Hello Bjorn,

On 9/26/24 05:51, Bjorn Andersson wrote:

On Fri, Aug 30, 2024 at 11:51:44AM GMT, Arnaud Pouliquen wrote:

quoted

Add support for releasing remote processor firmware through
the Trusted Execution Environment (TEE) interface.

The tee_rproc_release_fw() function is called in the following cases:

- An error occurs in rproc_start() between the loading of the segments and
  the start of the remote processor.
- When rproc_release_fw is called on error or after stopping the remote
  processor.

Signed-off-by: Arnaud Pouliquen <arnaud.pouliquen@foss.st.com>
---
 drivers/remoteproc/remoteproc_core.c | 10 ++++++++--
 1 file changed, 8 insertions(+), 2 deletions(-)

diff --git a/drivers/remoteproc/remoteproc_core.c b/drivers/remoteproc/remoteproc_core.c
index 7694817f25d4..32052dedc149 100644
--- a/drivers/remoteproc/remoteproc_core.c
+++ b/drivers/remoteproc/remoteproc_core.c

@@ -29,6 +29,7 @@
 #include <linux/debugfs.h>
 #include <linux/rculist.h>
 #include <linux/remoteproc.h>
+#include <linux/remoteproc_tee.h>
 #include <linux/iommu.h>
 #include <linux/idr.h>
 #include <linux/elf.h>

@@ -1258,6 +1259,9 @@ static int rproc_alloc_registered_carveouts(struct rproc *rproc)
 
 static void rproc_release_fw(struct rproc *rproc)
 {
+	if (rproc->state == RPROC_OFFLINE && rproc->tee_interface)
+		tee_rproc_release_fw(rproc);

I don't like the idea of having op-tee specific calls made from the
core. If the problem is that we need to unroll something we did at load,
can we instead come up with a more generic mechanism to unload that? Or

As proposed in [1] an alternative could be to define a new rproc_ops->release_fw
operation that will be initialized to tee_rproc_release_fw in the platform driver.

can we perhaps postpone the tee interaction until start() to avoid the
gap?

In such a case, the management of the resource table should also be postponed
as the firmware has to be authenticated first.

The OP-TEE implementation authenticates the firmware during the load
(in-destination memory authentication), so the sequence is:
1) Load the firmware.
2) Get the resource table and initialize resources.
3) Start the firmware.

The tee_rproc_release_fw() is used if something goes wrong during step 2 an3.

From my perspective, this would result in an alternative boot sequence, as we
have today for "attach". I proposed this approach in my V3 [2]. But this add
complexity in remote proc core.


Please, could you align with Mathieu to define how we should move forward to
address your concerns?

[1]https://lkml.org/lkml/2024/9/18/612
[2]https://lore.kernel.org/lkml/8af59b01-53cf-4fc4-9946-6c630fb7b38e@quicinc.com/T/ (local)

Thanks and Regards,
Arnaud

PS. Most of the Qualcomm drivers are TEE-based...so the "tee_interface"
boolean check here is not very nice.

Regards,
Bjorn

quoted

+
 	/* Free the copy of the resource table */
 	kfree(rproc->cached_table);
 	rproc->cached_table = NULL;

@@ -1348,7 +1352,7 @@ static int rproc_start(struct rproc *rproc, const struct firmware *fw)
 	if (ret) {
 		dev_err(dev, "failed to prepare subdevices for %s: %d\n",
 			rproc->name, ret);
-		goto reset_table_ptr;
+		goto release_fw;
 	}
 
 	/* power up the remote processor */

@@ -1376,7 +1380,9 @@ static int rproc_start(struct rproc *rproc, const struct firmware *fw)
 	rproc->ops->stop(rproc);
 unprepare_subdevices:
 	rproc_unprepare_subdevices(rproc);
-reset_table_ptr:
+release_fw:
+	if (rproc->tee_interface)
+		tee_rproc_release_fw(rproc);
 	rproc->table_ptr = rproc->cached_table;
 
 	return ret;

-- 
2.25.1