Age | Commit message (Collapse) | Author | Files | Lines |
|
KSM Mkey is KLM Mkey with a fixed buffer size. Due to this fact,
it is a faster mechanism than KLM.
SHAMPO feature used KLMs Mkeys for memory mappings of its headers buffer.
As it used KLMs with the same buffer size for each entry,
we can use KSMs instead.
This commit changes the Mkeys that map the SHAMPO headers buffer
from KLMs to KSMs.
Signed-off-by: Yoray Zack <yorayz@nvidia.com>
Signed-off-by: Tariq Toukan <tariqt@nvidia.com>
Link: https://lore.kernel.org/r/20240603212219.1037656-13-tariqt@nvidia.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
|
|
When CQE mode or DIM state is changed, gracefully reconfigure channels to
handle new configuration. Previously, would create new channels that would
reflect the changes rather than update the original channels.
Co-developed-by: Nabil S. Alramli <dev@nalramli.com>
Signed-off-by: Nabil S. Alramli <dev@nalramli.com>
Co-developed-by: Joe Damato <jdamato@fastly.com>
Signed-off-by: Joe Damato <jdamato@fastly.com>
Signed-off-by: Rahul Rameshbabu <rrameshbabu@nvidia.com>
Signed-off-by: Tariq Toukan <tariqt@nvidia.com>
Link: https://lore.kernel.org/r/20240419080445.417574-5-tariqt@nvidia.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
|
|
Use core DIM CQ period mode enum values for the CQ parameter for the period
mode. Translate the value to the specific mlx5 device constant for the
selected period mode when creating a CQ. Avoid needing to translate mlx5
device constants to DIM constants for core DIM functionality.
Co-developed-by: Nabil S. Alramli <dev@nalramli.com>
Signed-off-by: Nabil S. Alramli <dev@nalramli.com>
Co-developed-by: Joe Damato <jdamato@fastly.com>
Signed-off-by: Joe Damato <jdamato@fastly.com>
Signed-off-by: Rahul Rameshbabu <rrameshbabu@nvidia.com>
Signed-off-by: Tariq Toukan <tariqt@nvidia.com>
Link: https://lore.kernel.org/r/20240419080445.417574-3-tariqt@nvidia.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
|
|
Each queue counter object counts some events (in hardware) for the RQs
that are attached to it, like events of packet drops due to no receive
WQE (rx_out_of_buffer).
Each RQ can be attached to a queue counter only within the same vhca. To
still cover all RQs with these counters, we create multiple instances,
one per vhca.
The result that's shown to the user is now the sum of all instances.
Signed-off-by: Tariq Toukan <tariqt@nvidia.com>
Signed-off-by: Saeed Mahameed <saeedm@nvidia.com>
|
|
Distribute the channels between the different SD-devices to acheive
local numa node performance on multiple numas.
Each channel works against one specific mdev, creating all datapath
queues against it.
We distribute channels to mdevs in a round-robin policy.
Example for 2 mdevs and 6 channels:
+-------+---------+
| ch ix | mdev ix |
+-------+---------+
| 0 | 0 |
| 1 | 1 |
| 2 | 0 |
| 3 | 1 |
| 4 | 0 |
| 5 | 1 |
+-------+---------+
This round-robin distribution policy is preferred over another suggested
intuitive distribution, in which we first distribute one half of the
channels to mdev #0 and then the second half to mdev #1.
We prefer round-robin for a reason: it is less influenced by changes in
the number of channels. The mapping between channel index and mdev is
fixed, no matter how many channels the user configures. As the channel
stats are persistent to channels closure, changing the mapping every
single time would turn the accumulative stats less representing of the
channel's history.
Per-channel objects should stop using the primary mdev (priv->mdev)
directly, and instead move to using their own channel's mdev.
Signed-off-by: Tariq Toukan <tariqt@nvidia.com>
Reviewed-by: Gal Pressman <gal@nvidia.com>
Signed-off-by: Saeed Mahameed <saeedm@nvidia.com>
|
|
In the case of XDP Multi-Buffer with Striding RQ, an extra
page is allocated for the linear part of non-linear SKBs.
Including headroom and tailroom in the calculation may
result in an unnecessary increase in the amount of memory
allocated. This could be critical, particularly for large
MTUs (e.g. 7975B) and large RQ sizes (e.g. 8192).
In this case, the requested page pool size is 64K, but
32K would be sufficient. This causes a failure due to
exceeding the page pool size limit of 32K.
Exclude headroom and tailroom from SKB size calculations
to reduce page pool size.
Signed-off-by: Carolina Jubran <cjubran@nvidia.com>
Reviewed-by: Tariq Toukan <tariqt@nvidia.com>
Signed-off-by: Saeed Mahameed <saeedm@nvidia.com>
|
|
Packet data buffers lack reserved headroom or tailroom,
and SKBs are allocated on a side memory when needed.
Exclude the tailroom from the SKB size calculations.
Signed-off-by: Carolina Jubran <cjubran@nvidia.com>
Reviewed-by: Tariq Toukan <tariqt@nvidia.com>
Signed-off-by: Saeed Mahameed <saeedm@nvidia.com>
|
|
All ConnectX devices have software parsing capability enabled, but it is
more correct to set allow_swp only if capability exists, which for IPsec
means that crypto offload is supported.
Fixes: 2451da081a34 ("net/mlx5: Unify device IPsec capabilities check")
Signed-off-by: Leon Romanovsky <leonro@nvidia.com>
Signed-off-by: Saeed Mahameed <saeedm@nvidia.com>
|
|
Revert "net/mlx5: Implement management PF Ethernet profile"
This reverts commit 22c4640698a1d47606b5a4264a584e8046641784.
Revert "net/mlx5: Enable SD feature"
This reverts commit c88c49ac9c18fb7c3fa431126de1d8f8f555e912.
Revert "net/mlx5e: Block TLS device offload on combined SD netdev"
This reverts commit 83a59ce0057b7753d7fbece194b89622c663b2a6.
Revert "net/mlx5e: Support per-mdev queue counter"
This reverts commit d72baceb92539a178d2610b0e9ceb75706a75b55.
Revert "net/mlx5e: Support cross-vhca RSS"
This reverts commit c73a3ab8fa6e93a783bd563938d7cf00d62d5d34.
Revert "net/mlx5e: Let channels be SD-aware"
This reverts commit e4f9686bdee7b4dd89e0ed63cd03606e4bda4ced.
Revert "net/mlx5e: Create EN core HW resources for all secondary devices"
This reverts commit c4fb94aa822d6c9d05fc3c5aee35c7e339061dc1.
Revert "net/mlx5e: Create single netdev per SD group"
This reverts commit e2578b4f983cfcd47837bbe3bcdbf5920e50b2ad.
Revert "net/mlx5: SD, Add informative prints in kernel log"
This reverts commit c82d360325112ccc512fc11a3b68cdcdf04a1478.
Revert "net/mlx5: SD, Implement steering for primary and secondaries"
This reverts commit 605fcce33b2d1beb0139b6e5913fa0b2062116b2.
Revert "net/mlx5: SD, Implement devcom communication and primary election"
This reverts commit a45af9a96740873db9a4b5bb493ce2ad81ccb4d5.
Revert "net/mlx5: SD, Implement basic query and instantiation"
This reverts commit 63b9ce944c0e26c44c42cdd5095c2e9851c1a8ff.
Revert "net/mlx5: SD, Introduce SD lib"
This reverts commit 4a04a31f49320d078b8078e1da4b0e2faca5dfa3.
Revert "net/mlx5: Fix query of sd_group field"
This reverts commit e04984a37398b3f4f5a79c993b94c6b1224184cc.
Revert "net/mlx5e: Use the correct lag ports number when creating TISes"
This reverts commit a7e7b40c4bc115dbf2a2bb453d7bbb2e0ea99703.
There are some unanswered questions on the list, and we don't
have any docs. Given the lack of replies so far and the fact
that v6.8 merge window has started - let's revert this and
revisit for v6.9.
Link: https://lore.kernel.org/all/20231221005721.186607-1-saeed@kernel.org/
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
|
|
Each queue counter object counts some events (in hardware) for the RQs
that are attached to it, like events of packet drops due to no receive
WQE (rx_out_of_buffer).
Each RQ can be attached to a queue counter only within the same vhca. To
still cover all RQs with these counters, we create multiple instances,
one per vhca.
The result that's shown to the user is now the sum of all instances.
Signed-off-by: Tariq Toukan <tariqt@nvidia.com>
Signed-off-by: Saeed Mahameed <saeedm@nvidia.com>
|
|
Distribute the channels between the different SD-devices to acheive
local numa node performance on multiple numas.
Each channel works against one specific mdev, creating all datapath
queues against it.
We distribute channels to mdevs in a round-robin policy.
Example for 2 mdevs and 6 channels:
+-------+---------+
| ch ix | mdev ix |
+-------+---------+
| 0 | 0 |
| 1 | 1 |
| 2 | 0 |
| 3 | 1 |
| 4 | 0 |
| 5 | 1 |
+-------+---------+
This round-robin distribution policy is preferred over another suggested
intuitive distribution, in which we first distribute one half of the
channels to mdev #0 and then the second half to mdev #1.
We prefer round-robin for a reason: it is less influenced by changes in
the number of channels. The mapping between channel index and mdev is
fixed, no matter how many channels the user configures. As the channel
stats are persistent to channels closure, changing the mapping every
single time would turn the accumulative stats less representing of the
channel's history.
Per-channel objects should stop using the primary mdev (priv->mdev)
directly, and instead move to using their own channel's mdev.
Signed-off-by: Tariq Toukan <tariqt@nvidia.com>
Reviewed-by: Gal Pressman <gal@nvidia.com>
Signed-off-by: Saeed Mahameed <saeedm@nvidia.com>
|
|
Make CQ struct and methods independent of "priv", use more basic
arguments instead.
This will ease the transition to netdev with multiple mdevs.
Signed-off-by: Tariq Toukan <tariqt@nvidia.com>
Reviewed-by: Gal Pressman <gal@nvidia.com>
Signed-off-by: Saeed Mahameed <saeedm@nvidia.com>
|
|
Split types and pure function declarations from page_pool.h
and add them in page_page/types.h, so that C sources can
include page_pool.h and headers should generally only include
page_pool/types.h as suggested by jakub.
Rename page_pool.h to page_pool/helpers.h to have both in
one place.
Signed-off-by: Yunsheng Lin <linyunsheng@huawei.com>
Suggested-by: Jakub Kicinski <kuba@kernel.org>
Signed-off-by: Alexander Lobakin <aleksander.lobakin@intel.com>
Reviewed-by: Alexander Duyck <alexanderduyck@fb.com>
Link: https://lore.kernel.org/r/20230804180529.2483231-2-aleksander.lobakin@intel.com
[Jakub: change microsoft/mana, fix kdoc paths in Documentation]
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
|
|
The cited commits missed passing frag_size to __xdp_rxq_info_reg, which
is required by bpf_xdp_adjust_tail to support growing the tail pointer
in fragmented packets. Pass the missing parameter when the current RQ
mode allows XDP multi buffer.
Fixes: ea5d49bdae8b ("net/mlx5e: Add XDP multi buffer support to the non-linear legacy RQ")
Fixes: 9cb9482ef10e ("net/mlx5e: Use fragments of the same size in non-linear legacy RQ with XDP")
Signed-off-by: Maxim Mikityanskiy <maxtram95@gmail.com>
Cc: Tariq Toukan <tariqt@nvidia.com>
Reviewed-by: Tariq Toukan <tariqt@nvidia.com>
Signed-off-by: Saeed Mahameed <saeedm@nvidia.com>
|
|
Pull rdma updates from Jason Gunthorpe:
"Usual wide collection of unrelated items in drivers:
- Driver bug fixes and treewide cleanups in hfi1, siw, qib, mlx5,
rxe, usnic, usnic, bnxt_re, ocrdma, iser:
- remove unnecessary NULL checks
- kmap obsolescence
- pci_enable_pcie_error_reporting() obsolescence
- unused variables and macros
- trace event related warnings
- casting warnings
- Code cleanups for irdm and erdma
- EFA reporting of 128 byte PCIe TLP support
- mlx5 more agressively uses the out of order HW feature
- Big rework of how state machines and tasks work in rxe
- Fix a syzkaller found crash netdev refcount leak in siw
- bnxt_re revises their HW description header
- Congestion control for bnxt_re
- Use mmu_notifiers more safely in hfi1
- mlx5 gets better support for PCIe relaxed ordering inside VMs"
* tag 'for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/rdma/rdma: (81 commits)
RDMA/efa: Add rdma write capability to device caps
RDMA/mlx5: Use correct device num_ports when modify DC
RDMA/irdma: Drop spurious WQ_UNBOUND from alloc_ordered_workqueue() call
RDMA/rxe: Fix spinlock recursion deadlock on requester
RDMA/mlx5: Fix flow counter query via DEVX
RDMA/rxe: Protect QP state with qp->state_lock
RDMA/rxe: Move code to check if drained to subroutine
RDMA/rxe: Remove qp->req.state
RDMA/rxe: Remove qp->comp.state
RDMA/rxe: Remove qp->resp.state
RDMA/mlx5: Allow relaxed ordering read in VFs and VMs
net/mlx5: Update relaxed ordering read HCA capabilities
RDMA/mlx5: Check pcie_relaxed_ordering_enabled() in UMR
RDMA/mlx5: Remove pcie_relaxed_ordering_enabled() check for RO write
RDMA: Add ib_virt_dma_to_page()
RDMA/rxe: Fix the error "trying to register non-static key in rxe_cleanup_task"
RDMA/irdma: Slightly optimize irdma_form_ah_cm_frame()
RDMA/rxe: Fix incorrect TASKLET_STATE_SCHED check in rxe_task.c
IB/hfi1: Place struct mmu_rb_handler on cache line start
IB/hfi1: Fix bugs with non-PAGE_SIZE-end multi-iovec user SDMA requests
...
|
|
Here we add support for multi-buffer XDP handling in Striding RQ, which
is our default out-of-the-box RQ type. Before this series, loading such
an XDP program would fail, until you switch to the legacy RQ (by
unsetting the rx_striding_rq priv-flag).
To overcome the lack of headroom and tailroom between the strides, we
allocate a side page to be used for the descriptor (xdp_buff / skb) and
the linear part. When an XDP program is attached, we structure the
xdp_buff so that it contains no data in the linear part, and the whole
packet resides in the fragments.
In case of XDP_PASS, where an SKB still needs to be created, we copy up
to 256 bytes to its linear part, to match the current behavior, and
satisfy functions that assume finding the packet headers in the SKB
linear part (like eth_type_trans).
Performance testing:
Packet rate test, 64 bytes, 32 channels, MTU 9000 bytes.
CPU: Intel(R) Xeon(R) Platinum 8380 CPU @ 2.30GHz.
NIC: ConnectX-6 Dx, at 100 Gbps.
+----------+-------------+-------------+---------+
| Test | Legacy RQ | Striding RQ | Speedup |
+----------+-------------+-------------+---------+
| XDP_DROP | 101,615,544 | 117,191,020 | +15% |
+----------+-------------+-------------+---------+
| XDP_TX | 95,608,169 | 117,043,422 | +22% |
+----------+-------------+-------------+---------+
Reviewed-by: Saeed Mahameed <saeedm@nvidia.com>
Signed-off-by: Tariq Toukan <tariqt@nvidia.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
|
|
calculations
Function mlx5e_rx_get_linear_stride_sz() returns PAGE_SIZE immediately
in case an XDP program is attached. The more accurate formula is
ALIGN(sz, PAGE_SIZE), to prevent two packets from residing on the same
page.
The assumption behind the current code is that sz <= PAGE_SIZE holds for
all cases with XDP program set.
This is true because it is being called from:
- 3 times from Striding RQ flows, in which XDP is not supported for such
large packets.
- 1 time from Legacy RQ flow, under the condition
mlx5e_rx_is_linear_skb().
No functional change here, just removing the implied assumption in
preparation for supporting XDP multi-buffer in Striding RQ.
Reviewed-by: Saeed Mahameed <saeedm@nvidia.com>
Signed-off-by: Tariq Toukan <tariqt@nvidia.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
|
|
pcie_relaxed_ordering_enabled() check was added to avoid a syndrome when
creating a MKey with relaxed ordering (RO) enabled when the driver's
relaxed_ordering_{read,write} HCA capabilities are out of sync with FW.
While this can happen with relaxed_ordering_read, it can't happen with
relaxed_ordering_write as it's set if the device supports RO write,
regardless of RO in PCI config space, and thus can't change during
runtime.
Therefore, drop the pcie_relaxed_ordering_enabled() check for
relaxed_ordering_write while keeping it for relaxed_ordering_read.
Doing so will also allow the usage of RO write in VFs and VMs (where RO
in PCI config space is not reported/emulated properly).
Signed-off-by: Avihai Horon <avihaih@nvidia.com>
Reviewed-by: Shay Drory <shayd@nvidia.com>
Link: https://lore.kernel.org/r/7e8f55e31572c1702d69cae015a395d3a824a38a.1681131553.git.leon@kernel.org
Reviewed-by: Jacob Keller <jacob.e.keller@intel.com>
Signed-off-by: Leon Romanovsky <leon@kernel.org>
|
|
To avoid overflowing the page pool's cache, don't release the
whole bulk which is usually larger than the cache refill size.
Group release+alloc instead into cache refill units that
allow releasing to the cache and then allocating from the cache.
A refill_unit variable is added as a iteration unit over the
wqe_bulk when doing release+alloc.
For a single ring, single core, default MTU (1500) TCP stream
test the number of pages allocated from the cache directly
(rx_pp_recycle_cached) increases from 0% to 52%:
+---------------------------------------------+
| Page Pool stats (/sec) | Before | After |
+-------------------------+---------+---------+
|rx_pp_alloc_fast | 2145422 | 2193802 |
|rx_pp_alloc_slow | 2 | 0 |
|rx_pp_alloc_empty | 2 | 0 |
|rx_pp_alloc_refill | 34059 | 16634 |
|rx_pp_alloc_waive | 0 | 0 |
|rx_pp_recycle_cached | 0 | 1145818 |
|rx_pp_recycle_cache_full | 0 | 0 |
|rx_pp_recycle_ring | 2179361 | 1064616 |
|rx_pp_recycle_ring_full | 121 | 0 |
+---------------------------------------------+
With this patch, the performance for legacy rq for the above test is
back to baseline.
Signed-off-by: Dragos Tatulea <dtatulea@nvidia.com>
Signed-off-by: Saeed Mahameed <saeedm@nvidia.com>
|
|
Deferred page release was added to legacy rq but its desired effect
(driver releases last fragment to page pool cache) is not yet visible
due to the WQE bulks being too small.
This patch increases the WQE bulk size to span 512 KB and clip it to
one quarter of the rx queue size.
Signed-off-by: Dragos Tatulea <dtatulea@nvidia.com>
Signed-off-by: Saeed Mahameed <saeedm@nvidia.com>
|
|
Downstream patches require devlink params to access the PTYS register,
move the needed functions from mlx5e to the core layer.
Signed-off-by: Gal Pressman <gal@nvidia.com>
Reviewed-by: Tariq Toukan <tariqt@nvidia.com>
Signed-off-by: Saeed Mahameed <saeedm@nvidia.com>
Link: https://lore.kernel.org/r/20230314054234.267365-11-saeed@kernel.org
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
|
|
drivers/net/ipa/ipa_interrupt.c
drivers/net/ipa/ipa_interrupt.h
9ec9b2a30853 ("net: ipa: disable ipa interrupt during suspend")
8e461e1f092b ("net: ipa: introduce ipa_interrupt_enable()")
d50ed3558719 ("net: ipa: enable IPA interrupt handlers separate from registration")
https://lore.kernel.org/all/20230119114125.5182c7ab@canb.auug.org.au/
https://lore.kernel.org/all/79e46152-8043-a512-79d9-c3b905462774@tessares.net/
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
|
|
Add warning macro in the function mlx5e_mpwqe_get_log_num_strides()
when log WQE size is smaller than log stride size. Theoretically this
should not happen.
Signed-off-by: Adham Faris <afaris@nvidia.com>
Reviewed-by: Tariq Toukan <tariqt@nvidia.com>
Signed-off-by: Saeed Mahameed <saeedm@nvidia.com>
|
|
Current XSK prerequisites validation implementation
(setup.c/mlx5e_validate_xsk_param()) fails silently when xsk
prerequisites are not fulfilled.
Add error messages to the kernel log to help the user understand what
went wrong when params are not valid for XSK.
Signed-off-by: Adham Faris <afaris@nvidia.com>
Reviewed-by: Tariq Toukan <tariqt@nvidia.com>
Signed-off-by: Saeed Mahameed <saeedm@nvidia.com>
|
|
This validation function is relevant only for XSK cases, hence it
assumes to be called only with xsk != NULL.
Thus checking for invalid xsk pointer is redundant and misleads static
code analyzers.
This commit removes redundant xsk pointer check.
This solves the following smatch warning:
drivers/net/ethernet/mellanox/mlx5/core/en/params.c:481
mlx5e_mpwrq_validate_xsk() error: we previously assumed 'xsk' could be
null (see line 478)
Fixes: 6470d2e7e8ed ("net/mlx5e: xsk: Use KSM for unaligned XSK")
Signed-off-by: Adham Faris <afaris@nvidia.com>
Reported-by: kernel test robot <lkp@intel.com>
Reported-by: Dan Carpenter <error27@gmail.com>
Reviewed-by: Tariq Toukan <tariqt@nvidia.com>
Signed-off-by: Saeed Mahameed <saeedm@nvidia.com>
|
|
Per the device spec, MLX5_UMR_MTT_ALIGNMENT is good not only for UMR MTT
entries, but for all other entries as well, like KLMs and KSMs.
Signed-off-by: Tariq Toukan <tariqt@nvidia.com>
Reviewed-by: Gal Pressman <gal@nvidia.com>
Signed-off-by: Saeed Mahameed <saeedm@nvidia.com>
|
|
Params info print was meant to be printed on load.
With time, new calls to mlx5e_init_rq_type_params and
mlx5e_build_rq_params were added, mistakenly printing
the params once again.
Move the print to were it belongs, in mlx5e_probe.
Signed-off-by: Tariq Toukan <tariqt@nvidia.com>
Reviewed-by: Gal Pressman <gal@nvidia.com>
Signed-off-by: Saeed Mahameed <saeedm@nvidia.com>
|
|
CQE compression feature improves performance by reducing PCI bandwidth
bottleneck on CQEs write.
Enhanced CQE compression introduced in ConnectX-6 and it aims to reduce
CPU utilization of SW side packets decompression by eliminating the
need to rewrite ownership bit, which is likely to cost a cache-miss, is
replaced by validity byte handled solely by HW.
Another advantage of the enhanced feature is that session packets are
available to SW as soon as a single CQE slot is filled, instead of
waiting for session to close, this improves packet latency from NIC to
host.
Performance:
Following are tested scenarios and reults comparing basic and enahnced
CQE compression.
setup: IXIA 100GbE connected directly to port 0 and port 1 of
ConnectX-6 Dx 100GbE dual port.
Case #1 RX only, single flow goes to single queue:
IRQ rate reduced by ~ 30%, CPU utilization improved by 2%.
Case #2 IP forwarding from port 1 to port 0 single flow goes to
single queue:
Avg latency improved from 60us to 21us, frame loss improved from 0.5% to 0.0%.
Case #3 IP forwarding from port 1 to port 0 Max Throughput IXIA sends
100%, 8192 UDP flows, goes to 24 queues:
Enhanced is equal or slightly better than basic.
Testing the basic compression feature with this patch shows there is
no perfrormance degradation of the basic compression feature.
Signed-off-by: Ofer Levi <oferle@nvidia.com>
Reviewed-by: Tariq Toukan <tariqt@nvidia.com>
Signed-off-by: Saeed Mahameed <saeedm@nvidia.com>
|
|
When XSK frame size is 3072 (or another power of two multiplied by 3),
KLM mechanism for NIC virtual memory page mapping can be optimized by
replacing it with KSM.
Before this change, two KLM entries were needed to map an XSK frame that
is not a power of two: one entry maps the UMEM memory up to the frame
length, the other maps the rest of the stride to the garbage page.
When the frame length divided by 3 is a power of two, it can be mapped
using 3 KSM entries, and the fourth will map the rest of the stride to
the garbage page. All 4 KSM entries are of the same size, which allows
for a much faster lookup.
Frame size 3072 is useful in certain use cases, because it allows
packing 4 frames into 3 pages. Generally speaking, other frame sizes
equal to PAGE_SIZE minus a power of two can be optimized in a similar
way, but it will require many more KSMs per frame, which slows down UMRs
a little bit, but more importantly may hit the limit for the maximum
number of KSM entries.
Signed-off-by: Maxim Mikityanskiy <maximmi@nvidia.com>
Reviewed-by: Tariq Toukan <tariqt@nvidia.com>
Signed-off-by: Saeed Mahameed <saeedm@nvidia.com>
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
|
|
XSK RQs support striding RQ linear mode, but the stride size may be
bigger than the XSK frame size, because:
1. The stride size must be a power of two.
2. The stride size must be equal to the UMR page size. Each XSK frame is
treated as a separate page, because they aren't necessarily adjacent in
physical memory, so the driver can't put more than one stride per page.
3. The minimal MTT page size is 4096 on older firmware.
That means that if XSK frame size is 2048 or not a power of two, the
strides may be bigger than XSK frames. Normally, it's not a problem if
the hardware enforces the MTU. However, traffic between vports skips the
hardware MTU check, and oversized packets may be received.
If an oversized packet is bigger than the XSK frame but not bigger than
the stride, it will cause overwriting of the adjacent UMEM region. If
the packet takes more than one stride, they can be recycled for reuse,
so it's not a problem when the XSK frame size matches the stride size.
Work around the above issue by leveraging KLM to make a more
fine-grained mapping. The beginning of each stride is mapped to the
frame memory, and the padding up to the closest power of two is mapped
to the overflow page that doesn't belong to UMEM. This way, application
data corruption won't happen upon receiving packets bigger than MTU.
Signed-off-by: Maxim Mikityanskiy <maximmi@nvidia.com>
Reviewed-by: Tariq Toukan <tariqt@nvidia.com>
Signed-off-by: Saeed Mahameed <saeedm@nvidia.com>
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
|
|
Make mlx5e_mpwrq_mtts_per_wqe take into account that KSM requires
smaller alignment than MTT.
Ensure that there is always an even amount of MTTs in a UMR WQE, so that
complete octwords are formed, and no garbage is mapped.
Drop extra alignment in MLX5_MTT_OCTW that may cause setting too big
ucseg->xlt_octowords, also leading to mapping garbage.
Generalize some calculations by introducing the MLX5_OCTWORD constant.
Signed-off-by: Maxim Mikityanskiy <maximmi@nvidia.com>
Reviewed-by: Tariq Toukan <tariqt@nvidia.com>
Signed-off-by: Saeed Mahameed <saeedm@nvidia.com>
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
|
|
Instead of passing the unaligned flag, pass an enum that indicates the
UMR mode. The next commit will add the third mode (KLM for certain
configurations of XSK), which will be added to this enum instead of
adding another bool flag everywhere.
Signed-off-by: Maxim Mikityanskiy <maximmi@nvidia.com>
Reviewed-by: Tariq Toukan <tariqt@nvidia.com>
Signed-off-by: Saeed Mahameed <saeedm@nvidia.com>
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
|
|
The old calculation of wqe_index_mask may give false positives, i.e.
request bulking of pairs of WQEs when not strictly needed, for example,
when the first fragment size is equal to the PAGE_SIZE, bulking is not
needed, even if the number of fragments is odd.
Make the calculation more exact to cut false positives.
Signed-off-by: Maxim Mikityanskiy <maximmi@nvidia.com>
Reviewed-by: Tariq Toukan <tariqt@nvidia.com>
Signed-off-by: Saeed Mahameed <saeedm@nvidia.com>
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
|
|
When fragments of different WQEs share the same page, mlx5e_post_rx_wqes
must wait until the old WQE stops using the page, only then the new WQE
can allocate the new page. Essentially, it means that if WQE index i is
still in use, the allocation must stop before `i % bulk`, where bulk is
the number of WQEs that may share the same page.
As bulk is always a power of two, `i % bulk = i & (bulk - 1)`, and the
new wqe_index_mask field will be equal to `bulk - 1`.
At the same time, wqe_bulk remains for optimization purposes and stores
`max(bulk, 8)`, which allows to skip the allocation until we have at
least 8 WQEs free.
Signed-off-by: Maxim Mikityanskiy <maximmi@nvidia.com>
Reviewed-by: Tariq Toukan <tariqt@nvidia.com>
Signed-off-by: Saeed Mahameed <saeedm@nvidia.com>
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
|
|
WQEs must not cross page boundaries, they are padded with NOPs if they
don't fit the page. mlx5e_mpwrq_total_umr_wqebbs doesn't take into
account this padding, risking reserving not enough space.
The padding is not straightforward to add to this calculation, because
WQEs of different sizes may be mixed together in the queue. If each page
ends with a big WQE that doesn't fit and requires at most its size minus
1 WQEBB of padding, the total space can be much bigger than in case when
smaller WQEs take advantage of this padding.
Replace the wrong exact calculation by the following estimation. Each
padding can be at most the size of the maximum WQE used in the queue
minus one WQEBB. Let's call the rest of the page "useful space". If we
divide the total size of all needed WQEs by this useful space, rounding
up, we'll get the number of pages, which is enough to contain all these
WQEs. It's correct, because every WQE that appeared on the boundary
between two blocks of useful space would start in the useful space of
one page and end in the padding of the same page, while our estimation
reserved space for its tail in the next space, making the estimation not
smaller than the real space occupied in the queue.
The code actually uses a looser estimation: instead of taking the
maximum size of all used WQE types minus 1 WQEBB, it takes the maximum
hardware size of a WQE. It's made for simplicity and extensibility.
Signed-off-by: Maxim Mikityanskiy <maximmi@nvidia.com>
Reviewed-by: Saeed Mahameed <saeedm@nvidia.com>
Reviewed-by: Tariq Toukan <tariqt@nvidia.com>
Signed-off-by: Saeed Mahameed <saeedm@nvidia.com>
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
|
|
UMR MTTs used in striding RQ have certain alignment requirements. While
it's guaranteed to work when UMR pages are aligned to the UMR page size,
in practice it works then UMR pages are aligned to 8 bytes. However,
it's still not enough flexibility for the unaligned mode of XSK. This
patch leverages KSM to map UMR pages without alignment requirements,
when unaligned XSK is active. The downside is that KSM entries are twice
as big as MTTs, which limits the maximum WQE size, so regular RQs and
aligned XSK continue using MTTs.
Signed-off-by: Maxim Mikityanskiy <maximmi@nvidia.com>
Reviewed-by: Tariq Toukan <tariqt@nvidia.com>
Signed-off-by: Saeed Mahameed <saeedm@nvidia.com>
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
|
|
XSK RQs support striding RQ linear mode, but the stride size is always
set to PAGE_SIZE. It may be larger than the XSK frame size,
unnecessarily reducing the useful space in a WQE, but more importantly
causing UMEM data corruption in certain cases.
Normally, stride size bigger than XSK frame size is not a problem if the
hardware enforces the MTU. However, traffic between vports skips the
hardware MTU check, and oversized packets may be received.
If an oversized packet is bigger than the XSK frame but not bigger than
the stride, it will cause overwriting of the adjacent UMEM region. If
the packet takes more than one stride, they can be recycled for reuse
so it's not a problem when the XSK frame size matches the stride size.
To reduce the impact of the above issue, attempt to use the MTT page
size for striding RQ that matches the XSK frame size, allowing to safely
use 2048-byte frames on an up-to-date firmware.
Signed-off-by: Maxim Mikityanskiy <maximmi@nvidia.com>
Reviewed-by: Tariq Toukan <tariqt@nvidia.com>
Signed-off-by: Saeed Mahameed <saeedm@nvidia.com>
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
|
|
This commit allows striding RQ to determine MTT page size at runtime,
instead of sticking to the compile-time PAGE_SIZE. This functionality
will be used by a following commit that adjusts the MTT page size to the
XSK frame size.
Stick with PAGE_SIZE for XSK on legacy RQ, as frag_stride is not used in
data path, it only helps calculate how pages are partitioned into
fragments, and PAGE_SIZE will ensure each fragment starts at the
beginning of a new allocation unit (XSK frame).
Signed-off-by: Maxim Mikityanskiy <maximmi@nvidia.com>
Reviewed-by: Tariq Toukan <tariqt@nvidia.com>
Signed-off-by: Saeed Mahameed <saeedm@nvidia.com>
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
|
|
Some of the parameters of striding RQ are compile-time constants, but
they are going to become dynamically calculated at runtime in a
following commit. This commit prepares the datapath to take cached
runtime parameters, prefilled at queue creation.
New fields added to struct mlx5e_rq fit into an existing 7-byte hole.
Signed-off-by: Maxim Mikityanskiy <maximmi@nvidia.com>
Reviewed-by: Tariq Toukan <tariqt@nvidia.com>
Reviewed-by: Saeed Mahameed <saeedm@nvidia.com>
Signed-off-by: Saeed Mahameed <saeedm@nvidia.com>
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
|
|
Normally, the MTU change requires reopening the channels, but it can be
skipped if the new MTU doesn't change any of the queue parameters and if
MTU is not used in the data path.
The shortcut is applicable to the non-linear mode of striding RQ,
because the only thing affected by MTU is the queue length. As ethtool
sets the queue length in packets, but striding RQ length is defined in
strides or bytes, we estimate the RQ length to be at least as big as the
requested number of MTU-sized packets, that's why it depends on MTU.
Improve the shortcut by actually checking whether the RQ length stayed
the same, instead of an intermediate step in the calculation.
As MTU also affects the SHAMPO parameters, skip the shortcut if SHAMPO
is in use.
Signed-off-by: Maxim Mikityanskiy <maximmi@nvidia.com>
Reviewed-by: Tariq Toukan <tariqt@nvidia.com>
Reviewed-by: Saeed Mahameed <saeedm@nvidia.com>
Signed-off-by: Saeed Mahameed <saeedm@nvidia.com>
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
|
|
In a typical scenario, if an XSK socket is opened first, then an XDP
program is attached, mlx5e_validate_xsk_param will be called twice:
first on XSK bind, second on channel restart caused by enabling XDP. The
validation includes a call to mlx5e_rx_is_linear_skb, which checks the
presence of the XDP program.
The above means that mlx5e_rx_is_linear_skb might return true the first
time, but false the second time, as mlx5e_rx_get_linear_sz_skb's return
value will increase, because of a different headroom used with XDP.
As XSK RQs never exist without XDP, it would make sense to trick
mlx5e_rx_get_linear_sz_skb into thinking XDP is enabled at the first
check as well. This way, if MTU is too big, it would be detected on XSK
bind, without giving false hope to the userspace application.
However, it turns out that this check is too restrictive in the first
place. SKBs created on XDP_PASS on XSK RQs don't have any headroom. That
means that big MTUs filtered out on the first and the second checks
might actually work.
So, address this issue in the proper way, but taking into account the
absence of the SKB headroom on XSK RQs, when calculating the buffer
size.
Signed-off-by: Maxim Mikityanskiy <maximmi@nvidia.com>
Reviewed-by: Tariq Toukan <tariqt@nvidia.com>
Reviewed-by: Saeed Mahameed <saeedm@nvidia.com>
Signed-off-by: Saeed Mahameed <saeedm@nvidia.com>
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
|
|
One of the checks in mlx5e_rx_is_linear_skb verifies that the RX buffer
fits into the XSK frame size. Remove the duplicating check from
mlx5e_validate_xsk_param. It allows to make mlx5e_rx_get_min_frag_sz
static.
Remove mlx5e_rx_is_xdp altogether, as its only usage is located in a
branch where xsk == NULL.
Signed-off-by: Maxim Mikityanskiy <maximmi@nvidia.com>
Reviewed-by: Tariq Toukan <tariqt@nvidia.com>
Reviewed-by: Saeed Mahameed <saeedm@nvidia.com>
Signed-off-by: Saeed Mahameed <saeedm@nvidia.com>
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
|
|
Linear RX buffers must be big enough to fit the MTU-sized packet along
with the headroom. On the other hand, they must be small enough to fit
into a page (or into an XSK frame). A straightforward way to check
whether the linear mode is possible would be comparing the required
buffer size to PAGE_SIZE or XSK frame size.
Stride size in the linear mode is defined by the following constraints:
1. A stride is at least as big as the buffer size, and it's a power of
two.
2. If non-XSK XDP is enabled, the stride size is PAGE_SIZE, because
mlx5e requires each packet to be in its own page when XDP is in use. The
previous constraint is automatically fulfilled, because buffer size
can't be bigger than PAGE_SIZE.
3. XSK uses stride size equal to PAGE_SIZE, but the following commits
will allow it to use roundup_pow_of_two(XSK frame size), by allowing the
NIC's MMU to use page sizes not equal to the CPU page size.
This commit puts the above requirements and constraints straight to the
code in an attempt to simplify it and to prepare it for changes made in
the next patches.
For the reference, the old code uses an equivalent, but trickier
calculation (high-level simplified pseudocode):
if XDP or XSK:
mlx5e_rx_get_linear_frag_sz := max(buffer size, PAGE_SIZE)
else:
mlx5e_rx_get_linear_frag_sz := buffer size
mlx5e_rx_is_linear_skb := mlx5e_rx_get_linear_frag_sz <= PAGE_SIZE
stride size := roundup_pow_of_two(mlx5e_rx_get_linear_frag_sz)
The new code effectively removes mlx5e_rx_get_linear_frag_sz that used
to return either buffer size or stride size, depending on the situation,
making it hard to work with and to make changes:
if XDP or XSK:
mlx5e_rx_get_linear_stride_sz := PAGE_SIZE
else
mlx5e_rx_get_linear_stride_sz := roundup_pow_of_two(buffer size)
mlx5e_rx_is_linear_skb := buffer size <= (PAGE_SIZE or XSK frame sz)
stride size := mlx5e_rx_get_linear_stride_sz
Signed-off-by: Maxim Mikityanskiy <maximmi@nvidia.com>
Reviewed-by: Tariq Toukan <tariqt@nvidia.com>
Reviewed-by: Saeed Mahameed <saeedm@nvidia.com>
Signed-off-by: Saeed Mahameed <saeedm@nvidia.com>
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
|
|
TX MPWQE size is limited to the cacheline-aligned maximum. Use the same
value for the stop room and the capability check.
Signed-off-by: Maxim Mikityanskiy <maximmi@nvidia.com>
Reviewed-by: Tariq Toukan <tariqt@nvidia.com>
Reviewed-by: Saeed Mahameed <saeedm@nvidia.com>
Signed-off-by: Saeed Mahameed <saeedm@nvidia.com>
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
|
|
Currently, the driver can silently fall back to legacy RQ after enabling
XDP, even if striding RQ was active before. It happens when PAGE_SIZE is
bigger than the maximum supported stride size. This commit changes this
behavior to more straightforward: if an operation (enabling XDP) doesn't
support the current parameters (striding RQ mode), it fails.
Signed-off-by: Maxim Mikityanskiy <maximmi@nvidia.com>
Reviewed-by: Tariq Toukan <tariqt@nvidia.com>
Reviewed-by: Saeed Mahameed <saeedm@nvidia.com>
Signed-off-by: Saeed Mahameed <saeedm@nvidia.com>
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
|
|
mlx5e_verify_rx_mpwqe_strides is only used in en/params.c, so it can be
made static and removed from en/params.h.
Signed-off-by: Maxim Mikityanskiy <maximmi@nvidia.com>
Reviewed-by: Tariq Toukan <tariqt@nvidia.com>
Reviewed-by: Saeed Mahameed <saeedm@nvidia.com>
Signed-off-by: Saeed Mahameed <saeedm@nvidia.com>
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
|
|
ICOSQ is used to post UMR WQEs for both regular RQ and XSK RQ. However,
space in ICOSQ is reserved only for the regular RQ, which may cause
ICOSQ overflows when using XSK (the most risk is on activating
channels).
This commit fixes the issue by reserving space for XSK UMR WQEs as well.
As XSK may be enabled without restarting the channel and recreating the
ICOSQ, this space is reserved unconditionally.
Fixes: db05815b36cb ("net/mlx5e: Add XSK zero-copy support")
Signed-off-by: Maxim Mikityanskiy <maximmi@nvidia.com>
Reviewed-by: Tariq Toukan <tariqt@nvidia.com>
Signed-off-by: Saeed Mahameed <saeedm@nvidia.com>
|
|
FW is not ready, fix was sent too soon.
This reverts commit f05ec8d9d0d62367b6e1f2cb50d7d2a45e7747cf.
Fixes: f05ec8d9d0d6 ("net/mlx5e: Allow relaxed ordering over VFs")
Signed-off-by: Saeed Mahameed <saeedm@nvidia.com>
|
|
By PCI spec, the config space of the VF always report relaxed ordering
not supported while it inherits this property from its PF. Hence using
pcie_relaxed_ordering_enable(), always disables the relaxed ordering on
all VFs. Remove this check and rely on the firmware which queries the
config space of the PF and set the capability bit accordingly.
Signed-off-by: Aya Levin <ayal@nvidia.com>
Reviewed-by: Gal Pressman <gal@nvidia.com>
Reviewed-by: Marina Varshaver <marinav@nvidia.com>
Reviewed-by: Gal Shalom <galshalom@nvidia.com>
Signed-off-by: Saeed Mahameed <saeedm@nvidia.com>
|
|
The mlx5 IPsec code has logical separation between code that operates
with XFRM objects (ipsec.c), HW objects (ipsec_offload.c), flow steering
logic (ipsec_fs.c) and data path (ipsec_rxtx.c).
Such separation makes sense for C-files, but isn't needed at all for
H-files as they are included in batch anyway.
Reviewed-by: Raed Salem <raeds@nvidia.com>
Signed-off-by: Leon Romanovsky <leonro@nvidia.com>
Signed-off-by: Saeed Mahameed <saeedm@nvidia.com>
|