slight additional remark below.
On Fri, Mar 10, 2017 at 06:11:59PM +0100, Gaëtan Rivet wrote:
Hey, thanks for reading.
On Fri, Mar 10, 2017 at 04:24:32PM +0000, Legacy, Allain wrote:
-----Original Message-----
From: dev [mailto:dev-boun...@dpdk.org] On Behalf Of Gaetan Rivet
Sent: Friday, March 03, 2017 10:40 AM
...
+ errno = 0;
+ tmp = strtoul(val, NULL, 0);
The robustness of the strtoul() could be improved with something like the
following to catch non-integer characters following the port number.
char *end = NULL;
tmp = strtoull(val, &end, 0);
if ((val[0] == '\0') || (end == NULL) || (*end != '\0') || (errno != 0))
Thanks for the suggestion, I'd keep the strtoul though ;).
I will see into it for the v2, keeping in mind Stephen's remarks as
well.
+ if (errno) {
+ WARN("%s: \"%s\" is not a valid integer", key, val);
+ return -errno;
+ }
+ if (strcmp(MLX4_PMD_PORT_KVARG, key) == 0) {
+ if (tmp >= MLX4_PMD_MAX_PHYS_PORTS) {
+ ERROR("invalid port index %lu (max: %u)",
+ tmp, MLX4_PMD_MAX_PHYS_PORTS - 1);
+ return -EINVAL;
+ }
+ conf->active_ports |= 1 << tmp;
+ } else {
+ WARN("%s: unknown parameter", key);
+ return -EINVAL;
+ }
+ return 0;
+}
The usage of strtoul() should be moved to be within the strcmp(MLX4_PMD_PORT_KVARG, key) IF statement. That way the
"val" would only be parsed if "key" is "port" and it is expected that "val" is
an integer.
This function was aimed at being generic.
If we consider that no other parameter would ever be added, then the
strcmp should be scraped altogether, as this callback is only called
upon parsing this parameter in the kvlist in the first place.
But we are in the control path, avoiding a useless strtoul at the
price of making the function less useful seems an unnecessary tradeoff
to me.
+ if (mlx4_args(pci_dev->device.devargs, &conf)) {
+ ERROR("failed to process device arguments");
+ goto error;
+ }
It would be helpful for debugging if the error message included the devargs so
that we can see what is wrong with the input.
Agreed.
Actually, on second thought, here the devargs that was problematic has
already been shown with a warning while it was being parsed.
+ /* Use all ports when none are defined */
+ if (conf.active_ports == 0) {
+ for (i = 0; i < MLX4_PMD_MAX_PHYS_PORTS; i++)
+ conf.active_ports |= 1 << i;
+ }
Rather than use a loop to populate all active fields would a #define with an
all ports mask be better suited to this. Or alternatively just change the IF
statement below to use the following and avoid the need for this loop
altogether:
if (conf.active_ports & !(conf.active_ports & (1 << i)))
continue;
I do not agree with removing this loop.
Your second solution will scatter the relevant bits concerning the
default value of the active_port configuration option. While being
slightly slicker it hides it unnecessarily from the reader.
The first solution might be interesting, however it makes this option
dependent on two defines instead of one. If one had to change the
default MAX_PHYS_PORT value for mlx4 (however unlikely it might be),
then they would have to change the valid ALL_PORTS mask as well. In
principle this contradicts DRY[1].
[1]: https://en.wikipedia.org/wiki/Don't_repeat_yourself
--
Gaëtan Rivet
6WIND
--
Gaëtan Rivet
6WIND
