NVIDIA/open-gpu-kernel-modules
1
0
Fork 0
mirror of https://github.com/NVIDIA/open-gpu-kernel-modules.git synced 2026-02-22 16:04:00 +00:00
Code Issues Packages Projects Releases Wiki Activity

535.43.02

Browse Source
This commit is contained in:
Andy Ritger
2023-05-30 10:11:36 -07:00
parent 6dd092ddb7
commit eb5c7665a1
1403 changed files with 295367 additions and 86235 deletions
Download Patch File Download Diff File Expand all files Collapse all files

50
kernel-open/nvidia-uvm/uvm_perf_thrashing.c
View File

@@ -22,6 +22,7 @@
*******************************************************************************/
#include "uvm_api.h"
#include "uvm_conf_computing.h"
#include "uvm_perf_events.h"
#include "uvm_perf_module.h"
#include "uvm_perf_thrashing.h"
@@ -262,6 +263,7 @@ static unsigned uvm_perf_thrashing_pin_threshold = UVM_PERF_THRASHING_PIN_THRESH
// detection/prevention parameters
#define UVM_PERF_THRASHING_LAPSE_USEC_DEFAULT 500
#define UVM_PERF_THRASHING_LAPSE_USEC_DEFAULT_EMULATION (UVM_PERF_THRASHING_LAPSE_USEC_DEFAULT * 800)
#define UVM_PERF_THRASHING_LAPSE_USEC_DEFAULT_HCC (UVM_PERF_THRASHING_LAPSE_USEC_DEFAULT * 10)
// Lapse of time in microseconds that determines if two consecutive events on
// the same page can be considered thrashing
@@ -532,18 +534,20 @@ static void gpu_thrashing_stats_destroy(uvm_gpu_t *gpu)
// Get the thrashing detection struct for the given VA space if it exists
//
// VA space lock needs to be held
// The caller must ensure that the va_space cannot be deleted, for the
// duration of this call. Holding either the va_block or va_space lock will do
// that.
static va_space_thrashing_info_t *va_space_thrashing_info_get_or_null(uvm_va_space_t *va_space)
{
// TODO: Bug 3898454: check locking requirement for UVM-HMM.
return uvm_perf_module_type_data(va_space->perf_modules_data, UVM_PERF_MODULE_TYPE_THRASHING);
}
// Get the thrashing detection struct for the given VA space. It asserts that
// the information has been previously created.
//
// VA space lock needs to be held
// The caller must ensure that the va_space cannot be deleted, for the
// duration of this call. Holding either the va_block or va_space lock will do
// that.
static va_space_thrashing_info_t *va_space_thrashing_info_get(uvm_va_space_t *va_space)
{
va_space_thrashing_info_t *va_space_thrashing = va_space_thrashing_info_get_or_null(va_space);
@@ -1783,22 +1787,6 @@ const uvm_page_mask_t *uvm_perf_thrashing_get_thrashing_pages(uvm_va_block_t *va
return &block_thrashing->thrashing_pages;
}
bool uvm_perf_thrashing_is_block_thrashing(uvm_va_block_t *va_block)
{
uvm_va_space_t *va_space = uvm_va_block_get_va_space(va_block);
va_space_thrashing_info_t *va_space_thrashing = va_space_thrashing_info_get(va_space);
block_thrashing_info_t *block_thrashing = NULL;
if (!va_space_thrashing->params.enable)
return false;
block_thrashing = thrashing_info_get(va_block);
if (!block_thrashing)
return false;
return block_thrashing->num_thrashing_pages > 0;
}
#define TIMER_GRANULARITY_NS 20000ULL
static void thrashing_unpin_pages(struct work_struct *work)
{
@@ -1854,6 +1842,8 @@ static void thrashing_unpin_pages(struct work_struct *work)
break;
va_block = pinned_page->va_block;
if (uvm_va_block_is_hmm(va_block))
uvm_hmm_migrate_begin_wait(va_block);
uvm_mutex_lock(&va_block->lock);
// Only operate if the pinned page's tracking state isn't already
@@ -1876,6 +1866,8 @@ static void thrashing_unpin_pages(struct work_struct *work)
}
uvm_mutex_unlock(&va_block->lock);
if (uvm_va_block_is_hmm(va_block))
uvm_hmm_migrate_finish(va_block);
kmem_cache_free(g_pinned_page_cache, pinned_page);
}
@@ -1947,12 +1939,24 @@ void uvm_perf_thrashing_unload(uvm_va_space_t *va_space)
NV_STATUS uvm_perf_thrashing_register_gpu(uvm_va_space_t *va_space, uvm_gpu_t *gpu)
{
// If a simulated GPU is registered, re-initialize thrashing parameters in
// case they need to be adjusted
if (g_uvm_global.num_simulated_devices > 0) {
// case they need to be adjusted.
bool params_need_readjusting = g_uvm_global.num_simulated_devices > 0;
// Likewise, when the Confidential Computing feature is enabled, the DMA
// path is slower due to cryptographic operations & other associated
// overhead. Enforce a larger window to allow the thrashing mitigation
// mechanisms to work properly.
params_need_readjusting = params_need_readjusting || uvm_conf_computing_mode_enabled(gpu);
if (params_need_readjusting) {
va_space_thrashing_info_t *va_space_thrashing = va_space_thrashing_info_get(va_space);
if (!va_space_thrashing->params.test_overrides)
if (!va_space_thrashing->params.test_overrides) {
if (uvm_conf_computing_mode_enabled(gpu))
g_uvm_perf_thrashing_lapse_usec = UVM_PERF_THRASHING_LAPSE_USEC_DEFAULT_HCC;
va_space_thrashing_info_init_params(va_space_thrashing);
}
}
return NV_OK;