CA-422187: create an emergency reserve of pages (workaround)

Do not let domains fully use up all available memory on the host,
we have too many unexplained bugs in this area.

As a workaround try to reserve some amount (e.g. 256MiB) that domains cannot
normally use from XAPI's point of view.
Then during parallel domain construction this emergency reserve can be used by
Xen.

Signed-off-by: Edwin Török <[email protected]>

Author: edwintorok

doc/content/design/numa.md

@@ -112,7 +112,7 @@ This function receives as arguments a domain ID and the number of nodes
 this domain is using (acquired using `domain_get_numa_info_node_pages`)
 
 The number of NUMA nodes of the host (not domain) is reported by
-`Xenctrl.physinfo` which returns a value of type `physinfo`.
+`Xenctrlext.physinfo` which returns a value of type `physinfo`.
 
 ```diff
     index b4579862ff..491bd3fc73 100644

doc/content/xenopsd/walkthroughs/VM.build/Domain.build.md

@@ -64,7 +64,7 @@ to call:
     [wait_xen_free_mem](https://github.com/xapi-project/xen-api/blob/master/ocaml/xenopsd/xc/domain.ml#L236-L272)
     to wait (if necessary), for the Xen memory scrubber to catch up reclaiming memory.
     It
-    1. calls `Xenctrl.physinfo` which returns:
+    1. calls `Xenctrlext.physinfo` which returns:
        - `hostinfo.free_pages` - the free and already scrubbed pages (available)
        - `host.scrub_pages` - the not yet scrubbed pages (not yet available)
     2. repeats this until a timeout as long as `free_pages` is *lower*

ocaml/libs/xenctrl-ext/xenctrlext.ml

@@ -131,3 +131,8 @@ let domain_claim_pages handle domid ?(numa_node = NumaNode.none) nr_pages =
 let get_nr_nodes handle =
   let info = numainfo handle in
   Array.length info.memory
+
+let physinfo xc =
+  let info = Xenctrl.physinfo xc in
+  let emergency_reserve_pages = Nativeint.shift_left 1n 16 in
+  {info with free_pages= Nativeint.sub info.free_pages emergency_reserve_pages}

ocaml/libs/xenctrl-ext/xenctrlext.mli

@@ -107,3 +107,5 @@ val domain_claim_pages : handle -> domid -> ?numa_node:NumaNode.t -> int -> unit
 
 val get_nr_nodes : handle -> int
 (** Returns the count of NUMA nodes available in the system. *)
+
+val physinfo : Xenctrl.handle -> Xenctrl.physinfo

ocaml/squeezed/src/dune

@@ -13,6 +13,7 @@
     threads.posix
     unix
     xenctrl
+    xenctrl_ext
     xenstore
     xenstore.unix
     xenstore_transport

ocaml/squeezed/src/squeeze_xen.ml

@@ -579,7 +579,7 @@ let make_host ~verbose ~xc =
      pages -- this might cause something else to fail (eg domain builder?) *)
   while
     Int64.div
-      ((Xenctrl.physinfo xc).Xenctrl.scrub_pages |> Int64.of_nativeint)
+      ((Xenctrlext.physinfo xc).Xenctrl.scrub_pages |> Int64.of_nativeint)
       1024L
     <> 0L
   do
@@ -762,7 +762,7 @@ let make_host ~verbose ~xc =
   (* For the host free memory we sum the free pages and the pages needing
      scrubbing: we don't want to adjust targets simply because the scrubber is
      slow. *)
-  let physinfo = Xenctrl.physinfo xc in
+  let physinfo = Xenctrlext.physinfo xc in
   let free_pages_kib =
     Xenctrl.pages_to_kib (Int64.of_nativeint physinfo.Xenctrl.free_pages)
   and scrub_pages_kib =

ocaml/xcp-rrdd/bin/rrdp-cpu/dune

@@ -11,6 +11,7 @@
     xapi-rrd
     xapi-stdext-unix
     xenctrl
+    xenctrl_ext
   )
 )
 

ocaml/xcp-rrdd/bin/rrdp-cpu/rrdp_cpu.ml

@@ -184,7 +184,7 @@ let dss_pcpus xc =
   let len = Array.length !physcpus in
   let newinfos =
     if len = 0 then (
-      let physinfo = Xenctrl.physinfo xc in
+      let physinfo = Xenctrlext.physinfo xc in
       let pcpus = physinfo.Xenctrl.nr_cpus in
       physcpus := if pcpus > 0 then Array.make pcpus 0L else [||] ;
       Xenctrl.pcpu_info xc pcpus
@@ -237,7 +237,7 @@ let count_power_state_running_domains domains =
     0 domains
 
 let dss_hostload xc domains =
-  let physinfo = Xenctrl.physinfo xc in
+  let physinfo = Xenctrlext.physinfo xc in
   let pcpus = physinfo.Xenctrl.nr_cpus in
   let rec sum acc n f =
     match n with n when n >= 0 -> sum (acc + f n) (n - 1) f | _ -> acc
@@ -298,7 +298,7 @@ let _ =
       let _, domains, _ = Xenctrl_lib.domain_snapshot xc in
       Process.initialise () ;
       (* Share one page per PCPU and dom each *)
-      let physinfo = Xenctrl.physinfo xc in
+      let physinfo = Xenctrlext.physinfo xc in
       let shared_page_count =
         physinfo.Xenctrl.nr_cpus
         + Int.max Rrd_interface.max_supported_vms (List.length domains)

ocaml/xcp-rrdd/bin/rrdp-squeezed/dune

@@ -13,6 +13,7 @@
     xapi-log
     xapi-rrd
     xenctrl
+    xenctrl_ext
     xenstore
     xenstore.unix
     xenstore_transport

ocaml/xcp-rrdd/bin/rrdp-squeezed/rrdp_squeezed.ml

@@ -169,7 +169,7 @@ let generate_host_sources xc counters =
   in
   let memory_reclaimed = bytes_of_kib memory_reclaimed in
   let memory_possibly_reclaimed = bytes_of_kib memory_possibly_reclaimed in
-  let physinfo = Xenctrl.physinfo xc in
+  let physinfo = Xenctrlext.physinfo xc in
   let total_kib =
     Xenctrl.pages_to_kib (Int64.of_nativeint physinfo.Xenctrl.total_pages)
   in