On Mon, Oct 28, 2019 at 03:37:48PM -0400, Jeff Moyer wrote:
> Ira Weiny <[email protected]> writes:
>
> >> (Watching the unit test run fall into an infinite loop..) Nope, the
> >> break is in the switch scope, the while loop needs the 'goto out'.
> >>
> >> Yes this bit definitely needs to be refactored :)
> >
> > How about this patch instead? Untested.
>
> I'm not a fan of the looping with gotos.
Me either... But... the logic here is not the same.
>
> I think separating out the
> waiting for busy to its own function would make this more clear.
> Looking more closely, there are other issues. The timeout code looks at
> the seconds, but ignores the fractions, so you could be off by almost an
> entire second, there.
For this operation that is probably not a big deal. We should be waiting much
longer than the operation should take anyway.
>
> It also doens't retry the sleep if interrupted.
This could be an issue.
> Finally, I find the variables names to be highly confusing.
>
> I've decided not to fix those last two bugs just yet, but here's a patch
> that shows the dirction I think it should go. Compile-tested only for
> now. Let me know what you think.
I thought about doing something similar but to make the logic the same it
becomes a bit awkward.
>
> Ira, I used the same base as you. If you updated ndctl, you'll have to
> revert 9e0391e057b36 to apply this patch.
>
> Cheers,
> Jeff
>
> diff --git a/ndctl/dimm.c b/ndctl/dimm.c
> index c8821d6..701f58b 100644
> --- a/ndctl/dimm.c
> +++ b/ndctl/dimm.c
> @@ -674,6 +674,41 @@ out:
> return rc;
> }
>
> +static void wait_for_cmd_completion(struct ndctl_cmd *cmd, struct fw_info
> *fw,
> + struct timespec *start)
> +{
> + enum ND_FW_STATUS status;
> + struct timespec sleeptime, now;
> + int rc;
> +
> + sleeptime.tv_nsec = fw->query_interval / 1000;
> + sleeptime.tv_sec = 0;
> +
> + while ((status = ndctl_cmd_fw_xlat_firmware_status(cmd)) == FW_EBUSY) {
> +
> + rc = clock_gettime(CLOCK_MONOTONIC, &now);
> + if (rc < 0)
> + break;
> +
> + /*
> + * If we expire max query time, we timed out
> + */
> + if (now.tv_sec - start->tv_sec > fw->max_query / 1000000)
> + break;
> +
> + /*
> + * Sleep the interval dictated by firmware before
> + * query again.
> + */
> + rc = nanosleep(&sleeptime, NULL);
> + if (rc < 0)
> + break;
You need ndctl_cmd_submit() here to be the same logic.
> +
> + }
> +
> + return;
> +}
> +
> static int query_fw_finish_status(struct ndctl_dimm *dimm,
> struct action_context *actx)
> {
> @@ -682,98 +717,65 @@ static int query_fw_finish_status(struct ndctl_dimm
> *dimm,
> struct ndctl_cmd *cmd;
> int rc;
> enum ND_FW_STATUS status;
> - bool done = false;
> - struct timespec now, before, after;
> + struct timespec start;
> uint64_t ver;
>
> cmd = ndctl_dimm_cmd_new_fw_finish_query(uctx->start);
> if (!cmd)
> return -ENXIO;
>
> - rc = clock_gettime(CLOCK_MONOTONIC, &before);
> + rc = clock_gettime(CLOCK_MONOTONIC, &start);
> if (rc < 0)
> - goto out;
> -
> - now.tv_nsec = fw->query_interval / 1000;
> - now.tv_sec = 0;
> -
> - do {
> - rc = ndctl_cmd_submit(cmd);
> - if (rc < 0)
> - break;
> + goto unref;
>
> - status = ndctl_cmd_fw_xlat_firmware_status(cmd);
> - switch (status) {
> - case FW_SUCCESS:
> - ver = ndctl_cmd_fw_fquery_get_fw_rev(cmd);
> - if (ver == 0) {
> - fprintf(stderr, "No firmware updated.\n");
> - rc = -ENXIO;
> - goto out;
> - }
> -
> - printf("Image updated successfully to DIMM %s.\n",
> - ndctl_dimm_get_devname(dimm));
> - printf("Firmware version %#lx.\n", ver);
> - printf("Cold reboot to activate.\n");
> - done = true;
> - rc = 0;
> - break;
> - case FW_EBUSY:
> - /* Still on going, continue */
> - rc = clock_gettime(CLOCK_MONOTONIC, &after);
> - if (rc < 0) {
> - rc = -errno;
> - goto out;
> - }
> + rc = ndctl_cmd_submit(cmd);
> + if (rc < 0)
> + goto unref;
>
> - /*
> - * If we expire max query time,
> - * we timed out
> - */
> - if (after.tv_sec - before.tv_sec >
> - fw->max_query / 1000000) {
> - rc = -ETIMEDOUT;
> - goto out;
> - }
> + wait_for_cmd_completion(cmd, fw, &start);
wait_for_cmd_completion() does not call ndctl_cmd_submit()
Now I find it odd that we need to resubmit the command but I assume the logic
is correct. Therefore we need to go back and call ndctl_cmd_submit() again.
Or is this not required?
anyway that is why I went ahead and used the goto...
Ira
_______________________________________________
Linux-nvdimm mailing list -- [email protected]
To unsubscribe send an email to [email protected]
