diff mbox series

[RESEND,V6,2/3] Documentation: remoteproc: add overview section

Message ID 20241106051016.89113-3-yesanishhere@gmail.com (mailing list archive)
State New
Headers show
Series remoteproc documentation update | expand

Commit Message

anish kumar Nov. 6, 2024, 5:10 a.m. UTC
Added overview section which details
how the remote processor framework works and
how it handles crashes.

Signed-off-by: anish kumar <yesanishhere@gmail.com>
---
 Documentation/staging/remoteproc.rst | 43 ++++++++++++++++++++++++++++
 1 file changed, 43 insertions(+)
diff mbox series

Patch

diff --git a/Documentation/staging/remoteproc.rst b/Documentation/staging/remoteproc.rst
index eeebbeca71de..e0bf68ceade8 100644
--- a/Documentation/staging/remoteproc.rst
+++ b/Documentation/staging/remoteproc.rst
@@ -46,6 +46,49 @@  of several key components:
 - **Virtio Support**: Facilitates interaction with the virtio and
   rpmsg bus.
 
+Overview
+========
+
+The framework begins by gathering information about the firmware file
+to be downloaded through the request_firmware function. It supports
+the ELF format and parses the firmware image to identify the physical
+addresses that need to be populated from the corresponding ELF sections.
+Once this information is obtained from the driver, the framework transfers
+the data to the specified addresses and starts the remote processor,
+along with subdevices.
+
+Dependent devices, referred to as `subdevices` within the framework,
+are also managed post-registration by their respective drivers.
+Subdevices can register themselves using `rproc_(add/remove)_subdev`.
+Non-remoteproc drivers can use subdevices as a way to logically connect
+to remote and get lifecycle notifications of the remote.
+
+The framework oversees the lifecycle of the remote and
+provides the `rproc_report_crash` function, which the driver invokes
+upon receiving a crash notification from the remote. The
+notification method can differ based on the design of the remote
+processor and its communication with the application processor. For
+instance, if the remote is a DSP equipped with a watchdog,
+unresponsive behavior triggers the watchdog, generating an interrupt
+that routes to the application processor, allowing it to call
+`rproc_report_crash` in the driver's interrupt context.
+
+During crash handling, the framework performs the following actions:
+
+a. Sends a request to stop the remote and any connected or
+   dependent subdevices.
+b. Generates a coredump, dumping all `resources` requested by the
+   remote alongside relevant debugging information. Resources are
+   explained below.
+c. Reloads the firmware and restarts the remote processor.
+
+If the `RPROC_FEAT_ATTACH_ON_RECOVERY` flag is set, the detach and
+attach callbacks of the driver are invoked without reloading the
+firmware. This is useful when the remote requires no
+assistance for recovery, or when the application processor can restart
+independently. After recovery, the application processor can reattach
+to the remote.
+
 User API
 ========