Add get_mempolicy, set_mempolicy NUMA syscalls

syscall/linux/c.lua

@@ -347,6 +347,13 @@ function C.sched_setparam(pid, param)
   return syscall(sys.sched_setparam, int(pid), void(param))
 end
 
+function C.get_mempolicy(mode, mask, maxnode, addr, flags)
+  return syscall(sys.get_mempolicy, void(mode), void(mask), ulong(maxnode), ulong(addr), ulong(flags))
+end
+function C.set_mempolicy(mode, mask, maxnode)
+  return syscall(sys.set_mempolicy, int(mode), void(mask), ulong(maxnode))
+end
+
 -- in librt for glibc but use syscalls instead of loading another library
 function C.clock_nanosleep(clk_id, flags, req, rem)
   return syscall(sys.clock_nanosleep, int(clk_id), int(flags), void(req), void(rem))

syscall/linux/constants.lua

@@ -3360,6 +3360,23 @@ c.IPT_SO_GET = strflag {
   REVISION_TARGET      = IPT_BASE_CTL + 3,
 }
 
+c.MPOL_MODE = multiflags {
+  DEFAULT = 0,
+  PREFERRED = 1,
+  BIND = 2,
+  INTERLEAVE = 3,
+  LOCAL = 4,
+  -- TODO: Only the following two flags can be ORed.
+  STATIC_NODES     = 0x80000000,
+  RELATIVE_NODES   = 0x40000000,
+}
+
+c.MPOL_FLAG = multiflags {
+  NODE	           = 1,
+  ADDR	           = 2,
+  MEMS_ALLOWED     = 4
+}
+
 c.SCHED = multiflags {
   NORMAL           = 0,
   OTHER            = 0,

syscall/linux/syscalls.lua

@@ -457,6 +457,18 @@ function S.sched_setaffinity(pid, mask, len) -- note len last as rarely used
   return retbool(C.sched_setaffinity(pid or 0, len or s.cpu_set, mktype(t.cpu_set, mask)))
 end
 
+function S.get_mempolicy(mode, mask, addr, flags)
+  mode = mode or t.int1()
+  mask = mktype(t.bitmask, mask)
+  local ret, err = C.get_mempolicy(mode, mask.mask, mask.size, addr, c.MPOL_FLAG[flags])
+  if ret == -1 then return nil, t.error(err or errno()) end
+  return { mode=mode[0], mask=mask }
+end
+function S.set_mempolicy(mode, mask)
+  mask = mktype(t.bitmask, mask)
+  return retbool(C.set_mempolicy(c.MPOL_MODE[mode], mask.mask, mask.size))
+end
+
 function S.sched_get_priority_max(policy) return retnum(C.sched_get_priority_max(c.SCHED[policy])) end
 function S.sched_get_priority_min(policy) return retnum(C.sched_get_priority_min(c.SCHED[policy])) end
 

syscall/linux/types.lua

@@ -1016,6 +1016,74 @@ mt.cpu_set = {
 
 addtype(types, "cpu_set", "struct cpu_set_t", mt.cpu_set)
 
+local ulong_bit_count = ffi.sizeof('unsigned long') * 8
+local function ulong_index_and_bit(n)
+  local i = math.floor(n / ulong_bit_count)
+  local b = bit.lshift(1ULL, n - i * ulong_bit_count)
+  return i, b
+end
+
+mt.bitmask = {
+  index = {
+    zero = function(mask) ffi.fill(mask, s.bitmask) end,
+    set = function(mask, node)
+      if type(node) == "table" then -- table is an array of node numbers eg {1, 2, 4}
+        for i = 1, #node do mask:set(node[i]) end
+        return mask
+      end
+      if node >= mask.size then error("numa node too large " .. node) end
+      local i, b = ulong_index_and_bit(node)
+      mask.mask[i] = bit.bor(mask.mask[i], b)
+      return mask
+    end,
+    clear = function(mask, node)
+      if type(node) == "table" then -- table is an array of node numbers eg {1, 2, 4}
+        for i = 1, #node do mask:clear(node[i]) end
+        return mask
+      end
+      if node < mask.size then
+         local i, b = ulong_index_and_bit(node)
+         mask.mask[i] = bit.band(mask.mask[i], bit.bnot(b))
+      end
+      return mask
+    end,
+    get = function(mask, node)
+      local i, b = ulong_index_and_bit(node)
+      if node >= mask.size then return false end
+      return bit.band(mask.mask[i], b) ~= 0
+    end,
+  },
+  __index = function(mask, k)
+    if mt.bitmask.index[k] then return mt.bitmask.index[k] end
+    if type(k) == "number" then return mask:get(k) end
+    error("invalid index " .. k)
+  end,
+  __newindex = function(mask, k, v)
+    if type(k) ~= "number" then error("invalid index " .. k) end
+    if v then mask:set(k) else mask:clear(k) end
+  end,
+  __new = function(tp, tab, size)
+    -- Round size to multiple of ulong bit count.
+    if size then
+      size = bit.band(size + ulong_bit_count - 1, bit.bnot(ulong_bit_count - 1))
+    else
+      size = ulong_bit_count
+    end
+    local mask = ffi.new(tp, size / ulong_bit_count, size)
+    if tab then mask:set(tab) end
+    return mask
+  end,
+  __tostring = function(mask)
+    local tab = {}
+    for i = 0, tonumber(mask.size - 1) do
+       if mask:get(i) then tab[#tab + 1] = i end
+    end
+    return "{" .. table.concat(tab, ",") .. "}"
+  end,
+}
+
+addtype_var(types, "bitmask", "struct {unsigned long size; unsigned long mask[?];}", mt.bitmask)
+
 mt.mq_attr = {
   index = {
     flags = function(mqa) return tonumber(mqa.mq_flags) end,