libhsakmt: Enable invisible debug VRAM mappings by default

Remove the HSA_DEBUG environment variable that controlled the
creation of these mappings.

This should allow the debugger to attach to a running process and
access VRAM buffers through ptrace without having to do anything
special.

On processes that create many small VRAM mappings, this may cause
regressions due to the per-process mmap limit. However, the
sub-allocator in ROCr should consolidate most small allocations
into 2MB blocks nowadays, for good TLB efficiency. So this is
unlikely to cause problems.

Change-Id: I929da1be0f6cb51ec00a02f3f241d16083e4d95f
Signed-off-by: Felix Kuehling <Felix.Kuehling@amd.com>

src/fmm.c

@@ -221,7 +221,6 @@ static gpu_mem_t *gpu_mem;
 static unsigned int gpu_mem_count;
 static gpu_mem_t *g_first_gpu_mem;
 
-static bool hsa_debug;
 static void *dgpu_shared_aperture_base;
 static void *dgpu_shared_aperture_limit;
 
@@ -1351,7 +1350,7 @@ void *fmm_allocate_device(uint32_t gpu_id, void *address, uint64_t MemorySizeInB
 		pthread_mutex_unlock(&aperture->fmm_mutex);
 	}
 
-	if (mem && (flags.ui32.HostAccess || hsa_debug)) {
+	if (mem) {
 		int map_fd = mmap_offset >= (1ULL<<40) ? kfd_fd :
 					gpu_mem[gpu_mem_id].drm_render_fd;
 		int prot = flags.ui32.HostAccess ? PROT_READ | PROT_WRITE :
@@ -2055,15 +2054,11 @@ HSAKMT_STATUS fmm_init_process_apertures(unsigned int NumNodes)
 	uint32_t num_of_sysfs_nodes;
 	HSAKMT_STATUS ret = HSAKMT_STATUS_SUCCESS;
 	char *disableCache, *pagedUserptr, *checkUserptr, *guardPagesStr, *reserveSvm;
-	char *hsaDebug;
 	unsigned int guardPages = 1;
 	struct pci_access *pacc;
 	uint64_t svm_base = 0, svm_limit = 0;
 	uint32_t svm_alignment = 0;
 
-	hsaDebug = getenv("HSA_DEBUG");
-	hsa_debug = hsaDebug && strcmp(hsaDebug, "0");
-
 	/* If HSA_DISABLE_CACHE is set to a non-0 value, disable caching */
 	disableCache = getenv("HSA_DISABLE_CACHE");
 	svm.disable_cache = (disableCache && strcmp(disableCache, "0"));