Hi all,
This adds support for asynchronous notifications from OP-TEE in secure world to the OP-TEE driver. This allows a design with a top half and bottom half type of driver where the top half runs in secure interrupt context and a notifications tells normal world to schedule a yielding call to do the bottom half processing.
An interrupt is used to notify the driver that there are asynchronous notifications pending.
v2->v3: * Rebased on v5.14-rc2 which made the patch "dt-bindings: arm: Convert optee binding to json-schema" from the V2 patch set obsolete. * Applied Ard's Acked-by on "optee: add asynchronous notifications"
v1->v2: * Added documentation * Converted optee bindings to json-schema and added interrupt property * Configure notification interrupt from DT instead of getting it from secure world, suggested by Ard Biesheuvel ardb@kernel.org.
Thanks, Jens
Jens Wiklander (6): docs: staging/tee.rst: add a section on OP-TEE notifications dt-bindings: arm: optee: add interrupt property tee: fix put order in teedev_close_context() tee: add tee_dev_open_helper() primitive optee: separate notification functions optee: add asynchronous notifications
.../arm/firmware/linaro,optee-tz.yaml | 4 + Documentation/staging/tee.rst | 27 +++ drivers/tee/optee/Makefile | 1 + drivers/tee/optee/call.c | 27 +++ drivers/tee/optee/core.c | 87 +++++-- drivers/tee/optee/notif.c | 226 ++++++++++++++++++ drivers/tee/optee/optee_msg.h | 9 + drivers/tee/optee/optee_private.h | 23 +- drivers/tee/optee/optee_rpc_cmd.h | 31 +-- drivers/tee/optee/optee_smc.h | 75 +++++- drivers/tee/optee/rpc.c | 73 +----- drivers/tee/tee_core.c | 37 ++- include/linux/tee_drv.h | 27 +++ 13 files changed, 523 insertions(+), 124 deletions(-) create mode 100644 drivers/tee/optee/notif.c
Adds a section on notifications used by OP-TEE, synchronous and asynchronous.
Signed-off-by: Jens Wiklander jens.wiklander@linaro.org --- Documentation/staging/tee.rst | 27 +++++++++++++++++++++++++++ 1 file changed, 27 insertions(+)
diff --git a/Documentation/staging/tee.rst b/Documentation/staging/tee.rst index 4d4b5f889603..37bdd097336f 100644 --- a/Documentation/staging/tee.rst +++ b/Documentation/staging/tee.rst @@ -184,6 +184,33 @@ order to support device enumeration. In other words, OP-TEE driver invokes this application to retrieve a list of Trusted Applications which can be registered as devices on the TEE bus.
+OP-TEE notifications +-------------------- + +There are two kinds of notifications that secure world can use to make +normal world aware of some event. + +1. Synchronous notifications delivered with ``OPTEE_RPC_CMD_NOTIFICATION`` + using the ``OPTEE_RPC_NOTIFICATION_SEND`` parameter. +2. Asynchronous notifications delivered with a combination of a non-secure + interrupt and a fast call from the non-secure interrupt handler. + +Synchronous notifications are limited by depending on RPC for delivery, +this is only usable when secure world is entered with a yielding call via +``OPTEE_SMC_CALL_WITH_ARG``. This excludes such notifications from secure +world interrupt handlers. + +An asynchronous notification is delivered via a non-secure interrupt to an +interrupt handler registered in the OP-TEE driver. The actual notification +value are retrieved with the fast call ``OPTEE_SMC_GET_ASYNC_NOTIF_VALUE``. + +One notification value ``OPTEE_SMC_ASYNC_NOTIF_VALUE_DO_BOTTOM_HALF`` has a +special meaning. When this value is received it means that normal world is +supposed to make a yielding call ``OPTEE_MSG_CMD_DO_BOTTOM_HALF``. This +call is done from the thread assisting the interrupt handler. This is a +building block for OP-TEE OS in secure world to implement the top half and +bottom half style of device drivers. + AMD-TEE driver ==============
On Fri, 23 Jul 2021 10:44:17 +0100, Jens Wiklander jens.wiklander@linaro.org wrote:
Adds a section on notifications used by OP-TEE, synchronous and asynchronous.
Signed-off-by: Jens Wiklander jens.wiklander@linaro.org
Documentation/staging/tee.rst | 27 +++++++++++++++++++++++++++ 1 file changed, 27 insertions(+)
diff --git a/Documentation/staging/tee.rst b/Documentation/staging/tee.rst index 4d4b5f889603..37bdd097336f 100644 --- a/Documentation/staging/tee.rst +++ b/Documentation/staging/tee.rst @@ -184,6 +184,33 @@ order to support device enumeration. In other words, OP-TEE driver invokes this application to retrieve a list of Trusted Applications which can be registered as devices on the TEE bus. +OP-TEE notifications +--------------------
+There are two kinds of notifications that secure world can use to make +normal world aware of some event.
+1. Synchronous notifications delivered with ``OPTEE_RPC_CMD_NOTIFICATION``
- using the ``OPTEE_RPC_NOTIFICATION_SEND`` parameter.
+2. Asynchronous notifications delivered with a combination of a non-secure
- interrupt and a fast call from the non-secure interrupt handler.
+Synchronous notifications are limited by depending on RPC for delivery, +this is only usable when secure world is entered with a yielding call via +``OPTEE_SMC_CALL_WITH_ARG``. This excludes such notifications from secure +world interrupt handlers.
+An asynchronous notification is delivered via a non-secure interrupt to an +interrupt handler registered in the OP-TEE driver. The actual notification +value are retrieved with the fast call ``OPTEE_SMC_GET_ASYNC_NOTIF_VALUE``.
+One notification value ``OPTEE_SMC_ASYNC_NOTIF_VALUE_DO_BOTTOM_HALF`` has a +special meaning. When this value is received it means that normal world is +supposed to make a yielding call ``OPTEE_MSG_CMD_DO_BOTTOM_HALF``. This +call is done from the thread assisting the interrupt handler. This is a +building block for OP-TEE OS in secure world to implement the top half and +bottom half style of device drivers.
What I find missing here is a description of the trigger for this interrupt, and how it influences the way the kernel drivers interacts with the secure side:
- if it is edge triggered, this is 'fire and forget'. The interrupt will be consumed by the kernel handler, and whether it eventually calls into the secure side has no impact on the interrupt flow.
- if it is level triggered, then the interrupt may be asserted until the kernel calls into the secure side, which may then drop the line level if no other requests are pending.
These are evidently two very different flows, and you need to pick a side. Note that not all interrupt controllers support both signalling modes, so you are likely to leave something behind. Or you can try and support both flows, but that may make the driver slightly more complex.
Either way, this needs specifying, here and in the DT binding.
Thanks,
M.
On Fri, Jul 23, 2021 at 12:16 PM Marc Zyngier maz@kernel.org wrote:
On Fri, 23 Jul 2021 10:44:17 +0100, Jens Wiklander jens.wiklander@linaro.org wrote:
Adds a section on notifications used by OP-TEE, synchronous and asynchronous.
Signed-off-by: Jens Wiklander jens.wiklander@linaro.org
Documentation/staging/tee.rst | 27 +++++++++++++++++++++++++++ 1 file changed, 27 insertions(+)
diff --git a/Documentation/staging/tee.rst b/Documentation/staging/tee.rst index 4d4b5f889603..37bdd097336f 100644 --- a/Documentation/staging/tee.rst +++ b/Documentation/staging/tee.rst @@ -184,6 +184,33 @@ order to support device enumeration. In other words, OP-TEE driver invokes this application to retrieve a list of Trusted Applications which can be registered as devices on the TEE bus.
+OP-TEE notifications +--------------------
+There are two kinds of notifications that secure world can use to make +normal world aware of some event.
+1. Synchronous notifications delivered with ``OPTEE_RPC_CMD_NOTIFICATION``
- using the ``OPTEE_RPC_NOTIFICATION_SEND`` parameter.
+2. Asynchronous notifications delivered with a combination of a non-secure
- interrupt and a fast call from the non-secure interrupt handler.
+Synchronous notifications are limited by depending on RPC for delivery, +this is only usable when secure world is entered with a yielding call via +``OPTEE_SMC_CALL_WITH_ARG``. This excludes such notifications from secure +world interrupt handlers.
+An asynchronous notification is delivered via a non-secure interrupt to an +interrupt handler registered in the OP-TEE driver. The actual notification +value are retrieved with the fast call ``OPTEE_SMC_GET_ASYNC_NOTIF_VALUE``.
+One notification value ``OPTEE_SMC_ASYNC_NOTIF_VALUE_DO_BOTTOM_HALF`` has a +special meaning. When this value is received it means that normal world is +supposed to make a yielding call ``OPTEE_MSG_CMD_DO_BOTTOM_HALF``. This +call is done from the thread assisting the interrupt handler. This is a +building block for OP-TEE OS in secure world to implement the top half and +bottom half style of device drivers.
What I find missing here is a description of the trigger for this interrupt, and how it influences the way the kernel drivers interacts with the secure side:
if it is edge triggered, this is 'fire and forget'. The interrupt will be consumed by the kernel handler, and whether it eventually calls into the secure side has no impact on the interrupt flow.
if it is level triggered, then the interrupt may be asserted until the kernel calls into the secure side, which may then drop the line level if no other requests are pending.
These are evidently two very different flows, and you need to pick a side. Note that not all interrupt controllers support both signalling modes, so you are likely to leave something behind. Or you can try and support both flows, but that may make the driver slightly more complex.
Either way, this needs specifying, here and in the DT binding.
In the example I'm using a level triggered interrupt which is triggered by writing to GICD_ISPENDR by secure world. Reading of GICC_IAR should clear the interrupt, the GICv2 reference manual is quite clear on that. So, if I understand it correctly, it will for this purpose work in the same way as an edge triggered interrupt. If this wouldn't be the case in some configuration and the interrupt must be cleared by some other action that would be a job for the receiver of OPTEE_SMC_GET_ASYNC_NOTIF_VALUE, that is, a secure world problem. The normal world flow should be the same.
Now that we describe the interrupt configuration in device tree it must use something that mirrors the secure world expectations. I don't see a point in restricting what's allowed as long it doesn't need code changes in the kernel too. Does this make any sense?
If I just expand a bit above explaining that the interrupt handler must call OPTEE_SMC_GET_ASYNC_NOTIF_VALUE as part of clearing the interrupt even if it might be cleared anyway in some configurations. Would that make it more clear, good enough even :-) ?
Thanks, Jens
On Tue, 27 Jul 2021 08:46:39 +0100, Jens Wiklander jens.wiklander@linaro.org wrote:
On Fri, Jul 23, 2021 at 12:16 PM Marc Zyngier maz@kernel.org wrote:
On Fri, 23 Jul 2021 10:44:17 +0100, Jens Wiklander jens.wiklander@linaro.org wrote:
Adds a section on notifications used by OP-TEE, synchronous and asynchronous.
Signed-off-by: Jens Wiklander jens.wiklander@linaro.org
Documentation/staging/tee.rst | 27 +++++++++++++++++++++++++++ 1 file changed, 27 insertions(+)
diff --git a/Documentation/staging/tee.rst b/Documentation/staging/tee.rst index 4d4b5f889603..37bdd097336f 100644 --- a/Documentation/staging/tee.rst +++ b/Documentation/staging/tee.rst @@ -184,6 +184,33 @@ order to support device enumeration. In other words, OP-TEE driver invokes this application to retrieve a list of Trusted Applications which can be registered as devices on the TEE bus.
+OP-TEE notifications +--------------------
+There are two kinds of notifications that secure world can use to make +normal world aware of some event.
+1. Synchronous notifications delivered with ``OPTEE_RPC_CMD_NOTIFICATION``
- using the ``OPTEE_RPC_NOTIFICATION_SEND`` parameter.
+2. Asynchronous notifications delivered with a combination of a non-secure
- interrupt and a fast call from the non-secure interrupt handler.
+Synchronous notifications are limited by depending on RPC for delivery, +this is only usable when secure world is entered with a yielding call via +``OPTEE_SMC_CALL_WITH_ARG``. This excludes such notifications from secure +world interrupt handlers.
+An asynchronous notification is delivered via a non-secure interrupt to an +interrupt handler registered in the OP-TEE driver. The actual notification +value are retrieved with the fast call ``OPTEE_SMC_GET_ASYNC_NOTIF_VALUE``.
+One notification value ``OPTEE_SMC_ASYNC_NOTIF_VALUE_DO_BOTTOM_HALF`` has a +special meaning. When this value is received it means that normal world is +supposed to make a yielding call ``OPTEE_MSG_CMD_DO_BOTTOM_HALF``. This +call is done from the thread assisting the interrupt handler. This is a +building block for OP-TEE OS in secure world to implement the top half and +bottom half style of device drivers.
What I find missing here is a description of the trigger for this interrupt, and how it influences the way the kernel drivers interacts with the secure side:
if it is edge triggered, this is 'fire and forget'. The interrupt will be consumed by the kernel handler, and whether it eventually calls into the secure side has no impact on the interrupt flow.
if it is level triggered, then the interrupt may be asserted until the kernel calls into the secure side, which may then drop the line level if no other requests are pending.
These are evidently two very different flows, and you need to pick a side. Note that not all interrupt controllers support both signalling modes, so you are likely to leave something behind. Or you can try and support both flows, but that may make the driver slightly more complex.
Either way, this needs specifying, here and in the DT binding.
In the example I'm using a level triggered interrupt which is triggered by writing to GICD_ISPENDR by secure world. Reading of GICC_IAR should clear the interrupt,the GICv2 reference manual is quite clear on that.
No, it merely activates it. You can't transition an interrupt from pending to inactive (unless you clear it using GICD_ICPENDR). If you have spotted something else in the GICv2 architecture manual, please say so and I'll get it fixed 15 years after the facts. The fact that GICC_IAR consumes a pending bit introduced by a write to ISPENDR is an implementation detail, see below.
It is also a flawed approach, as this behaviour is IMPDEF on GICv3 (see 4.5 "Shared Peripheral Interrupts" in the GICv3 arch spec). Given that GICv2 is pretty much a dead horse (TFFT!), I can't see this approach being successful in the long run.
So, if I understand it correctly, it will for this purpose work in the same way as an edge triggered interrupt. If this wouldn't be the case in some configuration and the interrupt must be cleared by some other action that would be a job for the receiver of OPTEE_SMC_GET_ASYNC_NOTIF_VALUE, that is, a secure world problem. The normal world flow should be the same.
You are assuming that the secure side will use GICD_ISPENDR, and that's a leap of faith. An implementation should use, say, a GPIO to drive the interrupt line and give it proper level semantics.
Now that we describe the interrupt configuration in device tree it must use something that mirrors the secure world expectations. I don't see a point in restricting what's allowed as long it doesn't need code changes in the kernel too. Does this make any sense?
And that's the crucial point: what *are* the expectations of the secure side? You seem to assume edge semantics, but that's unclear at best.
If I just expand a bit above explaining that the interrupt handler must call OPTEE_SMC_GET_ASYNC_NOTIF_VALUE as part of clearing the interrupt even if it might be cleared anyway in some configurations. Would that make it more clear, good enough even :-) ?
This is an interrupt, please document it in terms of interrupt signalling.
- If it is level, the handler has to call into secure to observe the level dropping. If the driver can observe the level being low before calling into secure, it is perfectly allowed to consider the interrupt being spurious and not perform the call. If you don't have a device actively driving the line, this doesn't work.
- It is edge, the handler can do anything it likes, including ignoring the request after consuming the interrupt, or call into secure from a kernel thread with interrupts enabled.
At the end of the day, only you can decide which of these two flows are appropriate. If you don't want to mandate actual HW driving the line, edge triggered is your only option.
Thanks,
M.
On Tue, Jul 27, 2021 at 10:32 AM Marc Zyngier maz@kernel.org wrote:
On Tue, 27 Jul 2021 08:46:39 +0100, Jens Wiklander jens.wiklander@linaro.org wrote:
On Fri, Jul 23, 2021 at 12:16 PM Marc Zyngier maz@kernel.org wrote:
On Fri, 23 Jul 2021 10:44:17 +0100, Jens Wiklander jens.wiklander@linaro.org wrote:
Adds a section on notifications used by OP-TEE, synchronous and asynchronous.
Signed-off-by: Jens Wiklander jens.wiklander@linaro.org
Documentation/staging/tee.rst | 27 +++++++++++++++++++++++++++ 1 file changed, 27 insertions(+)
diff --git a/Documentation/staging/tee.rst b/Documentation/staging/tee.rst index 4d4b5f889603..37bdd097336f 100644 --- a/Documentation/staging/tee.rst +++ b/Documentation/staging/tee.rst @@ -184,6 +184,33 @@ order to support device enumeration. In other words, OP-TEE driver invokes this application to retrieve a list of Trusted Applications which can be registered as devices on the TEE bus.
+OP-TEE notifications +--------------------
+There are two kinds of notifications that secure world can use to make +normal world aware of some event.
+1. Synchronous notifications delivered with ``OPTEE_RPC_CMD_NOTIFICATION``
- using the ``OPTEE_RPC_NOTIFICATION_SEND`` parameter.
+2. Asynchronous notifications delivered with a combination of a non-secure
- interrupt and a fast call from the non-secure interrupt handler.
+Synchronous notifications are limited by depending on RPC for delivery, +this is only usable when secure world is entered with a yielding call via +``OPTEE_SMC_CALL_WITH_ARG``. This excludes such notifications from secure +world interrupt handlers.
+An asynchronous notification is delivered via a non-secure interrupt to an +interrupt handler registered in the OP-TEE driver. The actual notification +value are retrieved with the fast call ``OPTEE_SMC_GET_ASYNC_NOTIF_VALUE``.
+One notification value ``OPTEE_SMC_ASYNC_NOTIF_VALUE_DO_BOTTOM_HALF`` has a +special meaning. When this value is received it means that normal world is +supposed to make a yielding call ``OPTEE_MSG_CMD_DO_BOTTOM_HALF``. This +call is done from the thread assisting the interrupt handler. This is a +building block for OP-TEE OS in secure world to implement the top half and +bottom half style of device drivers.
What I find missing here is a description of the trigger for this interrupt, and how it influences the way the kernel drivers interacts with the secure side:
if it is edge triggered, this is 'fire and forget'. The interrupt will be consumed by the kernel handler, and whether it eventually calls into the secure side has no impact on the interrupt flow.
if it is level triggered, then the interrupt may be asserted until the kernel calls into the secure side, which may then drop the line level if no other requests are pending.
These are evidently two very different flows, and you need to pick a side. Note that not all interrupt controllers support both signalling modes, so you are likely to leave something behind. Or you can try and support both flows, but that may make the driver slightly more complex.
Either way, this needs specifying, here and in the DT binding.
In the example I'm using a level triggered interrupt which is triggered by writing to GICD_ISPENDR by secure world. Reading of GICC_IAR should clear the interrupt,the GICv2 reference manual is quite clear on that.
No, it merely activates it. You can't transition an interrupt from pending to inactive (unless you clear it using GICD_ICPENDR). If you have spotted something else in the GICv2 architecture manual, please say so and I'll get it fixed 15 years after the facts. The fact that GICC_IAR consumes a pending bit introduced by a write to ISPENDR is an implementation detail, see below.
I was looking at figure 4-10 "Logic of the pending status of a level-sensitive interrupt".
It is also a flawed approach, as this behaviour is IMPDEF on GICv3 (see 4.5 "Shared Peripheral Interrupts" in the GICv3 arch spec). Given that GICv2 is pretty much a dead horse (TFFT!), I can't see this approach being successful in the long run.
OK, thanks.
So, if I understand it correctly, it will for this purpose work in the same way as an edge triggered interrupt. If this wouldn't be the case in some configuration and the interrupt must be cleared by some other action that would be a job for the receiver of OPTEE_SMC_GET_ASYNC_NOTIF_VALUE, that is, a secure world problem. The normal world flow should be the same.
You are assuming that the secure side will use GICD_ISPENDR, and that's a leap of faith.
Not in this case with upstream OP-TEE. If we need to signal in a different way we can do that instead. What happens downstream we have no control over, but that's perhaps not so different from the kernel.
An implementation should use, say, a GPIO to drive the interrupt line and give it proper level semantics.
I'm not so keen on that since we often don't touch GPIO at all in OP-TEE and this would then mean more platform specific code. We may even need to synchronize some hardware access with the normal world and then we'd be back at square one again.
Now that we describe the interrupt configuration in device tree it must use something that mirrors the secure world expectations. I don't see a point in restricting what's allowed as long it doesn't need code changes in the kernel too. Does this make any sense?
And that's the crucial point: what *are* the expectations of the secure side?
That should be up to the OP-TEE port of that particular platform to decide and advise which device tree configuration to use.
You seem to assume edge semantics, but that's unclear at best.
Fair enough, edge semantics solves the problem here.
If I just expand a bit above explaining that the interrupt handler must call OPTEE_SMC_GET_ASYNC_NOTIF_VALUE as part of clearing the interrupt even if it might be cleared anyway in some configurations. Would that make it more clear, good enough even :-) ?
This is an interrupt, please document it in terms of interrupt signalling.
In this file, Documentation/staging/tee.rst ? A pointer to a good example of what you're expecting would be much appreciated. There's also the file drivers/tee/optee/optee_smc.h, this is where the ABI to OP-TEE is defined.
- If it is level, the handler has to call into secure to observe the level dropping. If the driver can observe the level being low before calling into secure, it is perfectly allowed to consider the interrupt being spurious and not perform the call. If you don't have a device actively driving the line, this doesn't work.
Spurious calls to OPTEE_SMC_GET_ASYNC_NOTIF_VALUE are harmless, but it will of course save a few cycles if they can be avoided.
- It is edge, the handler can do anything it likes, including ignoring the request after consuming the interrupt, or call into secure from a kernel thread with interrupts enabled.
Yes, that's a bit more relaxed. Now I'm doing that part in the upper half handler. I don't suppose it matters much, not much time is spent there. The advantage is that it will work with a level triggered interrupt too, if it would ever come to that. Provided that OPTEE_SMC_GET_ASYNC_NOTIF_VALUE does what's needed to clear the interrupt in secure world, but that should be mandated in such configurations.
At the end of the day, only you can decide which of these two flows are appropriate. If you don't want to mandate actual HW driving the line, edge triggered is your only option.
I agree that in the GIC case we should use edge triggered interrupts. But It's not clear to me why this must be nailed down in any other case instead of being left open to configuration.
The configuration that is best for the platform should be used, but this is still a platform specific detail [1]. If device tree can give us the flexibility to use level triggered interrupts with some other interrupt controller and the kernel code already is capable of working with that, what's the harm in leaving that open?
Thanks for your patience, I'm still trying to understand the parameters.
Cheers, Jens [1] The OP-TEE firmware is always built per platform with platform specific configuration. With FF-A that may change, but then we'll also base these notifications on FF-A instead of playing directly with interrupts.
On Tue, 27 Jul 2021 15:57:07 +0100, Jens Wiklander jens.wiklander@linaro.org wrote:
On Tue, Jul 27, 2021 at 10:32 AM Marc Zyngier maz@kernel.org wrote:
On Tue, 27 Jul 2021 08:46:39 +0100, Jens Wiklander jens.wiklander@linaro.org wrote:
On Fri, Jul 23, 2021 at 12:16 PM Marc Zyngier maz@kernel.org wrote:
On Fri, 23 Jul 2021 10:44:17 +0100, Jens Wiklander jens.wiklander@linaro.org wrote:
Adds a section on notifications used by OP-TEE, synchronous and asynchronous.
Signed-off-by: Jens Wiklander jens.wiklander@linaro.org
Documentation/staging/tee.rst | 27 +++++++++++++++++++++++++++ 1 file changed, 27 insertions(+)
diff --git a/Documentation/staging/tee.rst b/Documentation/staging/tee.rst index 4d4b5f889603..37bdd097336f 100644 --- a/Documentation/staging/tee.rst +++ b/Documentation/staging/tee.rst @@ -184,6 +184,33 @@ order to support device enumeration. In other words, OP-TEE driver invokes this application to retrieve a list of Trusted Applications which can be registered as devices on the TEE bus.
+OP-TEE notifications +--------------------
+There are two kinds of notifications that secure world can use to make +normal world aware of some event.
+1. Synchronous notifications delivered with ``OPTEE_RPC_CMD_NOTIFICATION``
- using the ``OPTEE_RPC_NOTIFICATION_SEND`` parameter.
+2. Asynchronous notifications delivered with a combination of a non-secure
- interrupt and a fast call from the non-secure interrupt handler.
+Synchronous notifications are limited by depending on RPC for delivery, +this is only usable when secure world is entered with a yielding call via +``OPTEE_SMC_CALL_WITH_ARG``. This excludes such notifications from secure +world interrupt handlers.
+An asynchronous notification is delivered via a non-secure interrupt to an +interrupt handler registered in the OP-TEE driver. The actual notification +value are retrieved with the fast call ``OPTEE_SMC_GET_ASYNC_NOTIF_VALUE``.
+One notification value ``OPTEE_SMC_ASYNC_NOTIF_VALUE_DO_BOTTOM_HALF`` has a +special meaning. When this value is received it means that normal world is +supposed to make a yielding call ``OPTEE_MSG_CMD_DO_BOTTOM_HALF``. This +call is done from the thread assisting the interrupt handler. This is a +building block for OP-TEE OS in secure world to implement the top half and +bottom half style of device drivers.
What I find missing here is a description of the trigger for this interrupt, and how it influences the way the kernel drivers interacts with the secure side:
if it is edge triggered, this is 'fire and forget'. The interrupt will be consumed by the kernel handler, and whether it eventually calls into the secure side has no impact on the interrupt flow.
if it is level triggered, then the interrupt may be asserted until the kernel calls into the secure side, which may then drop the line level if no other requests are pending.
These are evidently two very different flows, and you need to pick a side. Note that not all interrupt controllers support both signalling modes, so you are likely to leave something behind. Or you can try and support both flows, but that may make the driver slightly more complex.
Either way, this needs specifying, here and in the DT binding.
In the example I'm using a level triggered interrupt which is triggered by writing to GICD_ISPENDR by secure world. Reading of GICC_IAR should clear the interrupt,the GICv2 reference manual is quite clear on that.
No, it merely activates it. You can't transition an interrupt from pending to inactive (unless you clear it using GICD_ICPENDR). If you have spotted something else in the GICv2 architecture manual, please say so and I'll get it fixed 15 years after the facts. The fact that GICC_IAR consumes a pending bit introduced by a write to ISPENDR is an implementation detail, see below.
I was looking at figure 4-10 "Logic of the pending status of a level-sensitive interrupt".
It is also a flawed approach, as this behaviour is IMPDEF on GICv3 (see 4.5 "Shared Peripheral Interrupts" in the GICv3 arch spec). Given that GICv2 is pretty much a dead horse (TFFT!), I can't see this approach being successful in the long run.
OK, thanks.
So, if I understand it correctly, it will for this purpose work in the same way as an edge triggered interrupt. If this wouldn't be the case in some configuration and the interrupt must be cleared by some other action that would be a job for the receiver of OPTEE_SMC_GET_ASYNC_NOTIF_VALUE, that is, a secure world problem. The normal world flow should be the same.
You are assuming that the secure side will use GICD_ISPENDR, and that's a leap of faith.
Not in this case with upstream OP-TEE. If we need to signal in a different way we can do that instead. What happens downstream we have no control over, but that's perhaps not so different from the kernel.
An implementation should use, say, a GPIO to drive the interrupt line and give it proper level semantics.
I'm not so keen on that since we often don't touch GPIO at all in OP-TEE and this would then mean more platform specific code. We may even need to synchronize some hardware access with the normal world and then we'd be back at square one again.
I'm trying to make a general statement here. OP-TEE may not do any of this. Yet. Other TEE implementations may do it because there is no free SPI in the system, for example. The problem is that interrupt signalling is a *device* thing, even if the device is actually some SW running on the secure side, and I would like the TEE to behave as such.
Now that we describe the interrupt configuration in device tree it must use something that mirrors the secure world expectations. I don't see a point in restricting what's allowed as long it doesn't need code changes in the kernel too. Does this make any sense?
And that's the crucial point: what *are* the expectations of the secure side?
That should be up to the OP-TEE port of that particular platform to decide and advise which device tree configuration to use.
You seem to assume edge semantics, but that's unclear at best.
Fair enough, edge semantics solves the problem here.
If I just expand a bit above explaining that the interrupt handler must call OPTEE_SMC_GET_ASYNC_NOTIF_VALUE as part of clearing the interrupt even if it might be cleared anyway in some configurations. Would that make it more clear, good enough even :-) ?
This is an interrupt, please document it in terms of interrupt signalling.
In this file, Documentation/staging/tee.rst ? A pointer to a good example of what you're expecting would be much appreciated. There's also the file drivers/tee/optee/optee_smc.h, this is where the ABI to OP-TEE is defined.
This file is fine, but repeating in the DT binding that edge signalling is what is expected would be good.
Thanks,
M.
Adds an optional interrupt property to the optee binding.
Signed-off-by: Jens Wiklander jens.wiklander@linaro.org --- .../devicetree/bindings/arm/firmware/linaro,optee-tz.yaml | 4 ++++ 1 file changed, 4 insertions(+)
diff --git a/Documentation/devicetree/bindings/arm/firmware/linaro,optee-tz.yaml b/Documentation/devicetree/bindings/arm/firmware/linaro,optee-tz.yaml index c24047c1fdd5..73811a5d1714 100644 --- a/Documentation/devicetree/bindings/arm/firmware/linaro,optee-tz.yaml +++ b/Documentation/devicetree/bindings/arm/firmware/linaro,optee-tz.yaml @@ -24,6 +24,9 @@ properties: compatible: const: linaro,optee-tz
+ interrupts: + maxItems: 1 + method: enum: [smc, hvc] description: | @@ -46,6 +49,7 @@ examples: optee { compatible = "linaro,optee-tz"; method = "smc"; + interrupts = <0 187 4>; }; };
Prior to this patch was teedev_close_context() calling tee_device_put() before teedev_ctx_put() leading to teedev_ctx_release() accessing ctx->teedev just after the reference counter was decreased on the teedev. Fix this by calling teedev_ctx_put() before tee_device_put().
Fixes: 217e0250cccb ("tee: use reference counting for tee_context") Signed-off-by: Jens Wiklander jens.wiklander@linaro.org --- drivers/tee/tee_core.c | 4 +++- 1 file changed, 3 insertions(+), 1 deletion(-)
diff --git a/drivers/tee/tee_core.c b/drivers/tee/tee_core.c index 2b37bc408fc3..85102d12d716 100644 --- a/drivers/tee/tee_core.c +++ b/drivers/tee/tee_core.c @@ -98,8 +98,10 @@ void teedev_ctx_put(struct tee_context *ctx)
static void teedev_close_context(struct tee_context *ctx) { - tee_device_put(ctx->teedev); + struct tee_device *teedev = ctx->teedev; + teedev_ctx_put(ctx); + tee_device_put(teedev); }
static int tee_open(struct inode *inode, struct file *filp)
Unsuscribe
El vie., 23 de julio de 2021 6:46 a. m., Jens Wiklander < jens.wiklander@linaro.org> escribió:
Prior to this patch was teedev_close_context() calling tee_device_put() before teedev_ctx_put() leading to teedev_ctx_release() accessing ctx->teedev just after the reference counter was decreased on the teedev. Fix this by calling teedev_ctx_put() before tee_device_put().
Fixes: 217e0250cccb ("tee: use reference counting for tee_context") Signed-off-by: Jens Wiklander jens.wiklander@linaro.org
drivers/tee/tee_core.c | 4 +++- 1 file changed, 3 insertions(+), 1 deletion(-)
diff --git a/drivers/tee/tee_core.c b/drivers/tee/tee_core.c index 2b37bc408fc3..85102d12d716 100644 --- a/drivers/tee/tee_core.c +++ b/drivers/tee/tee_core.c @@ -98,8 +98,10 @@ void teedev_ctx_put(struct tee_context *ctx)
static void teedev_close_context(struct tee_context *ctx) {
tee_device_put(ctx->teedev);
struct tee_device *teedev = ctx->teedev;teedev_ctx_put(ctx);tee_device_put(teedev);}
static int tee_open(struct inode *inode, struct file *filp)
2.31.1
Adds tee_dev_open_helper() and tee_dev_ctx_put() to make it easier to create a driver internal struct tee_context without the usual tee_device_get() on the struct tee_device as that adds a circular reference counter dependency and would prevent the struct tee_device from ever being released again.
Signed-off-by: Jens Wiklander jens.wiklander@linaro.org --- drivers/tee/tee_core.c | 33 ++++++++++++++++++++++++--------- include/linux/tee_drv.h | 27 +++++++++++++++++++++++++++ 2 files changed, 51 insertions(+), 9 deletions(-)
diff --git a/drivers/tee/tee_core.c b/drivers/tee/tee_core.c index 85102d12d716..3beb682684a8 100644 --- a/drivers/tee/tee_core.c +++ b/drivers/tee/tee_core.c @@ -43,14 +43,11 @@ static DEFINE_SPINLOCK(driver_lock); static struct class *tee_class; static dev_t tee_devt;
-static struct tee_context *teedev_open(struct tee_device *teedev) +struct tee_context *tee_dev_open_helper(struct tee_device *teedev) { int rc; struct tee_context *ctx;
- if (!tee_device_get(teedev)) - return ERR_PTR(-EINVAL); - ctx = kzalloc(sizeof(*ctx), GFP_KERNEL); if (!ctx) { rc = -ENOMEM; @@ -66,10 +63,30 @@ static struct tee_context *teedev_open(struct tee_device *teedev) return ctx; err: kfree(ctx); - tee_device_put(teedev); return ERR_PTR(rc);
} +EXPORT_SYMBOL_GPL(tee_dev_open_helper); + +void tee_dev_ctx_put(struct tee_context *ctx) +{ + teedev_ctx_put(ctx); +} +EXPORT_SYMBOL_GPL(tee_dev_ctx_put); + +static struct tee_context *teedev_open(struct tee_device *teedev) +{ + struct tee_context *ctx; + + if (!tee_device_get(teedev)) + return ERR_PTR(-EINVAL); + + ctx = tee_dev_open_helper(teedev); + if (IS_ERR(ctx)) + tee_device_put(teedev); + + return ctx; +}
void teedev_ctx_get(struct tee_context *ctx) { @@ -90,10 +107,8 @@ static void teedev_ctx_release(struct kref *ref)
void teedev_ctx_put(struct tee_context *ctx) { - if (ctx->releasing) - return; - - kref_put(&ctx->refcount, teedev_ctx_release); + if (ctx && !ctx->releasing) + kref_put(&ctx->refcount, teedev_ctx_release); }
static void teedev_close_context(struct tee_context *ctx) diff --git a/include/linux/tee_drv.h b/include/linux/tee_drv.h index 54269e47ac9a..f592ba4e9561 100644 --- a/include/linux/tee_drv.h +++ b/include/linux/tee_drv.h @@ -456,6 +456,33 @@ static inline int tee_shm_get_id(struct tee_shm *shm) */ struct tee_shm *tee_shm_get_from_id(struct tee_context *ctx, int id);
+/** + * tee_dev_open_helper() - helper function to make a struct tee_context + * @teedev: Device to open + * + * Creates the struct tee_context without increasing the reference counter + * on @teedev. This is needed for instance when a driver need an internal + * struct tee_context to operate on. By skipping the reference counter + * the circular dependency is broken. + * + * Note that this struct tee_context need special care when freeing in + * order to avoid the normal put on the struct tee_device. + * tee_dev_ctx_put() is the best choice for this. + * + * @returns a pointer 'struct tee_context' on success or an ERR_PTR on failure + */ +struct tee_context *tee_dev_open_helper(struct tee_device *teedev); + +/** + * tee_dev_ctx_put() - helper function to release a struct tee_context + * @ctx: The struct tee_context to release + * + * Note that this function doesn't do a tee_device_put() on the internal + * struct tee_device so this function should normal only be used when + * releasing a struct tee_context obtained with tee_dev_open_helper(). + */ +void tee_dev_ctx_put(struct tee_context *ctx); + /** * tee_client_open_context() - Open a TEE context * @start: if not NULL, continue search after this context
Renames struct optee_wait_queue to struct optee_notif and all related functions to optee_notif_*().
The implementation is changed to allow sending a notification from an atomic state, that is from the top half of an interrupt handler.
Waiting for keys is currently only used when secure world is waiting for a mutex or condition variable. The old implementation could handle any 32-bit key while this new implementation is restricted to only 8 bits or the maximum value 255. A upper value is needed since a bitmap is allocated to allow an interrupt handler to only set a bit in case the waiter hasn't had the time yet to allocate and register a completion.
The keys are currently only representing secure world threads which number usually are never even close to 255 so it should be safe for now. In future ABI updates the maximum value of the key will be communicated while the driver is initializing.
Signed-off-by: Jens Wiklander jens.wiklander@linaro.org --- drivers/tee/optee/Makefile | 1 + drivers/tee/optee/core.c | 13 +++- drivers/tee/optee/notif.c | 125 ++++++++++++++++++++++++++++++ drivers/tee/optee/optee_private.h | 19 +++-- drivers/tee/optee/optee_rpc_cmd.h | 31 ++++---- drivers/tee/optee/rpc.c | 73 ++--------------- 6 files changed, 171 insertions(+), 91 deletions(-) create mode 100644 drivers/tee/optee/notif.c
diff --git a/drivers/tee/optee/Makefile b/drivers/tee/optee/Makefile index 3aa33ea9e6a6..df55e4ad5370 100644 --- a/drivers/tee/optee/Makefile +++ b/drivers/tee/optee/Makefile @@ -2,6 +2,7 @@ obj-$(CONFIG_OPTEE) += optee.o optee-objs += core.o optee-objs += call.o +optee-objs += notif.o optee-objs += rpc.o optee-objs += supp.o optee-objs += shm_pool.o diff --git a/drivers/tee/optee/core.c b/drivers/tee/optee/core.c index ddb8f9ecf307..2272696ac986 100644 --- a/drivers/tee/optee/core.c +++ b/drivers/tee/optee/core.c @@ -583,6 +583,7 @@ static int optee_remove(struct platform_device *pdev) */ optee_disable_shm_cache(optee);
+ optee_notif_uninit(optee); /* * The two devices have to be unregistered before we can free the * other resources. @@ -593,7 +594,6 @@ static int optee_remove(struct platform_device *pdev) tee_shm_pool_free(optee->pool); if (optee->memremaped_shm) memunmap(optee->memremaped_shm); - optee_wait_queue_exit(&optee->wait_queue); optee_supp_uninit(&optee->supp); mutex_destroy(&optee->call_queue.mutex);
@@ -681,18 +681,23 @@ static int optee_probe(struct platform_device *pdev)
mutex_init(&optee->call_queue.mutex); INIT_LIST_HEAD(&optee->call_queue.waiters); - optee_wait_queue_init(&optee->wait_queue); optee_supp_init(&optee->supp); optee->memremaped_shm = memremaped_shm; optee->pool = pool;
+ platform_set_drvdata(pdev, optee); + + rc = optee_notif_init(optee, 255); + if (rc) { + optee_remove(pdev); + return rc; + } + optee_enable_shm_cache(optee);
if (optee->sec_caps & OPTEE_SMC_SEC_CAP_DYNAMIC_SHM) pr_info("dynamic shared memory is enabled\n");
- platform_set_drvdata(pdev, optee); - rc = optee_enumerate_devices(PTA_CMD_GET_DEVICES); if (rc) { optee_remove(pdev); diff --git a/drivers/tee/optee/notif.c b/drivers/tee/optee/notif.c new file mode 100644 index 000000000000..a28fa03dcd0e --- /dev/null +++ b/drivers/tee/optee/notif.c @@ -0,0 +1,125 @@ +// SPDX-License-Identifier: GPL-2.0-only +/* + * Copyright (c) 2015-2021, Linaro Limited + */ + +#define pr_fmt(fmt) KBUILD_MODNAME ": " fmt + +#include <linux/arm-smccc.h> +#include <linux/errno.h> +#include <linux/slab.h> +#include <linux/spinlock.h> +#include <linux/tee_drv.h> +#include "optee_private.h" + +struct notif_entry { + struct list_head link; + struct completion c; + u_int key; +}; + +static bool have_key(struct optee *optee, u_int key) +{ + struct notif_entry *entry; + + list_for_each_entry(entry, &optee->notif.db, link) + if (entry->key == key) + return true; + + return false; +} + +int optee_notif_wait(struct optee *optee, u_int key) +{ + unsigned long flags; + struct notif_entry *entry; + int rc = 0; + + if (key > optee->notif.max_key) + return -EINVAL; + + entry = kmalloc(sizeof(*entry), GFP_KERNEL); + if (!entry) + return -ENOMEM; + init_completion(&entry->c); + entry->key = key; + + spin_lock_irqsave(&optee->notif.lock, flags); + + /* + * If the bit is already set it means that the key has already + * been posted and we must not wait. + */ + if (test_bit(key, optee->notif.bitmap)) { + clear_bit(key, optee->notif.bitmap); + goto out; + } + + /* + * Check if someone is already waiting for this key. If there is + * it's a programming error. + */ + if (have_key(optee, key)) { + rc = -EBUSY; + goto out; + } + + list_add_tail(&entry->link, &optee->notif.db); + + /* + * Unlock temporarily and wait for completion. + */ + spin_unlock_irqrestore(&optee->notif.lock, flags); + wait_for_completion(&entry->c); + spin_lock_irqsave(&optee->notif.lock, flags); + + list_del(&entry->link); +out: + spin_unlock_irqrestore(&optee->notif.lock, flags); + + kfree(entry); + + return rc; +} + +int optee_notif_send(struct optee *optee, u_int key) +{ + unsigned long flags; + struct notif_entry *entry; + + if (key > optee->notif.max_key) + return -EINVAL; + + spin_lock_irqsave(&optee->notif.lock, flags); + + list_for_each_entry(entry, &optee->notif.db, link) + if (entry->key == key) { + complete(&entry->c); + goto out; + } + + /* Only set the bit in case there where nobody waiting */ + set_bit(key, optee->notif.bitmap); +out: + spin_unlock_irqrestore(&optee->notif.lock, flags); + + return 0; +} + +int optee_notif_init(struct optee *optee, u_int max_key) +{ + spin_lock_init(&optee->notif.lock); + INIT_LIST_HEAD(&optee->notif.db); + optee->notif.bitmap = bitmap_zalloc(max_key, GFP_KERNEL); + if (!optee->notif.bitmap) + return -ENOMEM; + + optee->notif.max_key = max_key; + + return 0; +} + +void optee_notif_uninit(struct optee *optee) +{ + kfree(optee->notif.bitmap); +} diff --git a/drivers/tee/optee/optee_private.h b/drivers/tee/optee/optee_private.h index e25b216a14ef..7dc058d008b2 100644 --- a/drivers/tee/optee/optee_private.h +++ b/drivers/tee/optee/optee_private.h @@ -35,10 +35,12 @@ struct optee_call_queue { struct list_head waiters; };
-struct optee_wait_queue { - /* Serializes access to this struct */ - struct mutex mu; +struct optee_notif { + u_int max_key; + /* Serializes access to the elements below in this struct */ + spinlock_t lock; struct list_head db; + u_long *bitmap; };
/** @@ -72,8 +74,7 @@ struct optee_supp { * @teedev: client device * @invoke_fn: function to issue smc or hvc * @call_queue: queue of threads waiting to call @invoke_fn - * @wait_queue: queue of threads from secure world waiting for a - * secure world sync object + * @notif: notification synchronization struct * @supp: supplicant synchronization struct for RPC to supplicant * @pool: shared memory pool * @memremaped_shm virtual address of memory in shared memory pool @@ -88,7 +89,7 @@ struct optee { struct tee_device *teedev; optee_invoke_fn *invoke_fn; struct optee_call_queue call_queue; - struct optee_wait_queue wait_queue; + struct optee_notif notif; struct optee_supp supp; struct tee_shm_pool *pool; void *memremaped_shm; @@ -131,8 +132,10 @@ void optee_handle_rpc(struct tee_context *ctx, struct optee_rpc_param *param, struct optee_call_ctx *call_ctx); void optee_rpc_finalize_call(struct optee_call_ctx *call_ctx);
-void optee_wait_queue_init(struct optee_wait_queue *wq); -void optee_wait_queue_exit(struct optee_wait_queue *wq); +int optee_notif_init(struct optee *optee, u_int max_key); +void optee_notif_uninit(struct optee *optee); +int optee_notif_wait(struct optee *optee, u_int key); +int optee_notif_send(struct optee *optee, u_int key);
u32 optee_supp_thrd_req(struct tee_context *ctx, u32 func, size_t num_params, struct tee_param *param); diff --git a/drivers/tee/optee/optee_rpc_cmd.h b/drivers/tee/optee/optee_rpc_cmd.h index b8275140cef8..f3f06e0994a7 100644 --- a/drivers/tee/optee/optee_rpc_cmd.h +++ b/drivers/tee/optee/optee_rpc_cmd.h @@ -28,24 +28,27 @@ #define OPTEE_RPC_CMD_GET_TIME 3
/* - * Wait queue primitive, helper for secure world to implement a wait queue. + * Notification from/to secure world. * - * If secure world needs to wait for a secure world mutex it issues a sleep - * request instead of spinning in secure world. Conversely is a wakeup - * request issued when a secure world mutex with a thread waiting thread is - * unlocked. + * If secure world needs to wait for something, for instance a mutex, it + * does a notification wait request instead of spinning in secure world. + * Conversely can a synchronous notification can be sent when a secure + * world mutex with a thread waiting thread is unlocked. * - * Waiting on a key - * [in] value[0].a OPTEE_RPC_WAIT_QUEUE_SLEEP - * [in] value[0].b Wait key + * This interface can also be used to wait for a asynchronous notification + * which instead is sent via a non-secure interrupt. * - * Waking up a key - * [in] value[0].a OPTEE_RPC_WAIT_QUEUE_WAKEUP - * [in] value[0].b Wakeup key + * Waiting on notification + * [in] value[0].a OPTEE_RPC_NOTIFICATION_WAIT + * [in] value[0].b notification value + * + * Sending a synchronous notification + * [in] value[0].a OPTEE_RPC_NOTIFICATION_SEND + * [in] value[0].b notification value */ -#define OPTEE_RPC_CMD_WAIT_QUEUE 4 -#define OPTEE_RPC_WAIT_QUEUE_SLEEP 0 -#define OPTEE_RPC_WAIT_QUEUE_WAKEUP 1 +#define OPTEE_RPC_CMD_NOTIFICATION 4 +#define OPTEE_RPC_NOTIFICATION_WAIT 0 +#define OPTEE_RPC_NOTIFICATION_SEND 1
/* * Suspend execution diff --git a/drivers/tee/optee/rpc.c b/drivers/tee/optee/rpc.c index 1849180b0278..e5b931f50db2 100644 --- a/drivers/tee/optee/rpc.c +++ b/drivers/tee/optee/rpc.c @@ -1,6 +1,6 @@ // SPDX-License-Identifier: GPL-2.0-only /* - * Copyright (c) 2015-2016, Linaro Limited + * Copyright (c) 2015-2021, Linaro Limited */
#define pr_fmt(fmt) KBUILD_MODNAME ": " fmt @@ -14,23 +14,6 @@ #include "optee_smc.h" #include "optee_rpc_cmd.h"
-struct wq_entry { - struct list_head link; - struct completion c; - u32 key; -}; - -void optee_wait_queue_init(struct optee_wait_queue *priv) -{ - mutex_init(&priv->mu); - INIT_LIST_HEAD(&priv->db); -} - -void optee_wait_queue_exit(struct optee_wait_queue *priv) -{ - mutex_destroy(&priv->mu); -} - static void handle_rpc_func_cmd_get_time(struct optee_msg_arg *arg) { struct timespec64 ts; @@ -143,48 +126,6 @@ static void handle_rpc_func_cmd_i2c_transfer(struct tee_context *ctx, } #endif
-static struct wq_entry *wq_entry_get(struct optee_wait_queue *wq, u32 key) -{ - struct wq_entry *w; - - mutex_lock(&wq->mu); - - list_for_each_entry(w, &wq->db, link) - if (w->key == key) - goto out; - - w = kmalloc(sizeof(*w), GFP_KERNEL); - if (w) { - init_completion(&w->c); - w->key = key; - list_add_tail(&w->link, &wq->db); - } -out: - mutex_unlock(&wq->mu); - return w; -} - -static void wq_sleep(struct optee_wait_queue *wq, u32 key) -{ - struct wq_entry *w = wq_entry_get(wq, key); - - if (w) { - wait_for_completion(&w->c); - mutex_lock(&wq->mu); - list_del(&w->link); - mutex_unlock(&wq->mu); - kfree(w); - } -} - -static void wq_wakeup(struct optee_wait_queue *wq, u32 key) -{ - struct wq_entry *w = wq_entry_get(wq, key); - - if (w) - complete(&w->c); -} - static void handle_rpc_func_cmd_wq(struct optee *optee, struct optee_msg_arg *arg) { @@ -196,11 +137,13 @@ static void handle_rpc_func_cmd_wq(struct optee *optee, goto bad;
switch (arg->params[0].u.value.a) { - case OPTEE_RPC_WAIT_QUEUE_SLEEP: - wq_sleep(&optee->wait_queue, arg->params[0].u.value.b); + case OPTEE_RPC_NOTIFICATION_WAIT: + if (optee_notif_wait(optee, arg->params[0].u.value.b)) + goto bad; break; - case OPTEE_RPC_WAIT_QUEUE_WAKEUP: - wq_wakeup(&optee->wait_queue, arg->params[0].u.value.b); + case OPTEE_RPC_NOTIFICATION_SEND: + if (optee_notif_send(optee, arg->params[0].u.value.b)) + goto bad; break; default: goto bad; @@ -463,7 +406,7 @@ static void handle_rpc_func_cmd(struct tee_context *ctx, struct optee *optee, case OPTEE_RPC_CMD_GET_TIME: handle_rpc_func_cmd_get_time(arg); break; - case OPTEE_RPC_CMD_WAIT_QUEUE: + case OPTEE_RPC_CMD_NOTIFICATION: handle_rpc_func_cmd_wq(optee, arg); break; case OPTEE_RPC_CMD_SUSPEND:
Adds support for asynchronous notifications from secure world to normal world. This allows a design with a top half and bottom half type of driver where the top half runs in secure interrupt context and a notifications tells normal world to schedule a yielding call to do the bottom half processing.
The protocol is defined in optee_msg.h optee_rpc_cmd.h and optee_smc.h.
A notification consists of a 32-bit value which normal world can retrieve using a fastcall into secure world. The value OPTEE_SMC_ASYNC_NOTIF_VALUE_DO_BOTTOM_HALF (0) has a special meaning. When this value is sent it means that normal world is supposed to make a yielding call OPTEE_MSG_CMD_DO_BOTTOM_HALF.
Notification capability is negotiated while the driver is initialized. If both sides supports these notifications then they are enabled.
An interrupt is used to notify the driver that there are asynchronous notifications pending. The maximum needed notification value is communicated at this stage. This allows scaling up when needed.
Acked-by: Ard Biesheuvel ardb@kernel.org Signed-off-by: Jens Wiklander jens.wiklander@linaro.org --- drivers/tee/optee/call.c | 27 ++++++++ drivers/tee/optee/core.c | 82 +++++++++++++++------- drivers/tee/optee/notif.c | 109 ++++++++++++++++++++++++++++-- drivers/tee/optee/optee_msg.h | 9 +++ drivers/tee/optee/optee_private.h | 6 +- drivers/tee/optee/optee_smc.h | 75 +++++++++++++++++++- 6 files changed, 276 insertions(+), 32 deletions(-)
diff --git a/drivers/tee/optee/call.c b/drivers/tee/optee/call.c index 6e6eb836e9b6..3afd43b598f9 100644 --- a/drivers/tee/optee/call.c +++ b/drivers/tee/optee/call.c @@ -392,6 +392,33 @@ int optee_cancel_req(struct tee_context *ctx, u32 cancel_id, u32 session) return 0; }
+static int simple_call_with_arg(struct tee_context *ctx, u32 cmd) +{ + struct optee_msg_arg *msg_arg; + phys_addr_t msg_parg; + struct tee_shm *shm; + + shm = get_msg_arg(ctx, 0, &msg_arg, &msg_parg); + if (IS_ERR(shm)) + return PTR_ERR(shm); + + msg_arg->cmd = cmd; + optee_do_call_with_arg(ctx, msg_parg); + + tee_shm_free(shm); + return 0; +} + +int optee_do_bottom_half(struct tee_context *ctx) +{ + return simple_call_with_arg(ctx, OPTEE_MSG_CMD_DO_BOTTOM_HALF); +} + +int optee_stop_async_notif(struct tee_context *ctx) +{ + return simple_call_with_arg(ctx, OPTEE_MSG_CMD_STOP_ASYNC_NOTIF); +} + /** * optee_enable_shm_cache() - Enables caching of some shared memory allocation * in OP-TEE diff --git a/drivers/tee/optee/core.c b/drivers/tee/optee/core.c index 2272696ac986..e3c80505cc88 100644 --- a/drivers/tee/optee/core.c +++ b/drivers/tee/optee/core.c @@ -7,9 +7,12 @@
#include <linux/arm-smccc.h> #include <linux/errno.h> +#include <linux/interrupt.h> #include <linux/io.h> +#include <linux/irqdomain.h> #include <linux/module.h> #include <linux/of.h> +#include <linux/of_irq.h> #include <linux/of_platform.h> #include <linux/platform_device.h> #include <linux/slab.h> @@ -353,6 +356,17 @@ static const struct tee_desc optee_supp_desc = { .flags = TEE_DESC_PRIVILEGED, };
+static int enable_async_notif(optee_invoke_fn *invoke_fn) +{ + struct arm_smccc_res res; + + invoke_fn(OPTEE_SMC_ENABLE_ASYNC_NOTIF, 0, 0, 0, 0, 0, 0, 0, &res); + + if (res.a0) + return -EINVAL; + return 0; +} + static bool optee_msg_api_uid_is_optee_api(optee_invoke_fn *invoke_fn) { struct arm_smccc_res res; @@ -402,7 +416,7 @@ static bool optee_msg_api_revision_is_compatible(optee_invoke_fn *invoke_fn) }
static bool optee_msg_exchange_capabilities(optee_invoke_fn *invoke_fn, - u32 *sec_caps) + u32 *sec_caps, u32 *max_notif_value) { union { struct arm_smccc_res smccc; @@ -425,6 +439,7 @@ static bool optee_msg_exchange_capabilities(optee_invoke_fn *invoke_fn, return false;
*sec_caps = res.result.capabilities; + *max_notif_value = res.result.max_notif_value; return true; }
@@ -609,6 +624,7 @@ static int optee_probe(struct platform_device *pdev) struct optee *optee = NULL; void *memremaped_shm = NULL; struct tee_device *teedev; + u32 max_notif_value; u32 sec_caps; int rc;
@@ -628,7 +644,8 @@ static int optee_probe(struct platform_device *pdev) return -EINVAL; }
- if (!optee_msg_exchange_capabilities(invoke_fn, &sec_caps)) { + if (!optee_msg_exchange_capabilities(invoke_fn, &sec_caps, + &max_notif_value)) { pr_warn("capabilities mismatch\n"); return -EINVAL; } @@ -651,7 +668,7 @@ static int optee_probe(struct platform_device *pdev) optee = kzalloc(sizeof(*optee), GFP_KERNEL); if (!optee) { rc = -ENOMEM; - goto err; + goto err_free_pool; }
optee->invoke_fn = invoke_fn; @@ -660,24 +677,24 @@ static int optee_probe(struct platform_device *pdev) teedev = tee_device_alloc(&optee_desc, NULL, pool, optee); if (IS_ERR(teedev)) { rc = PTR_ERR(teedev); - goto err; + goto err_free_optee; } optee->teedev = teedev;
teedev = tee_device_alloc(&optee_supp_desc, NULL, pool, optee); if (IS_ERR(teedev)) { rc = PTR_ERR(teedev); - goto err; + goto err_unreg_teedev; } optee->supp_teedev = teedev;
rc = tee_device_register(optee->teedev); if (rc) - goto err; + goto err_unreg_supp_teedev;
rc = tee_device_register(optee->supp_teedev); if (rc) - goto err; + goto err_unreg_supp_teedev;
mutex_init(&optee->call_queue.mutex); INIT_LIST_HEAD(&optee->call_queue.waiters); @@ -687,10 +704,30 @@ static int optee_probe(struct platform_device *pdev)
platform_set_drvdata(pdev, optee);
- rc = optee_notif_init(optee, 255); - if (rc) { - optee_remove(pdev); - return rc; + if (sec_caps & OPTEE_SMC_SEC_CAP_ASYNC_NOTIF) { + unsigned int irq; + + rc = platform_get_irq(pdev, 0); + if (rc < 0) { + pr_err("platform_get_irq: ret %d\n", rc); + goto err_unreg_supp_teedev; + } + irq = rc; + + rc = optee_notif_init(optee, max_notif_value, irq); + if (rc) { + irq_dispose_mapping(irq); + optee_remove(pdev); + return rc; + } + enable_async_notif(optee->invoke_fn); + pr_info("Asynchronous notifications enabled\n"); + } else { + rc = optee_notif_init(optee, 255, 0); + if (rc) { + optee_remove(pdev); + return rc; + } }
optee_enable_shm_cache(optee); @@ -706,20 +743,15 @@ static int optee_probe(struct platform_device *pdev)
pr_info("initialized driver\n"); return 0; -err: - if (optee) { - /* - * tee_device_unregister() is safe to call even if the - * devices hasn't been registered with - * tee_device_register() yet. - */ - tee_device_unregister(optee->supp_teedev); - tee_device_unregister(optee->teedev); - kfree(optee); - } - if (pool) - tee_shm_pool_free(pool); - if (memremaped_shm) +err_unreg_supp_teedev: + tee_device_unregister(optee->supp_teedev); +err_unreg_teedev: + tee_device_unregister(optee->teedev); +err_free_optee: + kfree(optee); +err_free_pool: + tee_shm_pool_free(pool); + if (optee->memremaped_shm) memunmap(memremaped_shm); return rc; } diff --git a/drivers/tee/optee/notif.c b/drivers/tee/optee/notif.c index a28fa03dcd0e..ecfa82797695 100644 --- a/drivers/tee/optee/notif.c +++ b/drivers/tee/optee/notif.c @@ -7,10 +7,14 @@
#include <linux/arm-smccc.h> #include <linux/errno.h> +#include <linux/interrupt.h> +#include <linux/irqdomain.h> #include <linux/slab.h> #include <linux/spinlock.h> #include <linux/tee_drv.h> #include "optee_private.h" +#include "optee_smc.h" +#include "optee_rpc_cmd.h"
struct notif_entry { struct list_head link; @@ -18,6 +22,54 @@ struct notif_entry { u_int key; };
+static u32 get_async_notif_value(optee_invoke_fn *invoke_fn, bool *value_valid, + bool *value_pending) +{ + struct arm_smccc_res res; + + invoke_fn(OPTEE_SMC_GET_ASYNC_NOTIF_VALUE, 0, 0, 0, 0, 0, 0, 0, &res); + + if (res.a0) + return 0; + *value_valid = (res.a2 & OPTEE_SMC_ASYNC_NOTIF_VALUE_VALID); + *value_pending = (res.a2 & OPTEE_SMC_ASYNC_NOTIF_VALUE_PENDING); + return res.a1; +} + +static irqreturn_t notif_irq_handler(int irq, void *dev_id) +{ + struct optee *optee = dev_id; + bool do_bottom_half = false; + bool value_valid; + bool value_pending; + u32 value; + + do { + value = get_async_notif_value(optee->invoke_fn, &value_valid, + &value_pending); + if (!value_valid) + break; + + if (value == OPTEE_SMC_ASYNC_NOTIF_VALUE_DO_BOTTOM_HALF) + do_bottom_half = true; + else + optee_notif_send(optee, value); + } while (value_pending); + + if (do_bottom_half) + return IRQ_WAKE_THREAD; + return IRQ_HANDLED; +} + +static irqreturn_t notif_irq_thread_fn(int irq, void *dev_id) +{ + struct optee *optee = dev_id; + + optee_do_bottom_half(optee->notif.ctx); + + return IRQ_HANDLED; +} + static bool have_key(struct optee *optee, u_int key) { struct notif_entry *entry; @@ -106,20 +158,69 @@ int optee_notif_send(struct optee *optee, u_int key) return 0; }
-int optee_notif_init(struct optee *optee, u_int max_key) +int optee_notif_init(struct optee *optee, u_int max_key, u_int irq) { + struct tee_context *ctx; + int rc; + + if (irq) { + ctx = tee_dev_open_helper(optee->teedev); + if (IS_ERR(ctx)) + return PTR_ERR(ctx); + + optee->notif.ctx = ctx; + } + spin_lock_init(&optee->notif.lock); INIT_LIST_HEAD(&optee->notif.db); optee->notif.bitmap = bitmap_zalloc(max_key, GFP_KERNEL); - if (!optee->notif.bitmap) - return -ENOMEM; - + if (!optee->notif.bitmap) { + rc = -ENOMEM; + goto err_put_ctx; + } optee->notif.max_key = max_key;
+ if (irq) { + rc = request_threaded_irq(irq, notif_irq_handler, + notif_irq_thread_fn, + 0, "optee_notification", optee); + if (rc) + goto err_free_bitmap; + + optee->notif.irq = irq; + } + return 0; + +err_free_bitmap: + kfree(optee->notif.bitmap); +err_put_ctx: + tee_dev_ctx_put(optee->notif.ctx); + optee->notif.ctx = NULL; + + return rc; }
void optee_notif_uninit(struct optee *optee) { + if (optee->notif.ctx) { + optee_stop_async_notif(optee->notif.ctx); + if (optee->notif.irq) { + free_irq(optee->notif.irq, optee); + irq_dispose_mapping(optee->notif.irq); + } + + /* + * The thread normally working with optee->notif.ctx was + * stopped with free_irq() above. + * + * Note we're not using teedev_close_context() or + * tee_client_close_context() since we have already called + * tee_device_put() while initializing to avoid a circular + * reference counting. + */ + tee_dev_ctx_put(optee->notif.ctx); + } + kfree(optee->notif.bitmap); } diff --git a/drivers/tee/optee/optee_msg.h b/drivers/tee/optee/optee_msg.h index e3d72d09c484..3e09c8386e46 100644 --- a/drivers/tee/optee/optee_msg.h +++ b/drivers/tee/optee/optee_msg.h @@ -293,6 +293,13 @@ struct optee_msg_arg { * [in] param[0].u.rmem.shm_ref holds shared memory reference * [in] param[0].u.rmem.offs 0 * [in] param[0].u.rmem.size 0 + * + * OPTEE_MSG_CMD_DO_BOTTOM_HALF does the scheduled bottom half processing + * of a driver. + * + * OPTEE_MSG_CMD_STOP_ASYNC_NOTIF informs secure world that from now is + * normal world unable to process asynchronous notifications. Typically + * used when the driver is shut down. */ #define OPTEE_MSG_CMD_OPEN_SESSION 0 #define OPTEE_MSG_CMD_INVOKE_COMMAND 1 @@ -300,6 +307,8 @@ struct optee_msg_arg { #define OPTEE_MSG_CMD_CANCEL 3 #define OPTEE_MSG_CMD_REGISTER_SHM 4 #define OPTEE_MSG_CMD_UNREGISTER_SHM 5 +#define OPTEE_MSG_CMD_DO_BOTTOM_HALF 6 +#define OPTEE_MSG_CMD_STOP_ASYNC_NOTIF 7 #define OPTEE_MSG_FUNCID_CALL_WITH_ARG 0x0004
#endif /* _OPTEE_MSG_H */ diff --git a/drivers/tee/optee/optee_private.h b/drivers/tee/optee/optee_private.h index 7dc058d008b2..62365912a70b 100644 --- a/drivers/tee/optee/optee_private.h +++ b/drivers/tee/optee/optee_private.h @@ -37,6 +37,8 @@ struct optee_call_queue {
struct optee_notif { u_int max_key; + unsigned int irq; + struct tee_context *ctx; /* Serializes access to the elements below in this struct */ spinlock_t lock; struct list_head db; @@ -132,7 +134,7 @@ void optee_handle_rpc(struct tee_context *ctx, struct optee_rpc_param *param, struct optee_call_ctx *call_ctx); void optee_rpc_finalize_call(struct optee_call_ctx *call_ctx);
-int optee_notif_init(struct optee *optee, u_int max_key); +int optee_notif_init(struct optee *optee, u_int max_key, u_int irq); void optee_notif_uninit(struct optee *optee); int optee_notif_wait(struct optee *optee, u_int key); int optee_notif_send(struct optee *optee, u_int key); @@ -159,6 +161,8 @@ int optee_close_session(struct tee_context *ctx, u32 session); int optee_invoke_func(struct tee_context *ctx, struct tee_ioctl_invoke_arg *arg, struct tee_param *param); int optee_cancel_req(struct tee_context *ctx, u32 cancel_id, u32 session); +int optee_do_bottom_half(struct tee_context *ctx); +int optee_stop_async_notif(struct tee_context *ctx);
void optee_enable_shm_cache(struct optee *optee); void optee_disable_shm_cache(struct optee *optee); diff --git a/drivers/tee/optee/optee_smc.h b/drivers/tee/optee/optee_smc.h index 80eb763a8a80..c6eec6b6febf 100644 --- a/drivers/tee/optee/optee_smc.h +++ b/drivers/tee/optee/optee_smc.h @@ -107,6 +107,12 @@ struct optee_smc_call_get_os_revision_result { /* * Call with struct optee_msg_arg as argument * + * When calling this function normal world has a few responsibilities: + * 1. It must be able to handle eventual RPCs + * 2. Non-secure interrupts should not be masked + * 3. If asynchronous notifications has be negotiated successfully, then + * asynchronous notifications should be unmasked during this call. + * * Call register usage: * a0 SMC Function ID, OPTEE_SMC*CALL_WITH_ARG * a1 Upper 32 bits of a 64-bit physical pointer to a struct optee_msg_arg @@ -195,7 +201,8 @@ struct optee_smc_get_shm_config_result { * Normal return register usage: * a0 OPTEE_SMC_RETURN_OK * a1 bitfield of secure world capabilities OPTEE_SMC_SEC_CAP_* - * a2-7 Preserved + * a2 The maximum secure world notification number + * a3-7 Preserved * * Error return register usage: * a0 OPTEE_SMC_RETURN_ENOTAVAIL, can't use the capabilities from normal world @@ -218,6 +225,8 @@ struct optee_smc_get_shm_config_result { #define OPTEE_SMC_SEC_CAP_VIRTUALIZATION BIT(3) /* Secure world supports Shared Memory with a NULL reference */ #define OPTEE_SMC_SEC_CAP_MEMREF_NULL BIT(4) +/* Secure world supports asynchronous notification of normal world */ +#define OPTEE_SMC_SEC_CAP_ASYNC_NOTIF BIT(5)
#define OPTEE_SMC_FUNCID_EXCHANGE_CAPABILITIES 9 #define OPTEE_SMC_EXCHANGE_CAPABILITIES \ @@ -226,8 +235,8 @@ struct optee_smc_get_shm_config_result { struct optee_smc_exchange_capabilities_result { unsigned long status; unsigned long capabilities; + unsigned long max_notif_value; unsigned long reserved0; - unsigned long reserved1; };
/* @@ -319,6 +328,68 @@ struct optee_smc_disable_shm_cache_result { #define OPTEE_SMC_GET_THREAD_COUNT \ OPTEE_SMC_FAST_CALL_VAL(OPTEE_SMC_FUNCID_GET_THREAD_COUNT)
+/* + * Inform OP-TEE that normal world is able to receive asynchronous + * notifications. + * + * Call requests usage: + * a0 SMC Function ID, OPTEE_SMC_ENABLE_ASYNC_NOTIF + * a1-6 Not used + * a7 Hypervisor Client ID register + * + * Normal return register usage: + * a0 OPTEE_SMC_RETURN_OK + * a1-7 Preserved + * + * Not supported return register usage: + * a0 OPTEE_SMC_RETURN_ENOTAVAIL + * a1-7 Preserved + */ +#define OPTEE_SMC_FUNCID_ENABLE_ASYNC_NOTIF 16 +#define OPTEE_SMC_ENABLE_ASYNC_NOTIF \ + OPTEE_SMC_FAST_CALL_VAL(OPTEE_SMC_FUNCID_ENABLE_ASYNC_NOTIF) + +/* + * Retrieve a value of notifications pended since the last call of this + * function. + * + * OP-TEE keeps a records of all posted values. When an interrupts is + * received which indicates that there are posed values this function + * should be called until all pended values has been retrieved. When a + * value is retrieved it's cleared from the record in secure world. + * + * Call requests usage: + * a0 SMC Function ID, OPTEE_SMC_GET_ASYNC_NOTIF_VALUE + * a1-6 Not used + * a7 Hypervisor Client ID register + * + * Normal return register usage: + * a0 OPTEE_SMC_RETURN_OK + * a1 value + * a2 Bit[0]: OPTEE_SMC_ASYNC_NOTIF_VALUE_VALID if the value in a1 is + * valid, else 0 if no values where pending + * a2 Bit[1]: OPTEE_SMC_ASYNC_NOTIF_VALUE_PENDING if another value is + * pending, else 0. + * Bit[31:2]: MBZ + * a3-7 Preserved + * + * Not supported return register usage: + * a0 OPTEE_SMC_RETURN_ENOTAVAIL + * a1-7 Preserved + */ +#define OPTEE_SMC_ASYNC_NOTIF_VALUE_VALID BIT(0) +#define OPTEE_SMC_ASYNC_NOTIF_VALUE_PENDING BIT(1) + +/* + * Notification that OP-TEE expects a yielding call to do some bottom half + * work in a driver. + */ +#define OPTEE_SMC_ASYNC_NOTIF_VALUE_DO_BOTTOM_HALF 0 + +#define OPTEE_SMC_FUNCID_GET_ASYNC_NOTIF_VALUE 17 +#define OPTEE_SMC_GET_ASYNC_NOTIF_VALUE \ + OPTEE_SMC_FAST_CALL_VAL(OPTEE_SMC_FUNCID_GET_ASYNC_NOTIF_VALUE) + /* * Resume from RPC (for example after processing a foreign interrupt) *
op-tee@lists.trustedfirmware.org
-
Enrique Echeverria -
Jens Wiklander -
Marc Zyngier