vm: sparse sector map now extends to cover only all free pages by default

by default, the sector map created under the sparse model now only extends to the last non-reserved page frame, any reserved page frames afterwards are ignored.

vm/sparse.c

@@ -23,6 +23,7 @@
    overhead, and the flat memory model may be a better choice.
  */
 #include <socks/vm.h>
+#include <socks/arg.h>
 #include <socks/printk.h>
 #include <socks/panic.h>
 #include <socks/memblock.h>
@@ -32,6 +33,30 @@
 static struct vm_sector *sector_array = NULL;
 static size_t sector_array_count = 0;
 
+enum sector_coverage_mode {
+	SECTOR_COVERAGE_FREE,
+	SECTOR_COVERAGE_ALL,
+};
+
+static enum sector_coverage_mode get_sector_coverage_mode(void)
+{
+	const char *arg = arg_value("vm.sector-coverage-mode");
+	if (!arg) {
+		return SECTOR_COVERAGE_FREE;
+	}
+
+	if (!strcmp(arg, "free")) {
+		return SECTOR_COVERAGE_FREE;
+	}
+
+	if (!strcmp(arg, "all")) {
+		return SECTOR_COVERAGE_ALL;
+	}
+
+	printk("vm: [sparse] ignoring unknown sector coverage mode '%s', using FREE", arg);
+	return SECTOR_COVERAGE_FREE;
+}
+
 static struct vm_sector *phys_addr_to_sector_and_index(phys_addr_t addr, size_t *sector_id, size_t *index)
 {
 	/* all sectors have the same size */
@@ -98,13 +123,25 @@ static enum vm_page_order find_minimum_sector_size(phys_addr_t pmem_end)
    are in need of improvement to ensure that sparse works well on a wide
    range of systems. */
 static void calculate_sector_size_and_count(
-		phys_addr_t pmem_end, size_t reserved_size, size_t free_size,
+		size_t last_reserved_pfn, size_t last_free_pfn,
+		size_t reserved_size, size_t free_size,
 		unsigned int *out_sector_count, enum vm_page_order *out_sector_size)
 {
 	/* we can support up to VM_MAX_SECTORS memory sectors.
 	   the minimum sector size is what ever is required
 	   to cover all of physical memory in the maximum number of sectors */
-	enum vm_page_order sector_size = find_minimum_sector_size(pmem_end);
+
+	enum sector_coverage_mode mode = get_sector_coverage_mode();
+	phys_addr_t pmem_end = 0;
+
+	enum vm_page_order sector_size = find_minimum_sector_size(last_free_pfn);
+	if (mode == SECTOR_COVERAGE_FREE) {
+		pmem_end = last_free_pfn * VM_PAGE_SIZE;
+	} else {
+		pmem_end = MAX(last_free_pfn, last_reserved_pfn) * VM_PAGE_SIZE;
+	}
+
+	sector_size = find_minimum_sector_size(pmem_end);
 
 	if (sector_size <= VM_PAGE_2M) {
 		/* override really small sector sizes with something
@@ -152,6 +189,7 @@ static void calculate_sector_size_and_count(
 void vm_sparse_init(void)
 {
 	size_t pmem_limit = 0, reserved_size = 0, free_size = 0;
+	size_t last_reserved_pfn = 0, last_free_pfn = 0;
 
 	struct memblock_iter it;
 	for_each_mem_range (&it, 0x0, UINTPTR_MAX) {
@@ -162,17 +200,30 @@ void vm_sparse_init(void)
 
 	for_each_free_mem_range (&it, 0x0, UINTPTR_MAX) {
 		free_size += it.it_limit - it.it_base + 1;
+
+		size_t last_pfn = it.it_limit / VM_PAGE_SIZE;
+
+		if (last_pfn > last_free_pfn) {
+			last_free_pfn = last_pfn;
+		}
 	}
 
 	for_each_reserved_mem_range (&it, 0x0, UINTPTR_MAX) {
 		reserved_size += it.it_limit - it.it_base + 1;
+
+		size_t last_pfn = it.it_limit / VM_PAGE_SIZE;
+
+		if (last_pfn > last_reserved_pfn) {
+			last_reserved_pfn = last_pfn;
+		}
 	}
 
 	enum vm_page_order sector_size;
 	size_t sector_bytes = 0;
 	unsigned int nr_sectors = 0;
 	calculate_sector_size_and_count(
-			pmem_limit, reserved_size, free_size,
+			last_reserved_pfn, last_free_pfn,
+			reserved_size, free_size,
 			&nr_sectors, &sector_size);
 	sector_bytes = vm_page_order_to_bytes(sector_size);