963
Commit-Ready: ChromeOS CL Exonerator Bot <chromiumos-cl-exonerator@appspot.gserviceaccount.com>
Tested-by: kokoro <noreply+kokoro@google.com>
Reviewed-by: Zach Reizner <zachr@chromium.org>
Add capture support in AC'97 PCI device. Only capture at 48kHz is
supported.
BUG=chromium:932268
TEST=cargo test -p device start_capture
TEST=Run crosvm with `--cras-audio` option to run a guest vm then test
audio capture by command
$ arecord -D hw:0,0 -r 48000 -f dat -c 2 /tmp/test.raw
Change-Id: Ie3aab1004695f0df607fef8fc337fa58cb723b65
Reviewed-on: https://chromium-review.googlesource.com/1573600
Commit-Ready: Chih-Yang Hsia <paulhsia@chromium.org>
Tested-by: kokoro <noreply+kokoro@google.com>
Reviewed-by: Dylan Reid <dgreid@chromium.org>
AddressAllocator now maintains a HashMap<Alloc, (u64, u64, u64)>,
which uniquely maps a Allocation enum (e.g: PciBar(bus, dev, bar),
GpuRenderNode, etc...) to it's address, size, and human-readable tag
/ description.
The interface has also been modified to use Error instead of Option.
Aside from improving debugging, tracking allocations will have
numerous uses in the future. For example, when allocating guest memory
over VmControl sockets, it will be possible to restrict allocations to
pre-allocated slices of memory owned by the requesting device.
To plumb through PCI information to PCI devices, this CL necessitated
the addition of a PciDevice method called `assign_bus_dev`, which
notifies PCI devices of their uniquely assigned Bus and Device numbers.
BUG=chromium:936567
TEST=cargo test -p resources && cargo build --features="gpu gpu-forward"
Change-Id: I8b4b0e32c6f3168138739249ede53d03143ee5c3
Reviewed-on: https://chromium-review.googlesource.com/1536207
Commit-Ready: Daniel Prilik <prilik@google.com>
Tested-by: kokoro <noreply+kokoro@google.com>
Reviewed-by: Zach Reizner <zachr@chromium.org>
As described in:
https://doc.rust-lang.org/edition-guide/rust-2018/ownership-and-lifetimes/default-match-bindings.html
which also covers the new mental model that the Rust Book will use for
teaching binding modes and has been found to be more friendly for both
beginners and experienced users.
Before:
match *opt {
Some(ref v) => ...,
None => ...,
}
After:
match opt {
Some(v) => ...,
None => ...,
}
TEST=cargo check --all-features
TEST=local kokoro
Change-Id: I3c5800a9be36aaf5d3290ae3bd3116f699cb00b7
Reviewed-on: https://chromium-review.googlesource.com/1566669
Commit-Ready: David Tolnay <dtolnay@chromium.org>
Tested-by: David Tolnay <dtolnay@chromium.org>
Tested-by: kokoro <noreply+kokoro@google.com>
Reviewed-by: Daniel Verkamp <dverkamp@chromium.org>
Still TODO: Interrupt routing (and tests for that).
BUG=chromium:908689
TEST=Unit tests in file. Integration testing is blocked on rest of split-irqchip being implemented.
Change-Id: I08c187c44e49889ffc47322ede0f1223c6265757
Reviewed-on: https://chromium-review.googlesource.com/1565306
Commit-Ready: Miriam Zimmerman <mutexlox@chromium.org>
Tested-by: Miriam Zimmerman <mutexlox@chromium.org>
Reviewed-by: David Tolnay <dtolnay@chromium.org>
AddressRanges' name doesn't suggest that it's a SystemAllocator builder.
This CL renames it to SystemAllocatorBuilder, and adds a
SystemAllocator::builder() that removes the need to have a separate
import for the Builder.
A minor change, but it cleans up the interface a bit.
BUG=chromium:936567
TEST=cargo test -p resources && cargo build
Change-Id: I6d14368490c0d3c4018858f541e4ae5390995878
Reviewed-on: https://chromium-review.googlesource.com/1540398
Commit-Ready: Daniel Prilik <prilik@google.com>
Tested-by: kokoro <noreply+kokoro@google.com>
Reviewed-by: Zach Reizner <zachr@chromium.org>
Only allocate using minigbm if scanout bind is request (which can
indicate that the buffer is to be shared until a proper virgl bind flag
for shared is available), but if that fails because the format is not
supported for scanout, retry the minigbm allocation with the scanout
flag.
BUG=chromium:945033
TEST=None
Change-Id: Ib99bf01c8b9c2a98b1c0d1a8592d5f7c6e1aa859
Reviewed-on: https://chromium-review.googlesource.com/1569025
Commit-Ready: David Riley <davidriley@chromium.org>
Tested-by: David Riley <davidriley@chromium.org>
Tested-by: kokoro <noreply+kokoro@google.com>
Reviewed-by: Tomasz Figa <tfiga@chromium.org>
Reviewed-by: Zach Reizner <zachr@chromium.org>
Before the new borrow checker in the 2018 edition, we sometimes used to
have to manually insert curly braced blocks to limit the scope of
borrows. These are no longer needed.
Details in:
https://doc.rust-lang.org/edition-guide/rust-2018/ownership-and-lifetimes/non-lexical-lifetimes.html
TEST=cargo check --all-features
TEST=local kokoro
Change-Id: I59f9f98dcc03c8790c53e080a527ad9b68c8d6f3
Reviewed-on: https://chromium-review.googlesource.com/1568075
Commit-Ready: David Tolnay <dtolnay@chromium.org>
Tested-by: David Tolnay <dtolnay@chromium.org>
Tested-by: kokoro <noreply+kokoro@google.com>
Reviewed-by: Daniel Verkamp <dverkamp@chromium.org>
