Hi Babu,
On 8/5/25 4:30 PM, Babu Moger wrote:
> "io_alloc" feature enables direct insertion of data from I/O devices into
> the cache. By directly caching data from I/O devices rather than first
> storing the I/O data in DRAM, it reduces the demands on DRAM bandwidth and
> reduces latency to the processor consuming the I/O data.
>
> When CDP is enabled, io_alloc routes traffic using the highest CLOSID
> associated with the L3CODE resource. To ensure consistent cache allocation
> behavior, the L3CODE and L3DATA resources must remain synchronized.
> rdtgroup_init_cat() function takes both L3CODE and L3DATA into account when
> initializing CBMs for new groups. The io_alloc feature must maintain the
> same behavior, ensuring that the Cache Bit Masks (CBMs) for both L3CODE and
> L3DATA are updated together.
Please rework this copy&pasted text also and make specific to what this
patch actually does.
> Enable users to modify io_alloc CBMs (Capacity Bit Masks) via the
> io_alloc_cbm resctrl file when io_alloc is enabled.
Here the changelog can provide overview of what is done by this patch
when a user provides a new CBM. This can include that a CBM provided
to CDP enabled resource will copy the CBM to the CDP peer.
>
> Signed-off-by: Babu Moger <babu.moger@xxxxxxx>
> ---
> ---
> Documentation/filesystems/resctrl.rst | 8 +++
> fs/resctrl/ctrlmondata.c | 97 +++++++++++++++++++++++++++
> fs/resctrl/internal.h | 3 +
> fs/resctrl/rdtgroup.c | 3 +-
> 4 files changed, 110 insertions(+), 1 deletion(-)
>
> diff --git a/Documentation/filesystems/resctrl.rst b/Documentation/filesystems/resctrl.rst
> index 3002f7fdb2fe..d955e8525af0 100644
> --- a/Documentation/filesystems/resctrl.rst
> +++ b/Documentation/filesystems/resctrl.rst
> @@ -187,6 +187,14 @@ related to allocation:
> # cat /sys/fs/resctrl/info/L3/io_alloc_cbm
> 0=ffff;1=ffff
>
> + CBMs can be configured by writing to the interface.
> +
> + Example::
> +
> + # echo 1=FF > /sys/fs/resctrl/info/L3/io_alloc_cbm
It may be useful to demonstrate syntax when more than one CBM is modified.
> + # cat /sys/fs/resctrl/info/L3/io_alloc_cbm
> + 0=ffff;1=00ff
> +
> When CDP is enabled "io_alloc_cbm" associated with the DATA and CODE
> resources may reflect the same values. For example, values read from and
> written to /sys/fs/resctrl/info/L3DATA/io_alloc_cbm may be reflected by
> diff --git a/fs/resctrl/ctrlmondata.c b/fs/resctrl/ctrlmondata.c
> index 641094aac322..1f69117f96f8 100644
> --- a/fs/resctrl/ctrlmondata.c
> +++ b/fs/resctrl/ctrlmondata.c
> @@ -858,3 +858,100 @@ int resctrl_io_alloc_cbm_show(struct kernfs_open_file *of, struct seq_file *seq,
> cpus_read_unlock();
> return ret;
> }
> +
> +static int resctrl_io_alloc_parse_line(char *line, struct rdt_resource *r,
> + struct resctrl_schema *s, u32 closid)
> +{
> + enum resctrl_conf_type peer_type;
> + struct resctrl_schema *peer_s;
> + struct rdt_parse_data data;
> + struct rdt_ctrl_domain *d;
> + char *dom = NULL, *id;
> + unsigned long dom_id;
> +
> +next:
> + if (!line || line[0] == '\0')
> + return 0;
> +
> + dom = strsep(&line, ";");
> + id = strsep(&dom, "=");
> + if (!dom || kstrtoul(id, 10, &dom_id)) {
> + rdt_last_cmd_puts("Missing '=' or non-numeric domain\n");
> + return -EINVAL;
> + }
> +
> + dom = strim(dom);
> + list_for_each_entry(d, &r->ctrl_domains, hdr.list) {
> + if (d->hdr.id == dom_id) {
> + data.buf = dom;
> + data.mode = RDT_MODE_SHAREABLE;
> + data.closid = closid;
> + if (parse_cbm(&data, s, d))
> + return -EINVAL;
> + /*
> + * When CDP is enabled, update the schema for both CDP_DATA
> + * and CDP_CODE.
> + */
> + if (resctrl_arch_get_cdp_enabled(r->rid)) {
> + peer_type = resctrl_peer_type(s->conf_type);
> + peer_s = resctrl_get_schema(peer_type);
> + if (parse_cbm(&data, peer_s, d))
> + return -EINVAL;
CBM is still parsed twice. As I mentioned in v7 the parsing only need to be done
once and then the resulting CBM can be copied to CDP peer.
https://lore.kernel.org/lkml/82045638-2b26-4682-9374-1c3e400a580a@xxxxxxxxx/
> + }
> + goto next;
> + }
> + }
> +
> + return -EINVAL;
> +}
> +
> +ssize_t resctrl_io_alloc_cbm_write(struct kernfs_open_file *of, char *buf,
> + size_t nbytes, loff_t off)
> +{
> + struct resctrl_schema *s = rdt_kn_parent_priv(of->kn);
> + struct rdt_resource *r = s->res;
> + u32 io_alloc_closid;
> + int ret = 0;
> +
> + /* Valid input requires a trailing newline */
> + if (nbytes == 0 || buf[nbytes - 1] != '\n')
> + return -EINVAL;
> +
> + buf[nbytes - 1] = '\0';
> +
> + cpus_read_lock();
> + mutex_lock(&rdtgroup_mutex);
> +
> + rdt_last_cmd_clear();
> +
> + if (!r->cache.io_alloc_capable) {
> + rdt_last_cmd_printf("io_alloc is not supported on %s\n", s->name);
> + ret = -ENODEV;
> + goto out_unlock;
> + }
> +
> + rdt_last_cmd_clear();
Unnecessary rdt_last_cmd_clear().
> + rdt_staged_configs_clear();
Placement of this can be improved by putting it closer to the code that
populates the staged configs. That is, just before resctrl_io_alloc_parse_line().
The flow is not symmetrical in that the out_unlock exit code always
clears the buffer whether it was used or not. I think it will be easier to
understand if the out_unlock *only* unlocks the locks and there is a new
goto label, for example, "out_clear_configs" that calls rdt_staged_configs_clear()
and is used after resctrl_io_alloc_parse_line().
> +
> + if (!resctrl_arch_get_io_alloc_enabled(r)) {
> + rdt_last_cmd_printf("io_alloc is not enabled on %s\n", s->name);
> + ret = -ENODEV;
> + goto out_unlock;
> + }
> +
> + io_alloc_closid = resctrl_io_alloc_closid(r);
> +
> + ret = resctrl_io_alloc_parse_line(buf, r, s, io_alloc_closid);
> +
> + if (ret)
> + goto out_unlock;
> +
> + ret = resctrl_arch_update_domains(r, io_alloc_closid);
> +
> +out_unlock:
> + rdt_staged_configs_clear();
> + mutex_unlock(&rdtgroup_mutex);
> + cpus_read_unlock();
> +
> + return ret ?: nbytes;
> +}
Reinette
