diff --git a/docs/man/xl.cfg.5.pod.in b/docs/man/xl.cfg.5.pod.in
index b98d161398..eda1e77ebd 100644
--- a/docs/man/xl.cfg.5.pod.in
+++ b/docs/man/xl.cfg.5.pod.in
@@ -1768,6 +1768,11 @@ are not using hardware assisted paging (i.e. you are using shadow
 mode) and your guest workload consists of a very large number of
 similar processes then increasing this value may improve performance.
 
+On Arm, this field is used to determine the size of the guest P2M pages
+pool, and the default value is 1MB per vCPU plus 4KB per MB of RAM for
+the P2M map and additional 512KB for extended regions. Users should
+adjust this value if bigger P2M pool size is needed.
+
 =back
 
 =head3 Processor and Platform Features
diff --git a/tools/libs/light/libxl_arch.h b/tools/libs/light/libxl_arch.h
index 1522ecb97f..5a060c2c30 100644
--- a/tools/libs/light/libxl_arch.h
+++ b/tools/libs/light/libxl_arch.h
@@ -90,6 +90,10 @@ void libxl__arch_update_domain_config(libxl__gc *gc,
                                       libxl_domain_config *dst,
                                       const libxl_domain_config *src);
 
+_hidden
+unsigned long libxl__arch_get_required_paging_memory(unsigned long maxmem_kb,
+                                                     unsigned int smp_cpus);
+
 #if defined(__i386__) || defined(__x86_64__)
 
 #define LAPIC_BASE_ADDRESS  0xfee00000
diff --git a/tools/libs/light/libxl_arm.c b/tools/libs/light/libxl_arm.c
index eef1de0939..73a95e83af 100644
--- a/tools/libs/light/libxl_arm.c
+++ b/tools/libs/light/libxl_arm.c
@@ -154,6 +154,20 @@ out:
     return rc;
 }
 
+unsigned long libxl__arch_get_required_paging_memory(unsigned long maxmem_kb,
+                                                     unsigned int smp_cpus)
+{
+    /*
+     * 256 pages (1MB) per vcpu,
+     * plus 1 page per MiB of RAM for the P2M map,
+     * plus 1 page per MiB of extended region. This default value is 128 MiB
+     * which should be enough for domains that are not running backend.
+     * This is higher than the minimum that Xen would allocate if no value
+     * were given (but the Xen minimum is for safety, not performance).
+     */
+    return 4 * (256 * smp_cpus + maxmem_kb / 1024 + 128);
+}
+
 static struct arch_info {
     const char *guest_type;
     const char *timer_compat;
diff --git a/tools/libs/light/libxl_utils.c b/tools/libs/light/libxl_utils.c
index 4699c4a0a3..e276c0ee9c 100644
--- a/tools/libs/light/libxl_utils.c
+++ b/tools/libs/light/libxl_utils.c
@@ -18,6 +18,7 @@
 #include <ctype.h>
 
 #include "libxl_internal.h"
+#include "libxl_arch.h"
 #include "_paths.h"
 
 #ifndef LIBXL_HAVE_NONCONST_LIBXL_BASENAME_RETURN_VALUE
@@ -39,13 +40,7 @@ char *libxl_basename(const char *name)
 
 unsigned long libxl_get_required_shadow_memory(unsigned long maxmem_kb, unsigned int smp_cpus)
 {
-    /* 256 pages (1MB) per vcpu,
-       plus 1 page per MiB of RAM for the P2M map,
-       plus 1 page per MiB of RAM to shadow the resident processes.
-       This is higher than the minimum that Xen would allocate if no value
-       were given (but the Xen minimum is for safety, not performance).
-     */
-    return 4 * (256 * smp_cpus + 2 * (maxmem_kb / 1024));
+    return libxl__arch_get_required_paging_memory(maxmem_kb, smp_cpus);
 }
 
 char *libxl_domid_to_name(libxl_ctx *ctx, uint32_t domid)
diff --git a/tools/libs/light/libxl_x86.c b/tools/libs/light/libxl_x86.c
index 1feadebb18..51362893cf 100644
--- a/tools/libs/light/libxl_x86.c
+++ b/tools/libs/light/libxl_x86.c
@@ -882,6 +882,19 @@ void libxl__arch_update_domain_config(libxl__gc *gc,
                     libxl_defbool_val(src->b_info.arch_x86.msr_relaxed));
 }
 
+unsigned long libxl__arch_get_required_paging_memory(unsigned long maxmem_kb,
+                                                     unsigned int smp_cpus)
+{
+    /*
+     * 256 pages (1MB) per vcpu,
+     * plus 1 page per MiB of RAM for the P2M map,
+     * plus 1 page per MiB of RAM to shadow the resident processes.
+     * This is higher than the minimum that Xen would allocate if no value
+     * were given (but the Xen minimum is for safety, not performance).
+     */
+    return 4 * (256 * smp_cpus + 2 * (maxmem_kb / 1024));
+}
+
 /*
  * Local variables:
  * mode: C