Even though the STM32 TRNG hardware produces 2- or 4-byte sized words of
random data before triggering an interrupt, the driver currently discards
all but the bottom byte: 50/75% of the produced entropy goes to waste!
Make sure we consume all the random data from each word we read to improve
the entropy generation rate seen by users of the driver.
Signed-off-by: Mathieu Choplain <mathieu.choplain@st.com>
The low level function names in the stm32l4xx low level driver are
different than all the other stm32's even though the functionality is the
same. This breaks the entropy module for these parts. This patch fixes the
build and has been tested on a custom stm32l4p5 board.
This STM32CubeL4 issue has been reported (ST Internal Reference: 207828).
Fixes: https://github.com/zephyrproject-rtos/zephyr/issues/88621
Signed-off-by: Matt Ihnen <matt.ihnen@gmail.com>
The STM32WB09 TRNG has a prescaler that defaults (on reset) to 256, making
the entropy generation process unbearably slow. Change the prescaler value
to 1 instead, in order to make the IP about as fast as other STM32's.
Signed-off-by: Mathieu Choplain <mathieu.choplain@st.com>
Add support for TRNG peripherals that lack interrupt lines in the STM32
entropy driver.
This enables usage of the TRNG of STM32WB05/06/07 SoCs with the driver.
Signed-off-by: Mathieu Choplain <mathieu.choplain@st.com>
Add a semaphore to protect the disable access to the RNG peripheral
when the PM is going to suspend mode.
The change does not affect other stm32 than CONFIG_SOC_SERIES_STM32WBX
or CONFIG_STM32H7_DUAL_CORE
Signed-off-by: Francois Ramu <francois.ramu@st.com>
With PM, when resuming from low power mode, reenable the RNG register
clocks and check the health register. If it is not set to the desired
value, it means we exit Suspend to RAM mode, and that the RNG needs to be
reinitialized.
Signed-off-by: Guillaume Gautier <guillaume.gautier-ext@st.com>
STM32WBA controller uses a PKA driver to perform cyphering operations
on keys. Since PKA hardware block requires RNG clock to be enabled, a
synchronization with zephyr RNG driver is needed.
Use RNG enable status to check if RNG could be switched off or needs to
be switched on.
Similarly in entropy driver, don't cut RNG clock if PKA is enabled.
Signed-off-by: Erwan Gouriou <erwan.gouriou@st.com>
Enable the RNG module interrupt every time it is resumed.
It is done to make sure the interrupt is always enabled. The CR register
may not persist when the device clock is disabled on some chips.
Signed-off-by: Dawid Niedzwiecki <dawidn@google.com>
Move enabling/disabling the RNG module to acquire/release functions.
It causes enabling the RNG module for the get_entropy_isr function.
It fixes hanging in the get_entropy_isr function.
Signed-off-by: Dawid Niedzwiecki <dawidn@google.com>
Suspend the RNG module when the pool is full to save power. The
generated numbers aren't used anyway.
Signed-off-by: Dawid Niedzwiecki <dawidn@google.com>
rand32.h does not make much sense, since the random subsystem
provides more APIs than just getting a random 32 bits value.
Rename it to random.h and get consistently with other
subsystems.
Signed-off-by: Flavio Ceolin <flavio.ceolin@intel.com>
When the legacy LLCP implementation was removed this Kconfig option was
mistakenly left over. Remove it now with all its users.
Fixes#63212.
Signed-off-by: Carles Cufi <carles.cufi@nordicsemi.no>
Enable PM on STM32 entropy driver.
Only supports PM_DEVICE for now, runtime support will be added later.
Signed-off-by: Erwan Gouriou <erwan.gouriou@st.com>
Clang doesn't support push_options, pop_options and optimize pragmas.
Since "Ofast" is only applied to rng_pool_get() we can use function
attributes instead of pragmas.
Please note that clang doesn't support applying optimizations to
functions.
This change was verified by comparing assembler output.
Signed-off-by: Patryk Duda <pdk@semihalf.com>
Don't check clock configuration before kernel was started since it
will prevent message to be displayed.
After kernel initialization, message is still valid and then ASSERT
more valuable to the user.
Signed-off-by: Erwan Gouriou <erwan.gouriou@linaro.org>
RNG clock configuration constraints differ between each series.
Rather than providing complex build time code to verify RNG clock
configuration is correct, take advantage of CECS bit (Clock error
current status) to assess clock configuration.
This check is implemented under a specific ENTROPY_STM32_CLK_CHECK
Kconfig option. This allows user to disable this feature in specific
conditions:
- CED bit disabled in application (in which case CECS status is not valid)
- Clock configuration is deemed as correct by user. Note that RNG number
are always generated, whatever the clock status.
Signed-off-by: Erwan Gouriou <erwan.gouriou@linaro.org>
RNG domain source clock is now configured via call to clock_control
driver.
Besides, add static checks to verify domain clock configuration
is correct:
- If HSI48 is used because it is default domain clock config,
it should be enabled
- If no HSI48 is available, a specific domain clock should be set
- In L0 case, PLL could be used as domain clock only at a specific freq.
Signed-off-by: Erwan Gouriou <erwan.gouriou@linaro.org>
Use STM32_DT_INST_CLOCKS to populate device clock information.
This will allow to add clock source information in next commits.
Signed-off-by: Erwan Gouriou <erwan.gouriou@linaro.org>
The HSI48 is enabled by clock control driver.
It is no more done by each driver that requires this clock
However when using rng or sdmmc or bluetooth/ipm or usb,
the HSI48 clock must be present in the DTS.
Add a warning for this particular check but keep the deprecated
HSI48 clock enable : keeping for legacy but to remove later.
Signed-off-by: Francois Ramu <francois.ramu@st.com>
Change automated searching for files using "IRQ_CONNECT()" API not
including <zephyr/irq.h>.
Signed-off-by: Gerard Marull-Paretas <gerard.marull@nordicsemi.no>
The driver is writing the DTS nist-config to the RNG periph.
following the validation sequence described by the RefMan.
It depends on the he RNG IP version and is present on some
mcu devices : all with CONDRST bit.
Depends on the RNG CR Autoreset.
Takes the Health test control Register if property is given.
Signed-off-by: Francois Ramu <francois.ramu@st.com>
For the stm32 mcu with a CONDRST bit in its RNG_CR
for conditioning Soft Reset, the driver must wait for
the bit to be 0 after disabling. This could take
about 2 AHB clock cycles + 2RNG clock cycles.
Signed-off-by: Francois Ramu <francois.ramu@st.com>
On STM32WB and dual-core STM32H7 MCUs, the RNG peripheral is shared
between the cores and its access is protected by a hardware semaphore.
Locking was not performed in the current entropy driver, leading to a
race condition when multiple cores concurrently used the RNG. This
commit implements the necessary logic for locking the HSEM during entropy
generation on multi-core STM32 MCUs. It also reconfigures the RNG in case
the configuration was changed by the other core, as this can happen e.g
on STM32WB MCUs.
Signed-off-by: Thomas Altenbach <taltenbach@witekio.com>
In order to bring consistency in-tree, migrate all drivers to the new
prefix <zephyr/...>. Note that the conversion has been scripted, refer
to #45388 for more details.
Signed-off-by: Gerard Marull-Paretas <gerard.marull@nordicsemi.no>
The pm_constraint_* APIs were effectively used by the policy manager
only. This patch renames the API to the policy namespace and makes its
naming more explicit:
- pm_constraint_set -> pm_policy_state_lock_get()
- pm_constraint_release -> pm_policy_state_lock_put()
- pm_constraint_get -> pm_policy_state_lock_is_active()
The reason for these changes is that constraints can be of many types:
allow/disallow states, impose latency requirements, etc. The new naming
also makes explicit that the API calls will influence the PM policy
behavior.
All drivers and documentation have been updated accordingly.
Signed-off-by: Gerard Marull-Paretas <gerard.marull@nordicsemi.no>
When a Seed error occurs during the random nb generation,
the driver tries to recover and exit without providing a random data
This avoids looping endlessly on the DRDY bit of the RNG status reg
because it remains 0 in case of error.
Signed-off-by: Francois Ramu <francois.ramu@st.com>
Add power management constraints to the entropy driver.
This prevents the hardware block to lose it's clock when
going into any stop mode of the cpu, which would cause the
clock error flag to be set while filling the pool.
Signed-off-by: Wouter Cappelle <wouter.cappelle@crodeon.com>
The properties from dts to driver were not translated
correctly from - to _. This PR fixes this
Signed-off-by: Wouter Cappelle <wouter.cappelle@crodeon.com>
Stop using redundant DEV_DATA/DEV_CFG macros and use dev->data and
dev->config instead.
Signed-off-by: Gerard Marull-Paretas <gerard.marull@nordicsemi.no>
Some STM32 series need to configure health test register
for proper RNG behavior.
In addition, some also require to write a Magic number
before writing the configuration.
Signed-off-by: Alexandre Bourdiol <alexandre.bourdiol@st.com>
Introduce Kconfig for setting the driver initialization priority across
the entropy drivers and add a call to the logging template.
Signed-off-by: Henrik Brix Andersen <henrik@brixandersen.dk>
According to the Reference Manual of several series(G0,L5,WL,WB,...)
RNG_DR register value should only be used if it is different from 0:
"Because when it is the case a seed error occurred between RNG_SR
polling and RND_DR output reading (rare event)."
Signed-off-by: Thomas Stranger <thomas.stranger@outlook.com>
This commit introduces an automatic recovery procedure in cases an
entropy source error was detected.
- On Series with soft reset support a soft rest is executed.
- On Series w/o soft reset support the pipeline is cleared by reading
the RNG_DR 12 times.
With this changes the check for seed errors uses SEIS flag instead
of the SECS flag.
Signed-off-by: Thomas Stranger <thomas.stranger@outlook.com>
This commit adds entropy support for stm32wl and stm32g0.
Pll is used as clock source and has to be enabled,
other clock sources are not supported at the moment.
Signed-off-by: Thomas Stranger <thomas.stranger@outlook.com>
This commit selects LL_RCC_RNG_CLKSOURCE_CLK48 as a clock source
to rng peripheral. LL_RCC_RNG_CLKSOURCE_CLK48 is CLK48 divided by 3.
Signed-off-by: Krishna Mohan Dani <krishnamohan.d@hcl.com>
Function random_byte_get() returns only the least significant byte of
the 32-bit random datum, as this is the used value, so avoiding that
higher numbers are interpreted as negative error codes and their value
is not discarded.
Signed-off-by: Giancarlo Stasi <giancarlo.stasi.co@gmail.com>
Convert from device_get_binding to DEVICE_DT_GET. In doing this we
no longer need the label in the devicetree node so we remove that.
Removed all __ASSERT_NO_MSG(clk) since we'll get a build error if
DEVICE_DT_GET cant be satisfied, and the clock control api's will
handle reporting if the device_is_ready.
Signed-off-by: Kumar Gala <kumar.gala@linaro.org>
Mathieu Choplain
Matt Ihnen
Pieter De Gendt
Flavio Ceolin
Francois Ramu
John Johnson
Guillaume Gautier
Erwan Gouriou
Dawid Niedzwiecki
Erwan Gouriou
Carles Cufi
Patryk Duda
Carlo Caione
Gerard Marull-Paretas
Henrik Brix Andersen
Thomas Altenbach
Nazar Kazakov
Wouter Cappelle
Alexandre Bourdiol
Henrik Brix Andersen
Thomas Stranger
Krishna Mohan Dani
Anas Nashif
Giancarlo Stasi
Kumar Gala