Hi Chris,
Sorry for the inconvenience. The discussion to figure out the new guidance did cost some time.
The initial conclusion is that the weak symbol/function should be avoided in new code added to TF-M. Platform support code may use weak symbols as an exception to make platform specific implementation flexible.
The code guidance will be updated soon to address this issue formally.
Best regards,
Hu Ziji
-----Original Message-----
From: TF-M <tf-m-bounces(a)lists.trustedfirmware.org> On Behalf Of Christopher Brand via TF-M
Sent: Wednesday, October 30, 2019 7:50 PM
To: tf-m(a)lists.trustedfirmware.org
Subject: [TF-M] Weak symbols
Hi,
There are a couple of code reviews that I've been involved with that seem to be being held up due to the use of weak symbols.
https://review.trustedfirmware.org/c/trusted-firmware-m/+/1973https://review.trustedfirmware.org/c/trusted-firmware-m/+/2212
There seems to be a feeling among some reviewers that weak symbols are discouraged in TF-M, due to them not being in the C standard.
There's no mention of weak symbols in docs/coding_guide.rst. There is a brief mention in docs/user_guides/tfm_secure_irq_handling.rst:
In Library model a function with the name derived from the value of the ``line_num`` or ``line_name`` property is generated. This function will be put in the vector table by the linker (as the handlers in the startup assembly are defined as weak symbols). The code generated for this function will forward the call to the function with the name of the value of the ``signal`` property post-fixed with ``_isr``.
So that document explicitly states that weak symbols will be used in TF-M.
TFM already uses weak symbols fairly extensively. A quick grep shows two instances under "interface": interface/src/tfm_ns_interface.c:
__attribute__((weak)) int32_t tfm_ns_interface_dispatch(veneer_fn fn, [...]
__attribute__((weak)) enum tfm_status_e tfm_ns_interface_init(void) and several hundred under "platform":
$ grep -ri weak platform/ --exclude=tfm_mbedcrypto_config.h --exclude=cmsis_\*.h | wc
747 5902 109938
Looking at Phabricator, the only mention of weak symbols I can find is in https://developer.trustedfirmware.org/T363 and https://developer.trustedfirmware.org/T198, both of which explicitly state that new weak functions will be added, and https://developer.trustedfirmware.org/T463, which doesn't take a position either way.
Weak symbols have been discussed a little on the mailing list in the thread at https://lists.trustedfirmware.org/pipermail/tf-m/2019-October/000430.html but not very extensively.
My feeling is that it wouldn't be unreasonable to decide that weak symbols should be discouraged, and it might even make sense for somebody to take on the task of removing them from TF-M, although that looks like a significant amount of work. Of course it would be nice to see that documented, presumably in docs/coding_guide.rst (and presumably corresponding changes to docs/user_guides/tfm_secure_irq_handling.rst and tasks T198 and T463). As things stand right now, they do seem to be an accepted mechanism within TF-M, and worrying about adding a handful more to the nearly 750 we have now seems unreasonable - if we do decide at some point not to use weak symbols, the work involved in removing 760 or even 800 Is not significantly more than would be needed now.
The code reviews I cited have been out there for 6 weeks now. It would be really nice to find a way forward...
Thanks,
Chris
This message and any attachments may contain confidential information from Cypress or its subsidiaries. If it has been received in error, please advise the sender and immediately delete this message.
--
TF-M mailing list
TF-M(a)lists.trustedfirmware.org
https://lists.trustedfirmware.org/mailman/listinfo/tf-m
After further analysis, I found this to be true:
If the secure interrupt has a priority lower than the secure pendsv interrupt, then the associated SP code will run in thread mode at the exception priority level of the secure interrupt.
If the priority of the secure interrupt is equal to or greater than the secure pendsv interrupt then associated the SP code will run in thread mode with NO exception priority level.
Alan
-----Original Message-----
From: DeMars, Alan
Sent: Tuesday, October 29, 2019 8:54 AM
To: 'Mate Toth-Pal'
Cc: 'tf-m(a)lists.trustedfirmware.org'
Subject: RE: SP thread priority when servicing secure interrupts
Mate,
I'm using the IPC model. My secure interrupt has priority 0xe0, whereas the pendSV interrupt has priority 0x60.
I'll attempt to provide a procedure for reproducing my results.
In the meantime, is there a doubt-free way to determine the execution priority while the processor is in thread mode?
Alan
-----Original Message-----
From: TF-M [mailto:tf-m-bounces@lists.trustedfirmware.org] On Behalf Of Mate Toth-Pal via TF-M
Sent: Tuesday, October 29, 2019 5:50 AM
To: tf-m(a)lists.trustedfirmware.org
Cc: nd
Subject: [EXTERNAL] Re: [TF-M] SP thread priority when servicing secure interrupts
Hi Alan,
The expected behaviour depends on the Model used in the TF-M.
Library model:
The generated interrupt handler runs in Handler mode, and after doing an SVC, it starts executing the handler defined in the SP in thread mode. At that point the interrupt handler hadn't returned, so the interrupt is still active in the NVIC. This means that the execution priority of the CPU is equal to the interrupt's priority, even if the processor is executing in thread mode. At the end of the execution of the SP interrupt handler, an SVC is executed, which then return to the context of the generated interrupt handler. And only at this point does the interrupt handler return, and can a pending interrupt with lower priority be active.
IPC model:
The generated interrupt handler runs at the interrupt’s assigned priority, sets a signal for the partition that should handle the interrupt, and then returns. So the interrupt is cleared in the NVIC. After this, the secure scheduler can schedule the handling partition (the scheduling happens in a PendSV handler), and the signal handler runs in thread mode. In this case the priority of the thread mode is 256 (no priority boosting, and no active exceptions). So the execution can be interrupted. The generated stub is not run again, until the interrupt gets triggered again.
If you still find the behaviour of the system unexpected, could you please share some more information about the environment and sequence of steps for reproducing the behaviour?
Best would be if you could reproduce the behaviour with one of the platforms supported by the upstream TF-M, and share your patch, so Others could have a look at it in the debugger as well.
Thanks,
Mate
-----Original Message-----
From: TF-M <tf-m-bounces(a)lists.trustedfirmware.org> On Behalf Of DeMars, Alan via TF-M
Sent: 29 October 2019 01:52
To: 'tf-m(a)lists.trustedfirmware.org' <tf-m(a)lists.trustedfirmware.org>
Subject: [TF-M] SP thread priority when servicing secure interrupts
It appears that when a SP is servicing a manifest-specified secure interrupt configured with a priority equal to or lower than the pendsv interrupt, the processor is operating at the priority of the pendsv interrupt.
While the SP is servicing the secure interrupt's signal (ie: in thread mode), If another interrupt is triggered of the same priority or lower than the pendsv, that new interrupt is not immediately vectored to. Instead, it is vectored to AFTER the secure interrupt's stub function (generated by the template) returns.
Is this true and expected behavior?
Unless BASEPRI is set to some non-zero value, I thought the processor was always running with no exception priority while in thread mode.
Alan
--
TF-M mailing list
TF-M(a)lists.trustedfirmware.org
https://lists.trustedfirmware.org/mailman/listinfo/tf-m
--
TF-M mailing list
TF-M(a)lists.trustedfirmware.org
https://lists.trustedfirmware.org/mailman/listinfo/tf-m
Hi,
There are a couple of code reviews that I've been involved with that seem to be being held up due to the use of weak symbols.
https://review.trustedfirmware.org/c/trusted-firmware-m/+/1973https://review.trustedfirmware.org/c/trusted-firmware-m/+/2212
There seems to be a feeling among some reviewers that weak symbols are discouraged in TF-M, due to them not being in the C standard.
There's no mention of weak symbols in docs/coding_guide.rst. There is a brief mention in docs/user_guides/tfm_secure_irq_handling.rst:
In Library model a function with the name derived from the value of the
``line_num`` or ``line_name`` property is generated. This function will be put
in the vector table by the linker (as the handlers in the startup assembly are
defined as weak symbols). The code generated for this function will forward the
call to the function with the name of the value of the ``signal`` property
post-fixed with ``_isr``.
So that document explicitly states that weak symbols will be used in TF-M.
TFM already uses weak symbols fairly extensively. A quick grep shows two instances under "interface": interface/src/tfm_ns_interface.c:
__attribute__((weak)) int32_t tfm_ns_interface_dispatch(veneer_fn fn,
[...]
__attribute__((weak)) enum tfm_status_e tfm_ns_interface_init(void)
and several hundred under "platform":
$ grep -ri weak platform/ --exclude=tfm_mbedcrypto_config.h --exclude=cmsis_\*.h | wc
747 5902 109938
Looking at Phabricator, the only mention of weak symbols I can find is in https://developer.trustedfirmware.org/T363 and https://developer.trustedfirmware.org/T198, both of which explicitly state that new weak functions will be added, and https://developer.trustedfirmware.org/T463, which doesn't take a position either way.
Weak symbols have been discussed a little on the mailing list in the thread at https://lists.trustedfirmware.org/pipermail/tf-m/2019-October/000430.html but not very extensively.
My feeling is that it wouldn't be unreasonable to decide that weak symbols should be discouraged, and it might even make sense for somebody to take on the task of removing them from TF-M, although that looks like a significant amount of work. Of course it would be nice to see that documented, presumably in docs/coding_guide.rst (and presumably corresponding changes to docs/user_guides/tfm_secure_irq_handling.rst and tasks T198 and T463). As things stand right now, they do seem to be an accepted mechanism within TF-M, and worrying about adding a handful more to the nearly 750 we have now seems unreasonable - if we do decide at some point not to use weak symbols, the work involved in removing 760 or even 800 Is not significantly more than would be needed now.
The code reviews I cited have been out there for 6 weeks now. It would be really nice to find a way forward...
Thanks,
Chris
This message and any attachments may contain confidential information from Cypress or its subsidiaries. If it has been received in error, please advise the sender and immediately delete this message.
Is there a "search" interface for this mailing list somewhere? The archive page at https://lists.trustedfirmware.org/pipermail/tf-m/ doesn't seem to provide one (it's not too onerous at the moment to download the 12 gzipped files and search them, but that's going to get worse as time goes by).
Thanks,
Chris
This message and any attachments may contain confidential information from Cypress or its subsidiaries. If it has been received in error, please advise the sender and immediately delete this message.
Mate,
I'm using the IPC model. My secure interrupt has priority 0xe0, whereas the pendSV interrupt has priority 0x60.
I'll attempt to provide a procedure for reproducing my results.
In the meantime, is there a doubt-free way to determine the execution priority while the processor is in thread mode?
Alan
-----Original Message-----
From: TF-M [mailto:tf-m-bounces@lists.trustedfirmware.org] On Behalf Of Mate Toth-Pal via TF-M
Sent: Tuesday, October 29, 2019 5:50 AM
To: tf-m(a)lists.trustedfirmware.org
Cc: nd
Subject: [EXTERNAL] Re: [TF-M] SP thread priority when servicing secure interrupts
Hi Alan,
The expected behaviour depends on the Model used in the TF-M.
Library model:
The generated interrupt handler runs in Handler mode, and after doing an SVC, it starts executing the handler defined in the SP in thread mode. At that point the interrupt handler hadn't returned, so the interrupt is still active in the NVIC. This means that the execution priority of the CPU is equal to the interrupt's priority, even if the processor is executing in thread mode. At the end of the execution of the SP interrupt handler, an SVC is executed, which then return to the context of the generated interrupt handler. And only at this point does the interrupt handler return, and can a pending interrupt with lower priority be active.
IPC model:
The generated interrupt handler runs at the interrupt’s assigned priority, sets a signal for the partition that should handle the interrupt, and then returns. So the interrupt is cleared in the NVIC. After this, the secure scheduler can schedule the handling partition (the scheduling happens in a PendSV handler), and the signal handler runs in thread mode. In this case the priority of the thread mode is 256 (no priority boosting, and no active exceptions). So the execution can be interrupted. The generated stub is not run again, until the interrupt gets triggered again.
If you still find the behaviour of the system unexpected, could you please share some more information about the environment and sequence of steps for reproducing the behaviour?
Best would be if you could reproduce the behaviour with one of the platforms supported by the upstream TF-M, and share your patch, so Others could have a look at it in the debugger as well.
Thanks,
Mate
-----Original Message-----
From: TF-M <tf-m-bounces(a)lists.trustedfirmware.org> On Behalf Of DeMars, Alan via TF-M
Sent: 29 October 2019 01:52
To: 'tf-m(a)lists.trustedfirmware.org' <tf-m(a)lists.trustedfirmware.org>
Subject: [TF-M] SP thread priority when servicing secure interrupts
It appears that when a SP is servicing a manifest-specified secure interrupt configured with a priority equal to or lower than the pendsv interrupt, the processor is operating at the priority of the pendsv interrupt.
While the SP is servicing the secure interrupt's signal (ie: in thread mode), If another interrupt is triggered of the same priority or lower than the pendsv, that new interrupt is not immediately vectored to. Instead, it is vectored to AFTER the secure interrupt's stub function (generated by the template) returns.
Is this true and expected behavior?
Unless BASEPRI is set to some non-zero value, I thought the processor was always running with no exception priority while in thread mode.
Alan
--
TF-M mailing list
TF-M(a)lists.trustedfirmware.org
https://lists.trustedfirmware.org/mailman/listinfo/tf-m
--
TF-M mailing list
TF-M(a)lists.trustedfirmware.org
https://lists.trustedfirmware.org/mailman/listinfo/tf-m
Hi Alan,
The expected behaviour depends on the Model used in the TF-M.
Library model:
The generated interrupt handler runs in Handler mode, and after doing an SVC, it starts executing the handler defined in the SP in thread mode. At that point the interrupt handler hadn't returned, so the interrupt is still active in the NVIC. This means that the execution priority of the CPU is equal to the interrupt's priority, even if the processor is executing in thread mode. At the end of the execution of the SP interrupt handler, an SVC is executed, which then return to the context of the generated interrupt handler. And only at this point does the interrupt handler return, and can a pending interrupt with lower priority be active.
IPC model:
The generated interrupt handler runs at the interrupt’s assigned priority, sets a signal for the partition that should handle the interrupt, and then returns. So the interrupt is cleared in the NVIC. After this, the secure scheduler can schedule the handling partition (the scheduling happens in a PendSV handler), and the signal handler runs in thread mode. In this case the priority of the thread mode is 256 (no priority boosting, and no active exceptions). So the execution can be interrupted. The generated stub is not run again, until the interrupt gets triggered again.
If you still find the behaviour of the system unexpected, could you please share some more information about the environment and sequence of steps for reproducing the behaviour?
Best would be if you could reproduce the behaviour with one of the platforms supported by the upstream TF-M, and share your patch, so Others could have a look at it in the debugger as well.
Thanks,
Mate
-----Original Message-----
From: TF-M <tf-m-bounces(a)lists.trustedfirmware.org> On Behalf Of DeMars, Alan via TF-M
Sent: 29 October 2019 01:52
To: 'tf-m(a)lists.trustedfirmware.org' <tf-m(a)lists.trustedfirmware.org>
Subject: [TF-M] SP thread priority when servicing secure interrupts
It appears that when a SP is servicing a manifest-specified secure interrupt configured with a priority equal to or lower than the pendsv interrupt, the processor is operating at the priority of the pendsv interrupt.
While the SP is servicing the secure interrupt's signal (ie: in thread mode), If another interrupt is triggered of the same priority or lower than the pendsv, that new interrupt is not immediately vectored to. Instead, it is vectored to AFTER the secure interrupt's stub function (generated by the template) returns.
Is this true and expected behavior?
Unless BASEPRI is set to some non-zero value, I thought the processor was always running with no exception priority while in thread mode.
Alan
--
TF-M mailing list
TF-M(a)lists.trustedfirmware.org
https://lists.trustedfirmware.org/mailman/listinfo/tf-m
It appears that when a SP is servicing a manifest-specified secure interrupt configured with a priority equal to or lower than the pendsv interrupt, the processor is operating at the priority of the pendsv interrupt.
While the SP is servicing the secure interrupt's signal (ie: in thread mode), If another interrupt is triggered of the same priority or lower than the pendsv, that new interrupt is not immediately vectored to. Instead, it is vectored to AFTER the secure interrupt's stub function (generated by the template) returns.
Is this true and expected behavior?
Unless BASEPRI is set to some non-zero value, I thought the processor was always running with no exception priority while in thread mode.
Alan
Hi,
Back in March, I wrote the dual-core design document at
https://developer.trustedfirmware.org/w/tf_m/design/twin-cpu/bootloader/
It was discussed on the mailing list in a thread that starts at
https://lists.trustedfirmware.org/pipermail/tf-m/2019-March/000088.html
Since then, TF-M has defined a process for design reviews, and it would be
nice to get this design document into the docs/design directory, particularly
as the implementation of the design is already in the codebase.
Given that there has already been discussion on the mailing list, I've gone
ahead and created a review in Gerrit:
https://review.trustedfirmware.org/c/trusted-firmware-m/+/2384
I also added all the people who contributed to that thread as reviewers.
Please feel free to add yourself as a reviewer.
Also, comments on whether this is the right process to follow for design
documentation in this transitional state would be useful.
Thanks,
Chris
This message and any attachments may contain confidential information from Cypress or its subsidiaries. If it has been received in error, please advise the sender and immediately delete this message.
Hi Alan,
Andrej is right about PSA certification, for that you don't need to pass the regression test suite.
However for submitting a change to the upstream master branch on https://www.trustedfirmware.org/ it is a requirement to pass those tests.
The peripherals defined in the tfm_peripherals_def.h files are the peripherals that are mentioned in the 'mmio_regions' node in the partition manifest yaml files. It is a requirement that a macro is defined with this name that is substituted to a pointer type value. For more details see platform\readme.rst, chapter 'platformext/target/tfm_peripherals_def.h'.
The type 'tfm_spm_partition_platform_data_t' is defined by the platform implementation, so TF-M core has no knowledge about the content of an object of type 'tfm_spm_partition_platform_data_t', and doesn't use it. The pointer is only considered by it as a token, that is passed to the platform HAL functions.
This means that if the platform needs to do nothing related to the peripherals in the HAL functions either because
- no isolation is needed to be configured for that peripheral
- the peripheral doesn't exist, and the functionality is emulated (1)
it is OK to define the value of the macro to be NULL (Just as it is done in platform/ext/target/musca_a/tfm_peripherals_def.h).
The TFM_PERIPHERAL_FPGA_IO is used by the core test, to access button state and to blink leds. For the Musca-A platform these functionalities are emulated in the file platform/ext/target/musca_a/plat_test.c
TFM_PERIPHERAL_TIMER0 is used by the IRQ tests. The upstreamed platforms all support some kind of timer, so there is a real implementation for all the platforms. (platform/ext/target/musca_a/plat_test.c)
If it is not possible to create working implementations for the interfaces in platform/include/tfm_plat_test.h, then it is possible to disable the peripheral access and irq tests, by adding the following lines to the platform cmake file:
set(TFM_ENABLE_PERIPH_ACCESS_TEST Off)
set(TFM_ENABLE_IRQ_TEST Off)
Please note, that even in this case TFM_PERIPHERAL_FPGA_IO and TFM_PERIPHERAL_TIMER0 are still need to be defined (NULL is ok), and the plat_test interface functions also need to have at least an empty implementation.
Regards,
Mate
-----Original Message-----
From: TF-M <tf-m-bounces(a)lists.trustedfirmware.org> On Behalf Of Andrej Butok via TF-M
Sent: 24 October 2019 08:39
To: DeMars, Alan <ademars(a)ti.com>
Cc: tf-m(a)lists.trustedfirmware.org
Subject: Re: [TF-M] IPC Regressions reference platform specific resources
Hi Alan
> I assume that to obtain any kind of TF-M compliance certification for our platform, I will be required to demonstrate successful execution of these regression tests.
Regression tests are only to check if your TFM port is correct (not for certification).
You should to demonstrate a successful log from the PSA Test Suite, but only for the "PSA Certified Functional API". For the PSA Certified Level 1, it is not required.
Best regards,
Andrej Butok
NXP Semiconductors
-----Original Message-----
From: TF-M <tf-m-bounces(a)lists.trustedfirmware.org> On Behalf Of DeMars, Alan via TF-M
Sent: Wednesday, October 23, 2019 5:20 PM
To: 'tf-m(a)lists.trustedfirmware.org' <tf-m(a)lists.trustedfirmware.org>
Subject: [TF-M] IPC Regressions reference platform specific resources
I'm attempting to build the ConfigRegressionIPC image and have run into the issue that some of the tests rely on device specific resources (TFM_PERIPHERAL_TIMER0 and TFM_PERIPHERAL_FPGA_IO).
I assume that to obtain any kind of TF-M compliance certification for our platform, I will be required to demonstrate successful execution of these regression tests. If that is so, how should I proceed to resolve this conflict?
Alan
--
TF-M mailing list
TF-M(a)lists.trustedfirmware.org
https://eur01.safelinks.protection.outlook.com/?url=https%3A%2F%2Flists.tru…
--
TF-M mailing list
TF-M(a)lists.trustedfirmware.org
https://lists.trustedfirmware.org/mailman/listinfo/tf-m
Hi,
The process of contributing to TF-M patch is updated now: https://review.trustedfirmware.org/#/c/trusted-firmware-m/+/1007/
We can get an updated version after reviewing.
Regards,
Summer
On 10/23/19, 7:55 PM, "Shebu Varghese Kuriakose via TF-M" <tf-m(a)lists.trustedfirmware.org> wrote:
Hi Alan,
'Create Task' is the correct option to use. You can also add 'Trusted Firmware M' as the project name in the Tags field in the task.
Looks like contributing.rst can do with a few updates/corrections.
Regards,
Shebu
-----Original Message-----
From: TF-M <tf-m-bounces(a)lists.trustedfirmware.org> On Behalf Of DeMars, Alan via TF-M
Sent: Monday, October 21, 2019 7:37 PM
To: 'tf-m(a)lists.trustedfirmware.org' <tf-m(a)lists.trustedfirmware.org>
Subject: [TF-M] contributing to the TF-M project
I am trying to follow the instructions provided here:
https://git.trustedfirmware.org/trusted-firmware-m.git/tree/docs/contributi…
But I'm stumbling on the very first step:
- Create an issue in http://issues.trustedfirmware.org
to keep others informed about your ongoing work.
When I log into the sight, I have not been able to find how to create an issue.
I see links for "Create Task" and "Create Project" but nothing for "Create Issue".
What am I doing wrong?
Alan
--
TF-M mailing list
TF-M(a)lists.trustedfirmware.org
https://lists.trustedfirmware.org/mailman/listinfo/tf-m
--
TF-M mailing list
TF-M(a)lists.trustedfirmware.org
https://lists.trustedfirmware.org/mailman/listinfo/tf-m
