[PATCH v3] x86+libxl: correct p2m (shadow) memory pool size calculation

[PATCH v3] x86+libxl: correct p2m (shadow) memory pool size calculation

[Jan Beulich]

The reference "to shadow the resident processes" is applicable to
domains (potentially) running in shadow mode only. Adjust the
calculations accordingly. This, however, requires further parameters.
Since the original function is deprecated anyway, and since it can't be
changed (for being part of a stable ABI), introduce a new (internal
only) function, with the deprecated one simply becoming a wrapper.

In dom0_paging_pages() also take the opportunity and stop open-coding
DIV_ROUND_UP().

Signed-off-by: Jan Beulich <jbeulich@suse.com>
---
v3: Refine expression in dom0_paging_pages(). Update comment there as
    well.
v2: Introduce libxl__get_required_paging_memory().

--- a/tools/libs/light/libxl_create.c
+++ b/tools/libs/light/libxl_create.c
@@ -1017,7 +1017,7 @@ static bool ok_to_default_memkb_in_creat
      * The result is that the behaviour with old callers is the same
      * as in 4.13: no additional memory is allocated for shadow and
      * iommu (unless the caller set shadow_memkb, eg from a call to
-     * libxl_get_required_shadow_memory).
+     * libxl__get_required_paging_memory).
      */
     return !CTX->libxl_domain_need_memory_0x041200_called ||
             CTX->libxl_domain_need_memory_called;
@@ -1027,6 +1027,24 @@ static bool ok_to_default_memkb_in_creat
      */
 }
 
+unsigned long libxl__get_required_paging_memory(unsigned long maxmem_kb,
+                                                unsigned int smp_cpus,
+                                                libxl_domain_type type,
+                                                bool hap)
+{
+    /*
+     * 256 pages (1MB) per vcpu,
+     * plus 1 page per MiB of RAM for the P2M map (for non-PV guests),
+     * plus 1 page per MiB of RAM to shadow the resident processes (for shadow
+     * mode guests).
+     * 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 +
+                ((type != LIBXL_DOMAIN_TYPE_PV) + !hap) *
+                (maxmem_kb / 1024));
+}
+
 static unsigned long libxl__get_required_iommu_memory(unsigned long maxmem_kb)
 {
     unsigned long iommu_pages = 0, mem_pages = maxmem_kb / 4;
@@ -1194,10 +1212,16 @@ int libxl__domain_config_setdefault(libx
     }
 
     if (d_config->b_info.shadow_memkb == LIBXL_MEMKB_DEFAULT
-        && ok_to_default_memkb_in_create(gc))
+        && ok_to_default_memkb_in_create(gc)) {
+        bool hap = d_config->c_info.type != LIBXL_DOMAIN_TYPE_PV &&
+                   libxl_defbool_val(d_config->c_info.hap);
+
         d_config->b_info.shadow_memkb =
-            libxl_get_required_shadow_memory(d_config->b_info.max_memkb,
-                                             d_config->b_info.max_vcpus);
+            libxl__get_required_paging_memory(d_config->b_info.max_memkb,
+                                              d_config->b_info.max_vcpus,
+                                              d_config->c_info.type,
+                                              hap);
+    }
 
     /* No IOMMU reservation is needed if passthrough mode is not 'sync_pt' */
     if (d_config->b_info.iommu_memkb == LIBXL_MEMKB_DEFAULT
--- a/tools/libs/light/libxl_internal.h
+++ b/tools/libs/light/libxl_internal.h
@@ -1569,6 +1569,11 @@ _hidden int libxl__domain_need_memory_ca
                                       libxl_domain_build_info *b_info,
                                       uint64_t *need_memkb);
 
+_hidden unsigned long libxl__get_required_paging_memory(unsigned long maxmem_kb,
+                                                        unsigned int smp_cpus,
+                                                        libxl_domain_type type,
+                                                        bool hap);
+
 _hidden const char *libxl__device_nic_devname(libxl__gc *gc,
                                               uint32_t domid,
                                               uint32_t devid,
--- a/tools/libs/light/libxl_utils.c
+++ b/tools/libs/light/libxl_utils.c
@@ -38,13 +38,8 @@ 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__get_required_paging_memory(maxmem_kb, smp_cpus,
+                                             LIBXL_DOMAIN_TYPE_INVALID, false);
 }
 
 char *libxl_domid_to_name(libxl_ctx *ctx, uint32_t domid)
--- a/xen/arch/x86/dom0_build.c
+++ b/xen/arch/x86/dom0_build.c
@@ -314,12 +314,15 @@ unsigned int __initdata dom0_memflags =
 unsigned long __init dom0_paging_pages(const struct domain *d,
                                        unsigned long nr_pages)
 {
-    /* Copied from: libxl_get_required_shadow_memory() */
+    /* Keep in sync with libxl__get_required_paging_memory(). */
     unsigned long memkb = nr_pages * (PAGE_SIZE / 1024);
 
-    memkb = 4 * (256 * d->max_vcpus + 2 * (memkb / 1024));
+    memkb = 4 * (256 * d->max_vcpus +
+                 (is_pv_domain(d) ? opt_dom0_shadow || opt_pv_l1tf_hwdom
+                                  : 1 + opt_dom0_shadow) *
+                 (memkb / 1024));
 
-    return ((memkb + 1023) / 1024) << (20 - PAGE_SHIFT);
+    return DIV_ROUND_UP(memkb, 1024) << (20 - PAGE_SHIFT);
 }
 
 

Re: [PATCH v3] x86+libxl: correct p2m (shadow) memory pool size calculation

[Roger Pau Monné]

On Wed, Apr 27, 2022 at 12:57:17PM +0200, Jan Beulich wrote:
> The reference "to shadow the resident processes" is applicable to
> domains (potentially) running in shadow mode only. Adjust the
> calculations accordingly. This, however, requires further parameters.
> Since the original function is deprecated anyway, and since it can't be
> changed (for being part of a stable ABI), introduce a new (internal
> only) function, with the deprecated one simply becoming a wrapper.
> 
> In dom0_paging_pages() also take the opportunity and stop open-coding
> DIV_ROUND_UP().
> 
> Signed-off-by: Jan Beulich <jbeulich@suse.com>

Reviewed-by: Roger Pau Monné <roger.pau@citrix.com>

Thanks.

Re: [PATCH v3] x86+libxl: correct p2m (shadow) memory pool size calculation

[Anthony PERARD]

On Wed, Apr 27, 2022 at 12:57:17PM +0200, Jan Beulich wrote:
> The reference "to shadow the resident processes" is applicable to
> domains (potentially) running in shadow mode only. Adjust the
> calculations accordingly. This, however, requires further parameters.
> Since the original function is deprecated anyway, and since it can't be
> changed (for being part of a stable ABI), introduce a new (internal
> only) function, with the deprecated one simply becoming a wrapper.
> 
> In dom0_paging_pages() also take the opportunity and stop open-coding
> DIV_ROUND_UP().
> 
> Signed-off-by: Jan Beulich <jbeulich@suse.com>
> ---
> v3: Refine expression in dom0_paging_pages(). Update comment there as
>     well.
> v2: Introduce libxl__get_required_paging_memory().
> 
> --- a/tools/libs/light/libxl_create.c
> +++ b/tools/libs/light/libxl_create.c
> @@ -1017,7 +1017,7 @@ static bool ok_to_default_memkb_in_creat
>       * The result is that the behaviour with old callers is the same
>       * as in 4.13: no additional memory is allocated for shadow and
>       * iommu (unless the caller set shadow_memkb, eg from a call to
> -     * libxl_get_required_shadow_memory).
> +     * libxl__get_required_paging_memory).

I think in this comment, the "caller" is an application using libxl,
which might set shadow_memkb with a value from
libxl_get_required_shadow_memory(). So I don't think there's a need to
change the comment.


Otherwise, the patch looks good.
So with this chunk removed: Reviewed-by: Anthony PERARD <anthony.perard@citrix.com>

Thanks,

-- 
Anthony PERARD