Pull rdma updates from Jason Gunthorpe:
"Lighter that normal, but the now usual collection of driver fixes and
small improvements:
- Small fixes and minor improvements to cxgb4, bnxt_re, rxe, srp,
efa, cxgb4
- Update mlx4 to use the new umem APIs, avoiding direct use of
scatterlist
- Support ROCEv2 in erdma
- Remove various uncalled functions, constify bin_attribute
- Provide core infrastructure to catch netdev events and route them
to drivers, consolidating duplicated driver code
- Fix rare race condition crashes in mlx5 ODP flows"
* tag 'for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/rdma/rdma: (63 commits)
RDMA/mlx5: Fix implicit ODP use after free
RDMA/mlx5: Fix a race for an ODP MR which leads to CQE with error
RDMA/qib: Constify 'struct bin_attribute'
RDMA/hfi1: Constify 'struct bin_attribute'
RDMA/rxe: Fix the warning "__rxe_cleanup+0x12c/0x170 [rdma_rxe]"
RDMA/cxgb4: Notify rdma stack for IB_EVENT_QP_LAST_WQE_REACHED event
RDMA/bnxt_re: Allocate dev_attr information dynamically
RDMA/bnxt_re: Pass the context for ulp_irq_stop
RDMA/bnxt_re: Add support to handle DCB_CONFIG_CHANGE event
RDMA/bnxt_re: Query firmware defaults of CC params during probe
RDMA/bnxt_re: Add Async event handling support
bnxt_en: Add ULP call to notify async events
RDMA/mlx5: Fix indirect mkey ODP page count
MAINTAINERS: Update the bnxt_re maintainers
RDMA/hns: Clean up the legacy CONFIG_INFINIBAND_HNS
RDMA/rtrs: Add missing deinit() call
RDMA/efa: Align interrupt related fields to same type
RDMA/bnxt_re: Fix to drop reference to the mmap entry in case of error
RDMA/mlx5: Fix link status down event for MPV
RDMA/erdma: Support create_ah/destroy_ah in non-sleepable contexts
...
If RocE is supported on the device, set the number of RoCE MSIX vectors
to the number of online CPUs + 1 and capped at these maximums:
VF: 2
NPAR: 5
PF: 64
For the PF, the maximum is now increased from the previous value
of 9 to get better performance for kernel applications.
Remove the unnecessary check for BNXT_FLAG_ROCE_CAP.
bnxt_set_dflt_ulp_msix() will only be called if the flag is set.
Reviewed-by: Andy Gospodarek <andrew.gospodarek@broadcom.com>
Signed-off-by: Michael Chan <michael.chan@broadcom.com>
Reviewed-by: Simon Horman <horms@kernel.org>
Link: https://patch.msgid.link/20240909202737.93852-2-michael.chan@broadcom.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
The existing scheme sets aside a number of MSIX/NQs for the RoCE
driver whether the RoCE driver is registered or not. This scheme
is not flexible and limits the resources available for the L2 rings
if RoCE is never used.
Modify the scheme so that the RoCE MSIX/NQs can be used by the L2
driver if they are not used for RoCE. The MSIX/NQs are now
represented by 3 fields. bp->ulp_num_msix_want contains the
desired default value, edev->ulp_num_msix_vec contains the
available value (but not necessarily in use), and
ulp_tbl->msix_requested contains the actual value in use by RoCE.
The L2 driver can dip into edev->ulp_num_msix_vec if necessary.
We need to add rtnl_lock() back in bnxt_register_dev() and
bnxt_unregister_dev() to synchronize the MSIX usage between L2 and
RoCE.
Reviewed-by: Andy Gospodarek <andrew.gospodarek@broadcom.com>
Signed-off-by: Vikas Gupta <vikas.gupta@broadcom.com>
Signed-off-by: Michael Chan <michael.chan@broadcom.com>
Reviewed-by: Jacob Keller <jacob.e.keller@intel.com>
Link: https://lore.kernel.org/r/20240409215431.41424-6-michael.chan@broadcom.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
In its current form, bnxt_rdma_aux_device_init() not only initializes
the necessary data structures of the newly created aux device but also
adds the aux device into the aux bus subsytem. Refactor the logic into
separate functions, first function to initialize the aux device along
with the required resources and second, to actually add the device to
the aux bus subsytem.
This separation helps to create bnxt_en_dev much earlier and save its
resources separately.
Reviewed-by: Andy Gospodarek <andrew.gospodarek@broadcom.com>
Signed-off-by: Vikas Gupta <vikas.gupta@broadcom.com>
Signed-off-by: Michael Chan <michael.chan@broadcom.com>
Reviewed-by: Jacob Keller <jacob.e.keller@intel.com>
Link: https://lore.kernel.org/r/20240409215431.41424-5-michael.chan@broadcom.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
Since runtime MSIXs vector allocation/free has been removed,
the L2 driver needs to repopulate the MSIX entries for the
ulp client as the irq table may change during the recovery
process.
Fixes: 3034322113 ("bnxt_en: Remove runtime interrupt vector allocation")
Reviewed-by: Andy Gospodarek <andrew.gospodarek@broadcom.com>
Signed-off-by: Vikas Gupta <vikas.gupta@broadcom.com>
Signed-off-by: Michael Chan <michael.chan@broadcom.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
If ulp = kzalloc() fails, the allocated edev will leak because it is
not properly assigned and the cleanup path will not be able to free it.
Fix it by assigning it properly immediately after allocation.
Fixes: 3034322113 ("bnxt_en: Remove runtime interrupt vector allocation")
Reviewed-by: Andy Gospodarek <andrew.gospodarek@broadcom.com>
Signed-off-by: Vikas Gupta <vikas.gupta@broadcom.com>
Signed-off-by: Michael Chan <michael.chan@broadcom.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
The current VNIC structures are stored in an array bp->vnic_info[].
The index of the array (vnic_id) is passed to all the functions that
need to reference the VNIC.
This patch changes the scheme to pass the VNIC pointer instead of the
vnic index. Subsequent patches will create additional VNICs that
will not be stored in the bp->vnic_info[] array. Using the VNIC
pointer will work for all the VNICs.
Reviewed-by: Kalesh AP <kalesh-anakkur.purayil@broadcom.com>
Signed-off-by: Pavan Chebbi <pavan.chebbi@broadcom.com>
Signed-off-by: Michael Chan <michael.chan@broadcom.com>
Link: https://lore.kernel.org/r/20240325222902.220712-5-michael.chan@broadcom.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
In the driver unload path, the driver currently checks the valid
BNXT_FLAG_ROCE_CAP flag in bnxt_rdma_aux_device_uninit() before
proceeding. This is flawed because the flag may not be set initially
during driver load. It may be set later after the NVRAM setting is
changed followed by a firmware reset. Relying on the
BNXT_FLAG_ROCE_CAP flag may crash in bnxt_rdma_aux_device_uninit() if
the aux device was never initialized:
BUG: unable to handle kernel NULL pointer dereference at 0000000000000000
PGD 8ae6aa067 P4D 0
Oops: 0000 [#1] SMP NOPTI
CPU: 39 PID: 42558 Comm: rmmod Kdump: loaded Tainted: G OE --------- - - 4.18.0-348.el8.x86_64 #1
Hardware name: Dell Inc. PowerEdge R750/0WT8Y6, BIOS 1.5.4 12/17/2021
RIP: 0010:device_del+0x1b/0x410
Code: 89 a5 50 03 00 00 4c 89 a5 58 03 00 00 eb 89 0f 1f 44 00 00 41 56 41 55 41 54 4c 8d a7 80 00 00 00 55 53 48 89 fb 48 83 ec 18 <48> 8b 2f 4c 89 e7 65 48 8b 04 25 28 00 00 00 48 89 44 24 10 31 c0
RSP: 0018:ff7f82bf469a7dc8 EFLAGS: 00010292
RAX: 0000000000000000 RBX: 0000000000000000 RCX: 0000000000000000
RDX: 0000000000000000 RSI: 0000000000000206 RDI: 0000000000000000
RBP: ff31b7cd114b0ac0 R08: 0000000000000000 R09: ffffffff935c3400
R10: ff31b7cd45bc3440 R11: 0000000000000001 R12: 0000000000000080
R13: ffffffffc1069f40 R14: 0000000000000000 R15: 0000000000000000
FS: 00007fc9903ce740(0000) GS:ff31b7d4ffac0000(0000) knlGS:0000000000000000
CS: 0010 DS: 0000 ES: 0000 CR0: 0000000080050033
CR2: 0000000000000000 CR3: 0000000992fee004 CR4: 0000000000773ee0
DR0: 0000000000000000 DR1: 0000000000000000 DR2: 0000000000000000
DR3: 0000000000000000 DR6: 00000000fffe0ff0 DR7: 0000000000000400
PKRU: 55555554
Call Trace:
bnxt_rdma_aux_device_uninit+0x1f/0x30 [bnxt_en]
bnxt_remove_one+0x2f/0x1f0 [bnxt_en]
pci_device_remove+0x3b/0xc0
device_release_driver_internal+0x103/0x1f0
driver_detach+0x54/0x88
bus_remove_driver+0x77/0xc9
pci_unregister_driver+0x2d/0xb0
bnxt_exit+0x16/0x2c [bnxt_en]
__x64_sys_delete_module+0x139/0x280
do_syscall_64+0x5b/0x1a0
entry_SYSCALL_64_after_hwframe+0x65/0xca
RIP: 0033:0x7fc98f3af71b
Fix this by modifying the check inside bnxt_rdma_aux_device_uninit()
to check for bp->aux_priv instead. We also need to make some changes
in bnxt_rdma_aux_device_init() to make sure that bp->aux_priv is set
only when the aux device is fully initialized.
Fixes: d80d88b0df ("bnxt_en: Add auxiliary driver support")
Reviewed-by: Ajit Khaparde <ajit.khaparde@broadcom.com>
Signed-off-by: Kalesh AP <kalesh-anakkur.purayil@broadcom.com>
Signed-off-by: Michael Chan <michael.chan@broadcom.com>
Signed-off-by: Paolo Abeni <pabeni@redhat.com>
Modified the bnxt_en code to create and pre-configure RDMA devices
with the right MSI-X vector count for the ROCE driver to use.
This is to align the ROCE driver to the auxiliary device model which
will simply bind the driver without getting into PCI-related handling.
All PCI-related logic will now be in the bnxt_en driver.
Suggested-by: Leon Romanovsky <leonro@nvidia.com>
Signed-off-by: Ajit Khaparde <ajit.khaparde@broadcom.com>
Reviewed-by: Leon Romanovsky <leonro@nvidia.com>
Remove the SRIOV config callback which the bnxt_en was calling
to reconfigure the chip resources for a PF device when VFs are
created. The code is now modified to provision the VF resources
based on the total VF count instead of the actual VF count.
This allows the SRIOV config callback to be removed from the
list of ulp_ops.
Suggested-by: Leon Romanovsky <leonro@nvidia.com>
Signed-off-by: Ajit Khaparde <ajit.khaparde@broadcom.com>
Reviewed-by: Leon Romanovsky <leonro@nvidia.com>
Wherever possible use the function ops provided by auxiliary bus
instead of using proprietary ops.
Defined bnxt_re_suspend and bnxt_re_resume calls which can be
invoked by the bnxt_en driver instead of the ULP stop/start calls.
Signed-off-by: Ajit Khaparde <ajit.khaparde@broadcom.com>
Reviewed-by: Andy Gospodarek <andrew.gospodarek@broadcom.com>
Reviewed-by: Selvin Xavier <selvin.xavier@broadcom.com>
Reviewed-by: Leon Romanovsky <leonro@nvidia.com>
For a single ULP user there is no need for complicating function
indirection calls. Remove all this complexity in favour of direct
function calls exported by the bnxt_en driver. This allows to
simplify the code greatly. Also remove unused ulp_async_notifier.
Suggested-by: Leon Romanovsky <leonro@nvidia.com>
Signed-off-by: Ajit Khaparde <ajit.khaparde@broadcom.com>
Reviewed-by: Andy Gospodarek <andrew.gospodarek@broadcom.com>
Reviewed-by: Selvin Xavier <selvin.xavier@broadcom.com>
Reviewed-by: Leon Romanovsky <leonro@nvidia.com>
Use auxiliary driver interface for driver load, unload ROCE driver.
The driver does not need to register the interface using the netdev
notifier anymore. Removed the bnxt_re_dev_list which is not needed.
Currently probe, remove and shutdown ops have been implemented for
the auxiliary device.
Also remove exccessve validation checks for rdev.
Signed-off-by: Ajit Khaparde <ajit.khaparde@broadcom.com>
Reviewed-by: Andy Gospodarek <andrew.gospodarek@broadcom.com>
Reviewed-by: Selvin Xavier <selvin.xavier@broadcom.com>
Reviewed-by: Leon Romanovsky <leonro@nvidia.com>
Add auxiliary driver support.
An auxiliary device will be created if the hardware indicates
support for RDMA.
The bnxt_ulp_probe() function has been removed and a new
bnxt_rdma_aux_device_add() function has been added.
The bnxt_free_msix_vecs() and bnxt_req_msix_vecs() will now hold
the RTNL lock when they call the bnxt_close_nic()and bnxt_open_nic()
since the device close and open need to be protected under RTNL lock.
The operations between the bnxt_en and bnxt_re will be protected
using the en_ops_lock.
This will be used by the bnxt_re driver in a follow-on patch
to create ROCE interfaces.
Signed-off-by: Ajit Khaparde <ajit.khaparde@broadcom.com>
Reviewed-by: Andy Gospodarek <andrew.gospodarek@broadcom.com>
Reviewed-by: Selvin Xavier <selvin.xavier@broadcom.com>
Reviewed-by: Leon Romanovsky <leonro@nvidia.com>
The conversion follows this general pattern for most of the calls:
1. The input message is changed from a stack variable initialized
using bnxt_hwrm_cmd_hdr_init() to a pointer allocated and intialized
using hwrm_req_init().
2. If we don't need to read the firmware response, the hwrm_send_message()
call is replaced with hwrm_req_send().
3. If we need to read the firmware response, the mutex lock is replaced
by hwrm_req_hold() to hold the response. When the response is read, the
mutex unlock is replaced by hwrm_req_drop().
If additional DMA buffers are needed for firmware response data, the
hwrm_req_dma_slice() is used instead of calling dma_alloc_coherent().
Some minor refactoring is also done while doing these conversions.
v2: Fix unintialized variable warnings in __bnxt_hwrm_get_tx_rings()
and bnxt_approve_mac()
Signed-off-by: Edwin Peer <edwin.peer@broadcom.com>
Signed-off-by: Michael Chan <michael.chan@broadcom.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Move all firmware messaging functions and definitions to new
bnxt_hwrm.[ch]. The follow-on patches will make major modifications
to these APIs.
Signed-off-by: Edwin Peer <edwin.peer@broadcom.com>
Signed-off-by: Michael Chan <michael.chan@broadcom.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
The capabilities can change after firmware upgrade/downgrade, so we
should get the up-to-date RoCE capabilities everytime bnxt_ulp_probe()
is called.
Fixes: 2151fe0830 ("bnxt_en: Handle RESET_NOTIFY async event from firmware.")
Reviewed-by: Somnath Kotur <somnath.kotur@broadcom.com>
Reviewed-by: Edwin Peer <edwin.peer@broadcom.com>
Signed-off-by: Michael Chan <michael.chan@broadcom.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
The function bnxt_get_ulp_stat_ctxs() does not count the stats contexts
used by the RDMA driver correctly when the RDMA driver is freeing the
MSIX vectors. It assumes that if the RDMA driver is registered, the
additional stats contexts will be needed. This is not true when the
RDMA driver is about to unregister and frees the MSIX vectors.
This slight error leads to over accouting of the stats contexts needed
after the RDMA driver has unloaded. This will cause some firmware
warning and error messages in dmesg during subsequent config. changes
or ifdown/ifup.
Fix it by properly accouting for extra stats contexts only if the
RDMA driver is registered and MSIX vectors have been successfully
requested.
Fixes: c027c6b4e9 ("bnxt_en: get rid of num_stat_ctxs variable")
Reviewed-by: Yongping Zhang <yongping.zhang@broadcom.com>
Reviewed-by: Pavan Chebbi <pavan.chebbi@broadcom.com>
Signed-off-by: Michael Chan <michael.chan@broadcom.com>
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
The purpose of this is to inform the RDMA driver the size of the doorbell
BAR that the L2 driver has mapped and the portion that is mapped
uncacheable. The unchaeable portion is shared with the RoCE driver.
Any remaining unmapped doorbell BAR can be used by the RDMA driver for
its own purpose. Currently, the entire L2 portion is mapped uncacheable.
Signed-off-by: Michael Chan <michael.chan@broadcom.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
The driver provides completion ring or NQ doorbell offset for each
MSIX entry requested by the RDMA driver. The NQ offset on 57500
chips is different than legacy chips. Set it correctly based on
chip type for correctness. The RDMA driver is ignoring this field
for the 57500 chips so it is not causing any problem.
Signed-off-by: Michael Chan <michael.chan@broadcom.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
The logic needs to check both bp->total_irqs and the reserved IRQs in
hw_resc->resv_irqs if applicable and see if both are enough to cover
the L2 and RDMA requested vectors. The current code is only checking
bp->total_irqs and can fail in some code paths, such as the TX timeout
code path with the RDMA driver requesting vectors after recovery. In
this code path, we have not reserved enough MSIX resources for the
RDMA driver yet.
Fixes: 75720e6323 ("bnxt_en: Keep track of reserved IRQs.")
Signed-off-by: Michael Chan <michael.chan@broadcom.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Everytime driver registers with firmware, driver is required to
register for async event notifications as well. These 2 calls
are done using the same firmware command and can be combined.
We are also missing the 2nd step to register for async events
in the suspend/resume path and this will fix it. Prior to this,
we were getting only default notifications.
ULP can register for additional async events for the RDMA driver,
so we add a parameter to the new function to only do step 2 when
it is called from ULP.
Signed-off-by: Vasundhara Volam <vasundhara-v.volam@broadcom.com>
Signed-off-by: Michael Chan <michael.chan@broadcom.com>
Signed-off-by: Jakub Kicinski <jakub.kicinski@netronome.com>
We call bnxt_ulp_stop() to notify the RDMA driver that some error or
imminent reset is about to happen. After that we always call
some variants of bnxt_close().
In the next patch, we will integrate the recently added error
recovery with the RDMA driver. In response to ulp_stop, the
RDMA driver may free MSIX vectors and that will also trigger
bnxt_close(). To avoid bnxt_close() from being called twice,
we set a new flag after ulp_stop is called. If the RDMA driver
frees MSIX vectors while the new flag is set, we will not call
bnxt_close(), knowing that it will happen in due course.
With this change, we must make sure that the bnxt_close() call
after ulp_stop will reset IRQ. Modify bnxt_reset_task()
accordingly if we call ulp_stop.
Signed-off-by: Vasundhara Volam <vasundhara-v.volam@broadcom.com>
Signed-off-by: Michael Chan <michael.chan@broadcom.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Add the bnxt_fw_reset() main function to handle firmware reset. This
is triggered by firmware to initiate an orderly reset, for example
when a non-fatal exception condition has been detected. bnxt_fw_reset()
will first wait for all VFs to shutdown and then start the
bnxt_fw_reset_task() work queue to go through the sequence of reset,
re-probe, and re-initialization.
The next patch will add the devlink reporter to start the sequence and
call bnxt_fw_reset().
Signed-off-by: Michael Chan <michael.chan@broadcom.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
In an earlier commit to improve NQ reservations on 57500 chips, we
set the resv_irqs on the 57500 VFs to the fixed value assigned by
the PF regardless of how many are actually used. The current
code assumes that resv_irqs minus the ones used by the network driver
must be the ones for the RDMA driver. This is no longer true and
we may return more MSIX vectors than requested, causing inconsistency.
Fix it by capping the value.
Fixes: 01989c6b69 ("bnxt_en: Improve NQ reservations.")
Signed-off-by: Michael Chan <michael.chan@broadcom.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
When making configuration changes, the driver calls bnxt_close_nic()
and then bnxt_open_nic() for the changes to take effect. A parameter
irq_re_init is passed to the call sequence to indicate if IRQ
should be re-initialized. This irq_re_init parameter needs to
be included in the bnxt_reserve_rings() call. bnxt_reserve_rings()
can only call pci_disable_msix() if the irq_re_init parameter is
true, otherwise it may hit BUG() because some IRQs may not have been
freed yet.
Fixes: 41e8d79837 ("bnxt_en: Modify the ring reservation functions for 57500 series chips.")
Signed-off-by: Michael Chan <michael.chan@broadcom.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Calling bnxt_set_max_func_stat_ctxs() to modify max stat_ctxs requested
or freed by the RDMA driver is wrong. After introducing reservation of
resources recently, the driver has to keep track of all stat_ctxs
including the ones used by the RDMA driver. This will provide a better
foundation for accurate accounting of the stat_ctxs.
Signed-off-by: Vasundhara Volam <vasundhara-v.volam@broadcom.com>
Signed-off-by: Michael Chan <michael.chan@broadcom.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
For bnxt_en driver, stat_ctxs created will always be same as
cp_nr_rings. Remove extra variable that duplicates the value.
Also introduce bnxt_get_avail_stat_ctxs_for_en() helper to get
available stat_ctxs and bnxt_get_ulp_stat_ctxs() helper to return
number of stat_ctxs used by RDMA.
Signed-off-by: Vasundhara Volam <vasundhara-v.volam@broadcom.com>
Signed-off-by: Michael Chan <michael.chan@broadcom.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
