Age | Commit message (Collapse) | Author | Files | Lines |
|
Management firmware is used as an arbiter between the various PFs
in regard to loading - it causes the various PFs to load/unload
sequentially and informs each of its appropriate rule in the init.
But the existing flow is too weak to handle some scenarios where
PFs aren't properly cleaned prior to loading.
The significant scenarios falling under this criteria:
a. Preboot drivers in some environment can't properly unload.
b. Unexpected driver replacement [kdump, PDA].
Modern management firmware supports a more intricate loading flow,
where the driver has the ability to overcome previous limitations.
This moves qed into using this newer scheme.
Notice new scheme is backward compatible, so new drivers would
still be able to load properly on top of older management firmwares
and vice versa.
Signed-off-by: Tomer Tayar <[email protected]>
Signed-off-by: Yuval Mintz <[email protected]>
Signed-off-by: David S. Miller <[email protected]>
|
|
We'll soon need additional information, so start by changing
the infrastructure to receive the initializing variables
via a parameter struct.
Signed-off-by: Yuval Mintz <[email protected]>
Signed-off-by: David S. Miller <[email protected]>
|
|
Management firmware is used as arbiter between different PFs
which are loading/unloading, but in order to use the synchronization
it offers the contending configurations need to be applied either
between their LOAD_REQ <-> LOAD_DONE or UNLOAD_REQ <-> UNLOAD_DONE
management firmware commands.
Existing HW stop flow utilizes 2 different functions: qed_hw_stop() and
qed_hw_reset() which don't abide this requirement; Most of the closure
is doing outside the scope of the unload request.
This patch removes qed_hw_reset() and places the relevant stop
functionality underneath the management firmware protection.
Signed-off-by: Tomer Tayar <[email protected]>
Signed-off-by: Yuval Mintz <[email protected]>
Signed-off-by: David S. Miller <[email protected]>
|
|
Align the driver feature distribution with the flow utilized
by the management firmware - first reserve L2 queues for
VFs and use all the remaining for the PF.
The current distribution might lead to PFs with an enormous
amount of queues, but at the same time leave us with insufficient
resources for starting all VFs.
Signed-off-by: Yuval Mintz <[email protected]>
Signed-off-by: David S. Miller <[email protected]>
|
|
When RoCE is enabled on a given L2 interface, the interrupt lines
are divided equally between L2 and RoCE -
But in case number of lines needed for RoCE is limited by number
of available CNQs, we can utilize the additional lines for L2.
Signed-off-by: Yuval Mintz <[email protected]>
Signed-off-by: David S. Miller <[email protected]>
|
|
Management firmware and driver are meant to be both backward and forward
compatibile with each other.
If a new mangement firmware would work with an older driver,
it's possible that driver would receive indications which are meaningless
to it. That's perfectly acceptible from the firmware part - so no need to
log such messages at default verbosity; That would only serve to confuse
users.
Signed-off-by: Yuval Mintz <[email protected]>
Signed-off-by: David S. Miller <[email protected]>
|
|
The management firmware is running on a Big Endian processor,
and when running on LE platform HW is configured to swap access
to memory shared between management firmware and driver on
32-bit granulariy.
As a result, for matters of simplicity most of the APIs between
driver and management firmware are based on 32-bit variables.
MAC settings are one exception, as driver needs to fill a byte
array when indicating to management firmware that primary MAC
has changed.
Due to the swap, driver must make sure that the mac that was
provided in byte-order would be translated into native order,
otherwise after the swap the management firmware would read
it swapped.
Signed-off-by: Yuval Mintz <[email protected]>
Signed-off-by: David S. Miller <[email protected]>
|
|
The driver interaction with management firmware involves a union
of all the data-members relating to the commands the driver prepares.
Current interface assumes the caller always passes such a union -
but thats cumbersome as well as risky [chancing a stack corruption
in case caller accidentally passes a smaller member instead of union].
Change implementation so that caller could pass a pointer to any
of the members instead of the union.
Signed-off-by: Tomer Tayar <[email protected]>
Signed-off-by: Yuval Mintz <[email protected]>
Signed-off-by: David S. Miller <[email protected]>
|
|
Interaction of driver -> management firmware is based
on a one-pending mailbox [per interface], and various
mailbox commands need to be synchronized.
Current scheme is messy, and there's a difficulty extending
it as it deals differently with various commands as well as
making assumption on the required behavior for load/unload
requests.
Drop the current scheme into a completion-list-based approach;
Each flow would try sending the command when possible,
allowing one flow to complete another flow's completion and
relieve the mailbox before sending its own command.
Signed-off-by: Tomer Tayar <[email protected]>
Signed-off-by: Yuval Mintz <[email protected]>
Signed-off-by: David S. Miller <[email protected]>
|
|
The link information exists only on the leading hwfn,
but some of its derivatives [e.g., min/max rate] need to
be configured for each hwfn.
When re-basing the VF link view, use the leading hwfn
information as basis for all existing hwfns to allow
said configurations to stick.
Signed-off-by: Yuval Mintz <[email protected]>
Signed-off-by: David S. Miller <[email protected]>
|
|
Malicious VF existance should be interesting enough for the
hyperuser. Change the PF indication that one of its child VF
became malicious to appear by default.
Signed-off-by: Yuval Mintz <[email protected]>
Signed-off-by: David S. Miller <[email protected]>
|
|
Signed-off-by: Yuval Mintz <[email protected]>
Signed-off-by: David S. Miller <[email protected]>
|
|
The PF<->VF interface allows for the VF to request
multiple queues closure via a single message, but this has
never been used by any official driver.
We now deprecate this option, forcing each queue close
to arrive via a different command; This would be required
for future TLVs that are going to extend the queue TLVs with
additional information on a per-queue basis.
Signed-off-by: Yuval Mintz <[email protected]>
Signed-off-by: David S. Miller <[email protected]>
|
|
PF needs to validate the status of VF queues before asking firmware
to configure anything for them, but that validation is done in various
different forms - sometimes inadequate.
Add auxillary functions that can be used for testing of the queue
state and convert the various flows to use those instead of current
existing flows; Also, add missing validations where needed.
Signed-off-by: Yuval Mintz <[email protected]>
Signed-off-by: David S. Miller <[email protected]>
|
|
When starting the VF's vport, the PF would first configure
the status blocks of the VF and then reset them.
That would cause some of the configured information to be lost -
specifically it would mean that all the VFs queues would use
the Rx coalescing state-machine of the status block.
Signed-off-by: Yuval Mintz <[email protected]>
Signed-off-by: David S. Miller <[email protected]>
|
|
When PF responds to the VF requests it also cleans the HW-channel
indication in firmware to allow further VF messages to arrive,
but the order currently applied is wrong -
The PF is copying by DMAE the response the VF is polling on for
completion, and only afterwards sets the HW-channel to ready state.
This creates a race condition where the VF would be able to send
an additional message to the PF before the channel would get ready
again, causing the firmware to consider the VF as malicious.
Signed-off-by: Yuval Mintz <[email protected]>
Signed-off-by: David S. Miller <[email protected]>
|
|
When a VF is considered malicious, driver handling of the VF
FLR flow would clean said indication - but not if the FLR is
part of an sriov-disable flow.
That leads to further issues, as PF wouldn't re-enable the
previously malicious VF when sriov is re-enabled.
No reason for that - simply clean malicious indications in
the sriov-disable flow as well.
Signed-off-by: Yuval Mintz <[email protected]>
Signed-off-by: David S. Miller <[email protected]>
|
|
VFs are currently logging errors when communicating
with their PFs in a too-low verbosity that wouldn't
be shown by default. As timeouts and failed commands
are crucial for VF operability, make them appear by
default.
Signed-off-by: Yuval Mintz <[email protected]>
Signed-off-by: David S. Miller <[email protected]>
|
|
Conflicts:
drivers/net/ethernet/broadcom/genet/bcmgenet.c
net/core/sock.c
Conflicts were overlapping changes in bcmgenet and the
lockdep handling of sockets.
Signed-off-by: David S. Miller <[email protected]>
|
|
This adds the necessary infrastructure changes for initializing
and working with the new series of QL41xxx adapaters.
It also adds 2 new PCI device-IDs to qede:
- 0x8070 for QL41xxx PFs
- 0x8090 for VFs spawning from QL41xxx PFs
Signed-off-by: Tomer Tayar <[email protected]>
Signed-off-by: Yuval Mintz <[email protected]>
Signed-off-by: David S. Miller <[email protected]>
|
|
Missing in the initial submission, qed fails to propagate qedi's
request to enable OOO to firmware.
Fixes: fc831825f99e ("qed: Add support for hardware offloaded iSCSI")
Signed-off-by: Yuval Mintz <[email protected]>
Signed-off-by: David S. Miller <[email protected]>
|
|
Need to set the number of entries in database, otherwise the logic
would quickly surpass the array.
Fixes: 1d6cff4fca43 ("qed: Add iSCSI out of order packet handling")
Signed-off-by: Yuval Mintz <[email protected]>
Signed-off-by: David S. Miller <[email protected]>
|
|
Before iterating over the the LL2 Rx ring, the ring's
spinlock is taken via spin_lock_irqsave().
The actual processing of the packet [including handling
by the protocol driver] is done without said lock,
so qed releases the spinlock and re-claims it afterwards.
Problem is that the final spin_lock_irqrestore() at the end
of the iteration uses the original flags saved from the
initial irqsave() instead of the flags from the most recent
irqsave(). So it's possible that the interrupt status would
be incorrect at the end of the processing.
Fixes: 0a7fb11c23c0 ("qed: Add Light L2 support");
CC: Ram Amrani <[email protected]>
Signed-off-by: Yuval Mintz <[email protected]>
Signed-off-by: David S. Miller <[email protected]>
|
|
Fixes: fc831825f99e ("qed: Add support for hardware offloaded iSCSI")
Signed-off-by: Yuval Mintz <[email protected]>
Signed-off-by: David S. Miller <[email protected]>
|
|
When receiving an Rx LL2 packet, qed fails to unmap the previous buffer.
Fixes: 0a7fb11c23c0 ("qed: Add Light L2 support");
Signed-off-by: Yuval Mintz <[email protected]>
Signed-off-by: David S. Miller <[email protected]>
|
|
Current Logic would allow the creation of a chain with U32_MAX + 1
elements, when the actual maximum supported by the driver infrastructure
is U32_MAX.
Fixes: a91eb52abb50 ("qed: Revisit chain implementation")
Signed-off-by: Tomer Tayar <[email protected]>
Signed-off-by: Yuval Mintz <[email protected]>
Signed-off-by: David S. Miller <[email protected]>
|
|
The Doorbell HW block can be configured at a granularity
of 16 x CIDs, so we need to make sure that the actual number
of CIDs configured would be a multiplication of 16.
Today, when RoCE is enabled - given that the number is unaligned,
doorbelling the higher CIDs would fail to reach the firmware and
would eventually timeout.
Fixes: dbb799c39717 ("qed: Initialize hardware for new protocols")
Signed-off-by: Ram Amrani <[email protected]>
Signed-off-by: Yuval Mintz <[email protected]>
Signed-off-by: David S. Miller <[email protected]>
|
|
This patch advances the qed* drivers into using the newer firmware -
This solves several firmware bugs, mostly related [but not limited to]
various init/deinit issues in various offloaded protocols.
It also introduces a major 4-Cached SGE change in firmware, which can be
seen in the storage drivers' changes.
In addition, this firmware is required for supporting the new QL41xxx
series of adapters; While this patch doesn't add the actual support,
the firmware contains the necessary initialization & firmware logic to
operate such adapters [actual support would be added later on].
Changes from Previous versions:
-------------------------------
- V2 - fix kbuild-test robot warnings
Signed-off-by: Tomer Tayar <[email protected]>
Signed-off-by: Ram Amrani <[email protected]>
Signed-off-by: Manish Rangankar <[email protected]>
Signed-off-by: Chad Dupuis <[email protected]>
Signed-off-by: Yuval Mintz <[email protected]>
Signed-off-by: David S. Miller <[email protected]>
|
|
In qed_ll2_start_ooo() the ll2_info variable is uninitialized and then
passed to qed_ll2_acquire_connection() where it is copied into a new
memory space.
This shouldn't cause any issue as long as non of the copied memory is
every read.
But the potential for a bug being introduced by reading this memory
is real.
Detected by CoverityScan, CID#1399632 ("Uninitialized scalar variable")
Signed-off-by: Robert Foss <[email protected]>
Signed-off-by: David S. Miller <[email protected]>
|
|
Commit 653d2ffd6405 ("qed*: Fix link indication race") introduced another
race - one of the inner functions called from the link-change flow is
explicitly using the slowpath context dedicated PTT instead of gaining
that PTT from the caller. Since this flow can now be called from
a different context as well, we're in risk of the PTT breaking.
Fixes: 653d2ffd6405 ("qed*: Fix link indication race")
Signed-off-by: Yuval Mintz <[email protected]>
Signed-off-by: David S. Miller <[email protected]>
|
|
A PF syncronizes all IOV activity relating to its VFs
by using a single workqueue handling the work.
The workqueue would reach a bitmask of pending VF events
and act upon each in turn.
Problem is that the indication of a VF message [which sets
the 'vf event' bit for that VF] arrives and is set in
the slowpath attention context, which isn't syncronized with
the processing of the events.
When multiple VFs are present, it's possible that PF would
lose the indication of one of the VF's pending evens, leading
that VF to later timeout.
Instead of adding locks/barriers, simply move from a bitmask
into a per-VF indication inside that VF entry in the PF database.
Signed-off-by: Yuval Mintz <[email protected]>
Signed-off-by: David S. Miller <[email protected]>
|
|
Driver changes the link properties via communication with
the management firmware, and re-reads the resulting link status
when it receives an indication that the link has changed.
However, there are certain scenarios where such indications
might be missing, and so driver also re-reads the current link
results without attention in several places. Specifically, it
does so during load and when resetting the link.
This creates a race where driver might reflect incorrect
link status - e.g., when explicit reading of the link status is
switched by attention with the changed configuration.
Correct this flow by a lock syncronizing the handling of the
link indications [both explicit requests and attention].
Signed-off-by: Yuval Mintz <[email protected]>
Signed-off-by: David S. Miller <[email protected]>
|
|
Flows accessing registers require the flow to hold a PTT entry.
To protect 'major' load/unload flows a main_ptt is pre-allocated
to guarantee such flows wouldn't be blocked by PTT being
unavailable.
Status block initialization currently uses the main_ptt which
is incorrect, as this flow might run concurrently to others
[E.g., loading qedr while toggling qede]. That would have dire
effects as it means registers' access to device breaks and further
read/writes might access incorrect addresses.
Instead, when initializing status blocks acquire/release a PTT
as part of the flow.
Signed-off-by: Yuval Mintz <[email protected]>
Signed-off-by: David S. Miller <[email protected]>
|
|
VF learns of the current link state via its bulletin board,
which might reflect either the physical link state or some
user-configured logical state.
Whenever the physical link changes or whnever such a configuration
is explicitly made by user the PF driver would update the bulletin
that the VF reads. But if neither has happened - i.e., PF still
hasn't got a physical link up and no additional configuration was
done the VF wouldn't have a valid link information available.
Simply reflect the physical link state whenever the VF is
initialized. The user could then affect it however he wants.
Signed-off-by: Yuval Mintz <[email protected]>
Signed-off-by: David S. Miller <[email protected]>
|
|
Reserving doorbell BAR space according to the currently active CPUs
may result in a bug if disabled CPUs are later enabled but no
doorbell space was reserved for them.
Signed-off-by: Ram Amrani <[email protected]>
Signed-off-by: Yuval Mintz <[email protected]>
Signed-off-by: David S. Miller <[email protected]>
|
|
If QP is in reset state then there are no resources to free so avoid
freeing any.
Signed-off-by: Ram Amrani <[email protected]>
Signed-off-by: Yuval Mintz <[email protected]>
Signed-off-by: David S. Miller <[email protected]>
|
|
Currently the state is read only after the buffers are relesed.
Signed-off-by: Ram Amrani <[email protected]>
Signed-off-by: Yuval Mintz <[email protected]>
Signed-off-by: David S. Miller <[email protected]>
|
|
The CQ resource pool is protected by a spin lock. When a CQ creation
fails it now deallocates under that lock as well.
Signed-off-by: Ram Amrani <[email protected]>
Signed-off-by: Yuval Mintz <[email protected]>
Signed-off-by: David S. Miller <[email protected]>
|
|
This adds the backbone required for the various HW initalizations
which are necessary for the FCoE driver (qedf) for QLogic FastLinQ
4xxxx line of adapters - FW notification, resource initializations, etc.
Signed-off-by: Arun Easi <[email protected]>
Signed-off-by: Yuval Mintz <[email protected]>
Signed-off-by: David S. Miller <[email protected]>
|
|
The patch adds the required qed interfaces for configuring/reading
the PTP clock on the adapter.
Signed-off-by: Sudarsana Reddy Kalluru <[email protected]>
Signed-off-by: Yuval Mintz <[email protected]>
Signed-off-by: David S. Miller <[email protected]>
|
|
Two trivial overlapping changes conflicts in MPLS and mlx5.
Signed-off-by: David S. Miller <[email protected]>
|
|
struct qed_ll2_info is rather large, so putting it on the stack
can cause an overflow, as this warning tries to tell us:
drivers/net/ethernet/qlogic/qed/qed_ll2.c: In function 'qed_ll2_start':
drivers/net/ethernet/qlogic/qed/qed_ll2.c:2159:1: error: the frame size of 1056 bytes is larger than 1024 bytes [-Werror=frame-larger-than=]
qed_ll2_start_ooo() already uses a dynamic allocation for the structure
to work around that problem, and we could do the same in qed_ll2_start()
as well as qed_roce_ll2_start(), but since the structure is only
used to pass a couple of initialization values here, it seems nicer
to replace it with a different structure.
Lacking any idea for better naming, I'm adding 'struct qed_ll2_conn',
which now contains all the initialization data, and this now simply
gets copied into struct qed_ll2_info rather than assigning all members
one by one.
Signed-off-by: Arnd Bergmann <[email protected]>
Acked-by: Yuval Mintz <[email protected]>
Signed-off-by: David S. Miller <[email protected]>
|
|
Use eth_zero_addr to assign zero address to the given address array
instead of memset when the second argument in memset is address
of zero. Also, it makes the code clearer
Signed-off-by: Shyam Saini <[email protected]>
Signed-off-by: David S. Miller <[email protected]>
|
|
Signed-off-by: Yuval Mintz <[email protected]>
Signed-off-by: David S. Miller <[email protected]>
|
|
If qedr isn't part of the kernel then don't allocate RDMA resources
for it in qed.
Signed-off-by: Ram Amrani <[email protected]>
Signed-off-by: Yuval Mintz <[email protected]>
Signed-off-by: David S. Miller <[email protected]>
|
|
Currently multicast traffic wouldn't be routed internally to
listener; Instead it would only be sent to network via the
physical carrier.
Signed-off-by: Yuval Mintz <[email protected]>
Signed-off-by: David S. Miller <[email protected]>
|
|
Trusted VFs would be allowed to receive promiscuous and
multicast promiscuous data.
Signed-off-by: Yuval Mintz <[email protected]>
Signed-off-by: David S. Miller <[email protected]>
|
|
A step toward having qede agnostic to the queue configurations
in firmware/hardware - let the RSS indirections use queue handles
instead of actual queue indices.
Signed-off-by: Yuval Mintz <[email protected]>
Signed-off-by: David S. Miller <[email protected]>
|
|
Today qede requests contexts that would suffice for 64 'whole'
combined queues [192 meant for 64 rx, tx and xdp tx queues],
but registers netdev and limits the number of queues based on
information received by qed. In turn, qed doesn't take context
into account when informing qede how many queues it can support.
This would lead to a configuration problem in case user tries
configuring >64 combined queues to interface [or >96 in case
xdp isn't enabled]. Since we don't have a mangement firware
that actually provides so many interrupt lines to a single
device we're currently safe but that's about to change soon.
The new maximum is hence changed:
- For RoCE devices, the limit would remain 64.
- For non-RoCE devices, the limit might be higher [depending
on the actual configuration of the device].
qed would start enforcing that limit in both scenarios.
Signed-off-by: Yuval Mintz <[email protected]>
Signed-off-by: David S. Miller <[email protected]>
|
|
Since the submission of the qedr driver, there's inconsistency
in the licensing of the various qed/qede files - some are GPLv2
and some are dual-license.
Since qedr requires dual-license and it's dependent on both,
we're updating the licensing of all qed/qede source files.
Signed-off-by: Yuval Mintz <[email protected]>
Signed-off-by: David S. Miller <[email protected]>
|