As of prior to this patch, the controller checks the value of CC.IOCQES
and CC.IOSQES prior to enabling the controller. As reported by Ben in
GitLab issue #1691, this is not spec compliant. The controller should
only check these values when queues are created.
This patch moves these checks to nvme_create_cq(). We do not need to
check it in nvme_create_sq() since that will error out if the completion
queue is not already created.
Also, since the controller exclusively supports SQEs of size 64 bytes
and CQEs of size 16 bytes, hard code that.
Resolves: https://gitlab.com/qemu-project/qemu/-/issues/1691
Signed-off-by: Klaus Jensen <k.jensen@samsung.com>
As reported by Trend Micro's Zero Day Initiative, an oob memory read
vulnerability exists in nvme_fdp_events(). The host-provided offset is
not verified.
Fix this.
This is only exploitable when Flexible Data Placement mode (fdp=on) is
enabled.
Fixes: CVE-2023-4135
Fixes: 73064edfb8 ("hw/nvme: flexible data placement emulation")
Reported-by: Trend Micro's Zero Day Initiative
Signed-off-by: Klaus Jensen <k.jensen@samsung.com>
Use the stl/ldl pci dma api for writing/reading doorbells. This removes
the explicit endian conversions.
Reviewed-by: Philippe Mathieu-Daudé <philmd@linaro.org>
Reviewed-by: Cédric Le Goater <clg@redhat.com>
Tested-by: Cédric Le Goater <clg@redhat.com>
Reviewed-by: Thomas Huth <thuth@redhat.com>
Signed-off-by: Klaus Jensen <k.jensen@samsung.com>
In commit 2fda0726e5 ("hw/nvme: fix missing endian conversions for
doorbell buffers"), we fixed shadow doorbells for big-endian guests
running on little endian hosts. But I did not fix little-endian guests
on big-endian hosts. Fix this.
Resolves: https://gitlab.com/qemu-project/qemu/-/issues/1765
Fixes: 3f7fe8de3d ("hw/nvme: Implement shadow doorbell buffer support")
Cc: qemu-stable@nongnu.org
Reported-by: Thomas Huth <thuth@redhat.com>
Tested-by: Cédric Le Goater <clg@redhat.com>
Tested-by: Thomas Huth <thuth@redhat.com>
Reviewed-by: Keith Busch <kbusch@kernel.org>
Reviewed-by: Philippe Mathieu-Daudé <philmd@linaro.org>
Signed-off-by: Klaus Jensen <k.jensen@samsung.com>
Currently the only implementers of ARI is SR-IOV devices, and they
behave similar. Share the ARI next function number.
Signed-off-by: Akihiko Odaki <akihiko.odaki@daynix.com>
Reviewed-by: Ani Sinha <anisinha@redhat.com>
Message-Id: <20230710153838.33917-2-akihiko.odaki@daynix.com>
Reviewed-by: Michael S. Tsirkin <mst@redhat.com>
Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
Allow the placement handles to be specified as ranges, i.e.
`fdp.ruhs=1:3-5` will attempt to assign ruh 1, 3, 4 and 5 to the
namespace.
Reviewed-by: Jesper Wendel Devantier <j.devantier@samsung.com>
Signed-off-by: Klaus Jensen <k.jensen@samsung.com>
Verify that a reclaim unit handle identifier is only specified once in
fdp.ruhs.
Fixes: 73064edfb8 ("hw/nvme: flexible data placement emulation")
Reviewed-by: Jesper Wendel Devantier <j.devantier@samsung.com>
Signed-off-by: Klaus Jensen <k.jensen@samsung.com>
Fix a off-by-one error when verifying the number of reclaim unit handle
identifiers specified in fdp.ruhs. To make the fix nicer, move the
verification of the fdp.nruh parameter to an earlier point.
Fixes: 73064edfb8 ("hw/nvme: flexible data placement emulation")
Reviewed-by: Jesper Wendel Devantier <j.devantier@samsung.com>
Signed-off-by: Klaus Jensen <k.jensen@samsung.com>
MCL(Maximum Copy Length) in the Identify Namespace data structure limits
the number of LBAs to be copied inside of the controller. We've not
checked it at all, so added the check with returning the proper error
status.
Signed-off-by: Minwoo Im <minwoo.im@samsung.com>
Reviewed-by: Klaus Jensen <k.jensen@samsung.com>
Signed-off-by: Klaus Jensen <k.jensen@samsung.com>
If we don't have NVME_CMD_COPY consideration in the switch statement in
nvme_aio_err(), it will go to have NVME_INTERNAL_DEV_ERROR and
`req->status` will be ovewritten to it. During the aio context, it
might set the NVMe status field like NVME_CMD_SIZE_LIMIT, but it's
overwritten in the nvme_aio_err().
Add consideration for the NVME_CMD_COPY not to overwrite the status at
the end of the function.
Signed-off-by: Minwoo Im <minwoo.im@samsung.com>
Reviewed-by: Klaus Jensen <k.jensen@samsung.com>
Signed-off-by: Klaus Jensen <k.jensen@samsung.com>
Add more comments of existing properties for nvme-ns device.
Signed-off-by: Minwoo Im <minwoo.im@samsung.com>
Reviewed-by: Klaus Jensen <k.jensen@samsung.com>
Signed-off-by: Klaus Jensen <k.jensen@samsung.com>
We use the user_ss[] array to hold the user emulation sources,
and the softmmu_ss[] array to hold the system emulation ones.
Hold the latter in the 'system_ss[]' array for parity with user
emulation.
Mechanical change doing:
$ sed -i -e s/softmmu_ss/system_ss/g $(git grep -l softmmu_ss)
Signed-off-by: Philippe Mathieu-Daudé <philmd@linaro.org>
Reviewed-by: Richard Henderson <richard.henderson@linaro.org>
Message-Id: <20230613133347.82210-10-philmd@linaro.org>
Signed-off-by: Richard Henderson <richard.henderson@linaro.org>
This protects devices from bh->mmio reentrancy issues.
Thanks: Thomas Huth <thuth@redhat.com> for diagnosing OS X test failure.
Signed-off-by: Alexander Bulekov <alxndr@bu.edu>
Reviewed-by: Darren Kenny <darren.kenny@oracle.com>
Reviewed-by: Stefan Hajnoczi <stefanha@redhat.com>
Reviewed-by: Michael S. Tsirkin <mst@redhat.com>
Reviewed-by: Paul Durrant <paul@xen.org>
Reviewed-by: Thomas Huth <thuth@redhat.com>
Message-Id: <20230427211013.2994127-5-alxndr@bu.edu>
Signed-off-by: Thomas Huth <thuth@redhat.com>
The last argument to -pci_nvme_err_startfail_virt_state is always "OFFLINE"
due to the enclosing "if" condition requiring !sctrl->scs. Reported by
Coverity.
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
The iocb (and the allocated memory to hold LBA ranges) leaks if reading
the LBA ranges fails.
Fix this by adding a free and an unref of the iocb.
Reported-by: Coverity (CID 1508281)
Fixes: d7d1474fd8 ("hw/nvme: reimplement dsm to allow cancellation")
Reviewed-by: Philippe Mathieu-Daudé <philmd@linaro.org>
Signed-off-by: Klaus Jensen <k.jensen@samsung.com>
Coverity reports a memory leak of memory when parsing ruhids at
namespace initialization. Since this is just working memory, not needed
beyond the scope of the functions, fix this by adding a g_autofree
annotation.
Reported-by: Coverity (CID 1507979)
Fixes: 73064edfb8 ("hw/nvme: flexible data placement emulation")
Reviewed-by: Philippe Mathieu-Daudé <philmd@linaro.org>
Signed-off-by: Klaus Jensen <k.jensen@samsung.com>
Even if the host is somehow using compare to do compare-and-write, the
host should be notified immediately about the compare failure and not
have to wait for the driver to potentially retry the command.
Fixes: 0a384f923f ("hw/block/nvme: add compare command")
Reported-by: Jim Harris <james.r.harris@intel.com>
Signed-off-by: Klaus Jensen <k.jensen@samsung.com>
Since the nvme_blk_read/write are used by both the data and metadata
portions of the IO, it can't have the 512B alignment requirement.
Without this change any metadata transfer, which length isn't a multiple
of 512B and which is bigger than 512B, will result in only a partial
transfer.
Signed-off-by: Mateusz Kozlowski <kozlowski.mateuszpl@gmail.com>
Reviewed-by: Klaus Jensen <k.jensen@samsung.com>
Signed-off-by: Klaus Jensen <k.jensen@samsung.com>
Add emulation of TP4146 ("Flexible Data Placement").
Reviewed-by: Keith Busch <kbusch@kernel.org>
Signed-off-by: Jesper Devantier <j.devantier@samsung.com>
Signed-off-by: Klaus Jensen <k.jensen@samsung.com>
Add support for the Directive Send and Recv commands and the Identify
directive.
Reviewed-by: Keith Busch <kbusch@kernel.org>
Signed-off-by: Gollu Appalanaidu <anaidu.gollu@samsung.com>
Signed-off-by: Klaus Jensen <k.jensen@samsung.com>
Add the mandatory Endurance Group identify data structures and log
pages.
For now, all namespaces in a subsystem belongs to a single Endurance
Group.
Reviewed-by: Keith Busch <kbusch@kernel.org>
Signed-off-by: Klaus Jensen <k.jensen@samsung.com>
Each NvmeNamespace can be used by serveral controllers,
but a NvmeNamespace can at most belong to a single NvmeSubsystem.
Store a pointer to the NvmeSubsystem, if the namespace was realized
with a NvmeSubsystem.
Reviewed-by: Keith Busch <kbusch@kernel.org>
Signed-off-by: Niklas Cassel <niklas.cassel@wdc.com>
Signed-off-by: Klaus Jensen <k.jensen@samsung.com>
Move the rounding of bytes read/written into nvme_smart_log which
reports in units of 512 bytes, rounded up in thousands. This is in
preparation for adding the Endurance Group Information log page which
reports in units of billions, rounded up.
Reviewed-by: Keith Busch <kbusch@kernel.org>
Reviewed-by: Klaus Jensen <k.jensen@samsung.com>
Signed-off-by: Joel Granados <j.granados@samsung.com>
Signed-off-by: Klaus Jensen <k.jensen@samsung.com>
Replace the local Error variable with errp and ERRP_GUARD() and change
the return value to bool.
Reviewed-by: Philippe Mathieu-Daudé <philmd@linaro.org>
Signed-off-by: Klaus Jensen <k.jensen@samsung.com>
Remove an unnecessary local Error value in nvme_realize(). In the
process, change nvme_check_constraints() to return a bool.
Reviewed-by: Markus Armbruster <armbru@redhat.com>
Reviewed-by: Philippe Mathieu-Daudé <philmd@linaro.org>
Signed-off-by: Klaus Jensen <k.jensen@samsung.com>
Prior to reading the shadow doorbell cq head, we have to update the
eventidx. Otherwise, we risk that the driver will skip an mmio doorbell
write. This happens on riscv64, as reported by Guenter.
Adding the missing update to the cq eventidx fixes the issue.
Fixes: 3f7fe8de3d ("hw/nvme: Implement shadow doorbell buffer support")
Cc: qemu-stable@nongnu.org
Cc: qemu-riscv@nongnu.org
Reported-by: Guenter Roeck <linux@roeck-us.net>
Reviewed-by: Keith Busch <kbusch@kernel.org>
Signed-off-by: Klaus Jensen <k.jensen@samsung.com>
The eventidx and doorbell value are not handling endianness correctly.
Fix this.
Fixes: 3f7fe8de3d ("hw/nvme: Implement shadow doorbell buffer support")
Cc: qemu-stable@nongnu.org
Reported-by: Guenter Roeck <linux@roeck-us.net>
Reviewed-by: Keith Busch <kbusch@kernel.org>
Signed-off-by: Klaus Jensen <k.jensen@samsung.com>
Rename the trace events related to writing the event index and reading
the doorbell value to make it more clear that the event is associated
with an actual update (write or read respectively).
Reviewed-by: Philippe Mathieu-Daudé <philmd@linaro.org>
Reviewed-by: Keith Busch <kbusch@kernel.org>
Signed-off-by: Klaus Jensen <k.jensen@samsung.com>
Replace various ->parent_obj use with the equivalent QOM accessors.
Reviewed-by: Philippe Mathieu-Daudé <philmd@linaro.org>
Signed-off-by: Klaus Jensen <k.jensen@samsung.com>
PCIDeviceClass and PCIDevice are defined in pci.h. Many users of the
header don't actually need them. Similar structs live in their own
headers: PCIBusClass and PCIBus in pci_bus.h, PCIBridge in
pci_bridge.h, PCIHostBridgeClass and PCIHostState in pci_host.h,
PCIExpressHost in pcie_host.h, and PCIERootPortClass, PCIEPort, and
PCIESlot in pcie_port.h.
Move PCIDeviceClass and PCIDeviceClass to new pci_device.h, along with
the code that needs them. Adjust include directives.
This also enables the next commit.
Signed-off-by: Markus Armbruster <armbru@redhat.com>
Message-Id: <20221222100330.380143-6-armbru@redhat.com>
Reviewed-by: Michael S. Tsirkin <mst@redhat.com>
Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
Fix a potential use-after-free by removing the bottom half and enqueuing
the completion directly.
Fixes: 796d20681d ("hw/nvme: reimplement the copy command to allow aio cancellation")
Reviewed-by: Keith Busch <kbusch@kernel.org>
Signed-off-by: Klaus Jensen <k.jensen@samsung.com>
When the DSM operation is cancelled asynchronously, we set iocb->ret to
-ECANCELED. However, the callback function only checks the return value
of the completed aio, which may have completed succesfully prior to the
cancellation and thus the callback ends up continuing the dsm operation
instead of bailing out. Fix this.
Secondly, fix a potential use-after-free by removing the bottom half and
enqueuing the completion directly.
Fixes: d7d1474fd8 ("hw/nvme: reimplement dsm to allow cancellation")
Reviewed-by: Keith Busch <kbusch@kernel.org>
Signed-off-by: Klaus Jensen <k.jensen@samsung.com>
If the zone reset operation is cancelled but the block unmap operation
completes normally, the callback will continue resetting the next zone
since it neglects to check iocb->ret which will have been set to
-ECANCELED. Make sure that this is checked and bail out if an error is
present.
Secondly, fix a potential use-after-free by removing the bottom half and
enqueuing the completion directly.
Fixes: 63d96e4ffd ("hw/nvme: reimplement zone reset to allow cancellation")
Reviewed-by: Keith Busch <kbusch@kernel.org>
Signed-off-by: Klaus Jensen <k.jensen@samsung.com>
Make sure that iocb->aiocb is NULL'ed when cancelling.
Fix a potential use-after-free by removing the bottom half and enqueuing
the completion directly.
Fixes: 38f4ac65ac ("hw/nvme: reimplement flush to allow cancellation")
Reviewed-by: Keith Busch <kbusch@kernel.org>
Signed-off-by: Klaus Jensen <k.jensen@samsung.com>
There are several bugs in the async cancel code for the Format command.
Firstly, cancelling a format operation neglects to set iocb->ret as well
as clearing the iocb->aiocb after cancelling the underlying aiocb which
causes the aio callback to ignore the cancellation. Trivial fix.
Secondly, and worse, because the request is queued up for posting to the
CQ in a bottom half, if the cancellation is due to the submission queue
being deleted (which calls blk_aio_cancel), the req structure is
deallocated in nvme_del_sq prior to the bottom half being schedulued.
Fix this by simply removing the bottom half, there is no reason to defer
it anyway.
Fixes: 3bcf26d3d6 ("hw/nvme: reimplement format nvm to allow cancellation")
Reported-by: Jonathan Derrick <jonathan.derrick@linux.dev>
Reviewed-by: Keith Busch <kbusch@kernel.org>
Signed-off-by: Klaus Jensen <k.jensen@samsung.com>
lots of acpi rework
first version of biosbits infrastructure
ASID support in vhost-vdpa
core_count2 support in smbios
PCIe DOE emulation
virtio vq reset
HMAT support
part of infrastructure for viommu support in vhost-vdpa
VTD PASID support
fixes, tests all over the place
Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
-----BEGIN PGP SIGNATURE-----
iQFDBAABCAAtFiEEXQn9CHHI+FuUyooNKB8NuNKNVGkFAmNpXDkPHG1zdEByZWRo
YXQuY29tAAoJECgfDbjSjVRpD0AH/2G8ZPrgrxJC9y3uD5/5J6QRzO+TsDYbg5ut
uBf4rKSHHzcu6zdyAfsrhbAKKzyD4HrEGNXZrBjnKM1xCiB/SGBcDIWntwrca2+s
5Dpbi4xvd4tg6tVD4b47XNDCcn2uUbeI0e2M5QIbtCmzdi/xKbFAfl5G8DQp431X
Kmz79G4CdKWyjVlM0HoYmdCw/4FxkdjD02tE/Uc5YMrePNaEg5Bw4hjCHbx1b6ur
6gjeXAtncm9s4sO0l+sIdyiqlxiTry9FSr35WaQ0qPU+Og5zaf1EiWfdl8TRo4qU
EAATw5A4hyw11GfOGp7oOVkTGvcNB/H7aIxD7emdWZV8+BMRPKo=
=zTCn
-----END PGP SIGNATURE-----
Merge tag 'for_upstream' of https://git.kernel.org/pub/scm/virt/kvm/mst/qemu into staging
pci,pc,virtio: features, tests, fixes, cleanups
lots of acpi rework
first version of biosbits infrastructure
ASID support in vhost-vdpa
core_count2 support in smbios
PCIe DOE emulation
virtio vq reset
HMAT support
part of infrastructure for viommu support in vhost-vdpa
VTD PASID support
fixes, tests all over the place
Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
# -----BEGIN PGP SIGNATURE-----
#
# iQFDBAABCAAtFiEEXQn9CHHI+FuUyooNKB8NuNKNVGkFAmNpXDkPHG1zdEByZWRo
# YXQuY29tAAoJECgfDbjSjVRpD0AH/2G8ZPrgrxJC9y3uD5/5J6QRzO+TsDYbg5ut
# uBf4rKSHHzcu6zdyAfsrhbAKKzyD4HrEGNXZrBjnKM1xCiB/SGBcDIWntwrca2+s
# 5Dpbi4xvd4tg6tVD4b47XNDCcn2uUbeI0e2M5QIbtCmzdi/xKbFAfl5G8DQp431X
# Kmz79G4CdKWyjVlM0HoYmdCw/4FxkdjD02tE/Uc5YMrePNaEg5Bw4hjCHbx1b6ur
# 6gjeXAtncm9s4sO0l+sIdyiqlxiTry9FSr35WaQ0qPU+Og5zaf1EiWfdl8TRo4qU
# EAATw5A4hyw11GfOGp7oOVkTGvcNB/H7aIxD7emdWZV8+BMRPKo=
# =zTCn
# -----END PGP SIGNATURE-----
# gpg: Signature made Mon 07 Nov 2022 14:27:53 EST
# gpg: using RSA key 5D09FD0871C8F85B94CA8A0D281F0DB8D28D5469
# gpg: issuer "mst@redhat.com"
# gpg: Good signature from "Michael S. Tsirkin <mst@kernel.org>" [full]
# gpg: aka "Michael S. Tsirkin <mst@redhat.com>" [full]
# Primary key fingerprint: 0270 606B 6F3C DF3D 0B17 0970 C350 3912 AFBE 8E67
# Subkey fingerprint: 5D09 FD08 71C8 F85B 94CA 8A0D 281F 0DB8 D28D 5469
* tag 'for_upstream' of https://git.kernel.org/pub/scm/virt/kvm/mst/qemu: (83 commits)
checkpatch: better pattern for inline comments
hw/virtio: introduce virtio_device_should_start
tests/acpi: update tables for new core count test
bios-tables-test: add test for number of cores > 255
tests/acpi: allow changes for core_count2 test
bios-tables-test: teach test to use smbios 3.0 tables
hw/smbios: add core_count2 to smbios table type 4
vhost-user: Support vhost_dev_start
vhost: Change the sequence of device start
intel-iommu: PASID support
intel-iommu: convert VTD_PE_GET_FPD_ERR() to be a function
intel-iommu: drop VTDBus
intel-iommu: don't warn guest errors when getting rid2pasid entry
vfio: move implement of vfio_get_xlat_addr() to memory.c
tests: virt: Update expected *.acpihmatvirt tables
tests: acpi: aarch64/virt: add a test for hmat nodes with no initiators
hw/arm/virt: Enable HMAT on arm virt machine
tests: Add HMAT AArch64/virt empty table files
tests: acpi: q35: update expected blobs *.hmat-noinitiators expected HMAT:
tests: acpi: q35: add test for hmat nodes without initiators
...
Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>
There were several different ways to deal with the situation where the
vector specified for a msix function is out of bound:
- early return a function and keep progresssing
- propagate the error to the caller
- mark msix unusable
- assert it is in bound
- just ignore
An out-of-bound vector should not be specified if the device
implementation is correct so let msix functions always assert that the
specified vector is in range.
An exceptional case is virtio-pci, which allows the guest to configure
vectors. For virtio-pci, it is more appropriate to introduce its own
checks because it is sometimes too late to check the vector range in
msix functions.
Signed-off-by: Akihiko Odaki <akihiko.odaki@daynix.com>
Message-Id: <20220829083524.143640-1-akihiko.odaki@daynix.com>
Reviewed-by: Michael S. Tsirkin <mst@redhat.com>
Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
Reviewed-by: Yuval Shaia <yuval.shaia.ml@gmail.com>
Signed-off-by: Akihiko Odaki <<a href="mailto:akihiko.odaki@daynix.com" target="_blank">akihiko.odaki@daynix.com</a>><br>
As per the NVMe Command Set specification Section 3.2.2, if
i) The namespace is formatted to use 16b Guard Protection
Information (i.e., pif = 0) and
ii) The Descriptor Format is not cleared to 0h
Then the copy command should be aborted with the status code of Invalid
Namespace or Format
Fixes: 44219b6029 ("hw/nvme: 64-bit pi support")
Signed-off-by: Francis Pravin Antony Michael Raj <francis.michael@solidigm.com>
Signed-off-by: Jonathan Derrick <jonathan.derrick@solidigm.com>
Reviewed-by: Klaus Jensen <k.jensen@samsung.com>
Signed-off-by: Klaus Jensen <k.jensen@samsung.com>
Commit 2e53b0b450 ("hw/nvme: Use ioeventfd to handle doorbell
updates") had the unintended effect of disabling batching of CQEs.
This patch changes the sq/cq timers to bottom halfs and instead of
calling nvme_post_cqes() immediately (causing an interrupt per cqe), we
defer the call.
| iops
-----------------+------
baseline | 138k
+cqe batching | 233k
Fixes: 2e53b0b450 ("hw/nvme: Use ioeventfd to handle doorbell updates")
Reviewed-by: Keith Busch <kbusch@kernel.org>
Reviewed-by: Jinhao Fan <fanjinhao21s@ict.ac.cn>
Signed-off-by: Klaus Jensen <k.jensen@samsung.com>
Do not enable ioeventfd by default. Let the feature mature a bit before
we consider enabling it by default.
Fixes: 2e53b0b450 ("hw/nvme: Use ioeventfd to handle doorbell updates")
Reviewed-by: Keith Busch <kbusch@kernel.org>
Reviewed-by: Jinhao Fan <fanjinhao21s@ict.ac.cn>
Signed-off-by: Klaus Jensen <k.jensen@samsung.com>
Make sure the notifier handler is unregistered in the main loop prior to
cleaning it up.
Fixes: 2e53b0b450 ("hw/nvme: Use ioeventfd to handle doorbell updates")
Reviewed-by: Keith Busch <kbusch@kernel.org>
Reviewed-by: Jinhao Fan <fanjinhao21s@ict.ac.cn>
Signed-off-by: Klaus Jensen <k.jensen@samsung.com>
While it is safe to process the queues when they are empty, skip it if
the event notifier callback was invoked spuriously.
Reviewed-by: Keith Busch <kbusch@kernel.org>
Reviewed-by: Jinhao Fan <fanjinhao21s@ict.ac.cn>
Signed-off-by: Klaus Jensen <k.jensen@samsung.com>
Add property "ioeventfd" which is enabled by default. When this is
enabled, updates on the doorbell registers will cause KVM to signal
an event to the QEMU main loop to handle the doorbell updates.
Therefore, instead of letting the vcpu thread run both guest VM and
IO emulation, we now use the main loop thread to do IO emulation and
thus the vcpu thread has more cycles for the guest VM.
Since ioeventfd does not tell us the exact value that is written, it is
only useful when shadow doorbell buffer is enabled, where we check
for the value in the shadow doorbell buffer when we get the doorbell
update event.
IOPS comparison on Linux 5.19-rc2: (Unit: KIOPS)
qd 1 4 16 64
qemu 35 121 176 153
ioeventfd 41 133 258 313
Changes since v3:
- Do not deregister ioeventfd when it was not enabled on a SQ/CQ
Signed-off-by: Jinhao Fan <fanjinhao21s@ict.ac.cn>
Reviewed-by: Klaus Jensen <k.jensen@samsung.com>
Signed-off-by: Klaus Jensen <k.jensen@samsung.com>
Since commit 916b0f0b52 ("hw/nvme: change nvme-ns 'shared' default")
the default value of nvme-ns param 'shared' is set to true, regardless
if there is a nvme-subsys node or not.
On a system without a nvme-subsys node, a namespace will never be able
to be attached to more than one controller, so for this configuration,
it is counterintuitive for this parameter to be set by default.
Force the nvme-ns param 'shared' to false for configurations where
there is no nvme-subsys node, as the namespace will never be able to
attach to more than one controller anyway.
Signed-off-by: Niklas Cassel <niklas.cassel@wdc.com>
Reviewed-by: Klaus Jensen <k.jensen@samsung.com>
Signed-off-by: Klaus Jensen <k.jensen@samsung.com>
When shadow doorbell buffer is enabled, doorbell registers are lazily
updated. The actual queue head and tail pointers are stored in Shadow
Doorbell buffers.
Add trace events for updates on the Shadow Doorbell buffers and EventIdx
buffers. Also add trace event for the Doorbell Buffer Config command.
Signed-off-by: Jinhao Fan <fanjinhao21s@ict.ac.cn>
Reviewed-by: Klaus Jensen <k.jensen@samsung.com>
Reviewed-by: Keith Busch <kbusch@kernel.org>
[k.jensen: rebased]
Signed-off-by: Klaus Jensen <k.jensen@samsung.com>
Implement Doorbel Buffer Config command (Section 5.7 in NVMe Spec 1.3)
and Shadow Doorbel buffer & EventIdx buffer handling logic (Section 7.13
in NVMe Spec 1.3). For queues created before the Doorbell Buffer Config
command, the nvme_dbbuf_config function tries to associate each existing
SQ and CQ with its Shadow Doorbel buffer and EventIdx buffer address.
Queues created after the Doorbell Buffer Config command will have the
doorbell buffers associated with them when they are initialized.
In nvme_process_sq and nvme_post_cqe, proactively check for Shadow
Doorbell buffer changes instead of wait for doorbell register changes.
This reduces the number of MMIOs.
In nvme_process_db(), update the shadow doorbell buffer value with
the doorbell register value if it is the admin queue. This is a hack
since hosts like Linux NVMe driver and SPDK do not use shadow
doorbell buffer for the admin queue. Copying the doorbell register
value to the shadow doorbell buffer allows us to support these hosts
as well as spec-compliant hosts that use shadow doorbell buffer for
the admin queue.
Signed-off-by: Jinhao Fan <fanjinhao21s@ict.ac.cn>
Reviewed-by: Klaus Jensen <k.jensen@samsung.com>
Reviewed-by: Keith Busch <kbusch@kernel.org>
[k.jensen: rebased]
Signed-off-by: Klaus Jensen <k.jensen@samsung.com>
'namespace' is misspelled in a bunch of places.
Signed-off-by: Dr. David Alan Gilbert <dgilbert@redhat.com>
Reviewed-by: Klaus Jensen <k.jensen@samsung.com>
Message-Id: <20220614104045.85728-3-dgilbert@redhat.com>
Signed-off-by: Laurent Vivier <laurent@vivier.eu>
The internally maintained AEN mask is not cleared on reset. Fix this.
Reviewed-by: Keith Busch <kbusch@kernel.org>
Signed-off-by: Klaus Jensen <k.jensen@samsung.com>
