Increase memzone name size
http://bugs.dpdk.org/show_bug.cgi?id=1984
Bug ID: 1984
Summary: Increase memzone name size
Product: DPDK
Version: 26.07
Hardware: All
OS: All
Status: UNCONFIRMED
Severity: enhancement
Priority: Normal
Component: core
Assignee: [email protected]
Reporter: [email protected]Target Milestone: ---
"32 bytes ought to be enough for anybody."
Said nobody.
Many object types have names, and at object creation, they create sub-objects, adding a prefix to the name of the sub-object. The sub-objects may create its own sub-objects (with another prefix added to the name) etc..
E.g. when creating a mempool named "example", the mempool lib creates a memzone named "MP_example", and the mempool ring driver (a sub-object of the mempool) creates a memzone named "RG_MP_example".
Assuming the FIB and RIB libs also added proper prefixes (which the currently don't), the names would go:
"example" (FIB name),
"FIB_example" (RIB name),
"RIB_FIB_example" (mempool name),
"MP_RIB_FIB_example" (memzone name created by the mempool), "RG_MP_RIB_FIB_example" (memzone name created by the ring driver).
One of the FIB6 test cases creates a FIB with the name "test_drift_compression" (22 chars). Adding prefixes, it should end up being "RG_MP_RIB_FIB_test_drift_compression" (36 chars).
I suggest increasing RTE_MEMZONE_NAMESIZE from 32 to 64. This seems simple on the surface, as it does no harm to the rte_memzone structure itself.
However, it has a ripple effect into other object types, whose name sizes are derived by this, e.g. the rte_ring and the rte_mempool structures. When the size of the name field in those structures is increased, other fields in the structures will move down, and may end up in a different cache line. So, all structures depending on RTE_MEMZONE_NAMESIZE must be carefully reviewed and adjusted, so their fields are grouped efficiently, considering cache lines.
Signature · 2 lines
You are receiving this mail because: You are the assignee for the bug.