It has a couple of meanings. The two usual ones are defined in the reference Marce100 provided although that reference's HOL blocking, isn't fully correct, such as its discussion of out-of-order packets.
Basically, a switch which does not have resources to support all its ports, concurrently, at full rate, will "block" some traffic. For example, the Cisco 3750G series had a 32 Gb fabric, but could provide 24 or 48 gig copper ports. For the fabric to not every block, it would need either 48 or 96 Gb fabric. (If you don't exceed 16 Gb, ingress, you wouldn't block, either. BTW, fabric bandwidths, at least by Cisco, are provided as full duplex bandwidths.) The (about the same time?) 4948, for comparison, with 48 gig copper ports, does have a 96 Gb fabric.
HOL blocking comes about because of the architecture of a switch. Without getting into switch architectures, a simple "real world" example might help explain the issue.
Ever been in a single line of traffic, where a vehicle at the "head of the line" is making a left, but is waiting on on-coming traffic? I.e. the road ahead is clear, but traffic is blocked. However, if the road had a left turn lane, other traffic would not be blocked. I.e. road architecture creates or mitigates this issue.
