Re: [PATCH v5 03/10] x86/resctrl: Parse ACPI ERDT table and save CACD cpumask for RMDD domains
From: Chen, Yu C
Date: Tue Jul 14 2026 - 11:28:01 EST
Hi Chenyu,
On 7/11/2026 7:42 AM, Reinette Chatre wrote:
Hi Chenyu,
On 7/1/26 6:45 AM, Chen Yu wrote:
From: Anil S Keshavamurthy <anil.s.keshavamurthy@xxxxxxxxx>
Parse the ERDT (Enhanced RDT) ACPI table so enhanced RDT features can
consume firmware-provided domain information.
The ERDT may contain these sub-tables:
- Resource Management Domain Description Structure (RMDD)
- CPU Agent Collection Description Structure (CACD)
- Cache Monitoring Registers for CPU Agents Description Structure
(CMRC)
How should "The ERDT may contain these sub-tables" be interpreted here?
This sounds like some high level partial description of ERDT that is
not specific to this patch.
I planned to add some background context on ERDT. Let me remove unrelated
content and change the sentence to:
Parse the RMDD subtables within the ERDT ACPI table and their nested
CACD entries to construct per-domain CPU masks.
There is one ERDT per platform. Each RMDD describes one resource
"one ERDT" -> "one ERDT table"?
OK.
diff --git a/arch/x86/kernel/cpu/resctrl/erdt.c b/arch/x86/kernel/cpu/resctrl/erdt.c
new file mode 100644
index 000000000000..6405df9be817
--- /dev/null
+++ b/arch/x86/kernel/cpu/resctrl/erdt.c
@@ -0,0 +1,271 @@
+// SPDX-License-Identifier: GPL-2.0-only
+/*
+ * Enhanced Resource Director Technology (ERDT)
+ *
+ * Copyright (C) 2026 Intel Corporation
+ *
+ */
+
+#define pr_fmt(fmt) "resctrl: " fmt
+
+#include <linux/acpi.h>
+#include <linux/cleanup.h>
+#include <linux/cpu.h>
Which parts of cpu.h are used?
It was used in previous version but not needed in this version.
Additionally, cleanup.h and err.h can be removed as well, I will
drop them.
+#include <linux/err.h>
+#include <linux/overflow.h>
+#include <linux/resctrl.h>
+#include <linux/sizes.h>
+#include <linux/xarray.h>
xarray is no longer needed?
Not needed, will remove it.
+
+#include <asm/apic.h>
+
+#include "internal.h"
+
+static LIST_HEAD(domain_info_list);
+
+static bool __erdt_enabled;
Is double underscore needed?
The double underscore was used to prevent name collisions with the
existing erdt_enabled(). Let me switch it to a single underscore instead.
Could you please add a comment above the variable to describe what it means when
"erdt is enabled"?
OK, let me add this:
/*
* Set when the ERDT ACPI table has been successfully parsed and at least
* one valid RMDD domain with a CACD cpumask exists. Cleared on teardown.
*/
+
+#define ERDT_VALID_VERSION 1
+#define RMDD_FLAG_CPU_L3_DOMAIN BIT(0)
+
+/* Bitmask of valid sub-tables found in the first RMDD, used to ensure all RMDDs match. */
+static u32 valid_subtbl_mask;
+
+int erdt_get_max_rmid(int cpu)
+{
+ struct erdt_domain_info *d;
+ struct list_head *pos;
+
+ if (!__erdt_enabled)
+ return 0;
+
+ list_for_each(pos, &domain_info_list) {
+ d = container_of(pos, struct erdt_domain_info, list);
(list_for_each_entry()?)
OK, will do.
+
+ if (cpumask_test_cpu(cpu, d->cpu_mask))
+ return d->max_rmid;
+ }
+
+ return -1;
+}
Using a CPU as parameter to determine the maximum RMID supported by ERDT is
unexpected. Looking ahead at how this function is used I find only one usage:
rdt_get_l3_mon_config() {
...
/*
* Currently assume all CPU domains share the same maximum RMID
* value from the RMDD table, use CPU0 domain's value.
*/
int erdt_max_rmid = erdt_get_max_rmid(0);
From this usage I do not see any reason why to do any CPU matching ... erdt_get_max_rmid()
could just return the max_rmid of any domain ... but that does not look right either since
the comment states an assumption that is never enforced in the code.
Could this be made more robust by replacing the "per-erdt-domain max_rmid" with one global
"ERDT max RMID" to which all RMDD's max RMID is compared to *ensure* they are all the same.
If there is no such guarantee then I expect this will be the minimum among all RMDD? All seems
to point to there only being one global value that is determined during ERDT enumeration and
then this helper can just return that value directly?
Got it, let me switch to one global max_rmid = min(all CPU domains's non-zero max_rmid ).
IO-domain RMID checks can be deferred for future support.
+
+static void __iomem *erdt_ioremap(phys_addr_t base, u32 num_pages, const char *desc)
+{
+ void __iomem *addr;
+ size_t size;
+
+ if (check_mul_overflow((size_t)num_pages, (size_t)SZ_4K, &size))
I do not think check_mul_overflow() requires size_t type, does it?
No need to stick to the size_t convention - there is no risk of overflow,
I’ll remove it.
+ return NULL;
+
+ addr = ioremap(base, size);
+ if (!addr) {
+ pr_err("ERDT: Failed to map %s at phys addr %pa (size: %u pages)\n",
+ desc, &base, num_pages);
+ }
(unnecessary braces)
Will remove it.
+ return addr;
+}
+
+static void erdt_iounmap_domain(struct erdt_domain_info *domain)
+{
+ for (int i = 0; i < ERDT_MMIO_NUM_TYPES; i++) {
+ if (domain->base[i]) {
+ iounmap(domain->base[i]);
+ domain->base[i] = NULL;
+ }
+ }
+}
+
+static void cleanup_one_domain(struct erdt_domain_info *d)
+{
+ erdt_iounmap_domain(d);
+ free_cpumask_var(d->cpu_mask);
free_cpumask_var() does not look right ... it is for stack usage, no? (more later)
If I understand correctly, cpumask_var_t can be used for both stack contexts and
dynamic allocation? Depending on whether OFFSTACK is defined, it may either be a
dynamically allocated pointer or an inline single-element struct cpumask array.
That said, since erdt_domain_info itself is dynamically allocated, embedding cpu_mask
inside erdt_domain_info would be clearer. Let me change it.
+static __init bool parse_rmdd_entry(struct acpi_subtbl_hdr_16 *rmdd_hdr)
nit: what is motivation for the "entry" term? this is only occurance of the
word "entry" in this patch while RMDD is more frequently referred to as "table" or
"sub-table".
I previously considered the RMDD table an entry within the parent ERDT table.
I’ll change the function name to parse_rmdd_table().
+{
+ struct erdt_domain_info *domain_info;
+ struct acpi_subtbl_hdr_16 *subtbl;
+ struct acpi_erdt_rmdd *rmdd;
+ u32 subtbl_mask = 0;
+
+ if (rmdd_hdr->length < sizeof(*rmdd)) {
+ pr_warn(FW_BUG "Invalid RMDD length %u\n", rmdd_hdr->length);
Please include unit, similar to equivalent ERDT message.
OK, will do.
+ return false;
+ }
+
+ rmdd = (struct acpi_erdt_rmdd *)rmdd_hdr;
+
+ /* Quietly ignore non-CPU-based L3 domains */
+ if (!(rmdd->flags & RMDD_FLAG_CPU_L3_DOMAIN))
+ return true;
+
+ domain_info = kzalloc_obj(*domain_info, GFP_KERNEL);
+ if (!domain_info)
+ return false;
+
+ if (!zalloc_cpumask_var(&domain_info->cpu_mask, GFP_KERNEL))
+ goto cleanup;
Similar to free_cpumask_var() this does not look right since the cpu_mask is not on stack here ...
(more later)
Ok, let me switch to struct cpumask instead.
+
+ domain_info->base[ERDT_MMIO_RMDD_CREG] =
+ erdt_ioremap(rmdd->creg_base, rmdd->creg_size, "RMDD ctrl base");
+ if (!domain_info->base[ERDT_MMIO_RMDD_CREG])
+ goto cleanup;
+
+ for (subtbl = rmdd_subtbl(rmdd);
+ subtbl_valid((void *)rmdd + rmdd->header.length, subtbl);
+ subtbl = next_subtbl(subtbl)) {
+ switch (subtbl->type) {
+ case ACPI_ERDT_TYPE_CACD:
It is quite subtle that there could be multiple CACD tables. Could this be highlighted with
a small comment? Something like:
/* An RMDD table has one or more CACD sub-table(s) */
OK, will do.
+ if (cacd_init(subtbl, domain_info))
+ goto cleanup;
+
+ subtbl_mask |= BIT(ACPI_ERDT_TYPE_CACD);
+ break;
+ default:
+ break;
+ }
+ }
+
+ if (!subtbl_mask)
+ goto cleanup;
+
+ /*
+ * Require all RMDDs to support same set of sub-tables
+ */
+ if (!valid_subtbl_mask) {
+ valid_subtbl_mask = subtbl_mask;
+ } else if (subtbl_mask != valid_subtbl_mask) {
+ pr_warn(FW_BUG "RMDD sub-table set does not match the first RMDD\n");
Would it be useful to print the domain ID to help diagnistics?
OK, let print the ID of the first RMDD and the "broken" RMDD.
+ goto cleanup;
+ }
+
+ if (!rmdd->max_rmid || rmdd->max_rmid > INT_MAX) {
rmdd->max_rmid is a u32 so INT_MAX test is not clear to me here
I overlooked this, let me remove the INT_MAX check.
+ pr_warn(FW_BUG "Unreasonable RMDD max_rmid %u\n", rmdd->max_rmid);
Is there a limit in the spec?
The spec defines it as a 4-byte value, so a u32 here should not overflow.
+ goto cleanup;
+ }
+ domain_info->max_rmid = rmdd->max_rmid;
As mentioned before, instead of a per-domain RMID, could there be a global ERDT
RMID that is initialized by first ERDT domain and updated/compared with every following
ERDT domain?
OK, will do.
+
+void erdt_exit(void)
+{
+ struct erdt_domain_info *d;
+ struct list_head *pos, *n;
+
+ list_for_each_safe(pos, n, &domain_info_list) {
list_for_each_entry_safe()?
OK.
+ d = container_of(pos, struct erdt_domain_info, list);
+ list_del(pos);
+ cleanup_one_domain(d);
+ }
+ __erdt_enabled = false;
+ valid_subtbl_mask = 0;
+}
+
+static __init int enumerate_erdt_table(struct acpi_table_header *table_hdr)
+{
+ struct acpi_table_erdt *erdt = (struct acpi_table_erdt *)table_hdr;
+ struct acpi_subtbl_hdr_16 *subtbl;
+ void *table_end;
+
+ if (erdt->header.revision != ERDT_VALID_VERSION) {
+ pr_info("Unsupported ERDT table revision %d\n", erdt->header.revision);
Would it be helpful to print what the ERDT table revision is to help diagnostics?
Yes, it will print the revision, do you mean print the expected revision?
pr_info("Unsupported ERDT table revision %d (expected %d)\n",
erdt->header.revision, ERDT_VALID_VERSION);
Please add documentation here that describes each member of struct erdt_domain_info.
OK, will do.
+struct erdt_domain_info {
+ void __iomem *base[ERDT_MMIO_NUM_TYPES];
+ cpumask_var_t cpu_mask;
Should this be struct cpumask instead? Compare with struct rdt_domain_hdr.
When evaluating cpumask_var_t, please read the detailed comments above its
definition in include/linux/cpumask_types.h - specifically note the start:
*cpumask_var_t: struct cpumask for stack usage*
Yes, let me switch to struct cpumask instead.
+ int max_rmid;
+ struct list_head list;
Could you please rename this to be "node" or "entry" to make it obvious that this
is an entry of a list? This is not consistent in resctrl but really helps
when reading the code.
OK, let me rename it to "entry".
thanks,
Chenyu