Fix the rescind handling. This patch addresses the following rescind
scenario that is currently not handled correctly:
If a rescind were to be received while the offer is still being
peocessed, we will be blocked indefinitely since the rescind message
is handled on the same work element as the offer message. Fix this
issue.
I would like to thank Dexuan Cui <decui@microsoft.com> and
Long Li <longli@microsoft.com> for working with me on this patch.
Signed-off-by: K. Y. Srinivasan <kys@microsoft.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
This table is immutable and should be const.
Cleanup indentation and whitespace for this as well.
Signed-off-by: Stephen Hemminger <sthemmin@microsoft.com>
Signed-off-by: K. Y. Srinivasan <kys@microsoft.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
If we cannot allocate memory for the channel, free the relid
associated with the channel.
Signed-off-by: K. Y. Srinivasan <kys@microsoft.com>
Cc: <stable@vger.kernel.org>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
With the recent introduction of per-channel tasklet, we need to update
the way we handle the 3 concurrency issues:
1. hv_process_channel_removal -> percpu_channel_deq vs.
vmbus_chan_sched -> list_for_each_entry(..., percpu_list);
2. vmbus_process_offer -> percpu_channel_enq/deq vs. vmbus_chan_sched.
3. vmbus_close_internal vs. the per-channel tasklet vmbus_on_event;
The first 2 issues can be handled by Stephen's recent patch
"vmbus: use rcu for per-cpu channel list", and the third issue
can be handled by calling tasklet_disable in vmbus_close_internal here.
We don't need the original hv_event_tasklet_disable/enable since we
now use per-channel tasklet instead of the previous per-CPU tasklet,
and actually we must remove them due to the side effect now:
vmbus_process_offer -> hv_event_tasklet_enable -> tasklet_schedule will
start the per-channel callback prematurely, cauing NULL dereferencing
(the channel may haven't been properly configured to run the callback yet).
Fixes: 631e63a9f3 ("vmbus: change to per channel tasklet")
Signed-off-by: Dexuan Cui <decui@microsoft.com>
Cc: "K. Y. Srinivasan" <kys@microsoft.com>
Cc: Haiyang Zhang <haiyangz@microsoft.com>
Cc: Stephen Hemminger <sthemmin@microsoft.com>
Tested-by: Vitaly Kuznetsov <vkuznets@redhat.com>
Signed-off-by: K. Y. Srinivasan <kys@microsoft.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
The per-cpu channel list is now referred to in the interrupt
routine. This is mostly safe since the host will not normally generate
an interrupt when channel is being deleted but if it did then there
would be a use after free problem.
To solve, this use RCU protection on ther per-cpu list.
Fixes: 631e63a9f3 ("vmbus: change to per channel tasklet")
Signed-off-by: Stephen Hemminger <sthemmin@microsoft.com>
Signed-off-by: K. Y. Srinivasan <kys@microsoft.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
All current usage of vmbus write uses the acquire_lock flag, therefore
having it be optional is unnecessary. This also fixes a sparse warning
since sparse doesn't like when a function has conditional locking.
Signed-off-by: Stephen Hemminger <sthemmin@microsoft.com>
Signed-off-by: K. Y. Srinivasan <kys@microsoft.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
Change the simple boolean batched_reading into a tri-value.
For future NAPI support in netvsc driver, the callback needs to
occur directly in interrupt handler.
Batched mode is also changed to disable host interrupts immediately
in interrupt routine (to avoid unnecessary host signals), and the
tasklet is rescheduled if more data is detected.
Signed-off-by: Stephen Hemminger <sthemmin@microsoft.com>
Signed-off-by: K. Y. Srinivasan <kys@microsoft.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
Make the event handling tasklet per channel rather than per-cpu.
This allows for better fairness when getting lots of data on the same
cpu.
Signed-off-by: Stephen Hemminger <sthemmin@microsoft.com>
Signed-off-by: K. Y. Srinivasan <kys@microsoft.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
The hv_context structure had several arrays which were per-cpu
and was allocating small structures (tasklet_struct). Instead use
a single per-cpu array.
Signed-off-by: Stephen Hemminger <sthemmin@microsoft.com>
Signed-off-by: K. Y. Srinivasan <kys@microsoft.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
Previously, we were assuming that each IC protocol version was tied to a
specific host version. For example, some Windows 10 preview hosts only
support v3 TimeSync even though driver assumes v4 is supported by all
Windows 10 hosts.
The guest will stop trying to negotiate even though older supported
versions may still be offered by the host.
Make IC version negotiation more robust by going through all versions
that are supported by the guest.
Fixes: 3da0401b4d ("Drivers: hv: utils: Fix the mapping between host
version and protocol to use")
Reported-by: Rolf Neugebauer <rolf.neugebauer@docker.com>
Signed-off-by: Alex Ng <alexng@messages.microsoft.com>
Signed-off-by: K. Y. Srinivasan <kys@microsoft.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
As part of the effort to separate out architecture specific code, move the
code for signaling end of message.
Signed-off-by: K. Y. Srinivasan <kys@microsoft.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
The host can rescind a channel that has been offered to the
guest and once the channel is rescinded, the host does not
respond to any requests on that channel. Deal with the case where
the guest may be blocked waiting for a response from the host.
Signed-off-by: K. Y. Srinivasan <kys@microsoft.com>
Cc: <stable@vger.kernel.org>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
DoS protection conditions were altered in WS2016 and now it's easy to get
-EAGAIN returned from vmbus_post_msg() (e.g. when we try changing MTU on a
netvsc device in a loop). All vmbus_post_msg() callers don't retry the
operation and we usually end up with a non-functional device or crash.
While host's DoS protection conditions are unknown to me my tests show that
it can take up to 10 seconds before the message is sent so doing udelay()
is not an option, we really need to sleep. Almost all vmbus_post_msg()
callers are ready to sleep but there is one special case:
vmbus_initiate_unload() which can be called from interrupt/NMI context and
we can't sleep there. I'm also not sure about the lonely
vmbus_send_tl_connect_request() which has no in-tree users but its external
users are most likely waiting for the host to reply so sleeping there is
also appropriate.
Signed-off-by: Vitaly Kuznetsov <vkuznets@redhat.com>
Signed-off-by: K. Y. Srinivasan <kys@microsoft.com>
Cc: <stable@vger.kernel.org>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
"kernel BUG at drivers/hv/channel_mgmt.c:350!" is observed when hv_vmbus
module is unloaded. BUG_ON() was introduced in commit 85d9aa7051
("Drivers: hv: vmbus: add an API vmbus_hvsock_device_unregister()") as
vmbus_free_channels() codepath was apparently forgotten.
Fixes: 85d9aa7051 ("Drivers: hv: vmbus: add an API vmbus_hvsock_device_unregister()")
Signed-off-by: Vitaly Kuznetsov <vkuznets@redhat.com>
Signed-off-by: K. Y. Srinivasan <kys@microsoft.com>
Cc: <Stable@vger.kernel.org>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
Changed it to HV_UNKNOWN
Signed-off-by: Haiyang Zhang <haiyangz@microsoft.com>
Signed-off-by: K. Y. Srinivasan <kys@microsoft.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
One of the factors that can result in the host concluding that a given
guest in mounting a DOS attack is if the guest generates interrupts
to the host when the host is not expecting it. If these "spurious"
interrupts reach a certain rate, the host can throttle the guest to
minimize the impact. The host computation of the "expected number
of interrupts" is strictly based on the ring transitions. Until
the host logic is fixed, base the guest logic to interrupt solely
on the ring state.
Signed-off-by: K. Y. Srinivasan <kys@microsoft.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
The auto incremented counter is not being used anymore, get rid of it.
Signed-off-by: Vitaly Kuznetsov <vkuznets@redhat.com>
Signed-off-by: K. Y. Srinivasan <kys@microsoft.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
Introduce a mechanism to control how channels will be affinitized. We will
support two policies:
1. HV_BALANCED: All performance critical channels will be dstributed
evenly amongst all the available NUMA nodes. Once the Node is assigned,
we will assign the CPU based on a simple round robin scheme.
2. HV_LOCALIZED: Only the primary channels are distributed across all
NUMA nodes. Sub-channels will be in the same NUMA node as the primary
channel. This is the current behaviour.
The default policy will be the HV_BALANCED as it can minimize the remote
memory access on NUMA machines with applications that span NUMA nodes.
Signed-off-by: K. Y. Srinivasan <kys@microsoft.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
For synthetic NIC channels, enable explicit signaling policy as netvsc wants to
explicitly control when the host is to be signaled.
Signed-off-by: K. Y. Srinivasan <kys@microsoft.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
There is a rare race when we remove an entry from the global list
hv_context.percpu_list[cpu] in hv_process_channel_removal() ->
percpu_channel_deq() -> list_del(): at this time, if vmbus_on_event() ->
process_chn_event() -> pcpu_relid2channel() is trying to query the list,
we can get the kernel fault.
Similarly, we also have the issue in the code path: vmbus_process_offer() ->
percpu_channel_enq().
We can resolve the issue by disabling the tasklet when updating the list.
The patch also moves vmbus_release_relid() to a later place where
the channel has been removed from the per-cpu and the global lists.
Reported-by: Rolf Neugebauer <rolf.neugebauer@docker.com>
Signed-off-by: Dexuan Cui <decui@microsoft.com>
Signed-off-by: K. Y. Srinivasan <kys@microsoft.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
Kdump keeps biting. Turns out CHANNELMSG_UNLOAD_RESPONSE is always
delivered to the CPU which was used for initial contact or to CPU0
depending on host version. vmbus_wait_for_unload() doesn't account for
the fact that in case we're crashing on some other CPU we won't get the
CHANNELMSG_UNLOAD_RESPONSE message and our wait on the current CPU will
never end.
Do the following:
1) Check for completion_done() in the loop. In case interrupt handler is
still alive we'll get the confirmation we need.
2) Read message pages for all CPUs message page as we're unsure where
CHANNELMSG_UNLOAD_RESPONSE is going to be delivered to. We can race with
still-alive interrupt handler doing the same, add cmpxchg() to
vmbus_signal_eom() to not lose CHANNELMSG_UNLOAD_RESPONSE message.
3) Cleanup message pages on all CPUs. This is required (at least for the
current CPU as we're clearing CPU0 messages now but we may want to bring
up additional CPUs on crash) as new messages won't be delivered till we
consume what's pending. On boot we'll place message pages somewhere else
and we won't be able to read stale messages.
Signed-off-by: Vitaly Kuznetsov <vkuznets@redhat.com>
Signed-off-by: K. Y. Srinivasan <kys@microsoft.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
Message header is modified by the hypervisor and we read it in a loop,
we need to prevent compilers from optimizing accesses. There are no such
optimizations at this moment, this is just a future proof.
Suggested-by: Radim Krcmar <rkrcmar@redhat.com>
Signed-off-by: Vitaly Kuznetsov <vkuznets@redhat.com>
Reviewed-by: Radim Kr.má<rkrcmar@redhat.com>
Signed-off-by: K. Y. Srinivasan <kys@microsoft.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
We have 3 functions dealing with messages and they all implement
the same logic to finalize reads, move it to vmbus_signal_eom().
Suggested-by: Radim Krcmar <rkrcmar@redhat.com>
Signed-off-by: Vitaly Kuznetsov <vkuznets@redhat.com>
Reviewed-by: Radim Kr.má<rkrcmar@redhat.com>
Signed-off-by: K. Y. Srinivasan <kys@microsoft.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
wait_for_completion() may sleep, it enables interrupts and this
is something we really want to avoid on crashes because interrupt
handlers can cause other crashes. Switch to the recently introduced
vmbus_wait_for_unload() doing busy wait instead.
Reported-by: Radim Krcmar <rkrcmar@redhat.com>
Signed-off-by: Vitaly Kuznetsov <vkuznets@redhat.com>
Reviewed-by: Radim Kr.má<rkrcmar@redhat.com>
Signed-off-by: K. Y. Srinivasan <kys@microsoft.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
On the channel send side, many of the VMBUS
device drivers explicity serialize access to the
outgoing ring buffer. Give more control to the
VMBUS device drivers in terms how to serialize
accesss to the outgoing ring buffer.
The default behavior will be to aquire the
ring lock to preserve the current behavior.
Signed-off-by: K. Y. Srinivasan <kys@microsoft.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
The hvsock driver needs this API to release all the resources related
to the channel.
Signed-off-by: Dexuan Cui <decui@microsoft.com>
Signed-off-by: K. Y. Srinivasan <kys@microsoft.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
This will be used by the coming hv_sock driver.
Signed-off-by: Dexuan Cui <decui@microsoft.com>
Signed-off-by: K. Y. Srinivasan <kys@microsoft.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
A function to send the type of message is also added.
The coming net/hvsock driver will use this function to proactively request
the host to offer a VMBus channel for a new hvsock connection.
Signed-off-by: Dexuan Cui <decui@microsoft.com>
Signed-off-by: K. Y. Srinivasan <kys@microsoft.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
We have to call vmbus_initiate_unload() on crash to make kdump work but
the crash can also be happening in interrupt (e.g. Sysrq + c results in
such) where we can't schedule or the following will happen:
[ 314.905786] bad: scheduling from the idle thread!
Just skipping the wait (and even adding some random wait here) won't help:
to make host-side magic working we're supposed to receive CHANNELMSG_UNLOAD
(and actually confirm the fact that we received it) but we can't use
interrupt-base path (vmbus_isr()-> vmbus_on_msg_dpc()). Implement a simple
busy wait ignoring all the other messages and use it if we're in an
interrupt context.
Signed-off-by: Vitaly Kuznetsov <vkuznets@redhat.com>
Signed-off-by: K. Y. Srinivasan <kys@microsoft.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
When we pick a CPU to use for a new subchannel we try find a non-used one
on the appropriate NUMA node, we keep track of them with the
primary->alloced_cpus_in_node mask. Under normal circumstances we don't run
out of available CPUs but it is possible when we we don't initialize some
cpus in Linux, e.g. when we boot with 'nr_cpus=' limitation.
Avoid the infinite loop in init_vp_index() by checking that we still have
non-used CPUs in the alloced_cpus_in_node mask and resetting it in case
we don't.
Signed-off-by: Vitaly Kuznetsov <vkuznets@redhat.com>
Signed-off-by: K. Y. Srinivasan <kys@microsoft.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
Add vendor and device attributes to VMBUS devices. These will be used
by Hyper-V tools as well user-level RDMA libraries that will use the
vendor/device tuple to discover the RDMA device.
Signed-off-by: K. Y. Srinivasan <kys@microsoft.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
For performance critical devices, we distribute the incoming
channel interrupt load across available CPUs in the guest.
Include Fibre channel devices in the set of devices for which
we would distribute the interrupt load.
Signed-off-by: K. Y. Srinivasan <kys@microsoft.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
spinlock is unnecessary here.
mutex is enough.
Signed-off-by: Dexuan Cui <decui@microsoft.com>
Signed-off-by: K. Y. Srinivasan <kys@microsoft.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
We want to simplify vmbus_onoffer_rescind() by not invoking
hv_process_channel_removal(NULL, ...).
Signed-off-by: Dexuan Cui <decui@microsoft.com>
Signed-off-by: K. Y. Srinivasan <kys@microsoft.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
In the path vmbus_onoffer_rescind() -> vmbus_device_unregister() ->
device_unregister() -> ... -> __device_release_driver(), we can see for a
device without a driver loaded: dev->driver is NULL, so
dev->bus->remove(dev), namely vmbus_remove(), isn't invoked.
As a result, vmbus_remove() -> hv_process_channel_removal() isn't invoked
and some cleanups(like sending a CHANNELMSG_RELID_RELEASED message to the
host) aren't done.
We can demo the issue this way:
1. rmmod hv_utils;
2. disable the Heartbeat Integration Service in Hyper-V Manager and lsvmbus
shows the device disappears.
3. re-enable the Heartbeat in Hyper-V Manager and modprobe hv_utils, but
lsvmbus shows the device can't appear again.
This is because, the host thinks the VM hasn't released the relid, so can't
re-offer the device to the VM.
We can fix the issue by moving hv_process_channel_removal()
from vmbus_close_internal() to vmbus_device_release(), since the latter is
always invoked on device_unregister(), whether or not the dev has a driver
loaded.
Signed-off-by: Dexuan Cui <decui@microsoft.com>
Signed-off-by: K. Y. Srinivasan <kys@microsoft.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
Use uuid_le_cmp() for comparing GUIDs.
Signed-off-by: K. Y. Srinivasan <kys@microsoft.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
Consistently use uuid_le type in the Hyper-V driver code.
Signed-off-by: K. Y. Srinivasan <kys@microsoft.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
This defines the channel type for PCI front-ends in Hyper-V VMs.
Signed-off-by: Jake Oshins <jakeo@microsoft.com>
Signed-off-by: K. Y. Srinivasan <kys@microsoft.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
This fixes the recent commit 3b71107d73:
Drivers: hv: vmbus: Further improve CPU affiliation logic
Without the fix, reloading hv_netvsc hangs the guest.
Signed-off-by: Dexuan Cui <decui@microsoft.com>
Signed-off-by: K. Y. Srinivasan <kys@microsoft.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
Keep track of CPU affiliations of sub-channels within the scope of the primary
channel. This will allow us to better distribute the load amongst available
CPUs.
Signed-off-by: Dexuan Cui <decui@microsoft.com>
Signed-off-by: K. Y. Srinivasan <kys@microsoft.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
The current code tracks the assigned CPUs within a NUMA node in the context of
the primary channel. So, if we have a VM with a single NUMA node with 8 VCPUs, we may
end up unevenly distributing the channel load. Fix the issue by tracking affiliations
globally.
Signed-off-by: K. Y. Srinivasan <kys@microsoft.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
We cycle through all the "high performance" channels to distribute
load across the available CPUs. Process the NetworkDirect as a
high performance device.
Signed-off-by: K. Y. Srinivasan <kys@microsoft.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
Pre-Win2012R2 hosts don't properly handle CHANNELMSG_UNLOAD and
wait_for_completion() hangs. Avoid sending such request on old hosts.
Signed-off-by: Vitaly Kuznetsov <vkuznets@redhat.com>
Signed-off-by: K. Y. Srinivasan <kys@microsoft.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
Channels/sub-channels can be affinitized to VCPUs in the guest. Implement
this affinity in a way that is NUMA aware. The current protocol distributed
the primary channels uniformly across all available CPUs. The new protocol
is NUMA aware: primary channels are distributed across the available NUMA
nodes while the sub-channels within a primary channel are distributed amongst
CPUs within the NUMA node assigned to the primary channel.
Signed-off-by: K. Y. Srinivasan <kys@microsoft.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
Map target_cpu to target_vcpu using the mapping table.
We should use the mapping table to transform guest CPU ID to VP Index
as is done for the non-performance critical channels.
While the value CPU 0 is special and will
map to VP index 0, it is good to be consistent.
Signed-off-by: K. Y. Srinivasan <kys@microsoft.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
Primary channels are distributed evenly across all vcpus we have. When the host
asks us to create subchannels it usually makes us num_cpus-1 offers and we are
supposed to distribute the work evenly among the channel itself and all its
subchannels. Make sure they are all assigned to different vcpus.
Signed-off-by: Vitaly Kuznetsov <vkuznets@redhat.com>
Signed-off-by: K. Y. Srinivasan <kys@microsoft.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
We need to call init_vp_index() after we added the channel to the appropriate
list (global or subchannel) to be able to use this information when assigning
the channel to the particular vcpu. To do so we need to move a couple of
functions around. The only real change is the init_vp_index() call. This is a
small refactoring without a functional change.
Signed-off-by: Vitaly Kuznetsov <vkuznets@redhat.com>
Signed-off-by: K. Y. Srinivasan <kys@microsoft.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
It is unlikely that that host will ask us to close only one subchannel for a
device but let's be consistent. Do both num_sc++ and num_sc-- with
channel->lock to be on the safe side.
Signed-off-by: Vitaly Kuznetsov <vkuznets@redhat.com>
Signed-off-by: K. Y. Srinivasan <kys@microsoft.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
