On Thu Oct 30, 2025 at 7:16 PM IST, Andrew Davis wrote:
> On 10/30/25 12:19 AM, Anshul Dalal wrote:
>> We use the spl_board_prepare_for_boot hook to call k3_r5_falcon_prep
>> which is ran after tispl is loaded but before jump_to_image.
>> 
>> In k3_r5_falcon_prep, we find the boot media and load the kernel FIT
>> just as standard secure falcon mode (since spl_start_uboot returns 0
>> now). Once the kernel and args are loaded.
>> 
>> Now when the flow goes to jump_to_image, we do the regular pre-jump
>> procedure and jump to TFA which jumps to the kernel directly since we
>> have already loaded the kernel and dtb at their respective addresses
>> (PRELOADED_BL33_BASE and K3_HW_CONFIG_BASE).
>> 
>> Signed-off-by: Anshul Dalal <[email protected]>
>> ---
>>   arch/arm/mach-k3/common.c    |  8 +++++++
>>   arch/arm/mach-k3/common.h    |  4 ++++
>>   arch/arm/mach-k3/r5/common.c | 45 ++++++++++++++++++++++++++++++++++++
>>   3 files changed, 57 insertions(+)
>> 
>> diff --git a/arch/arm/mach-k3/common.c b/arch/arm/mach-k3/common.c
>> index 04a0f232d84..c446d1c47fd 100644
>> --- a/arch/arm/mach-k3/common.c
>> +++ b/arch/arm/mach-k3/common.c
>> @@ -360,6 +360,14 @@ void spl_perform_arch_fixups(struct spl_image_info 
>> *spl_image)
>>   
>>   void spl_board_prepare_for_boot(void)
>>   {
>> +#if IS_ENABLED(CONFIG_SPL_OS_BOOT_SECURE) && !IS_ENABLED(CONFIG_ARM64)
>> +    int ret;
>> +
>> +    ret = k3_r5_falcon_prep();
>> +    if (ret)
>> +            panic("%s: Failed to boot in falcon mode: %d\n", __func__, ret);
>> +#endif /* falcon mode on R5 SPL */
>> +
>>   #if !(defined(CONFIG_SYS_ICACHE_OFF) && defined(CONFIG_SYS_DCACHE_OFF))
>>      dcache_disable();
>>   #endif
>> diff --git a/arch/arm/mach-k3/common.h b/arch/arm/mach-k3/common.h
>> index 52d3faaab5c..5807d358464 100644
>> --- a/arch/arm/mach-k3/common.h
>> +++ b/arch/arm/mach-k3/common.h
>> @@ -52,6 +52,10 @@ void do_board_detect(void);
>>   void ti_secure_image_check_binary(void **p_image, size_t *p_size);
>>   int shutdown_mcu_r5_core1(void);
>>   
>> +#if IS_ENABLED(CONFIG_SPL_OS_BOOT_SECURE) && !IS_ENABLED(CONFIG_ARM64)
>> +int k3_r5_falcon_bootmode(void);
>> +#endif
>> +
>>   #if (IS_ENABLED(CONFIG_K3_QOS))
>>   void setup_qos(void);
>>   #else
>> diff --git a/arch/arm/mach-k3/r5/common.c b/arch/arm/mach-k3/r5/common.c
>> index 6269b33f66b..439dd92ef54 100644
>> --- a/arch/arm/mach-k3/r5/common.c
>> +++ b/arch/arm/mach-k3/r5/common.c
>> @@ -376,3 +376,48 @@ void board_fit_image_post_process(const void *fit, int 
>> node, void **p_image,
>>      }
>>   }
>>   #endif
>> +
>> +#ifdef CONFIG_SPL_OS_BOOT_SECURE
>> +
>> +static bool tifalcon_loaded;
>
> Not strictly needed as statics will default to 0, assigning false
> here makes it more clear what the starting value should be,
>
> static bool tifalcon_loaded = false;
>
>> +
>> +int spl_start_uboot(void)
>> +{
>> +    /* If tifalcon.bin is not loaded, proceed to regular boot */
>> +    if (!tifalcon_loaded)
>> +            return 1;
>> +
>> +    /* Boot to linux on R5 SPL with tifalcon.bin loaded */
>> +            return 0;
>
> Extra indent tab here.
>
>> +}
>> +
>> +int k3_r5_falcon_prep(void)
>> +{
>> +    struct spl_image_loader *loader, *drv;
>> +    struct spl_image_info kernel_image;
>> +    struct spl_boot_device bootdev;
>> +    int ret = -ENXIO, n_ents;
>> +
>> +    tifalcon_loaded = true;
>
> Should this be set down after we make sure loading doesn't fail?
>

That wouldn't work actually, tifalcon_loaded is used to detect if
tifalcon.bin has been loaded or not. We then in-turn use this
information in our two step boot process to decide whether or not to do
falcon boot.

The call to load_image below will internally calls spl_start_uboot to
check for falcon mode status, which will return 0 now allowing us to
load fitImage.

So, the overall boot flow looks as follows:

        board_init_r
        |-> boot_from_devices
        |   +-> load_image (we load tifalcon.bin here since spl_start_uboot 
returns 1)
        |
        +-> spl_prepare_for_boot
        |   +-> k3_falcon_prep
        |       +-> load_image (we load fitImage here since spl_start_uboot 
returns 0 now)
        |
        +-> jump_to_image

>> +    memset(&kernel_image, '\0', sizeof(kernel_image));
>> +    drv = ll_entry_start(struct spl_image_loader, spl_image_loader);
>> +    n_ents = ll_entry_count(struct spl_image_loader, spl_image_loader);
>> +    bootdev.boot_device = spl_boot_device();
>> +
>> +    for (loader = drv; loader != drv + n_ents; loader++) {
>> +            if (bootdev.boot_device != loader->boot_device)
>> +                    continue;
>> +            if (loader) {
>
> If loader is NULL, the if above containing loader->boot_device would have
> already null pointer deref crashed. You might want to move this check up
> into the if check above.

You're right, this needs to be fixed. I was following a similar
structure to boot_from_devices (common/spl/spl.c) which also seems to
have the same issue.

Regards,
Anshul

>
> Andrew
>
>> +                    printf("Loading falcon payload from %s\n",
>> +                           spl_loader_name(loader));
>> +                    ret = loader->load_image(&kernel_image, &bootdev);
>> +                    if (ret)
>> +                            continue;
>> +
>> +                    return 0;
>> +            }
>> +    }
>> +
>> +    return ret;
>> +}
>> +#endif

Reply via email to