On 21.03.2019 17:10, Horia Geanta wrote:
On 3/21/2019 2:42 PM, Tudor Laurentiu-B10716 wrote:
Hi Horia,
On 21.03.2019 12:36, Horia Geanta wrote:
On 3/20/2019 4:31 PM, laurentiu.tudor@nxp.com wrote:
#define SET_SEC_QI_ICID(streamid) \
- SET_ICID_ENTRY("fsl,sec-v4.0", streamid, \
- SET_ICID_ENTRY("fsl,sec-v4.0", SEC_ICID_REG_VAL(streamid), \
Is this a fix for LS104x? Then it should be a separate patch.
Not really. I added an intermediate macro, SEC_ICID_REG_VAL(streamid) that forms the correct register value starting from ICID, depending on the chip version (the register layouts are different between the two).
This is the define #define SEC_ICID_REG_VAL(streamid) (((streamid) << 16) | (streamid))
Thus the QI ICID for LS104x changes.
Ah, alright. Good point. I checked in the manuals and this unintended change does indeed fix this register setup. I'm leaning towards mentioning this collateral fix in the commit message.
+#else /* CONFIG_FSL_LSCH2 */
[...]
+#define SEC_ICID_REG_VAL(streamid) ((streamid) << 24)
ICID is in lower 6:0 bits, not in 31:24.
That was also my initial impression but it didn't work (smmu global faults with icid 0). Probably there's an ambiguity related to endianness in the documentation.
Note that on DPAA 2.x both the core and device (CAAM) are little endian. Probably the problem is with the I/O accessors writing the CAAM registers.
Instead of out_be32() use either sec_out32() (but this means splitting CAAM-specific code) or out_le32().
CAAM endianness can be detected using CONFIG_SYS_FSL_SEC_LE / CONFIG_SYS_FSL_SEC_BE.
Thanks for the explanation. I think I can change the macros to take into consideration the CONFIG_*s you pointed.
--- Best Regards, Laurentiu