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>
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>
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>
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>
With the new command one can:
- assign flexible resources (queues, interrupts) to primary and
secondary controllers,
- toggle the online/offline state of given controller.
Signed-off-by: Łukasz Gieryk <lukasz.gieryk@linux.intel.com>
Acked-by: Michael S. Tsirkin <mst@redhat.com>
Reviewed-by: Klaus Jensen <k.jensen@samsung.com>
Signed-off-by: Klaus Jensen <k.jensen@samsung.com>
With four new properties:
- sriov_v{i,q}_flexible,
- sriov_max_v{i,q}_per_vf,
one can configure the number of available flexible resources, as well as
the limits. The primary and secondary controller capability structures
are initialized accordingly.
Since the number of available queues (interrupts) now varies between
VF/PF, BAR size calculation is also adjusted.
Signed-off-by: Łukasz Gieryk <lukasz.gieryk@linux.intel.com>
Acked-by: Michael S. Tsirkin <mst@redhat.com>
Reviewed-by: Klaus Jensen <k.jensen@samsung.com>
Signed-off-by: Klaus Jensen <k.jensen@samsung.com>
The n->reg_size parameter unnecessarily splits the BAR0 size calculation
in two phases; removed to simplify the code.
With all the calculations done in one place, it seems the pow2ceil,
applied originally to reg_size, is unnecessary. The rounding should
happen as the last step, when BAR size includes Nvme registers, queue
registers, and MSIX-related space.
Finally, the size of the mmio memory region is extended to cover the 1st
4KiB padding (see the map below). Access to this range is handled as
interaction with a non-existing queue and generates an error trace, so
actually nothing changes, while the reg_size variable is no longer needed.
--------------------
| BAR0 |
--------------------
[Nvme Registers ]
[Queues ]
[power-of-2 padding] - removed in this patch
[4KiB padding (1) ]
[MSIX TABLE ]
[4KiB padding (2) ]
[MSIX PBA ]
[power-of-2 padding]
Signed-off-by: Łukasz Gieryk <lukasz.gieryk@linux.intel.com>
Reviewed-by: Klaus Jensen <k.jensen@samsung.com>
Acked-by: Michael S. Tsirkin <mst@redhat.com>
Signed-off-by: Klaus Jensen <k.jensen@samsung.com>
The NVMe device defines two properties: max_ioqpairs, msix_qsize. Having
them as constants is problematic for SR-IOV support.
SR-IOV introduces virtual resources (queues, interrupts) that can be
assigned to PF and its dependent VFs. Each device, following a reset,
should work with the configured number of queues. A single constant is
no longer sufficient to hold the whole state.
This patch tries to solve the problem by introducing additional
variables in NvmeCtrl’s state. The variables for, e.g., managing queues
are therefore organized as:
- n->params.max_ioqpairs – no changes, constant set by the user
- n->(mutable_state) – (not a part of this patch) user-configurable,
specifies number of queues available _after_
reset
- n->conf_ioqpairs - (new) used in all the places instead of the ‘old’
n->params.max_ioqpairs; initialized in realize()
and updated during reset() to reflect user’s
changes to the mutable state
Since the number of available i/o queues and interrupts can change in
runtime, buffers for sq/cqs and the MSIX-related structures are
allocated big enough to handle the limits, to completely avoid the
complicated reallocation. A helper function (nvme_update_msixcap_ts)
updates the corresponding capability register, to signal configuration
changes.
Signed-off-by: Łukasz Gieryk <lukasz.gieryk@linux.intel.com>
Reviewed-by: Klaus Jensen <k.jensen@samsung.com>
Acked-by: Michael S. Tsirkin <mst@redhat.com>
Signed-off-by: Klaus Jensen <k.jensen@samsung.com>
This patch implements the Function Level Reset, a feature currently not
implemented for the Nvme device, while listed as a mandatory ("shall")
in the 1.4 spec.
The implementation reuses FLR-related building blocks defined for the
pci-bridge module, and follows the same logic:
- FLR capability is advertised in the PCIE config,
- custom pci_write_config callback detects a write to the trigger
register and performs the PCI reset,
- which, eventually, calls the custom dc->reset handler.
Depending on reset type, parts of the state should (or should not) be
cleared. To distinguish the type of reset, an additional parameter is
passed to the reset function.
This patch also enables advertisement of the Power Management PCI
capability. The main reason behind it is to announce the no_soft_reset=1
bit, to signal SR-IOV support where each VF can be reset individually.
The implementation purposedly ignores writes to the PMCS.PS register,
as even such naïve behavior is enough to correctly handle the D3->D0
transition.
It’s worth to note, that the power state transition back to to D3, with
all the corresponding side effects, wasn't and stil isn't handled
properly.
Signed-off-by: Łukasz Gieryk <lukasz.gieryk@linux.intel.com>
Reviewed-by: Klaus Jensen <k.jensen@samsung.com>
Acked-by: Michael S. Tsirkin <mst@redhat.com>
Signed-off-by: Klaus Jensen <k.jensen@samsung.com>
Introduce handling for Secondary Controller List (Identify command with
CNS value of 15h).
Secondary controller ids are unique in the subsystem, hence they are
reserved by it upon initialization of the primary controller to the
number of sriov_max_vfs.
ID reservation requires the addition of an intermediate controller slot
state, so the reserved controller has the address 0xFFFF.
A secondary controller is in the reserved state when it has no virtual
function assigned, but its primary controller is realized.
Secondary controller reservations are released to NULL when its primary
controller is unregistered.
Signed-off-by: Lukasz Maniak <lukasz.maniak@linux.intel.com>
Acked-by: Michael S. Tsirkin <mst@redhat.com>
Reviewed-by: Klaus Jensen <k.jensen@samsung.com>
Signed-off-by: Klaus Jensen <k.jensen@samsung.com>
Implementation of Primary Controller Capabilities data
structure (Identify command with CNS value of 14h).
Currently, the command returns only ID of a primary controller.
Handling of remaining fields are added in subsequent patches
implementing virtualization enhancements.
Signed-off-by: Lukasz Maniak <lukasz.maniak@linux.intel.com>
Reviewed-by: Klaus Jensen <k.jensen@samsung.com>
Acked-by: Michael S. Tsirkin <mst@redhat.com>
Signed-off-by: Klaus Jensen <k.jensen@samsung.com>
This patch implements initial support for Single Root I/O Virtualization
on an NVMe device.
Essentially, it allows to define the maximum number of virtual functions
supported by the NVMe controller via sriov_max_vfs parameter.
Passing a non-zero value to sriov_max_vfs triggers reporting of SR-IOV
capability by a physical controller and ARI capability by both the
physical and virtual function devices.
NVMe controllers created via virtual functions mirror functionally
the physical controller, which may not entirely be the case, thus
consideration would be needed on the way to limit the capabilities of
the VF.
NVMe subsystem is required for the use of SR-IOV.
Signed-off-by: Lukasz Maniak <lukasz.maniak@linux.intel.com>
Reviewed-by: Klaus Jensen <k.jensen@samsung.com>
Acked-by: Michael S. Tsirkin <mst@redhat.com>
Signed-off-by: Klaus Jensen <k.jensen@samsung.com>
The Identify Controller Serial Number (SN) is the serial number for the
NVM subsystem and must be the same across all controller in the NVM
subsystem.
Enforce this.
Reviewed-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Keith Busch <kbusch@kernel.org>
Signed-off-by: Klaus Jensen <k.jensen@samsung.com>
Header guard symbols should match their file name to make guard
collisions less likely.
Cleaned up with scripts/clean-header-guards.pl, followed by some
renaming of new guard symbols picked by the script to better ones.
Signed-off-by: Markus Armbruster <armbru@redhat.com>
Message-Id: <20220506134911.2856099-2-armbru@redhat.com>
Reviewed-by: Richard Henderson <richard.henderson@linaro.org>
[Change to generated file ebpf/rss.bpf.skeleton.h backed out]
This adds support for one possible new protection information format
introduced in TP4068 (and integrated in NVMe 2.0): the 64-bit CRC guard
and 48-bit reference tag. This version does not support storage tags.
Like the CRC16 support already present, this uses a software
implementation of CRC64 (so it is naturally pretty slow). But its good
enough for verification purposes.
This may go nicely hand-in-hand with the support that Keith submitted
for the Linux kernel[1].
[1]: https://lore.kernel.org/linux-nvme/20220126165214.GA1782352@dhcp-10-100-145-180.wdc.com/T/
Reviewed-by: Keith Busch <kbusch@kernel.org>
Signed-off-by: Naveen Nagar <naveen.n1@samsung.com>
Signed-off-by: Klaus Jensen <k.jensen@samsung.com>
Add support for up to 64 LBA formats through the LBAFEE field of the
Host Behavior Support feature.
Reviewed-by: Keith Busch <kbusch@kernel.org>
Signed-off-by: Naveen Nagar <naveen.n1@samsung.com>
Signed-off-by: Klaus Jensen <k.jensen@samsung.com>
Add support for getting and setting the Host Behavior Support feature.
Reviewed-by: Keith Busch <kbusch@kernel.org>
Signed-off-by: Naveen Nagar <naveen.n1@samsung.com>
Signed-off-by: Klaus Jensen <k.jensen@samsung.com>
Add support for TP 4076 ("Zoned Random Write Area"), v2021.08.23
("Ratified").
This adds three new namespace parameters: "zoned.numzrwa" (number of
zrwa resources, i.e. number of zones that can have a zrwa),
"zoned.zrwas" (zrwa size in LBAs), "zoned.zrwafg" (granularity in LBAs
for flushes).
Reviewed-by: Keith Busch <kbusch@kernel.org>
Signed-off-by: Klaus Jensen <k.jensen@samsung.com>
These buffers can be anything, not an array of chars,
so use the 'void *' type for them.
Signed-off-by: Philippe Mathieu-Daudé <philmd@redhat.com>
Reviewed-by: Klaus Jensen <k.jensen@samsung.com>
Signed-off-by: Klaus Jensen <k.jensen@samsung.com>
Prior to this patch the nvme-ns devices are always children of the
NvmeBus owned by the NvmeCtrl. This causes the namespaces to be
unrealized when the parent device is removed. However, when subsystems
are involved, this is not what we want since the namespaces may be
attached to other controllers as well.
This patch adds an additional NvmeBus on the subsystem device. When
nvme-ns devices are realized, if the parent controller device is linked
to a subsystem, the parent bus is set to the subsystem one instead. This
makes sure that namespaces are kept alive and not unrealized.
Reviewed-by: Hannes Reinecke <hare@suse.de>
Signed-off-by: Klaus Jensen <k.jensen@samsung.com>
Make sure the controller is unregistered from the subsystem when device
is removed.
Reviewed-by: Hannes Reinecke <hare@suse.de>
Signed-off-by: Klaus Jensen <k.jensen@samsung.com>
The nvme_ns_setup and nvme_ns_check_constraints should not depend on the
controller state. Refactor and remove it.
Reviewed-by: Hannes Reinecke <hare@suse.de>
Signed-off-by: Klaus Jensen <k.jensen@samsung.com>
Jakub noticed[1] that, when using pin-based interrupts, the device will
unconditionally deasssert when any CQEs are acknowledged. However, the
pin should not be deasserted if other completion queues still holds
unacknowledged CQEs.
The bug is an artifact of commit ca247d3509 ("hw/block/nvme: fix
pin-based interrupt behavior") which fixed one bug but introduced
another. This is the third time someone tries to fix pin-based
interrupts (see commit 5e9aa92eb1 ("hw/block: Fix pin-based interrupt
behaviour of NVMe"))...
Third time's the charm, so fix it, again, by keeping track of how many
CQs have unacknowledged CQEs and only deassert when all are cleared.
[1]: <20210610114624.304681-1-jakub.jermar@kernkonzept.com>
Cc: qemu-stable@nongnu.org
Fixes: ca247d3509 ("hw/block/nvme: fix pin-based interrupt behavior")
Reported-by: Jakub Jermář <jakub.jermar@kernkonzept.com>
Signed-off-by: Klaus Jensen <k.jensen@samsung.com>
Reviewed-by: Keith Busch <kbusch@kernel.org>
The nvme_check_prinfo() and nvme_dif_check() functions operate on the
16 bit "control" member of the NvmeCmd. These functions do not otherwise
operate on an NvmeCmd or an NvmeRequest, so change them to expect the
actual 4 bit PRINFO field and add constants that work on this field as
well.
Signed-off-by: Klaus Jensen <k.jensen@samsung.com>
Reviewed-by: Keith Busch <kbusch@kernel.org>
Prepare nvme_dif_pract_generate_dif() and nvme_dif_check() to be
callable in smaller increments by making the reftag a pointer parameter
updated by the function.
Signed-off-by: Klaus Jensen <k.jensen@samsung.com>
Reviewed-by: Keith Busch <kbusch@kernel.org>
Prior to this patch, a broadcast flush would result in submitting
multiple "fire and forget" aios (no reference saved to the aiocbs
returned from the blk_aio_flush calls).
Fix this by issuing the flushes one after another.
Signed-off-by: Klaus Jensen <k.jensen@samsung.com>
Reviewed-by: Keith Busch <kbusch@kernel.org>
On machines with version > 6.0 replace a missing EUI-64 by a generated
value.
Signed-off-by: Heinrich Schuchardt <xypron.glpk@gmx.de>
Reviewed-by: Klaus Jensen <k.jensen@samsung.com>
The EUI-64 field is the only identifier for NVMe namespaces in UEFI device
paths. Add a new namespace property "eui64", that provides the user the
option to specify the EUI-64.
Signed-off-by: Heinrich Schuchardt <xypron.glpk@gmx.de>
Acked-by: Klaus Jensen <k.jensen@samsung.com>
Reviewed-by: Klaus Jensen <k.jensen@samsung.com>
In the Zoned Namespace Command Set Specification, chapter
2.5.1 Managing resources
"The controller may transition zones in the ZSIO:Implicitly Opened state
to the ZSC:Closed state for resource management purposes."
The word may in this sentence means that automatically transitioning
an implicitly opened zone to closed is completely optional.
Add a new parameter so that the user can control if this automatic
transitioning should be performed or not.
Being able to control this can help with verifying that e.g. a user-space
program behaves properly even without this optional ZNS feature.
The default value is set to true, in order to not change the existing
behavior.
Signed-off-by: Niklas Cassel <niklas.cassel@wdc.com>
[k.jensen: moved parameter to controller]
Signed-off-by: Klaus Jensen <k.jensen@samsung.com>
With the introduction of the nvme-subsystem device we are really
cluttering up the hw/block directory.
As suggested by Philippe previously, move the nvme emulation to hw/nvme.
Suggested-by: Philippe Mathieu-Daudé <philmd@redhat.com>
Signed-off-by: Klaus Jensen <k.jensen@samsung.com>
Reviewed-by: Keith Busch <kbusch@kernel.org>
