[PATCH 0/6] K3 R5F & DSP IPC-only mode support
Suman Anna
s-anna at ti.com
Fri May 21 17:03:03 PDT 2021
Hi All,
The following series adds the IPC-only mode support for the TI K3 R5F
and DSP (C66x and C71x) remoteprocs covering AM65x, J721E, J7200 and
AM64x SoCs. Patches are on top of 5.13-rc1 + Minor remoteproc cleanup
series [1] + TI K3 R5F remoteproc support on AM64x series [2].
The 'IPC-only' mode terminology essentially means establishing only the
IPC rpmsg stack for remoteprocs that are booted early by bootloaders and
supporting only 'attach' and 'detach' of the remoteprocs. The existing
remoteproc infrastructure does support 'stop' of an early-booted remote
processor, and the TI K3 remoteprocs are intentionally being limited to
'detach', and are designed to only establish and tear apart IPC (the
virtio devices that provide the IPC transport) without ever shutting
down the core. This is done by introducing a new flag in remoteproc core
in Patch 2. Support for the regular 'stop' can be enhanced in the future
easily by adding a new sysfs or configfs file that changes this flag,
but IPC-only is the expected usage model for now on K3 SoCs.
Following is a summary of some of the design details:
- The TI K3 SoCs use a dedicated system processor for Power and Clocks
and the IPC-only mode is detected by communicating with this central
processor and checking on the power status of remoteprocs.
- The driver support is provided through the recently added .attach(),
.detach() and .get_loaded_rsc_table() rproc ops, and new RPROC_ATTACHED
and RPROC_DETACHED state flags.
- The default kernel dts cluster modes and TCM configuration for R5Fs
are ignored and overridden by querying this config from the system
processor in IPC-only mode. Any core not booted earlier will fall back
to using the kernel dts (eg: R5F Core0 can be booted by bootloader, and
R5F Core1 will be booted using kernel).
- The remoteproc firmwares are not loaded again in kernel to retrieve
the resource table. This allows early-boot to be done from different
boot media such as OSPI or eMMC, and be completely independent of the
rootfs in SDCard or NFS.
- The IPC-only mode support follows a design-by-contract (DBC) approach
to achieve the above. The resource tables are always expected to be at
the base on the DDR memory region reserved for firmwares (base of the
second memory-region property in the remoteproc dts node). This also
eliminates the need for dts properties being dynamically updated to
provide the resource table address (there are no dedicated h/w
registers either like with the ST SoCs). This scales to designs where
either U-Boot is not involved or a different processor is used to
load the various remoteprocs.
- The "stop" command results in a failure when the remoteproc is in
IPC-only mode, and "detach" serves as the corresponding replacement.
Note that there is no equivalent "attach" command, and "start" command
also serves as "attach" for remoteprocs that are in detached state.
- The remoteproc sysfs 'state' file shows the mode as "detached"
when the early-booted remoteproc is detached. "offline" continues
to be shown when the remoteproc is powered off.
Following is the summary of patches:
- Patch 1 is a minor cleanup patch in remoteproc core to make the
helper functions look symmetric.
- Patch 2 introduces a new flag 'detach_on_shutdown' and enhances
the remoteproc core to restrict stop on early-booted remoteprocs.
This patch provides the required behavior for K3 IPC-only mode (no
stopping of early-booted rprocs). The default behavior continues
to support stopping of remoteproc like with the existing code.
- Patches 3 and 5 refactor the mailbox request code out of start
for reuse in the new attach callbacks.
- Patch 4 adds the IPC-only mode support for R5F.
- Patch 6 adds the IPC-only mode support for both K3 C66x and C71x
DSPs.
Patches 1, 3 and 5 can be considered cleanup, so I suggest these be
picked up if there are any major comments on the others.
I have verified the different combinations on J721E, J7200 and AM65x
SoCs. AM64x currently lacks early-boot support, but the logic is ready
for Single-CPU and Split modes that are specific to AM64x SoCs.
regards
Suman
[1] https://patchwork.kernel.org/project/linux-remoteproc/list/?series=485235
[2] https://patchwork.kernel.org/project/linux-remoteproc/list/?series=456755
Suman Anna (6):
remoteproc: Introduce rproc_detach_device() wrapper
remoteproc: Add support for detach-only during shutdown
remoteproc: k3-r5: Refactor mbox request code in start
remoteproc: k3-r5: Add support for IPC-only mode for all R5Fs
remoteproc: k3-dsp: Refactor mbox request code in start
remoteproc: k3-dsp: Add support for IPC-only mode for all K3 DSPs
drivers/remoteproc/remoteproc_cdev.c | 7 +
drivers/remoteproc/remoteproc_core.c | 7 +-
drivers/remoteproc/remoteproc_internal.h | 8 +
drivers/remoteproc/remoteproc_sysfs.c | 6 +
drivers/remoteproc/ti_k3_dsp_remoteproc.c | 214 ++++++++++++++---
drivers/remoteproc/ti_k3_r5_remoteproc.c | 279 ++++++++++++++++++++--
include/linux/remoteproc.h | 3 +
7 files changed, 459 insertions(+), 65 deletions(-)
--
2.30.1
More information about the linux-arm-kernel
mailing list