fdo-mirrors/mesa
1
0
Fork 0
mirror of https://gitlab.freedesktop.org/mesa/mesa.git synced 2026-05-06 11:38:05 +02:00
Code Issues Projects Releases Packages Wiki Activity Actions 1

anv: Rework init_meminfo

Browse source 
Instead of making LMEM the special case, unify the two paths by setting
up a fake drm_i915_query_memory_regions struct and filling it out based
on OS queries.  The important functional change here is that we now pass
system memory through the same GTT size and 3/4 filter that we were
using with the OS queries.  This should make behavior consistent on
integrated GPUs regardless of whether or not we have the memory region
query API.

Reviewed-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/12433>
This commit is contained in:
Jason Ekstrand 2021-08-18 09:40:37 -05:00 • committed by Marge Bot
parent be216ae9d9
commit 5c79c545e3
1 changed files with 78 additions and 49 deletions
Show stats Download patch file Download diff file Expand all files Collapse all files

127
src/intel/vulkan/anv_device.c
View file

@ -305,13 +305,9 @@ get_device_extensions(const struct anv_physical_device *device,
}
static uint64_t
anv_compute_heap_size(int fd, uint64_t gtt_size)
anv_compute_sys_heap_size(struct anv_physical_device *device,
uint64_t total_ram)
{
/* Query the total ram from the system */
uint64_t total_ram;
if (!os_get_total_physical_memory(&total_ram))
return 0;
/* We don't want to burn too much ram with the GPU. If the user has 4GiB
* or less, we use at most half. If they have more than 4GiB, we use 3/4.
*/
@ -324,47 +320,9 @@ anv_compute_heap_size(int fd, uint64_t gtt_size)
/* We also want to leave some padding for things we allocate in the driver,
* so don't go over 3/4 of the GTT either.
*/
uint64_t available_gtt = gtt_size * 3 / 4;
available_ram = MIN2(available_ram, device->gtt_size * 3 / 4);
return MIN2(available_ram, available_gtt);
}
static bool
anv_get_query_meminfo(struct anv_physical_device *device, int fd)
{
struct drm_i915_query_memory_regions *mem_regions =
intel_i915_query_alloc(fd, DRM_I915_QUERY_MEMORY_REGIONS);
if (mem_regions == NULL)
return false;
for(int i = 0; i < mem_regions->num_regions; i++) {
switch(mem_regions->regions[i].region.memory_class) {
case I915_MEMORY_CLASS_SYSTEM:
device->sys.region = mem_regions->regions[i].region;
device->sys.size = mem_regions->regions[i].probed_size;
break;
case I915_MEMORY_CLASS_DEVICE:
device->vram.region = mem_regions->regions[i].region;
device->vram.size = mem_regions->regions[i].probed_size;
break;
default:
break;
}
}
free(mem_regions);
return true;
}
static void
anv_init_meminfo(struct anv_physical_device *device, int fd)
{
if (anv_get_query_meminfo(device, fd))
return;
uint64_t heap_size = anv_compute_heap_size(fd, device->gtt_size);
if (heap_size > (2ull << 30) && !device->supports_48bit_addresses) {
if (available_ram > (2ull << 30) && !device->supports_48bit_addresses) {
/* When running with an overridden PCI ID, we may get a GTT size from
* the kernel that is greater than 2 GiB but the execbuf check for 48bit
* address support can still fail. Just clamp the address space size to
@ -373,10 +331,78 @@ anv_init_meminfo(struct anv_physical_device *device, int fd)
mesa_logw("%s:%d: The kernel reported a GTT size larger than 2 GiB but "
"not support for 48-bit addresses",
__FILE__, __LINE__);
heap_size = 2ull << 30;
available_ram = 2ull << 30;
}
device->sys.size = heap_size;
return available_ram;
}
static VkResult MUST_CHECK
anv_init_meminfo(struct anv_physical_device *device, int fd)
{
char sys_mem_regions[sizeof(struct drm_i915_query_memory_regions) +
sizeof(struct drm_i915_memory_region_info)];
struct drm_i915_query_memory_regions *mem_regions =
intel_i915_query_alloc(fd, DRM_I915_QUERY_MEMORY_REGIONS);
if (mem_regions == NULL) {
if (device->info.has_local_mem) {
return vk_errorfi(device->instance, NULL,
VK_ERROR_INCOMPATIBLE_DRIVER,
"failed to memory regions: %m");
}
uint64_t total_phys;
if (!os_get_total_physical_memory(&total_phys)) {
return vk_errorfi(device->instance, NULL,
VK_ERROR_INITIALIZATION_FAILED,
"failed to get total physical memory: %m");
}
uint64_t available;
if (!os_get_available_system_memory(&available))
available = 0; /* Silently disable VK_EXT_memory_budget */
/* The kernel query failed. Fake it using OS memory queries. This
* should be roughly the same for integrated GPUs.
*/
mem_regions = (void *)sys_mem_regions;
mem_regions->num_regions = 1;
mem_regions->regions[0] = (struct drm_i915_memory_region_info) {
.region.memory_class = I915_MEMORY_CLASS_SYSTEM,
.probed_size = total_phys,
.unallocated_size = available,
};
}
for(int i = 0; i < mem_regions->num_regions; i++) {
struct drm_i915_memory_region_info *info = &mem_regions->regions[i];
struct anv_memregion *region;
switch (info->region.memory_class) {
case I915_MEMORY_CLASS_SYSTEM:
region = &device->sys;
break;
case I915_MEMORY_CLASS_DEVICE:
region = &device->vram;
break;
default:
/* We don't know what kind of memory this is */
continue;
}
uint64_t size = info->probed_size;
if (info->region.memory_class == I915_MEMORY_CLASS_SYSTEM)
size = anv_compute_sys_heap_size(device, size);
region->region = info->region;
region->size = size;
}
if (mem_regions != (void *)sys_mem_regions)
free(mem_regions);
return VK_SUCCESS;
}
static VkResult
@ -403,7 +429,10 @@ anv_physical_device_init_heaps(struct anv_physical_device *device, int fd)
device->supports_48bit_addresses = (device->info.ver >= 8) &&
device->gtt_size > (4ULL << 30 /* GiB */);
anv_init_meminfo(device, fd);
VkResult result = anv_init_meminfo(device, fd);
if (result != VK_SUCCESS)
return result;
assert(device->sys.size != 0);
if (device->vram.size > 0) {