HV Syscall Reference

HV Syscalls

lv1_allocate_memory (0)

Create a memory region in the Hypervisor Virtual Address Space (vas)

Kernel Call

result = lv1_allocate_memory( /*IN*/ size, page_size_exp, 0, flags, /*OUT*/ &addr, &muid );

Parameters

Inputs
Register	Description
R3	size - of the region to allocate, must be a multiple of page_size
R4	page_size_exp - where required page_size = 2 ^ page_size_exp
R5	0 - Unknown, see notes
R6	flags - (from linux/include/asm-powerpc/lv1call.h) bit 63: transferability: TF_NO = 0×00, TF_YES = 0×01 bit 62: destruction_scheme: DS_NO_CONNECTIONS = 0×00, DS_ANYTIME = 0×02 bit 61: fail or alternative: FA_FAIL = 0×00, FA_ALTERNATIVE = 0×04 bit 60: need LPAR address 0: ADDR_ANY = 0×00, ADDR_0 = 0×08 function unknown.
Outputs
Register	Description
R3	Status - 0 = OK, LV1_RESOURCE_SHORTAGE (-2), LV1_NO_ENTRY (-6), LV1_DUPLICATE_ENTRY (-7)
R4	addr - LPAR Address of region
R5	muid - Unknown, unused by Kernel

Notes:

page_size_exp takes values of 12 (page_size = 4K) to 21 (page_size = 2M) before LV1_RESOURCE_SHORTAGE (-2) is returned under a fully booted Linux OS. Higher values (24, page_size = 16M) can be found in the actual kernel source and can presumably be made before the OS has fully booted. page_size_exp values below 12 cause a return status of LV1_ILLEGAL_PARAMETER_VALUE (-17).

Input R5 was speculated to be the initialization value for the allocated region, but appears not to be the case. Values other than 0 or 1 appear to return a status of LV1_NO_ENTRY (-6), though a valid value of page_size_exp appears to be checked first (-17 is returned for invalid values of page_size_exp, regardless of the value of R5).

Allocations with flags = 0×00, 0×01, 0×02, 0×03 and 0×04 were successful though the effects of the flags could not be tested at this point. Allocations with flags >= 0×400 return LV1_ILLEGAL_PARAMETER_VALUE.

Initial tests allocating memory with flags = 0×08 (ADDR_0, presumably request physical address rather than logical partition address) result in a status of LV1_DUPLICATE_ENTRY (-7). This and the previous return value of -6 suggest an association with a database of some kind (repository values or memory maps?). It appears that some form of allocation may be taking place as LV1_ILLEGAL_PARAMETER_VALUE and LV1_RESOURCE_SHORTAGE are reported for invalid input parameters, rather than LV1_DUPLICATE_ENTRY.

For all successful allocations so far, output muid (R5) = 1

---

lv1_write_htab_entry (1)

Write an entry to the hash page table.

Kernel Call

result = lv1_write_htab_entry( /*IN*/ vas_id, slot, va, pa );

Parameters

Inputs
Register	Description
R3	vas_id - virtual address space id (0 for current)
R4	slot - table slot to write entry to
R5	va - first half of PTE
R6	pa - second half of PTE, except RPN is replaced with LPAR address
Outputs
Register	Description
R3	Status - 0 = OK, Other values are unknown, but indicate failure.

---

lv1_construct_virtual_address_space (2)

Construct a PPE virtual address space.

Kernel Call

result = lv1_construct_virtual_address_space( /*IN*/ htab_size, number_of_sizes, page_sizes, /*OUT*/ &vas_id, &act_htab_size );

Parameters

Inputs
Register	Description
R3	htab_size - must be 18, 19 or 20 (256KB, 512KB or 1MB)
R4	number_of_sizes - How many page sizes are specified in page_sizes
R5	page_sizes - see notes
Outputs
Register	Description
R3	Status - 0 = OK, Other values are unknown, but indicate failure.
R4	vas_id - virtual address space id
R5	act_htab_size - actual hash table size?

Notes:

Page sizes are specified as the power of two for the desired sizes. Each power of two is stored as an 8 bit field in page_sizes, starting from the MSB.

The “pages_sizes” parameter is set in “mm.c” using the following function:

page_sizes = make_page_sizes(PAGE_SHIFT_16M, PAGE_SHIFT_64K);

static unsigned long make_page_sizes(unsigned long a, unsigned long b)
{
	return (a << 56) | (b << 48);
}

---