ROM configures certain firewalls based on its usage, which includes
the one in front of boot peripherals. In specific case of boot
peripherals, ROM does not open up the full address space corresponding
to the peripherals. Like in OSPI, ROM only configures the firewall region
for 32 bit address space and mark 64bit address space flash regions
as in-accessible.
When security-cfg is initialized by sysfw, all the non-configured
firewalls are kept in bypass state using a global setting. Since ROM
configured firewalls for certain peripherals, these will not be touched.
So when bootloader touches any of the address space that ROM marked as
in-accessible, system raises a firewall exception causing boot hang.
It would have been ideal if sysfw cleans up the ROM configured boot
peripheral firewalls. Given the memory overhead to store this
information provided by ROM and the boot time increase in re configuring
the firewalls, it is concluded to clean this up in bootloaders.
So disable all the firewalls that ROM doesn't open up the full address
space.
Signed-off-by: Andrew F. Davis <afd@ti.com>
Signed-off-by: Venkateswara Rao Mandela <venkat.mandela@ti.com>
Signed-off-by: Lokesh Vutla <lokeshvutla@ti.com>
Not finding a node that we try to disable does not always need to be
fatal to boot but should at least print out a warning. Return error
from fdt_disable_node as it did fail to disable the node, but only
warn in the case of disabling the TRNG as this will not prevent boot.
Signed-off-by: Andrew F. Davis <afd@ti.com>
Reviewed-by: Lokesh Vutla <lokeshvutla@ti.com>
Signed-off-by: Lokesh Vutla <lokeshvutla@ti.com>
On K3 devices there are 2 conditions where R5F can deadlock:
1.When software is performing series of store operations to
cacheable write back/write allocate memory region and later
on software execute barrier operation (DSB or DMB). R5F may
hang at the barrier instruction.
2.When software is performing a mix of load and store operations
within a tight loop and store operations are all writing to
cacheable write back/write allocates memory regions, R5F may
hang at one of the load instruction.
To avoid the above two conditions disable linefill optimization
inside Cortex R5F which will make R5F to only issue up to 2 cache
line fills at any point of time.
Signed-off-by: Lokesh Vutla <lokeshvutla@ti.com>
U-Boot cleans and invalidate L1 and L2 caches before jumping to Linux
by set/way in cleanup_before_linux(). Additionally there is a custom
hook provided to clean and invalidate L3 cache.
Unfortunately on K3 devices(having a coherent architecture), there is no
easy way to quickly clean all the cache lines for L3. The entire address
range needs to be cleaned and invalidated by Virtual Address. This can
be implemented using the L3 custom hook but it take lot of time to clean
the entire address range. In the interest of boot time this might not be
a viable solution.
The best hit is to make sure the loaded Linux image is flushed so that
the entire image is written to DDR from L3. When Linux starts running with
caches disabled the full image is available from DDR.
Reported-by: Andrew F. Davis <afd@ti.com>
Reported-by: Faiz Abbas <faiz_abbas@ti.com>
Signed-off-by: Lokesh Vutla <lokeshvutla@ti.com>
Implement an early console functionality in SPL that can be used before
the main console is being brought up. This helps in situations where the
main console is dependent on System Firmware (SYSFW) being up and running,
which is usually not the case during the very early stages of boot. Using
this early console functionality will allow for an alternate serial port
to be used to support things like UART-based boot and early diagnostic
messages until the main console is ready to get activated.
Signed-off-by: Andreas Dannenberg <dannenberg@ti.com>
On HS devices the access to TRNG is restricted on the non-secure
ARM side, disable the node in DT to prevent firewall violations.
Signed-off-by: Andrew F. Davis <afd@ti.com>
Reviewed-by: Lokesh Vutla <lokeshvutla@ti.com>
API get_ti_sci_handle() is relying on the device-tree node name
to be "dmsc" for probing the ti_sci device. But with the introduction
of debug messages for dmsc, the node name changed to dmsc@44083000.
Because of this ti_sci is never probed cause a boot failure. Instead
of relying on device-tree node name, use the first available firmware
node for probing ti_sci.
Signed-off-by: Lokesh Vutla <lokeshvutla@ti.com>
The J721E SoC belongs to the K3 Multicore SoC architecture platform,
providing advanced system integration to enable lower system costs
of automotive applications such as infotainment, cluster, premium
Audio, Gateway, industrial and a range of broad market applications.
This SoC is designed around reducing the system cost by eliminating
the need of an external system MCU and is targeted towards ASIL-B/C
certification/requirements in addition to allowing complex software
and system use-cases.
Some highlights of this SoC are:
* Dual Cortex-A72s in a single cluster, three clusters of lockstep
capable dual Cortex-R5F MCUs, Deep-learning Matrix Multiply Accelerator(MMA),
C7x floating point Vector DSP, Two C66x floating point DSPs.
* 3D GPU PowerVR Rogue 8XE GE8430
* Vision Processing Accelerator (VPAC) with image signal processor and Depth
and Motion Processing Accelerator (DMPAC)
* Two Gigabit Industrial Communication Subsystems (ICSSG), each with dual
PRUs and dual RTUs
* Two CSI2.0 4L RX plus one CSI2.0 4L TX, one eDP/DP, One DSI Tx, and
up to two DPI interfaces.
* Integrated Ethernet switch supporting up to a total of 8 external ports in
addition to legacy Ethernet switch of up to 2 ports.
* System MMU (SMMU) Version 3.0 and advanced virtualisation
capabilities.
* Upto 4 PCIe-GEN3 controllers, 2 USB3.0 Dual-role device subsystems,
16 MCANs, 12 McASP, eMMC and SD, UFS, OSPI/HyperBus memory controller, QSPI,
I3C and I2C, eCAP/eQEP, eHRPWM, MLB among other peripherals.
* Two hardware accelerator block containing AES/DES/SHA/MD5 called SA2UL
management.
* Configurable L3 Cache and IO-coherent architecture with high data throughput
capable distributed DMA architecture under NAVSS
* Centralized System Controller for Security, Power, and Resource
Management (DMSC)
See J721E Technical Reference Manual (SPRUIL1, May 2019)
for further details: http://www.ti.com/lit/pdf/spruil1
Add base support for J721E SoC
Signed-off-by: Lokesh Vutla <lokeshvutla@ti.com>
Signed-off-by: Andreas Dannenberg <dannenberg@ti.com>
Signed-off-by: Nishanth Menon <nm@ti.com>
Rather than simply parking the R5 core in WFE after starting up ATF
on A53 instead use SYSFW API to properly shut down the R5 CPU cores
as well as associated timer resources that were pre-allocated. This
allows software further downstream to properly and gracefully bring
the R5 cores back online if desired.
Signed-off-by: Andreas Dannenberg <dannenberg@ti.com>
Signed-off-by: Lokesh Vutla <lokeshvutla@ti.com>
Certain parts of msmc sram can be used by DMSC or can be
marked as L3 cache. Since the available size can vary, changing
DT every time the size varies might be painful. So, query this
information using TISCI cmd and fixup the DT for kernel.
Fixing up DT does the following:
- Create a sram node if not available
- update the reg property with available size
- update ranges property
- loop through available sub nodes and delete it if:
- mentioned size is out if available range
- subnode represents l3 cache or dmsc usage.
Signed-off-by: Lokesh Vutla <lokeshvutla@ti.com>
Perform some cosmetic cleanup of the ATF image startup function, namely
fixing a spelling mistake, capitalization of a few words, spacing, as
well aligning how errors are printed and as using panic() for cases that
were using a combination of printf() + hang().
Signed-off-by: Andreas Dannenberg <dannenberg@ti.com>
Reviewed-by: Lokesh Vutla <lokeshvutla@ti.com>
Considering the boot time requirements, Cortex-A core
should be able to start immediately after SPL on R5.
Add support for the same.
Reviewed-by: Tom Rini <trini@konsulko.com>
Signed-off-by: Lokesh Vutla <lokeshvutla@ti.com>
