Hi Michal,
On 20 April 2016 at 22:47, Michal Simek michal.simek@xilinx.com wrote:
Hi Simon,
On 20.4.2016 16:41, Simon Glass wrote:
Hi Michal,
On 14 April 2016 at 07:02, Michal Simek michal.simek@xilinx.com wrote:
Fix fdtdec_get_alias_seq() which is not checking full path to certain node and it incorrectly provides incorrect seq number. Checking full path ensure that if alias is present correct seq number is return. This problem was found on ZynqMP zcu102 where are several I2C muxes where the first bus name is i2c@0 and for following muxes incorrect seq numbers are return.
Signed-off-by: Michal Simek michal.simek@xilinx.com
I am not sure if there is any macro for max alias length which I can use instead of new macro.
Before: ZynqMP> i2c bus Bus 0: i2c@ff020000 20: gpio@20, offset len 1, flags 0 21: gpio@21, offset len 1, flags 0 75: i2cswitch@75, offset len 1, flags 0 Bus 750: i2c@0 Bus 751: i2c@1 Bus 752: i2c@2 Bus 1: i2c@ff030000 74: i2cswitch@74, offset len 1, flags 0 75: i2cswitch@75, offset len 1, flags 0 Bus 750: i2c@0
Sorry I'm a bit confused by this. Do you have two i2c@0 buses?
Yes there are two i2c IP cores. It means 2 main i2c busses.
Here is full i2c description arch/arm/dts/zynqmp-zcu102.dts
I have pushed my hacky/testing tree here with these changes. u-boot-microblaze.git xnext/i2c
Bus 751: i2c@1 Bus 752: i2c@2 Bus 1743: i2c@3 Bus 1744: i2c@4 Bus 750: i2c@0 Bus 751: i2c@1 Bus 752: i2c@2 Bus 1743: i2c@3 Bus 1744: i2c@4 Bus 1755: i2c@5 Bus 1756: i2c@6 Bus 1757: i2c@7
After: ZynqMP> i2c bus Bus 0: i2c@ff020000 20: gpio@20, offset len 1, flags 0 21: gpio@21, offset len 1, flags 0 75: i2cswitch@75, offset len 1, flags 0 Bus 750: i2c@0 Bus 751: i2c@1 Bus 752: i2c@2 Bus 1: i2c@ff030000 74: i2cswitch@74, offset len 1, flags 0 75: i2cswitch@75, offset len 1, flags 0 Bus 1740: i2c@0 Bus 1741: i2c@1 Bus 1742: i2c@2 Bus 1743: i2c@3 Bus 1744: i2c@4 Bus 1750: i2c@0 Bus 1751: i2c@1 Bus 1752: i2c@2 Bus 1753: i2c@3 Bus 1754: i2c@4 Bus 1755: i2c@5 Bus 1756: i2c@6 Bus 1757: i2c@7
Which reflects my aliases i2c0 = &i2c0; i2c750 = &i2c0_75_0; i2c751 = &i2c0_75_1; i2c752 = &i2c0_75_2; i2c1 = &i2c1; i2c1740 = &i2c1_74_0; i2c1741 = &i2c1_74_1; i2c1742 = &i2c1_74_2; i2c1743 = &i2c1_74_3; i2c1744 = &i2c1_74_4; i2c1750 = &i2c1_75_0; i2c1751 = &i2c1_75_1; i2c1752 = &i2c1_75_2; i2c1753 = &i2c1_75_3; i2c1754 = &i2c1_75_4; i2c1755 = &i2c1_75_5; i2c1756 = &i2c1_75_6; i2c1757 = &i2c1_75_7;
I only see one in you aliases above.
i2c0 and i2c1 are that main one.
lib/fdtdec.c | 15 ++++++++++++++- 1 file changed, 14 insertions(+), 1 deletion(-)
diff --git a/lib/fdtdec.c b/lib/fdtdec.c index 70acc29c924d..79efcf0cd6b0 100644 --- a/lib/fdtdec.c +++ b/lib/fdtdec.c @@ -506,6 +506,8 @@ int fdtdec_add_aliases_for_id(const void *blob, const char *name, return num_found; }
+#define FDT_ALIAS_PATH_LEN 64
int fdtdec_get_alias_seq(const void *blob, const char *base, int offset, int *seqp) { @@ -514,9 +516,13 @@ int fdtdec_get_alias_seq(const void *blob, const char *base, int offset, int find_namelen; int prop_offset; int aliases;
char buf[FDT_ALIAS_PATH_LEN];fdt_get_path(blob, offset, buf, sizeof(buf));Eek! That is pretty expensive.
How to do it better?
This is so confusing. It would help to have the 'dm tree' output.
From what I can see i2c1_74_1 is the problem. It is supposed to get a
sequence number of 1741 but is detected as 1, since the understore before '1' stops trailing_strtol() from moving left to see additional digits.
So when the algorithm looks at the i2c1_74_1 alias, and sees that it points to /path/to/i2c1_74_1, it decides that the sequence number for this is 1, and returns that.
This is not what you want, since you already have another i2c1 alias, which needs to get that sequence number.
Your solution relies on skipping that answer and finding another one. But for the skipping the function would still returns 1 in *seqp. Wouldn't it be better to adjust the function to detect 1741 in this case? This would involve adding something like trailing_strto_skip_underscoresl() to skip underscores.
Whatever we do here, we need a unit test to cover the cases (could be added to test/dm/core.c with a few strange i2c nodes added to test.dts), and the header file comment needs to be expanded a little.
BTW at some point I think we will want to introduce a little more device tree support in driver model. For example, we might want to add a table of aliases built up in dm_scan_fdt_node() or similar. We might want a table of phandles. I haven't looked into it in detail, but this fdtdec function predates driver model and I suspect we can create something more efficient now.
find_name = fdt_get_name(blob, offset, &find_namelen);
debug("Looking for '%s' at %d, name %s\n", base, offset, find_name);
debug("Looking for '%s' at %d, name %s, buf %s\n",base, offset, find_name, buf); aliases = fdt_path_offset(blob, "/aliases"); for (prop_offset = fdt_first_property_offset(blob, aliases);@@ -533,6 +539,13 @@ int fdtdec_get_alias_seq(const void *blob, const char *base, int offset, strncmp(name, base, base_len)) continue;
if (len >= FDT_ALIAS_PATH_LEN)printf("Too long path in aliases node\n");Can you return an error and make it debug() here?
Error code?
/* Check full path first not to point to incorrect alias */if (strncmp(prop, buf, min(len, FDT_ALIAS_PATH_LEN)))continue;slash = strrchr(prop, '/'); if (strcmp(slash + 1, find_name)) continue;-- 1.9.1
Thanks, Michal
Regards, Simon