Hi Breno,
On 19/09/19 03:31, Breno Matheus Lima wrote:
HI Stefano and Jagan,
Em qua, 18 de set de 2019 às 04:59, Stefano Babic sbabic@denx.de escreveu:
Hi Jagan, Breno,
On 17/09/19 09:13, Jagan Teki wrote:
Hi Breno,
On Thu, Jul 18, 2019 at 6:06 PM Breno Matheus Lima breno.lima@nxp.com wrote:
In case CONFIG_SECURE_BOOT is enabled we need to limit the SPL size to avoid a possible HAB failure event:
--------- HAB Event 1 ----------------- event data: 0xdb 0x00 0x14 0x42 0x33 0x22 0x33 0x00 0x00 0x00 0x00 0x0f 0x00 0x90 0x70 0x00 0x00 0x01 0x10 0x00 STS = HAB_FAILURE (0x33) RSN = HAB_INV_ADDRESS (0x22) CTX = HAB_CTX_TARGET (0x33) ENG = HAB_ENG_ANY (0x00)
As explained in Commit 23612534fe0f ("spl: imx6: Provide a SPL_SIZE_LIMIT default") the i.MX6 SPL size limit is 68KB.
The ROM code is copying the image size defined in boot data to its respective load address, in case we exceed the OCRAM free region a HAB invalid address failure event is generated.
The maximum CSF size is defined in CONFIG_CSF_SIZE, reduce SPL size limit based on this configuration.
Signed-off-by: Breno Lima breno.lima@nxp.com
tools/spl_size_limit.c | 3 +++ 1 file changed, 3 insertions(+)
diff --git a/tools/spl_size_limit.c b/tools/spl_size_limit.c index 98ff491867..8902e30129 100644 --- a/tools/spl_size_limit.c +++ b/tools/spl_size_limit.c @@ -14,6 +14,9 @@ int main(int argc, char *argv[])
#ifdef CONFIG_SPL_SIZE_LIMIT spl_size_limit = CONFIG_SPL_SIZE_LIMIT; +#if defined(CONFIG_SECURE_BOOT) && defined(CONFIG_CSF_SIZE)
spl_size_limit -= CONFIG_CSF_SIZE;+#endif
But, if the target enable HAB on SPL the size would be part of SPL limit, isn't ?
Indeed - it is not clear to me, too, if it is correct, even if CSF is added later by the NXP signing tools. The patch reduces significantly the available space for SPL, I just wondering why just mamoj is affected. Jagan, does it work without this patch applied ?
When enabling CONFIG_SECURE_BOOT we increase the image length in boot data by the size defined in CONFIG_CSF_SIZE. The HAB code will parse the boot data structure and copy the image length defined (SPL image plus CSF appended) to its respective load address.
HAB code is checking if the image length defined can fit in OCRAM free region, and logs the following HAB event in case not:
--------- HAB Event 1 ----------------- event data: 0xdb 0x00 0x14 0x42 0x33 0x22 0x33 0x00 0x00 0x00 0x00 0x0f 0x00 0x90 0x70 0x00 0x00 0x01 0x10 0x00 STS = HAB_FAILURE (0x33) RSN = HAB_INV_ADDRESS (0x22) CTX = HAB_CTX_TARGET (0x33) ENG = HAB_ENG_ANY (0x00)
HAB closed targets would then fail to boot, so for that reason we added CONFIG_CSF_SIZE into consideration.
Clear - thanks for detailed explanation.
We can reduce the default CONFIG_CSF_SIZE but it depends on the user specific HAB setup. I did a quick test with RSA 4K keys and couldn't achieve 0x2000 length.
That is much less as we have now.
Do you think we should decrease default CONFIG_CSF_SIZE?
I think yes - if we set it for the worst case, we reduce the SPL size so much that most boards, if they enable SECURE_BOOT, won't build. I cannot say that imx6dl_mamoj has dead code in its SPL, it is one of the board with the "state of art" in U-Boot, with DM and OF in SPL. But this is also something we decided to push into U-Boot. Anyway, every board maintainer can change it and add it to the own defconfig.
Jagan, after setting CONFIG_CSF_SIZE to 0x2060 as suggested by Breno, board builds fine - but I have no idea if it can boots. Can you check this ?
Perhaps 0x2000 plus the maximum dek blob size (0x60) would be enough for most uses cases, users requiring more space can modify their CONFIG_CSF_SIZE.
Best regards, Stefano