-----Original Message-----
From: Bob Cochran [mailto:[email protected]]
Sent: Saturday, September 27, 2014 11:36 PM
To: Luo Zhenhua-B19537; [email protected]
Cc: Wood Scott-B07421; Wrobel Heinz-R39252
Subject: Re: [meta-freescale] Should all T4240 errata be applied to T1040?
On 09/19/2014 06:04 AM, [email protected] wrote:
Hi Bob,
Thanks for the great finding.
Defining the CONFIG_FMAN_T4240 for T1 is inaccurate and potentially
dangerous for the correct function. The manual explicitly states for TNTSKS that
exceeding the max supported values is not permitted, but it doesn't actually
state that the reset value is the maximum permissible value.
As for the correct way to define things, we probably need to share
integrations across common SoCs per, e.g., Table A-5 in the T1040 DPAA RM to
make SW match our docs properly. A configuration "T4240" is likely a bad one
anyway if we should name it effectively "FMAN_V3H_384KiB" vs.
"FMAN_V3H_512KiB" for B4860 rev 2. So it is good to change "FMAN_V3L" for
T1 to match the documentation and part behavior.
Also the correct way is to stop configuring such things at compile time to
avoid separate kernel build for every FMan revision.
Zhenhua,
I believe a similar issue exists with the dts files.
fsl/qoriq-fman3-0.dtsi defines two O/H ports that don't exist on the T1040.
After setting CONFIG_FMAN_V3L, FM_MAX_NUM_OF_OH_PORTS was
defined as 4 in T4240/dpaa_integration_ext.h. This matches what I see in the
T1040 SoC RM.
However, during boot, I see errors reported on the command line from the FM
module:
Freescale FM module (Sep 26 2014:14:25:29), FMD API version 21.1.0
cpu1/1: ! MAJOR FM Error [CPU01,
drivers/net/ethernet/freescale/fman/src/wrapper/lnxwrp_fm_port.c:272
ReadFmPortDevTreeNode]: Invalid Value;
cpu1/1: of_get_property(/soc@ffe000000/fman@400000/port@86000,
cell-index) failedcpu1/1:
fsl-fman-port: probe of ffe486000.port failed with error -5
cpu1/1: ! MAJOR FM Error [CPU01,
drivers/net/ethernet/freescale/fman/src/wrapper/lnxwrp_fm_port.c:272
ReadFmPortDevTreeNode]: Invalid Value;
cpu1/1: of_get_property(/soc@ffe000000/fman@400000/port@87000,
cell-index) failedcpu1/1:
If I remove the extra O/H ports in my dts file, the errors go away.
So, if we follow your suggestion, I believe we should add a qoriq-fman3l-0.dtsi
file that excludes the non-existent O/H ports.
However, I need some feedback because my T1040 DPAA Reference Manual
Table A-47 states that the number of O/H ports are SoC specific, so maybe
some V3L devices actually have these extra O/H ports???
And fsl/t1040si-post.dtsi also specifies the extra two O/H ports, but that's a
straight forward patch.
Finally, who will do these patches for proper V3H/V3L support (me or FSL)? If I
do it, I'm just going to work through the issues with the
T1040 / V3L (I can't address B4860 rev 2 - don't have the specs & don't want
them).
Thanks,
Bob
Best Regards,
Zhenhua
-----Original Message-----
From: [email protected] [mailto:meta-freescale-
[email protected]] On Behalf Of Bob Cochran
Sent: Thursday, September 18, 2014 11:44 AM
To: [email protected]
Subject: [meta-freescale] Should all T4240 errata be applied to T1040?
I find that during initialization of my T1040rdb-64b using the kernel
built with meta-fsl-ppc master, the FMBM_CFG2.TNTASKS field (total
number of BMI
tasks) is being overwritten to 0x7b (the reset value is documented as
0x3b in the T1040 reference manual).
I believe this is being overwritten due to sharing the errata
definitions with the T4240, but this particular erratum may not apply
to the T1040 and may cause undesirable side effects.
The T4240 uses DPAA FMAN_v3H, and the T1040 uses DPAA FMAN_v3L.
Here is my understanding of how this erratum (A005127) is being
applied to the
T1040:
1) t1040_64bit_smp_defconfig defines CONFIG_FMAN_T4240
2) By defining CONFIG_FMAN_T4240,
drivers/net/ethernet/freescale/fman/ncsw_config.mk adds
-I$(FMAN)/inc/integrations/T4240 to EXTRA_CFLAGS.
3) dpaa_integration_ext.h is included throughout the fman source
files from
drivers/net/ethernet/freescale/fman/inc/integrations/T4240
4) This defines FM_WRONG_RESET_VALUES_ERRATA_FMAN_A005127, but
it is
my understanding that this erratum does not apply to the T1040.
5) This results in FMBM_CFG2.TNTASKS being overwritten with a value
approximately twice its default value, and my concern is that the
FMAN_V3L does not have the resources to support this many tasks.
In general, I'm also wondering whether the other errata in the
integration file is appropriate for the T1040 and whether the
FMAN_v3L devices should have their own integration tree.
--
_______________________________________________
meta-freescale mailing list
[email protected]
https://lists.yoctoproject.org/listinfo/meta-freescale