| Age | Commit message (Collapse) | Author |
|---|
|
All ena files should now use SPDX format in their license string. This
doesn't change the license of the files, but rather states the same
license in fewer words.
Also update the license years in some of the files.
Signed-off-by: Shay Agroskin <shayagr@amazon.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
|
|
When using XDP every ingress packet is passed to an eBPF (xdp) program
which returns an action for this packet.
This patch adds counters for the number of times each such action was
received. It also counts all the invalid actions received from the eBPF
program.
Signed-off-by: Shay Agroskin <shayagr@amazon.com>
Signed-off-by: Sameeh Jubran <sameehj@amazon.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
|
|
The new metrics provide granular visibility along multiple network
dimensions and enable troubleshooting and remediation of issues caused
by instances exceeding network performance allowances.
The new statistics can be queried using ethtool command.
Signed-off-by: Guy Tzalik <gtzalik@amazon.com>
Signed-off-by: Sameeh Jubran <sameehj@amazon.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
|
|
New devices add a new hardware acceleration engine, which adds some
restrictions to the driver.
Metadata descriptor must be present for each packet and the maximum
burst size between two doorbells is now limited to a number
advertised by the device.
This patch adds:
1. A handshake protocol between the driver and the device, so the
device will enable the accelerated queues only when both sides
support it.
2. The driver support for the new acceleration engine:
2.1. Send metadata descriptor for each Tx packet.
2.2. Limit the number of packets sent between doorbells.(*)
(*) A previous driver implementation of this feature was comitted in
commit 05d62ca218f8 ("net: ena: add handling of llq max tx burst size")
however the design of the interface between the driver and device
changed since then. This change is reflected in this commit.
Signed-off-by: Netanel Belgazal <netanel@amazon.com>
Signed-off-by: Arthur Kiyanovski <akiyano@amazon.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
|
|
For an overview of the race created by this patch goto synchronization
label.
In napi busy-poll mode, the kernel invokes the napi handler of the
device repeatedly to poll the NIC's receive queues. This process
repeats until a timeout, specific for each connection, is up.
By polling packets in busy-poll mode the user may gain lower latency
and higher throughput (since the kernel no longer waits for interrupts
to poll the queues) in expense of CPU usage.
Upon completing a napi routine, the driver checks whether
the routine was called by an interrupt handler. If so, the driver
re-enables interrupts for the device. This is needed since an
interrupt routine invocation disables future invocations until
explicitly re-enabled.
The driver avoids re-enabling the interrupts if they were not disabled
in the first place (e.g. if driver in busy mode).
Originally, the driver checked whether interrupt re-enabling is needed
by reading the 'ena_napi->unmask_interrupt' variable. This atomic
variable was set upon interrupt and cleared after re-enabling it.
In the 4.10 Linux version, the 'napi_complete_done' call was changed
so that it returns 'false' when device should not re-enable
interrupts, and 'true' otherwise. The change includes reading the
"NAPIF_STATE_IN_BUSY_POLL" flag to check if the napi call is in
busy-poll mode, and if so, return 'false'.
The driver was changed to re-enable interrupts according to this
routine's return value.
The Linux community rejected the use of the
'ena_napi->unmaunmask_interrupt' variable to determine whether
unmasking is needed, and urged to use napi_napi_complete_done()
return value solely.
See https://lore.kernel.org/patchwork/patch/741149/ for more details
As explained, a busy-poll session exists for a specified timeout
value, after which it exits the busy-poll mode and re-enters it later.
This leads to many invocations of the napi handler where
napi_complete_done() false indicates that interrupts should be
re-enabled.
This creates a bug in which the interrupts are re-enabled
unnecessarily.
To reproduce this bug:
1) echo 50 | sudo tee /proc/sys/net/core/busy_poll
2) echo 50 | sudo tee /proc/sys/net/core/busy_read
3) Add counters that check whether
'ena_unmask_interrupt(tx_ring, rx_ring);'
is called without disabling the interrupts in the first
place (i.e. with calling the interrupt routine
ena_intr_msix_io())
Steps 1+2 enable busy-poll as the default mode for new connections.
The busy poll routine rearms the interrupts after every session by
design, and so we need to add an extra check that the interrupts were
masked in the first place.
synchronization:
This patch introduces a race between the interrupt handler
ena_intr_msix_io() and the napi routine ena_io_poll().
Some macros and instruction were added to prevent this race from leaving
the interrupts masked. The following specifies the different race
scenarios in this patch:
1) interrupt handler and napi routine run sequentially
i) interrupt handler is called, sets 'interrupts_masked' flag and
successfully schedules the napi handler via softirq.
In this scenario the napi routine might not see the flag change
for several reasons:
a) The flag is stored in a register by the compiler. For this
case the WRITE_ONCE macro which prevents this.
b) The compiler might reorder the instruction. For this the
smp_wmb() instruction was used which implies a compiler memory
barrier.
c) On archs with weak consistency model (like ARM64) the napi
routine might be scheduled and start running before the flag
STORE instruction is committed to cache/memory. To ensure this
doesn't happen, the smp_wmb() instruction was added. It ensures
that the flag set instruction is committed before scheduling
napi.
ii) compiler reorders the flag's value check in the 'if' with
the flag set in the napi routine.
This scenario is prevented by smp_rmb() call after the flag check.
2) interrupt handler and napi routine run in parallel (can happen when
busy poll routine invokes the napi handler)
i) interrupt handler sets the flag in one core, while the napi
routine reads it in another core.
This scenario also is divided into two cases:
a) napi_complete_done() doesn't finish running, in which case
napi_sched() would just set NAPIF_STATE_MISSED and the napi
routine would reschedule itself without changing the flag's value.
b) napi_complete_done() finishes running. In this case the
napi routine might override the flag's value.
This doesn't present any rise since it later unmasks the
interrupt vector.
Signed-off-by: Shay Agroskin <shayagr@amazon.com>
Signed-off-by: Arthur Kiyanovski <akiyano@amazon.com>
Cc: Eric Dumazet <eric.dumazet@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
|
|
This commit reduces the driver load time by using usec resolution
instead of msec when polling for hardware state change.
Also add back-off mechanism to handle cases where minimal sleep
time is not enough.
Signed-off-by: Josh Triplett <josh@joshtriplett.org>
Signed-off-by: Sameeh Jubran <sameehj@amazon.com>
Signed-off-by: Arthur Kiyanovski <akiyano@amazon.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
|
|
1. Use BIT macro instead of shift operator for code clarity
2. Replace multiple flag assignments to a single assignment of multiple
flags in ena_com_add_single_rx_desc()
3. Move ENA_HASH_KEY_SIZE from ena_netdev.h to ena_com.h
Signed-off-by: Arthur Kiyanovski <akiyano@amazon.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
|
|
1. Remove unused definition of DRV_MODULE_VERSION
2. Remove {} from single line-of-code ifs
3. Remove unnecessary comments from ena_get/set_coalesce()
4. Remove unnecessary extra spaces and newlines
Signed-off-by: Arthur Kiyanovski <akiyano@amazon.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
|
|
Frame size ENA_PAGE_SIZE is limited to 16K on systems with larger
PAGE_SIZE than 16K. Change ENA_XDP_MAX_MTU to also take into account
the reserved tailroom.
Signed-off-by: Jesper Dangaard Brouer <brouer@redhat.com>
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
Acked-by: Sameeh Jubran <sameehj@amazon.com>
Cc: Arthur Kiyanovski <akiyano@amazon.com>
Link: https://lore.kernel.org/bpf/158945341384.97035.907403694833419456.stgit@firesoul
|
|
Conflicts were all overlapping changes.
Signed-off-by: David S. Miller <davem@davemloft.net>
|
|
1. Add support for getting tx drops from the device and saving them
in the driver.
2. Report tx via netdev stats.
Signed-off-by: Igor Chauskin <igorch@amazon.com>
Signed-off-by: Guy Tzalik <gtzalik@amazon.com>
Signed-off-by: Arthur Kiyanovski <akiyano@amazon.com>
Signed-off-by: Sameeh Jubran <sameehj@amazon.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
|
|
Add unmask interrupts statistics to ethtool.
Signed-off-by: Netanel Belgazal <netanel@amazon.com>
Signed-off-by: Arthur Kiyanovski <akiyano@amazon.com>
Signed-off-by: Sameeh Jubran <sameehj@amazon.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
|
|
This fixes the following build warning in ena_xdp_set(), which is
observed on aarch64 with 64KB page size.
In file included from ./include/net/inet_sock.h:19,
from ./include/net/ip.h:27,
from drivers/net/ethernet/amazon/ena/ena_netdev.c:46:
drivers/net/ethernet/amazon/ena/ena_netdev.c: In function \
‘ena_xdp_set’: \
drivers/net/ethernet/amazon/ena/ena_netdev.c:557:6: warning: \
format ‘%lu’ \
expects argument of type ‘long unsigned int’, but argument 4 \
has type ‘int’ \
[-Wformat=] "Failed to set xdp program, the current MTU (%d) is \
larger than the maximum allowed MTU (%lu) while xdp is on",
Signed-off-by: Gavin Shan <gshan@redhat.com>
Acked-by: Shay Agroskin <shayagr@amazon.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
|
|
In this commit we revert the part of
commit 1a63443afd70 ("net/amazon: Ensure that driver version is aligned to the linux kernel"),
which breaks the interface between the ENA driver and FW.
We also replace the use of DRIVER_VERSION with DRIVER_GENERATION
when we bring back the deleted constants that are used in interface with
ENA device FW.
This commit does not change the driver version reported to the user via
ethtool, which remains the kernel version.
Fixes: 1a63443afd70 ("net/amazon: Ensure that driver version is aligned to the linux kernel")
Signed-off-by: Arthur Kiyanovski <akiyano@amazon.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
|
|
Upstream drivers are managed inside global repository and released all
together, this ensure that driver version is the same as linux kernel,
so update amazon drivers to properly reflect it.
Signed-off-by: Leon Romanovsky <leonro@mellanox.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
|
|
table
The indirection table has the indices of the Rx queues. When we store it
during set indirection operation, we convert the indices to our internal
representation of the indices.
Our internal representation of the indices is: even indices for Tx and
uneven indices for Rx, where every Tx/Rx pair are in a consecutive order
starting from 0. For example if the driver has 3 queues (3 for Tx and 3
for Rx) then the indices are as follows:
0 1 2 3 4 5
Tx Rx Tx Rx Tx Rx
The BUG:
The issue is that when we satisfy a get request for the indirection
table, we don't convert the indices back to the original representation.
The FIX:
Simply apply the inverse function for the indices of the indirection
table after we set it.
Fixes: 1738cd3ed342 ("net: ena: Add a driver for Amazon Elastic Network Adapters (ENA)")
Signed-off-by: Sameeh Jubran <sameehj@amazon.com>
Signed-off-by: Arthur Kiyanovski <akiyano@amazon.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
|
|
The first_interrupt field is accessed in ena_intr_msix_io() upon
receiving an interrupt.The rx_ring and tx_ring fields of napi can
be NULL when receiving interrupt for xdp queues. This patch fixes
the issue by moving the field to the ena_napi struct.
Signed-off-by: Sameeh Jubran <sameehj@amazon.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
|
|
This commit implements the XDP_TX action in the ena driver. We allocate
separate tx queues for the XDP_TX. We currently allow xdp only when
there is enough queues to allocate for xdp.
Signed-off-by: Sameeh Jubran <sameehj@amazon.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
|
|
This commit implements the basic functionality of drop/pass logic in the
ena driver.
Signed-off-by: Sameeh Jubran <sameehj@amazon.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
|
|
Set channels callback enables the user to change the count of queues
used by the driver using ethtool. We decided to currently support only
equal number of rx and tx queues, this might change in the future.
Also rename dev_up to dev_was_up in ena_update_queue_count() to make
it clearer.
Signed-off-by: Sameeh Jubran <sameehj@amazon.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
|
|
- Update ena_ethtool:ena_get_channels() to return adapter->max_io_queues
so that ethtool -l returns the correct maximum queue number.
- Change the name of ena_calc_io_queue_num() to
ena_calc_max_io_queue_num() as it returns the maximum number of io
queues and actual number of queues can be smaller if changed
by ethtool -L which is implemented in a later commit.
- Change variable name from io_queue_num to max_num_io_queues in
ena_calc_max_io_queue_num() and ena_probe().
- Make all types of variables that convey the number and sizeof queues
to be u32, for consistency with the API between the driver and the
device.
Signed-off-by: Arthur Kiyanovski <akiyano@amazon.com>
Signed-off-by: Sameeh Jubran <sameehj@amazon.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
|
|
Most places in the code refer to the IO queues as io_queues and not
simply queues. Examples - max_io_queues_per_vf, ENA_MAX_NUM_IO_QUEUES,
ena_destroy_all_io_queues() etc..
We are also adding the new max_num_io_queues field to struct ena_adapter
in the following commit.
The changes included in this commit are:
struct ena_adapter->num_queues => struct ena_adapter->num_io_queues
Signed-off-by: Arthur Kiyanovski <akiyano@amazon.com>
Signed-off-by: Sameeh Jubran <sameehj@amazon.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
|
|
Deleted unused 4 fields from struct ena_adapter and their only user
ena_restore_ethtool_params().
Signed-off-by: Arthur Kiyanovski <akiyano@amazon.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
|
|
1. Out of the fields {per_napi_bytes, per_napi_packets} in struct ena_ring,
only rx_ring->per_napi_packets are used to determine if napi did work
for dim.
This commit removes all other uses of these fields.
2. Remove ena_ring->moder_tbl_idx, which is not used by dim.
3. Remove all calls to ena_com_destroy_interrupt_moderation(), since all it
did was to destroy the interrupt moderation table, which is removed as
part of removing old interrupt moderation code.
Signed-off-by: Arthur Kiyanovski <akiyano@amazon.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
|
|
Use the dim library for the rx adaptive interrupt moderation implementation
Signed-off-by: Arthur Kiyanovski <akiyano@amazon.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
|
|
This reverts commit 315c28d2b714 ("net: ena: ethtool: add extra properties retrieval via get_priv_flags").
As discussed at netconf and on the mailing list we can't allow
for the the abuse of private flags for exposing arbitrary device
labels.
Signed-off-by: Jakub Kicinski <jakub.kicinski@netronome.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
|
|
Update driver version to match device specification.
Signed-off-by: Sameeh Jubran <sameehj@amazon.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
|
|
Implement the set_ringparam() function of the ethtool interface
to enable the changing of io queue sizes.
Signed-off-by: Arthur Kiyanovski <akiyano@amazon.com>
Signed-off-by: Sameeh Jubran <sameehj@amazon.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
|
|
If there is not enough memory to allocate io queues the driver will
try to allocate smaller queues.
The backoff algorithm is as follows:
1. Try to allocate TX and RX and if successful.
1.1. return success
2. Divide by 2 the size of the larger of RX and TX queues (or both if their size is the same).
3. If TX or RX is smaller than 256
3.1. return failure.
4. else
4.1. go back to 1.
Also change the tx_queue_size, rx_queue_size field names in struct
adapter to requested_tx_queue_size and requested_rx_queue_size, and
use RX and TX queue 0 for actual queue sizes.
Explanation:
The original fields were useless as they were simply used to assign
values once from them to each of the queues in the adapter in ena_probe().
They could simply be deleted. However now that we have a backoff
feature, we have use for them. In case of backoff there is a difference
between the requested queue sizes and the actual sizes. Therefore there
is a need to save the requested queue size for future retries of queue
allocation (for example if allocation failed and then ifdown + ifup was
called we want to start the allocation from the original requested size of
the queues).
Signed-off-by: Arthur Kiyanovski <akiyano@amazon.com>
Signed-off-by: Sameeh Jubran <sameehj@amazon.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
|
|
Use MAX_QUEUES_EXT get feature capability to query the device.
Signed-off-by: Netanel Belgazal <netanel@amazon.com>
Signed-off-by: Sameeh Jubran <sameehj@amazon.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
|
|
Add a new statistics to ETHTOOL to specify if the device calculated
and validated the Rx csum.
Signed-off-by: Evgeny Shmeilin <evgeny@annapurnaLabs.com>
Signed-off-by: Sameeh Jubran <sameehj@amazon.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
|
|
struct ena_ring holds a union of free_rx_ids and free_tx_ids.
Both of the above fields mean the exact same thing and are used
exactly the same way.
Furthermore, these fields are always used with a prefix of the
type of ring. So for tx it will be tx_ring->free_tx_ids, and for
rx it will be rx_ring->free_rx_ids, which shows how redundant the
"_tx" and "_rx" parts are.
Furthermore still, this may lead to confusing code like where
tx_ring->free_rx_ids which works correctly but looks like a mess.
This commit removes the aforementioned redundancy by replacing the
free_rx/tx_ids union with a single free_ids field.
It also changes a single goto label name from err_free_tx_ids: to
err_tx_free_ids: for consistency with the above new notation.
Signed-off-by: Arthur Kiyanovski <akiyano@amazon.com>
Signed-off-by: Sameeh Jubran <sameehj@amazon.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
|
|
This commit adds a mechanism for exposing different device
properties via ethtool's priv_flags. The strings are provided
by the device and copied to user space through the driver.
In this commit we:
Add commands, structs and defines necessary for handling
extra properties
Add functions for:
Allocation/destruction of a buffer for extra properties strings.
Retreival of extra properties strings and flags from the network device.
Handle the allocation of a buffer for extra properties strings.
* Initialize buffer with extra properties strings from the
network device at driver startup.
Use ethtool's get_priv_flags to expose extra properties of
the ENA device
Signed-off-by: Arthur Kiyanovski <akiyano@amazon.com>
Signed-off-by: Sameeh Jubran <sameehj@amazon.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
|
|
Update driver version due to bug fix.
Signed-off-by: Arthur Kiyanovski <akiyano@amazon.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
|
|
Update driver version due to critical bug fixes.
Signed-off-by: Arthur Kiyanovski <akiyano@amazon.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
|
|
Signed-off-by: Arthur Kiyanovski <akiyano@amazon.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
|
|
Improves socket memory utilization when receiving packets larger
than 128 bytes (the previous rx copybreak) and smaller than 256 bytes.
Signed-off-by: Arthur Kiyanovski <akiyano@amazon.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
|
|
Currently Rx refill is done when the number of required descriptors is
above 1/8 queue size. With a default of 1024 entries per queue the
threshold is 128 descriptors.
There is intention to increase the queue size to 8196 entries.
In this case threshold of 1024 descriptors is too large and can hurt
latency.
Add another limitation to Rx threshold to be at most 256 descriptors.
Signed-off-by: Arthur Kiyanovski <akiyano@amazon.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
|
|
Signed-off-by: Arthur Kiyanovski <akiyano@amazon.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
|
|
Set skb->ip_summed to the correct value as reported by the device.
Add counter for the case where rx csum offload is enabled but
device didn't check it.
Signed-off-by: Arthur Kiyanovski <akiyano@amazon.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
|
|
This patch includes all code changes necessary in ena_netdev to enable
packet sending via the LLQ placemnt mode.
Signed-off-by: Arthur Kiyanovski <akiyano@amazon.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
|
|
The buffer length field in the ena rx descriptor is 16 bit, and the
current driver passes a full page in each ena rx descriptor.
When PAGE_SIZE equals 64kB or more, the buffer length field becomes
zero.
To solve this issue, limit the ena Rx descriptor to use 16kB even
when allocating 64kB kernel pages. This change would not impact ena
device functionality, as 16kB is still larger than maximum MTU.
Signed-off-by: Netanel Belgazal <netanel@amazon.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
|
|
Signed-off-by: Netanel Belgazal <netanel@amazon.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
|
|
MSI-X
A mechanism for detection of stuck Rx/Tx rings due to missed or
misrouted interrupts.
Check if there are unhandled completion descriptors before the first
MSI-X interrupt arrived.
The check is per queue and per interrupt vector.
Once such condition is detected, driver and device reset is scheduled.
Signed-off-by: Netanel Belgazal <netanel@amazon.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
|
|
In rare cases, ena driver would reset and re-start the device,
for example, in case of misbehaving application that causes
transmit timeout
The first step in the reset procedure is to stop the Tx traffic by
calling ena_carrier_off().
After the driver have just started the device reset procedure, device
happens to send an asynchronous notification (via AENQ) to the driver
than there was a link change (to link-up state).
This link change is mapped to a call to netif_carrier_on() which
re-activates the Tx queues, violating the assumption of no tx traffic
until device reset is completed, as the reset task might still be in
the process of queues initialization, leading to an access to
uninitialized memory.
Signed-off-by: Netanel Belgazal <netanel@amazon.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
|
|
Signed-off-by: Netanel Belgazal <netanel@amazon.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
|
|
Add a new statistic to ethtool stats that show the number of packets
without transmit acknowledgement from ENA device.
Signed-off-by: Netanel Belgazal <netanel@amazon.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
|
|
Signed-off-by: Netanel Belgazal <netanel@amazon.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
|
|
Remove ena_device_io_suspend/resume() methods
Those methods were intend to be used by the device to trigger
suspend/resume but eventually it was dropped.
Signed-off-by: Netanel Belgazal <netanel@amazon.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
|
|
Signed-off-by: Netanel Belgazal <netanel@amazon.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
|
