Using examples of other driver(s), add the ability to hot-swap an XDP
program without having to reconfigure the queues. To prevent the
q->xdp_prog to be read/written more than once use READ_ONCE() and
WRITE_ONCE() on the q->xdp_prog.
The q->xdp_prog was being checked in multiple different for loops in the
hot path. The change to allow xdp_prog hot swapping created the
possibility for many READ_ONCE(q->xdp_prog) calls during a single napi
callback. Refactor the Rx napi handling to allow a previous
READ_ONCE(q->xdp_prog) (or NULL for hwstamp_rxq) to be passed into the
relevant functions.
Also, move other Rx related hotpath handling into the newly created
ionic_rx_cq_service() function to reduce the scope of the xdp_prog
local variable and put all Rx handling in one function similar to Tx.
Signed-off-by: Brett Creeley <brett.creeley@amd.com>
Signed-off-by: Shannon Nelson <shannon.nelson@amd.com>
Link: https://patch.msgid.link/20240906232623.39651-8-brett.creeley@amd.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
Our home-grown buffer management needs to go away and we need
to be playing nicely with the page_pool infrastructure. This
converts the Rx traffic queues to use page_pool.
Also, since ionic_rx_buf_size() was removed, redefine
IONIC_PAGE_SIZE to account for IONIC_MAX_BUF_LEN being the
largest allowed buffer to prevent overflowing u16 variables,
which could happen when PAGE_SIZE is defined as >= 64KB.
include/linux/minmax.h:93:37: warning: conversion from 'long unsigned int' to 'u16' {aka 'short unsigned int'} changes value from '65536' to '0' [-Woverflow]
Signed-off-by: Shannon Nelson <shannon.nelson@amd.com>
Signed-off-by: Brett Creeley <brett.creeley@amd.com>
Link: https://patch.msgid.link/20240906232623.39651-7-brett.creeley@amd.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
Currently when going to/from a NULL XDP program the driver uses
ionic_stop_queues_reconfig() and then ionic_start_queues_reconfig() in
order to re-register the xdp_rxq_info and re-init the queues. This is
fine until page_pool(s) are used in an upcoming patch.
In preparation for adding page_pool support make sure to completely
rebuild the queues when going to/from a NULL XDP program. Without this
change the call to mem_allocator_disconnect() never happens when going
to a NULL XDP program, which eventually results in
xdp_rxq_info_reg_mem_model() failing with -ENOSPC due to the mem_id_pool
ida having no remaining space.
Signed-off-by: Brett Creeley <brett.creeley@amd.com>
Signed-off-by: Shannon Nelson <shannon.nelson@amd.com>
Link: https://patch.msgid.link/20240906232623.39651-6-brett.creeley@amd.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
Instead of setting up and tearing down the rxq_info only when the XDP
program is loaded or unloaded, we will build the rxq_info whether or not
XDP is in use. This is the more common use pattern and better supports
future conversion to page_pool. Since the rxq_info wants the napi_id
we re-order things slightly to tie this into the queue init and deinit
functions where we do the add and delete of napi.
Signed-off-by: Shannon Nelson <shannon.nelson@amd.com>
Signed-off-by: Brett Creeley <brett.creeley@amd.com>
Link: https://patch.msgid.link/20240906232623.39651-5-brett.creeley@amd.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
We originally were using a per-interface xdp_prog variable to track
a loaded XDP program since we knew there would never be support for a
per-queue XDP program. With that, we only built the per queue rxq_info
struct when an XDP program was loaded and removed it on XDP program unload,
and used the pointer as an indicator in the Rx hotpath to know to how build
the buffers. However, that's really not the model generally used, and
makes a conversion to page_pool Rx buffer cacheing a little problematic.
This patch converts the driver to use the more common approach of using
a per-queue xdp_prog pointer to work out buffer allocations and need
for bpf_prog_run_xdp(). We jostle a couple of fields in the queue struct
in order to keep the new xdp_prog pointer in a warm cacheline.
Signed-off-by: Shannon Nelson <shannon.nelson@amd.com>
Signed-off-by: Brett Creeley <brett.creeley@amd.com>
Link: https://patch.msgid.link/20240906232623.39651-4-brett.creeley@amd.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
With recent work to the doorbell workaround code a small hole was
introduced that could cause a tx_timeout. This happens if the rx
dbell_deadline goes beyond the netdev watchdog timeout set by the driver
(i.e. 2 seconds). Fix this by changing the netdev watchdog timeout to 5
seconds and reduce the max rx dbell_deadline to 4 seconds.
The test that can reproduce the issue being fixed is a multi-queue send
test via pktgen with the "burst" setting to 1. This causes the queue's
doorbell to be rung on every packet sent to the driver, which may result
in the device missing doorbells due to the high doorbell rate.
Cc: stable@vger.kernel.org
Fixes: 4ded136c78 ("ionic: add work item for missed-doorbell check")
Signed-off-by: Brett Creeley <brett.creeley@amd.com>
Reviewed-by: Shannon Nelson <shannon.nelson@amd.com>
Link: https://patch.msgid.link/20240822192557.9089-1-brett.creeley@amd.com
Signed-off-by: Paolo Abeni <pabeni@redhat.com>
Cross-merge networking fixes after downstream PR.
No conflicts.
Adjacent changes:
e3f02f32a0 ("ionic: fix kernel panic due to multi-buffer handling")
d9c0420999 ("ionic: Mark error paths in the data path as unlikely")
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
If we're not in a NAPI softirq context, we need to be careful
about how we call napi_consume_skb(), specifically we need to
call it with budget==0 to signal to it that we're not in a
safe context.
This was found while running some configuration stress testing
of traffic and a change queue config loop running, and this
curious note popped out:
[ 4371.402645] BUG: using smp_processor_id() in preemptible [00000000] code: ethtool/20545
[ 4371.402897] caller is napi_skb_cache_put+0x16/0x80
[ 4371.403120] CPU: 25 PID: 20545 Comm: ethtool Kdump: loaded Tainted: G OE 6.10.0-rc3-netnext+ #8
[ 4371.403302] Hardware name: HPE ProLiant DL360 Gen10/ProLiant DL360 Gen10, BIOS U32 01/23/2021
[ 4371.403460] Call Trace:
[ 4371.403613] <TASK>
[ 4371.403758] dump_stack_lvl+0x4f/0x70
[ 4371.403904] check_preemption_disabled+0xc1/0xe0
[ 4371.404051] napi_skb_cache_put+0x16/0x80
[ 4371.404199] ionic_tx_clean+0x18a/0x240 [ionic]
[ 4371.404354] ionic_tx_cq_service+0xc4/0x200 [ionic]
[ 4371.404505] ionic_tx_flush+0x15/0x70 [ionic]
[ 4371.404653] ? ionic_lif_qcq_deinit.isra.23+0x5b/0x70 [ionic]
[ 4371.404805] ionic_txrx_deinit+0x71/0x190 [ionic]
[ 4371.404956] ionic_reconfigure_queues+0x5f5/0xff0 [ionic]
[ 4371.405111] ionic_set_ringparam+0x2e8/0x3e0 [ionic]
[ 4371.405265] ethnl_set_rings+0x1f1/0x300
[ 4371.405418] ethnl_default_set_doit+0xbb/0x160
[ 4371.405571] genl_family_rcv_msg_doit+0xff/0x130
[...]
I found that ionic_tx_clean() calls napi_consume_skb() which calls
napi_skb_cache_put(), but before that last call is the note
/* Zero budget indicate non-NAPI context called us, like netpoll */
and
DEBUG_NET_WARN_ON_ONCE(!in_softirq());
Those are pretty big hints that we're doing it wrong. We can pass a
context hint down through the calls to let ionic_tx_clean() know what
we're doing so it can call napi_consume_skb() correctly.
Fixes: 386e698653 ("ionic: Make use napi_consume_skb")
Signed-off-by: Shannon Nelson <shannon.nelson@amd.com>
Link: https://patch.msgid.link/20240624175015.4520-1-shannon.nelson@amd.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
Add a work item for each queue that will be run on the queue's
preferred cpu and will schedule another napi. This napi is
run in case the device missed a doorbell and didn't process
a packet. This is a problem for the Elba asic that happens
very rarely.
Signed-off-by: Shannon Nelson <shannon.nelson@amd.com>
Link: https://lore.kernel.org/r/20240619003257.6138-6-shannon.nelson@amd.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
Instead of using the system's default workqueue, add a private
workqueue for the device to use for its little jobs. This is
to better support the new work items we will be adding in the
next patches for PF and VF specific jobs, without inundating
the system workqueue in a couple of customer cases where our
devices get scaled out to 100-200 VFs.
Signed-off-by: Brett Creeley <brett.creeley@amd.com>
Signed-off-by: Shannon Nelson <shannon.nelson@amd.com>
Link: https://lore.kernel.org/r/20240619003257.6138-4-shannon.nelson@amd.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
Currently the driver either sets the initial interrupt affinity for its
adminq and tx/rx queues on probe or resets it on various
down/up/reconfigure flows. If any user and/or user process
(i.e. irqbalance) changes IRQ affinity for any of the driver's interrupts
that will be reset to driver defaults whenever any down/up/reconfigure
operation happens. This is incorrect and is fixed by making 2 changes:
1. Allocate an array of cpumasks that's only allocated on probe and
destroyed on remove.
2. Update the cpumask(s) for interrupts that are in use by registering
for affinity notifiers.
Signed-off-by: Brett Creeley <brett.creeley@amd.com>
Signed-off-by: Shannon Nelson <shannon.nelson@amd.com>
Link: https://lore.kernel.org/r/20240619003257.6138-3-shannon.nelson@amd.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
Recovery from broken states can be hard. If the LIF reset in
the fw_down path didn't work because the PCI link was broken,
the FW won't be in the right state for proper restart. We can
fire another LIF reset in the fw_up path to be sure things
are clean on restart.
Signed-off-by: Shannon Nelson <shannon.nelson@amd.com>
Link: https://lore.kernel.org/r/20240529000259.25775-3-shannon.nelson@amd.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
Make desc_info structure specific to the queue type, which
allows us to cut down the Rx and AdminQ descriptor sizes by
not including all the fields needed for the Tx desriptors.
Before:
struct ionic_desc_info {
/* size: 464, cachelines: 8, members: 6 */
After:
struct ionic_tx_desc_info {
/* size: 464, cachelines: 8, members: 6 */
struct ionic_rx_desc_info {
/* size: 224, cachelines: 4, members: 2 */
struct ionic_admin_desc_info {
/* size: 8, cachelines: 1, members: 1 */
Suggested-by: Neel Patel <npatel2@amd.com>
Reviewed-by: Brett Creeley <brett.creeley@amd.com>
Signed-off-by: Shannon Nelson <shannon.nelson@amd.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
With a little simple math we don't need another struct array to
find the completion structs, so we can remove the ionic_cq_info
altogether. This doesn't really save anything in the ionic_cq
since it gets padded out to the cacheline, but it does remove
the parallel array allocation of 8 * num_descriptors, or about
8 Kbytes per queue in a default configuration.
Suggested-by: Neel Patel <npatel2@amd.com>
Reviewed-by: Brett Creeley <brett.creeley@amd.com>
Signed-off-by: Shannon Nelson <shannon.nelson@amd.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
By reworking the queue service routines to have their own
servicing loops we can remove the cb pointer from desc_info
to save another 8 bytes per descriptor,
This simplifies some of the queue handling indirection and makes
the code a little easier to follow, and keeps service code in
one place rather than jumping between code files.
struct ionic_desc_info
Before: /* size: 472, cachelines: 8, members: 7 */
After: /* size: 464, cachelines: 8, members: 6 */
Suggested-by: Neel Patel <npatel2@amd.com>
Reviewed-by: Brett Creeley <brett.creeley@amd.com>
Signed-off-by: Shannon Nelson <shannon.nelson@amd.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Now that we're not using desc_info pointers mapped in every q
we can simplify and drop the unnecessary utility functions.
Reviewed-by: Brett Creeley <brett.creeley@amd.com>
Signed-off-by: Shannon Nelson <shannon.nelson@amd.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Use the kernel's CQE dim table to align better with the
driver's use of completion queues, and use the tx moderation
when using Tx interrupts.
Signed-off-by: Brett Creeley <brett.creeley@amd.com>
Signed-off-by: Shannon Nelson <shannon.nelson@amd.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Currently the driver attempts to wake the Tx queue
for every descriptor processed. However, this is
overkill and can cause thrashing since Tx xmit can be
running concurrently on a different CPU than Tx clean.
Fix this by refactoring Tx cq servicing into its own
function so the Tx wake code can run after processing
all Tx descriptors.
The driver isn't using the expected memory barriers
to make sure the stop/start bits are coherent. Fix
this by making sure to use the correct memory barriers.
Also, the driver is using the wake API during Tx
xmit even though it's already scheduled. Fix this by
using the start API during Tx xmit.
Signed-off-by: Brett Creeley <brett.creeley@amd.com>
Signed-off-by: Shannon Nelson <shannon.nelson@amd.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
When rebuilding the lif after an FLR, be sure to restore the
current netdev features, not do the usual first time feature
init. This prevents losing user changes to things like TSO
or vlan tagging states.
Fixes: 45b84188a0 ("ionic: keep filters across FLR")
Reviewed-by: Brett Creeley <brett.creeley@amd.com>
Signed-off-by: Shannon Nelson <shannon.nelson@amd.com>
Signed-off-by: Paolo Abeni <pabeni@redhat.com>
The XDP_REDIRECT packets are given to the XDP stack and
we drop the use of the related page: it will get freed
by the driver that ends up doing the Tx. Because we have
some hardware configurations with limited queue resources,
we use the existing datapath Tx queues rather than creating
and managing a separate set of xdp_tx queues.
Co-developed-by: Brett Creeley <brett.creeley@amd.com>
Signed-off-by: Brett Creeley <brett.creeley@amd.com>
Signed-off-by: Shannon Nelson <shannon.nelson@amd.com>
Reviewed-by: Jacob Keller <jacob.e.keller@intel.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Set up the basics for running Rx packets through XDP programs.
Add new queue setup and teardown steps for adding/removing an
XDP program, and add the call to run the XDP on a packet.
The XDP frame size needs to be the MTU plus standard ethernet
header, plus head room for XDP scribblings and tail room for a
struct skb_shared_info. Also, at this point, we don't support
XDP frags, only a single contiguous Rx buffer. This means
that our page splitting is not very useful, so when XDP is in
use we need to use the full Rx buffer size and not do sharing.
Co-developed-by: Brett Creeley <brett.creeley@amd.com>
Signed-off-by: Brett Creeley <brett.creeley@amd.com>
Signed-off-by: Shannon Nelson <shannon.nelson@amd.com>
Reviewed-by: Jacob Keller <jacob.e.keller@intel.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
If the driver or firmware is stuck in reset state, don't bother
trying to use adminq commands. This speeds up shutdown and
prevents unnecessary timeouts and error messages.
This includes a bit of rework on ionic_adminq_post_wait()
and ionic_adminq_post_wait_nomsg() to both use
__ionic_adminq_post_wait() which can do the checks needed in
both cases.
Signed-off-by: Shannon Nelson <shannon.nelson@amd.com>
Reviewed-by: Brett Creeley <brett.creeley@amd.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Each time a VF attribute is set via iproute a call to get the VF
configuration is also made. This is currently problematic because for
each VF configuration call there are multiple commands sent to the
device. Unfortunately, this doesn't scale well. Fix this by reporting
the cached VF attributes.
The original change to query the device for getting the VF attributes
f16f5be310 ("ionic: Query FW when getting VF info via ndo_get_vf_config")
was made to remain consistent with device set VF attributes. However,
after further investigation there is no need to query the device.
Signed-off-by: Brett Creeley <brett.creeley@amd.com>
Signed-off-by: Shannon Nelson <shannon.nelson@amd.com>
Reviewed-by: Florian Fainelli <florian.fainelli@broadcom.com>
Reviewed-by: Rahul Rameshbabu <rrameshbabu@nvidia.com>
Link: https://lore.kernel.org/r/20231204210936.16587-2-shannon.nelson@amd.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
Add a check of the queue's max_sg_elems against the maximum frags we
expect to see per SKB and take the smaller of the two as our max for
the queues' descriptor buffer allocations.
Signed-off-by: Shannon Nelson <shannon.nelson@amd.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
ionic_start_queues_reconfig returns an error code if txrx_init fails.
Handle this error code in the relevant places.
This fixes a corner case where the device could get left in a detached
state if the CMB reconfig fails and the attempt to clean up the mess
also fails. Note that calling netif_device_attach when the netdev is
already attached does not lead to unexpected behavior.
Change goto name "errout" to "err_out" to maintain consistency across
goto statements.
Fixes: 40bc471dc7 ("ionic: add tx/rx-push support with device Component Memory Buffers")
Fixes: 6f7d6f0fd7 ("ionic: pull reset_queues into tx_timeout handler")
Signed-off-by: Nitya Sunkad <nitya.sunkad@amd.com>
Signed-off-by: Shannon Nelson <shannon.nelson@amd.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Add support for the PCI reset handlers in order to manage an FLR event.
Signed-off-by: Shannon Nelson <shannon.nelson@amd.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Pull out some code from ionic_lif_handle_fw_up() that can be
used in the coming FLR recovery patch.
Signed-off-by: Shannon Nelson <shannon.nelson@amd.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Remove unnecessary early code development check and the WARN_ON
that it uses. The irq alloc and free paths have long been
cleaned up and this check shouldn't have stuck around so long.
Fixes: 77ceb68e29 ("ionic: Add notifyq support")
Signed-off-by: Nitya Sunkad <nitya.sunkad@amd.com>
Signed-off-by: Shannon Nelson <shannon.nelson@amd.com>
Reviewed-by: Jacob Keller <jacob.e.keller@intel.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Following the example of 'commit 9a0f830f80 ("ethtool: linkstate:
add a statistic for PHY down events")', added support for link down
events.
Add callback ionic_get_link_ext_stats to ionic_ethtool.c to support
link_down_count, a property of netdev that gets reported exclusively
on physical link down events.
Run ethtool -I <devname> to display the device link down count.
Signed-off-by: Nitya Sunkad <nitya.sunkad@amd.com>
Signed-off-by: Shannon Nelson <shannon.nelson@amd.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
The ionic device has on-board memory (CMB) that can be used
for descriptors as a way to speed descriptor access for faster
packet processing. It is rumored to improve latency and/or
packets-per-second for some profiles of small packet traffic,
although your mileage may vary.
Signed-off-by: Shannon Nelson <shannon.nelson@amd.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
