How many write pointers are there in zns mode?

[yaOoo830]

Hi,i have some questions about zns mode.
I noticed in zns.c, every zone (structure NvmeZone) has its own writepointer: zone->w_ptr

FEMU/hw/femu/zns/zns.c

Line 834 in 1c2d358

static uint16_t zns_nvme_rw(FemuCtrl *n, NvmeNamespace *ns, NvmeCmd *cmd,

but there is no delay emulation in any operation.
in zftl.c,there is another function of advancing wp :zns_advance_write_pointer, and delay emulation is in zftl.c,

FEMU/hw/femu/zns/zftl.c

Line 54 in 1c2d358

static void zns_advance_write_pointer(struct zns_ssd *zns)

but it seems like there is only one wp of whole ssd.
it seems like the granularity of mapping table is page, but one of the contributions of zns is its coarse-grained mapping table,is this reasonable?

FEMU/hw/femu/zns/zftl.c

Line 7 in 1c2d358

static inline struct ppa get_maptbl_ent(struct zns_ssd *zns, uint64_t lpn)

In summary,my questions are as follows:
1.why not emulate delay in zns.c but in zftl.c?
2. why there is only one wp in zftl.c?
3. is page-level mapping table reasonable?
Looking forward to your reply,thanks! @huaicheng

[huaicheng]

1.why not emulate delay in zns.c but in zftl.c?

This is an implementation choice, both would work. Putting it in zftl is mainly for modular design.

2. why there is only one wp in zftl.c?

The WP used by FEMU is a superblock data structure for tracking the next available block for use, it's different from the write pointers per zone in ZNS spec.

3. is page-level mapping table reasonable?

FEMU only provides a page mapping as a baseline implementation, it serves as a platform where you can implement customized management policies.