On Tue, Sep 16, 2025 at 4:19 AM Paolo Abeni [off-list ref] wrote:

On 9/15/25 5:04 AM, Michael Chan wrote:

quoted

Optimize the loop inside this function that sends the FW message
to configure backing store memory for each instance of a memory
type.  It uses 2 loop counters i and j, but both counters advance
at the same time so we can eliminate one of them.

The above statement does not look correct.

quoted

@@ -9128,20 +9128,20 @@ static int bnxt_hwrm_func_backing_store_cfg_v2(struct bnxt *bp,
      req->subtype_valid_cnt = ctxm->split_entry_cnt;
      for (i = 0, p = &req->split_entry_0; i < ctxm->split_entry_cnt; i++)
              p[i] = cpu_to_le32(ctxm->split[i]);
-     for (i = 0, j = 0; j < n && !rc; i++) {
+     for (i = 0; i < n && !rc; i++) {
              struct bnxt_ctx_pg_info *ctx_pg;

              if (!(instance_bmap & (1 << i)))
                      continue;
              req->instance = cpu_to_le16(i);
-             ctx_pg = &ctxm->pg_info[j++];
+             ctx_pg = &ctxm->pg_info[i];

`j` is incremented only for bit set in `instance_bmap`, AFAICS this does
not introduces functional changes only if `instance_bmap` has all the
bit set.

Yes, you are absolutely right.  The original code will skip over zeros
in the instance_bmap without incrementing j.  Let me rethink this
patch or drop this patch.  Thanks for the review.