When operating at mtu 9000, qlge does order-1 allocations for rx buffers in
atomic context. This is especially unreliable when free memory is low or
fragmented. Add an approach similar to commit 3161e453e4 ("virtio: net
refill on out-of-memory") to qlge so that the device doesn't lock up if
there are allocation failures.
Signed-off-by: Benjamin Poirier <bpoirier@suse.com>
Link: https://lore.kernel.org/r/20190927101210.23856-18-bpoirier@suse.com
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
Reading the {s,l}bq_prod_idx registers on a running device, it appears that
the adapter will only use buffers up to prod_idx & 0xfff0. The driver
currently uses fixed-size guard zones (16 for sbq, 32 for lbq - don't know
why this difference). After the previous patch, this approach no longer
guarantees prod_idx values aligned on multiples of 16. While it appears
that we can write unaligned values to prod_idx without ill effects on
device operation, it makes more sense to change qlge_refill_bq() to refill
up to a limit that corresponds with the device's behavior.
Signed-off-by: Benjamin Poirier <bpoirier@suse.com>
Link: https://lore.kernel.org/r/20190927101210.23856-17-bpoirier@suse.com
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
Currently, if we repeatedly fail to allocate all of the buffers from the
desired batching budget, we will never update the prod_idx register.
Restructure code to always update prod_idx if new buffers could be
allocated. This eliminates the current two stage process (clean_idx ->
prod_idx) and some associated bookkeeping variables.
Signed-off-by: Benjamin Poirier <bpoirier@suse.com>
Link: https://lore.kernel.org/r/20190927101210.23856-16-bpoirier@suse.com
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
The qlge driver (and device) uses two kinds of buffers for reception,
so-called "small buffers" and "large buffers". The two are arranged in
rings, the sbq and lbq. These two share similar data structures and code.
Factor out data structures into a common struct qlge_bq, make required
adjustments to code and dedup the most obvious cases of copy/paste.
This patch should not introduce any functional change other than to some of
the printk format strings.
Signed-off-by: Benjamin Poirier <bpoirier@suse.com>
Link: https://lore.kernel.org/r/20190927101210.23856-9-bpoirier@suse.com
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
qlge uses an irq enable/disable refcounting scheme that is:
* poorly implemented
Uses a spin_lock to protect accesses to the irq_cnt atomic
variable.
* buggy
Breaks when there is not a 1:1 sequence of irq - napi_poll, such as
when using SO_BUSY_POLL.
* unnecessary
The purpose or irq_cnt is to reduce irq control writes when
multiple work items result from one irq: the irq is re-enabled
after all work is done.
Analysis of the irq handler shows that there is only one case where
there might be two workers scheduled at once, and those have
separate irq masking bits.
Therefore, remove irq_cnt.
Additionally, we get a performance improvement:
perf stat -e cycles -a -r5 super_netperf 100 -H 192.168.33.1 -t TCP_RR
Before:
628560
628056
622103
622744
627202
[...]
268,803,947,669 cycles ( +- 0.09% )
After:
636300
634106
634984
638555
634188
[...]
259,237,291,449 cycles ( +- 0.19% )
Signed-off-by: Benjamin Poirier <bpoirier@suse.com>
Link: https://lore.kernel.org/r/20190927101210.23856-3-bpoirier@suse.com
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
Tracing the driver operation reveals that the INTR_EN_EN bit (per-queue
interrupt control) does not immediately prevent rx completion interrupts
when the device is operating in INTx mode. This leads to interrupts being
raised while napi is scheduled/running. Those interrupts are ignored by
qlge_isr() and falsely reported as IRQ_NONE thanks to the irq_cnt scheme.
This in turn can cause frames to loiter in the receive queue until a later
frame leads to another rx interrupt that will schedule napi.
Use the INTR_EN_EI bit (master interrupt control) instead.
Signed-off-by: Benjamin Poirier <bpoirier@suse.com>
Link: https://lore.kernel.org/r/20190927101210.23856-2-bpoirier@suse.com
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
A few tasks remain to be done in order to finish chip initial
configuration:
- configure chip to use multi-tx confirmation (speed up data
transfer)
- configure chip to use wake-up feature (save power consumption
during runtime)
- set hardware configuration (clocks, RF, pinout, etc...) using a
Platform Data Set (PDS) file
On release, driver completely shutdown the chip to save power
consumption.
Documentation about PDS and PDS data for sample boards are available
here[1]. One day, PDS data may find a place in device tree but,
currently, PDS is too much linked with firmware to allowing that.
This patch also add "send_pds" file in debugfs to be able to dynamically
change PDS (only for debug, of course).
[1]: https://github.com/SiliconLabs/wfx-firmware/tree/master/PDS
Signed-off-by: Jérôme Pouiller <jerome.pouiller@silabs.com>
Link: https://lore.kernel.org/r/20190919142527.31797-15-Jerome.Pouiller@silabs.com
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
Chip has multiple input buffers and can handle multiple 802.11 frames
in parallel. However, other HIF command must be sent sequentially.
wsm_send_cmd() handles these requests.
This commit also add send_hif_cmd in debugfs. This file allows to send
arbitrary commands to chip. It can be used for debug and testing.
Signed-off-by: Jérôme Pouiller <jerome.pouiller@silabs.com>
Link: https://lore.kernel.org/r/20190919142527.31797-12-Jerome.Pouiller@silabs.com
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
bh_work() is in charge to schedule all HIF message from/to chip.
On normal operation, when an IRQ is received, driver can get size of
next message in control register. In order to save control register
access, when chip send a message, it also appends a copy of control
register after the message (this register is not accounted in message
length declared in message header, but must accounted in bus request).
This copy of control register is called "piggyback".
It also handles a power saving mechanism specific to WFxxx series. This
mechanism is based on a GPIO called "wakeup" GPIO. Obviously, this gpio
is not part of SPI/SDIO standard buses and must be declared
independently (this is the main reason for why SDIO mode try to get
parameters from DT).
When wakeup is enabled, host can communicate with chip only if it is
awake. To wake up chip, there are two cases:
- host receive an IRQ from chip (chip initiate communication): host
just have to set wakeup GPIO before reading data
- host want to send data to chip: host set wakeup GPIO, then wait
for an IRQ (in fact, wait for an empty message) and finally send data
bh_work() is also in charge to track usage of chip buffers. Normally
each request expect a confirmation. However, you can notice that special
"multi tx" confirmation can acknowledge multiple requests at time.
Finally, note that wfx_bh_request_rx() is not atomic (because of
control_reg_read()). So, in SPI mode, hard-irq handler only postpone all
processing to wfx_spi_request_rx().
Signed-off-by: Jérôme Pouiller <jerome.pouiller@silabs.com>
Link: https://lore.kernel.org/r/20190919142527.31797-8-Jerome.Pouiller@silabs.com
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
These files are shared with firmware sources. Only a subset of these
definitions are used by driver but, for now, it is easier to import all.
API defines 3 kinds of messages:
- Requests (req) are sent from host to chip
- Confirmations (cnf) are sent by chip and are always in reply to a
request
- Indications (ind) are spontaneous message from chip to host
One request normally generate one confirmation. There are a few
exceptions to this rule:
- "shutdown" request is not acknowledged
- multiple tx request can be acknowledged a unique "multi-tx"
confirmation
In add, API defines MIB. They are sub-structures for write_mib and
read_mib API.
Note that all numbers in API have to be little endian when sent/received
from/to chip (I didn't declared them with __le32 because driver also use
them internally).
Signed-off-by: Jérôme Pouiller <jerome.pouiller@silabs.com>
Link: https://lore.kernel.org/r/20190919142527.31797-7-Jerome.Pouiller@silabs.com
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
Instantiate build infrastructure WFx driver. This driver provides support
for Wifi chipset Silicon Labs WF200 and further:
https://www.silabs.com/documents/public/data-sheets/wf200-datasheet.pdf
This chip support SPI and SDIO bus.
SDIO interface has two particularities:
1. Some parameters may be useful for end user (I will talk about
gpio_wakeup later).
2. The SDIO VID and PID of WF200 are 0000:0001 which are too much
generic to rely on.
So, current code checks VID/PID and looks for a node in DT (since WF200
targets embedded platforms, I don't think it is a problem to rely on
DT). DT can also be used to define to parameters for driver. Currently,
if no node is found, a warning is emitted, but it could be changed in
error.
Signed-off-by: Jérôme Pouiller <jerome.pouiller@silabs.com>
Link: https://lore.kernel.org/r/20190919142527.31797-2-Jerome.Pouiller@silabs.com
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
We've seen several incorrect patches for fs_sync() calls in the exfat driver.
Add code to the TODO that explains this isn't just a delete code and refactor,
but that actual analysis of when the filesystem should be flushed to disk
needs to be done.
Signed-off-by: Valdis Kletnieks <valdis.kletnieks@vt.edu>
Link: https://lore.kernel.org/r/9837.1570042895@turing-police
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
The majority of them were totally backwards. Change the logic
so that if DELAYED_SYNC *isn't* in the config, we actually flush to disk
before flagging the file system as clean.
That leaves two calls in the DELAYED_SYNC case. More detailed
analysis is needed to make sure that's what's really needed, or if other
call sites also need a fs_sync() call. This patch is at least "less wrong"
than the code was, but further changes should be another patch.
Signed-off-by: Valdis Kletnieks <valdis.kletnieks@vt.edu>
Link: https://lore.kernel.org/r/11092.1570043784@turing-police
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
