Using a fixed 1s timeout for all commands is a bit problematic.
For some commands it means waiting longer than needed for the timeout to
expire, which may not a big issue, but still. For other commands, like for
an erase (CMD38) that uses a R1B response, may require longer timeouts than
1s. In these cases, we may end up treating the command as it failed, while
it just needed some more time to complete successfully.
Fix the problem by respecting the cmd->busy_timeout, which is provided by
the mmc core.
Cc: Rui Miguel Silva <rmfrfs@gmail.com>
Cc: Johan Hovold <johan@kernel.org>
Cc: Alex Elder <elder@kernel.org>
Cc: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
Cc: greybus-dev@lists.linaro.org
Signed-off-by: Ulf Hansson <ulf.hansson@linaro.org>
Acked-by: Rui Miguel Silva <rmfrfs@gmail.com>
Acked-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
Link: https://lore.kernel.org/r/20200414161413.3036-20-ulf.hansson@linaro.org
Signed-off-by: Ulf Hansson <ulf.hansson@linaro.org>
Some commands uses R1B responses, which means the card may assert the DAT0
line to signal busy for a period of time, after it has received the
command. The mmc core normally specifies the busy period for the command in
the cmd->busy_timeout. Ideally the driver should respect it, but that
requires quite some update of the code, so let's defer that to someone with
the HW at hand.
Instead, let's inform the mmc core about the maximum supported busy timeout
in ->max_busy_timeout during ->probe(). This value corresponds to the fixed
4s timeout used by usdhi6rol0. In this way, we let the mmc core validate
the needed timeout, which may lead to that it converts from a R1B into a R1
response and then use CMD13 to poll for busy completion.
In other words, this change enables support for commands with longer busy
periods than 4s, like erase (CMD38) for example.
Cc: Jesper Nilsson <jesper.nilsson@axis.com>
Cc: Lars Persson <lars.persson@axis.com>
Signed-off-by: Ulf Hansson <ulf.hansson@linaro.org>
Acked-by: Jesper Nilsson <jesper.nilsson@axis.com>
Link: https://lore.kernel.org/r/20200414161413.3036-6-ulf.hansson@linaro.org
Signed-off-by: Ulf Hansson <ulf.hansson@linaro.org>
Using a fixed 2s timeout for all commands is a bit problematic.
For some commands it means waiting longer than needed for the timer to
expire, which may not a big issue, but still. For other commands, like for
an erase (CMD38) that uses a R1B response, may require longer timeouts than
2s. In these cases, we may end up treating the command as it failed, while
it just needed some more time to complete successfully.
Fix the problem by respecting the cmd->busy_timeout, which is provided by
the mmc core.
Cc: Ludovic Desroches <ludovic.desroches@microchip.com>
Cc: Nicolas Ferre <nicolas.ferre@microchip.com>
Signed-off-by: Ulf Hansson <ulf.hansson@linaro.org>
Acked-by: Ludovic Desroches <ludovic.desroches@microchip.com>
Link: https://lore.kernel.org/r/20200414161413.3036-4-ulf.hansson@linaro.org
Signed-off-by: Ulf Hansson <ulf.hansson@linaro.org>
Setting the timer on a per request basis, is rather limiting as the timer
really depends on what commands that is to be sent as part of the request.
Therefore improve the behaviour by programming the timer per command basis
instead.
Cc: Ludovic Desroches <ludovic.desroches@microchip.com>
Cc: Nicolas Ferre <nicolas.ferre@microchip.com>
Signed-off-by: Ulf Hansson <ulf.hansson@linaro.org>
Acked-by: Ludovic Desroches <ludovic.desroches@microchip.com>
Link: https://lore.kernel.org/r/20200414161413.3036-3-ulf.hansson@linaro.org
Signed-off-by: Ulf Hansson <ulf.hansson@linaro.org>
When atmci_request_end() is about to finish a request for one slot, there
is a possibility that there is new request queued for another slot. If this
turns out to be the case, the new request is started and the timer is
re-programmed for it.
Although, a few lines below in atmci_request_end(), this timer becomes
deleted, likely corresponding to the other recently completed request. This
looks wrong, so let's fix it.
Cc: Ludovic Desroches <ludovic.desroches@microchip.com>
Cc: Nicolas Ferre <nicolas.ferre@microchip.com>
Signed-off-by: Ulf Hansson <ulf.hansson@linaro.org>
Acked-by: Ludovic Desroches <ludovic.desroches@microchip.com>
Link: https://lore.kernel.org/r/20200414161413.3036-2-ulf.hansson@linaro.org
Signed-off-by: Ulf Hansson <ulf.hansson@linaro.org>
ADMA_ERR_SIZE_EN bit of VENDOR_SPECIFIC_FUNC register controls
ADMA length mismatch error interrupt. Enable it by default.
And update all bit shift defines with BIT macro.
Signed-off-by: Veerabhadrarao Badiganti <vbadigan@codeaurora.org>
Acked-by: Adrian Hunter <adrian.hunter@intel.com>
Link: https://lore.kernel.org/r/1587363626-20413-4-git-send-email-vbadigan@codeaurora.org
Signed-off-by: Ulf Hansson <ulf.hansson@linaro.org>
sdhci-msm can support auto cmd12.
So enable SDHCI_QUIRK_MULTIBLOCK_READ_ACMD12 quirk.
Signed-off-by: Veerabhadrarao Badiganti <vbadigan@codeaurora.org>
Acked-by: Adrian Hunter <adrian.hunter@intel.com>
Link: https://lore.kernel.org/r/1587363626-20413-3-git-send-email-vbadigan@codeaurora.org
Signed-off-by: Ulf Hansson <ulf.hansson@linaro.org>
When use one SDIO wifi which enable the runtime PM feature on i.MX6SX,
we meet system hang. This hang happened during the usdhc runtime resume,
in sdhci_init(), when call the sdhci_set_default_irqs. One interrupt
(SDHCI_INT_CARD_INT) triggered just after the host->ier update and before
the write of register SDHCI_SIGNAL_ENABLE. So in sdhci_irq, it will skip
the call of sdio_signal_irq() because current host->ier do not set the
SDHCI_INT_CARD_INT. So this SDIO wifi interrupt always keep triggered,
let the system stuck in irq handle, can't response any other thread.
This patch add spin lock for the sdhci_set_default_irqs to fix this issue.
Signed-off-by: Haibo Chen <haibo.chen@nxp.com>
Acked-by: Adrian Hunter <adrian.hunter@intel.com>
Link: https://lore.kernel.org/r/1586941255-9237-1-git-send-email-haibo.chen@nxp.com
Signed-off-by: Ulf Hansson <ulf.hansson@linaro.org>
Implement the request_atomic() API for nonremovable cards, that means
we can submit next request in the irq hard handler context to reduce
latency.
Moreover factor out the AUTO CMD23 checking into a separate function
to reduce duplicate code.
Signed-off-by: Baolin Wang <baolin.wang7@gmail.com>
Acked-by: Adrian Hunter <adrian.hunter@intel.com>
Link: https://lore.kernel.org/r/60142fe6c6c1dbba2696e775564ae2166786f0bc.1586744073.git.baolin.wang7@gmail.com
Signed-off-by: Ulf Hansson <ulf.hansson@linaro.org>
Implement the request_atomic() ops for the sdhci driver to process
one request in the atomic context if the card is nonremovable.
Moreover, we should return BUSY flag if controller has not released
the inhibit bits to allow HSQ trying to send request again in non-atomic
context.
Suggested-by: Adrian Hunter <adrian.hunter@intel.com>
Signed-off-by: Baolin Wang <baolin.wang7@gmail.com>
Acked-by: Adrian Hunter <adrian.hunter@intel.com>
Link: https://lore.kernel.org/r/9ed34afa9fb42e0c234065cac5401d7826942b55.1586744073.git.baolin.wang7@gmail.com
Signed-off-by: Ulf Hansson <ulf.hansson@linaro.org>
The SD host controller can process one request in the atomic context if
the card is nonremovable, which means we can submit next request in the
irq hard handler when using the MMC host software queue to reduce the
latency. Thus this patch adds a new API request_atomic() for the host
controller, as well as adding support for host software queue to submit
a request by the new request_atomic() API.
Moreover there is an unusual case that the card is busy when trying to
send a command, and we can not polling the card status in interrupt
context by using request_atomic() to dispatch requests. Thus we should
queue a work to try again in the non-atomic context in case the host
releases the busy signal later.
Suggested-by: Adrian Hunter <adrian.hunter@intel.com>
Acked-by: Adrian Hunter <adrian.hunter@intel.com>
Signed-off-by: Baolin Wang <baolin.wang7@gmail.com>
Link: https://lore.kernel.org/r/a344e27e506cb2329073cbd5cf65e15cc3cbeba9.1586744073.git.baolin.wang7@gmail.com
Signed-off-by: Ulf Hansson <ulf.hansson@linaro.org>
Spending time under spinlock increases IRQ latencies and also
response times because preemption is disabled.
sdhci_send_command() waits up to 10 ms under spinlock for inhibit bits
to clear. In general inhibit bits will not be set, but there may be
corner cases, especially in the face of errors, where waiting helps.
There might also be dysfunctional hardware that needs the waiting. So
retain the legacy behaviour but do not wait for inhibit bits while under
spinlock. Instead adjust the logic to enable waiting while not under
spinlock. That is mostly straight forward, but in the interrupt handler
it requires deferring an "inhibited" command to the IRQ thread where
sleeping is allowed.
Signed-off-by: Adrian Hunter <adrian.hunter@intel.com>
Tested-by: Baolin Wang <baolin.wang7@gmail.com>
Link: https://lore.kernel.org/r/20200412090349.1607-6-adrian.hunter@intel.com
Signed-off-by: Ulf Hansson <ulf.hansson@linaro.org>
In preparation for further changes, tidy sdhci_request() a bit.
Signed-off-by: Adrian Hunter <adrian.hunter@intel.com>
Tested-by: Baolin Wang <baolin.wang7@gmail.com>
Link: https://lore.kernel.org/r/20200412090349.1607-5-adrian.hunter@intel.com
Signed-off-by: Ulf Hansson <ulf.hansson@linaro.org>
sdhci_finish_data() is defined before it is referenced, so forward
declaration is not necessary.
Signed-off-by: Adrian Hunter <adrian.hunter@intel.com>
Tested-by: Baolin Wang <baolin.wang7@gmail.com>
Link: https://lore.kernel.org/r/20200412090349.1607-4-adrian.hunter@intel.com
Signed-off-by: Ulf Hansson <ulf.hansson@linaro.org>
sdhci_send_command() has not been used outside of sdhci.c for many
years. Stop exporting it.
Signed-off-by: Adrian Hunter <adrian.hunter@intel.com>
Tested-by: Baolin Wang <baolin.wang7@gmail.com>
Link: https://lore.kernel.org/r/20200412090349.1607-3-adrian.hunter@intel.com
Signed-off-by: Ulf Hansson <ulf.hansson@linaro.org>
Add 2 helper functions to make the use of the auto-CMD23 flag more
readable.
Signed-off-by: Adrian Hunter <adrian.hunter@intel.com>
Tested-by: Baolin Wang <baolin.wang7@gmail.com>
Link: https://lore.kernel.org/r/20200412090349.1607-2-adrian.hunter@intel.com
Signed-off-by: Ulf Hansson <ulf.hansson@linaro.org>
Use the well defined HIGH_SPEED_BUS_SPEED macro in mmc_sd_switch_hs()
to make code more readable.
Signed-off-by: Jisheng Zhang <Jisheng.Zhang@synaptics.com>
Link: https://lore.kernel.org/r/20200410145643.630b0731@xhacker.debian
Signed-off-by: Ulf Hansson <ulf.hansson@linaro.org>
Instead of reimplementing the logic in mmc_regulator_set_vqmmc(), use the
mmc code function directly.
This also allows us to fix a related issue on STM32MP1, when a voltage
switch of 1.8V is done for the eMMC, but the current level is already set
to 1.8V. More precisely, in this scenario the call to the
->post_sig_volt_switch() hangs, indefinitely waiting for the voltage switch
to complete. Fix this problem by checking if mmc_regulator_set_vqmmc()
returned 1 and then skip invoking the callback.
Signed-off-by: Marek Vasut <marex@denx.de>
Link: https://lore.kernel.org/r/20200416163649.336967-3-marex@denx.de
[Ulf: Updated the commit message]
Signed-off-by: Ulf Hansson <ulf.hansson@linaro.org>
Adjust mmc_regulator_set_vqmmc() to return 1 if the voltage switch was
skipped because the regulator voltage was already correct. This allows
drivers to detect such condition and possibly skip various voltage
switching extras.
Signed-off-by: Marek Vasut <marex@denx.de>
Link: https://lore.kernel.org/r/20200416163649.336967-2-marex@denx.de
Signed-off-by: Ulf Hansson <ulf.hansson@linaro.org>
Patch all drivers which use mmc_regulator_set_vqmmc() and prepare them for
the fact that mmc_regulator_set_vqmmc() can return a value > 0, which would
happen if the signal voltage switch did NOT happen, because the voltage was
already set correctly.
Signed-off-by: Marek Vasut <marex@denx.de>
Link: https://lore.kernel.org/r/20200416163649.336967-1-marex@denx.de
[Ulf: Re-worked/simplified the code a bit]
Signed-off-by: Ulf Hansson <ulf.hansson@linaro.org>
Clang warns:
drivers/mmc/host/sdhci-of-arasan.c:784:9: warning: variable 'ret' is
uninitialized when used here [-Wuninitialized]
return ret;
^~~
drivers/mmc/host/sdhci-of-arasan.c:738:9: note: initialize the variable
'ret' to silence this warning
int ret;
^
= 0
drivers/mmc/host/sdhci-of-arasan.c:860:9: warning: variable 'ret' is
uninitialized when used here [-Wuninitialized]
return ret;
^~~
drivers/mmc/host/sdhci-of-arasan.c:810:9: note: initialize the variable
'ret' to silence this warning
int ret;
^
= 0
2 warnings generated.
This looks like a copy paste error. Neither function has handling that
needs ret so just remove it and return 0 directly.
Link: https://github.com/ClangBuiltLinux/linux/issues/996
Reported-by: kernelci.org bot <bot@kernelci.org>
Signed-off-by: Nathan Chancellor <natechancellor@gmail.com>
Reviewed-by: Nick Desaulniers <ndesaulniers@google.com>
Link: https://lore.kernel.org/r/20200416182402.16858-1-natechancellor@gmail.com
Signed-off-by: Ulf Hansson <ulf.hansson@linaro.org>
No need to call platform_get_resource twice when we still have the
pointer from before. Also, use '%pa' for a resource_size_t pointer.
Signed-off-by: Wolfram Sang <wsa+renesas@sang-engineering.com>
Reviewed-by: Niklas Söderlund <niklas.soderlund+renesas@ragnatech.se>
Link: https://lore.kernel.org/r/20200408142252.21958-1-wsa+renesas@sang-engineering.com
Signed-off-by: Ulf Hansson <ulf.hansson@linaro.org>
Reviewed-by: Geert Uytterhoeven <geert+renesas@glider.be>
When tuning HS400, if all TAPS are good, we can utilize the SMPCMP
register to select the optimal TAP. For that, we populate a second
bitmap with SMPCMP results and query it in case the regular bitmap is
full (= all good).
Signed-off-by: Masaharu Hayakawa <masaharu.hayakawa.ry@renesas.com>
Signed-off-by: Takeshi Saito <takeshi.saito.xv@renesas.com>
Signed-off-by: Wolfram Sang <wsa+renesas@sang-engineering.com>
Reviewed-by: Yoshihiro Shimoda <yoshihiro.shimoda.uh@renesas.com>
Tested-by: Yoshihiro Shimoda <yoshihiro.shimoda.uh@renesas.com>
Link: https://lore.kernel.org/r/20200408094638.10375-4-wsa+renesas@sang-engineering.com
Signed-off-by: Ulf Hansson <ulf.hansson@linaro.org>
The comment and the define about how TAPs are selected were confusing to
me because the good TAP was only valid if it was bigger than a *_MAX_*
value. Rename the define and adapt the comment to what really happens.
Signed-off-by: Wolfram Sang <wsa+renesas@sang-engineering.com>
Reviewed-by: Yoshihiro Shimoda <yoshihiro.shimoda.uh@renesas.com>
Link: https://lore.kernel.org/r/20200408094638.10375-3-wsa+renesas@sang-engineering.com
Signed-off-by: Ulf Hansson <ulf.hansson@linaro.org>
To select the best TAP, we need to find the longest stream of set bits
in a bit field. There is now a helper function for bitmaps which
iterates over all region of set bits. Using it makes the code much more
concise and easier to understand. Double so, because we need to handle
two bitmaps in the near future. Remove a superfluous comment while here.
Signed-off-by: Wolfram Sang <wsa+renesas@sang-engineering.com>
Reviewed-by: Yoshihiro Shimoda <yoshihiro.shimoda.uh@renesas.com>
Link: https://lore.kernel.org/r/20200408094638.10375-2-wsa+renesas@sang-engineering.com
Signed-off-by: Ulf Hansson <ulf.hansson@linaro.org>
Use FIELD_GET and FIELD_PREP to get access to the register fields. Delete
the shift macros and use GENMASK() for the touched macros.
Note that, this has the side-effect of changing the constants to 64-bit on
64-bit platforms.
Signed-off-by: Masahiro Yamada <yamada.masahiro@socionext.com>
Acked-by: Adrian Hunter <adrian.hunter@intel.com>
Link: https://lore.kernel.org/r/20200408072105.422-2-yamada.masahiro@socionext.com
Signed-off-by: Ulf Hansson <ulf.hansson@linaro.org>
In the SDHCI specification, the Capabilities Register (Offset 0x40h)
is the 64-bit width register, but in Linux, it is represented as two
registers, SDHCI_CAPABILITIES and SDHCI_CAPABILITIES_1 so that drivers
can use 32-bit register accessors.
The upper 32-bit field is associated with SDHCI_CAPABILITIES_1.
Move the definition of SDHCI_CAPABILITIES_1 to the correct place.
Signed-off-by: Masahiro Yamada <yamada.masahiro@socionext.com>
Acked-by: Adrian Hunter <adrian.hunter@intel.com>
Link: https://lore.kernel.org/r/20200408072105.422-1-yamada.masahiro@socionext.com
Signed-off-by: Ulf Hansson <ulf.hansson@linaro.org>
The SDHCI clock operations are platform specific. So it better to define
them separately for particular platform. This will prevent multiple
if..else conditions and will make it easy for user to add their own
clock operations handlers.
Signed-off-by: Manish Narani <manish.narani@xilinx.com>
Link: https://lore.kernel.org/r/1586195015-128992-6-git-send-email-manish.narani@xilinx.com
Signed-off-by: Ulf Hansson <ulf.hansson@linaro.org>
Existing driver code has the platform specific structures scattered
throughout the driver code. Rearrange the platform specific data
structures for more modularity and readability. This will help in adding
new static functions with more ease.
Signed-off-by: Manish Narani <manish.narani@xilinx.com>
Link: https://lore.kernel.org/r/1586195015-128992-5-git-send-email-manish.narani@xilinx.com
Signed-off-by: Ulf Hansson <ulf.hansson@linaro.org>
There is 'struct sdhci_arasan_data' but also
'struct sdhci_arasan_of_data sdhci_arasan_data'. Rename the latter to
avoid confusion with the name of the struct.
Reported-by: Adrian Hunter <adrian.hunter@intel.com>
Signed-off-by: Manish Narani <manish.narani@xilinx.com>
Link: https://lore.kernel.org/r/1586195015-128992-4-git-send-email-manish.narani@xilinx.com
Signed-off-by: Ulf Hansson <ulf.hansson@linaro.org>
Add support to set tap delays for Xilinx Versal SD controller. The tap
delay registers have moved to SD controller space in Versal. Make the
changes accordingly.
Signed-off-by: Manish Narani <manish.narani@xilinx.com>
Acked-by: Adrian Hunter <adrian.hunter@intel.com>
Link: https://lore.kernel.org/r/1586195015-128992-3-git-send-email-manish.narani@xilinx.com
Signed-off-by: Ulf Hansson <ulf.hansson@linaro.org>
The in-parameter struct mmc_data *data is never NULL, because the caller
always provides a valid pointer. Hence drop the corresponding redundant
code.
Signed-off-by: Ulf Hansson <ulf.hansson@linaro.org>
Acked-by: Krzysztof Kozlowski <krzk@kernel.org>
Link: https://lore.kernel.org/r/20200407143903.22477-1-ulf.hansson@linaro.org
The in-parameter "wait" is always set to 0 by the caller, hence just drop
it and its corresponding code.
Signed-off-by: Ulf Hansson <ulf.hansson@linaro.org>
Acked-by: Manuel Lauss <manuel.lauss@gmail.com>
Link: https://lore.kernel.org/r/20200406114337.8802-1-ulf.hansson@linaro.org
The MMC_CAP_ERASE and MMC_CAP_CMD23 flags are already being set in the
common sdhci_setup_host(). This makes it redundant to set them for
sdhci-sprd, so let's drop them.
Signed-off-by: Ulf Hansson <ulf.hansson@linaro.org>
Reviewed-by: Baolin Wang <baolin.wang7@gmail.com>
Acked-by: Adrian Hunter <adrian.hunter@intel.com>
Link: https://lore.kernel.org/r/20200406113724.8504-1-ulf.hansson@linaro.org
Instead of explicitly checking for SDIO specific requests and then
returning an error code, let's set MMC_CAP2_NO_SDIO to tell the mmc core to
prevent them altogether.
Signed-off-by: Ulf Hansson <ulf.hansson@linaro.org>
Link: https://lore.kernel.org/r/20200401145531.23247-1-ulf.hansson@linaro.org
Move the prototypes for sched_ttwu_pending() and send_call_function_single_ipi()
into the newly created kernel/sched/smp.h header, to make sure they are all
the same, and to architectures happy that use -Wmissing-prototypes.
Signed-off-by: Ingo Molnar <mingo@kernel.org>
Instead of warning when mutex_is_locked(), just use the lockdep
framework. The code is smaller and checks could be disabled for
production environments (it is useful only during development).
Put asserts at beginning of function, even before validating arguments.
The behavior of update_devfreq() is now changed because lockdep assert
will only print a warning, not return with EINVAL.
Signed-off-by: Krzysztof Kozlowski <krzk@kernel.org>
Signed-off-by: Chanwoo Choi <cw00.choi@samsung.com>
Fix inconsistent IS_ERR and PTR_ERR in imx_bus_init_icc().
The proper pointer to be passed as argument to PTR_ERR() is
priv->icc_pdev.
This bug was detected with the help of Coccinelle.
Fixes: 16c1d2f1b0bd ("PM / devfreq: imx: Register interconnect device")
Signed-off-by: Gustavo A. R. Silva <gustavoars@kernel.org>
Reviewed-by: Dong Aisheng <aisheng.dong@nxp.com>
[cw00.choi: Edit the patch title from 'imx' to 'imx-bus']
Signed-off-by: Chanwoo Choi <cw00.choi@samsung.com>
GCC produces this warning when kernel compiled using `make W=1`:
warning: ‘strncpy’ specified bound 16 equals destination size [-Wstringop-truncation]
772 | strncpy(devfreq->governor_name, governor_name, DEVFREQ_NAME_LEN);
The strncpy doesn't take care of NULL-termination of the destination
buffer, while the strscpy does.
Signed-off-by: Dmitry Osipenko <digetx@gmail.com>
Signed-off-by: Chanwoo Choi <cw00.choi@samsung.com>
There is no single device which can represent the imx interconnect.
Instead of adding a virtual one just make the main &noc act as the
global interconnect provider.
The imx interconnect provider driver will scale the NOC and DDRC based
on bandwidth request. More scalable nodes can be added in the future,
for example for audio/display/vpu/gpu NICs.
Signed-off-by: Leonard Crestez <leonard.crestez@nxp.com>
Tested-by: Martin Kepplinger <martin.kepplinger@puri.sm>
Acked-by: Chanwoo Choi <cw00.choi@samsung.com>
Signed-off-by: Chanwoo Choi <cw00.choi@samsung.com>
Add initial support for dynamic frequency switching on pieces of the imx
interconnect fabric.
All this driver does is set a clk rate based on an opp table, it does
not map register areas.
Signed-off-by: Leonard Crestez <leonard.crestez@nxp.com>
Tested-by: Martin Kepplinger <martin.kepplinger@puri.sm>
Acked-by: Chanwoo Choi <cw00.choi@samsung.com>
Signed-off-by: Chanwoo Choi <cw00.choi@samsung.com>
The function “platform_get_irq” can log an error already.
Thus omit a redundant message for the exception handling in the
calling function.
This issue was detected by using the Coccinelle software.
Signed-off-by: Markus Elfring <elfring@users.sourceforge.net>
Reviewed-by: Dmitry Osipenko <digetx@gmail.com>
Signed-off-by: Chanwoo Choi <cw00.choi@samsung.com>
We're taking into account both HW memory-accesses + CPU activity based on
current CPU's frequency. For memory-accesses there is a kind of hysteresis
in a form of "boosting" which is managed by the tegra30-devfreq driver.
If current HW memory activity is higher than activity judged based of the
CPU's frequency, then there is no need to schedule cpufreq_update_work
because the result of the work will be a NO-OP. And thus,
tegra_actmon_cpufreq_contribution() should return 0, meaning that at the
moment CPU frequency doesn't contribute anything to the final decision
about required memory clock rate.
Signed-off-by: Dmitry Osipenko <digetx@gmail.com>
Signed-off-by: Chanwoo Choi <cw00.choi@samsung.com>
We have a I915_REQUEST_NOPREEMPT flag that we set when we must prevent
the HW from preempting during the course of this request. We need to
honour this flag and protect the HW even if we have a heartbeat request,
or other maximum priority barrier, pending. As such, restrict the
timeslicing check to avoid preempting into the topmost priority band,
leaving the unpreemptable requests in blissful peace running
uninterrupted on the HW.
v2: Set the I915_PRIORITY_BARRIER to be less than
I915_PRIORITY_UNPREEMPTABLE so that we never submit a request
(heartbeat or barrier) that can legitimately preempt the current
non-premptable request.
Fixes: 2a98f4e65b ("drm/i915: add infrastructure to hold off preemption on a request")
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Cc: Tvrtko Ursulin <tvrtko.ursulin@intel.com>
Reviewed-by: Tvrtko Ursulin <tvrtko.ursulin@intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20200527162418.24755-1-chris@chris-wilson.co.uk
The recent commit: 90b5363acd ("sched: Clean up scheduler_ipi()")
got smp_call_function_single_async() subtly wrong. Even though it will
return -EBUSY when trying to re-use a csd, that condition is not
atomic and still requires external serialization.
The change in ttwu_queue_remote() got this wrong.
While on first reading ttwu_queue_remote() has an atomic test-and-set
that appears to serialize the use, the matching 'release' is not in
the right place to actually guarantee this serialization.
The actual race is vs the sched_ttwu_pending() call in the idle loop;
that can run the wakeup-list without consuming the CSD.
Instead of trying to chain the lists, merge them.
Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Signed-off-by: Ingo Molnar <mingo@kernel.org>
Link: https://lore.kernel.org/r/20200526161908.129371594@infradead.org
In preparation of removing rq->wake_list, replace the
!list_empty(rq->wake_list) with rq->ttwu_pending. This is not fully
equivalent as this new variable is racy.
Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Signed-off-by: Ingo Molnar <mingo@kernel.org>
Link: https://lore.kernel.org/r/20200526161908.070399698@infradead.org
Currently irq_work_queue_on() will issue an unconditional
arch_send_call_function_single_ipi() and has the handler do
irq_work_run().
This is unfortunate in that it makes the IPI handler look at a second
cacheline and it misses the opportunity to avoid the IPI. Instead note
that struct irq_work and struct __call_single_data are very similar in
layout, so use a few bits in the flags word to encode a type and stick
the irq_work on the call_single_queue list.
Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Signed-off-by: Ingo Molnar <mingo@kernel.org>
Link: https://lore.kernel.org/r/20200526161908.011635912@infradead.org
