core.memory: Add 'align' argument to dma_alloc()

Now it is possible to request specific alignment for DMA memory.

This is practical. For example, Mellanox ConnectX-4 requires specific
alignments (e.g. 4KB).

src/README.md

@@ -339,10 +339,13 @@ can be accessed directly by network cards. The important
 characteristic of DMA memory is being located in contiguous physical
 memory at a stable address.
 
-— Function **memory.dma_alloc** *bytes*
+— Function **memory.dma_alloc** *bytes*, *[alignment]*
 
 Returns a pointer to *bytes* of new DMA memory.
 
+Optionally a specific *alignment* requirement can be provided (in
+bytes). The default alignment is 128.
+
 — Function **memory.virtual_to_physical** *pointer*
 
 Returns the physical address (`uint64_t`) the DMA memory at *pointer*.

src/core/memory.lua

@@ -23,13 +23,20 @@ chunks = {}
 
 -- Allocate DMA-friendly memory.
 -- Return virtual memory pointer, physical address, and actual size.
-function dma_alloc (bytes)
+function dma_alloc (bytes,  align)
+   align = align or 128
    assert(bytes <= huge_page_size)
-   bytes = lib.align(bytes, 128)
-   if #chunks == 0 or bytes + chunks[#chunks].used > chunks[#chunks].size then
+   -- Get current chunk of memory to allocate from
+   if #chunks == 0 then allocate_next_chunk() end
+   local chunk = chunks[#chunks]
+   -- Skip allocation forward pointer to suit alignment
+   chunk.used = lib.align(chunk.used, align)
+   -- Need a new chunk to service this allocation?
+   if chunk.used + bytes > chunk.size then
       allocate_next_chunk()
+      chunk = chunks[#chunks]
    end
-   local chunk = chunks[#chunks]
+   -- Slice out the memory we need
    local where = chunk.used
    chunk.used = chunk.used + bytes
    return chunk.pointer + where, chunk.physical + where, bytes