v3d: Add support for handling OOM signals from the simulator.

I have v3d allocating enough initial allocation memory that we've been
passing tests without it, but to match kernel behavior more it would be
good to actually exercise the OOM path.

3 files changed, 78 insertions, 14 deletions

diff --git a/src/gallium/drivers/v3d/v3d_context.h b/src/gallium/drivers/v3d/v3d_context.h
index de902ece205..3b39d18145d 100644
--- a/src/gallium/drivers/v3d/v3d_context.h
+++ b/src/gallium/drivers/v3d/v3d_context.h
@@ -552,6 +552,7 @@ void v3d_query_init(struct pipe_context *pctx);
 
 void v3d_simulator_init(struct v3d_screen *screen);
 void v3d_simulator_destroy(struct v3d_screen *screen);
+uint32_t v3d_simulator_get_spill(uint32_t spill_size);
 int v3d_simulator_ioctl(int fd, unsigned long request, void *arg);
 void v3d_simulator_open_from_handle(int fd, int handle, uint32_t size);
 
diff --git a/src/gallium/drivers/v3d/v3d_simulator.c b/src/gallium/drivers/v3d/v3d_simulator.c
index c5ddaa03103..6e2d25e9448 100644
--- a/src/gallium/drivers/v3d/v3d_simulator.c
+++ b/src/gallium/drivers/v3d/v3d_simulator.c
@@ -52,6 +52,7 @@
 #include "util/hash_table.h"
 #include "util/ralloc.h"
 #include "util/set.h"
+#include "util/u_dynarray.h"
 #include "util/u_memory.h"
 #include "util/u_mm.h"
 #include "drm-uapi/i915_drm.h"
@@ -63,6 +64,7 @@
 /** Global (across GEM fds) state for the simulator */
 static struct v3d_simulator_state {
         mtx_t mutex;
+        mtx_t submit_lock;
 
         struct v3d_hw *v3d;
         int ver;
@@ -80,6 +82,7 @@ static struct v3d_simulator_state {
         /** Mapping from GEM fd to struct v3d_simulator_file * */
         struct hash_table *fd_map;
 
+        struct util_dynarray bin_oom;
         int refcount;
 } sim_state = {
         .mutex = _MTX_INITIALIZER_NP,
@@ -162,7 +165,7 @@ set_gmp_flags(struct v3d_simulator_file *file,
  * that also contains the drm_gem_cma_object struct.
  */
 static struct v3d_simulator_bo *
-v3d_create_simulator_bo(int fd, int handle, unsigned size)
+v3d_create_simulator_bo(int fd, unsigned size)
 {
         struct v3d_simulator_file *file = v3d_get_simulator_file_for_fd(fd);
         struct v3d_simulator_bo *sim_bo = rzalloc(file,
@@ -170,7 +173,6 @@ v3d_create_simulator_bo(int fd, int handle, unsigned size)
         size = align(size, 4096);
 
         sim_bo->file = file;
-        sim_bo->handle = handle;
 
         mtx_lock(&sim_state.mutex);
         sim_bo->block = u_mmAllocMem(sim_state.heap, size + 4, GMP_ALIGN2, 0);
@@ -187,6 +189,18 @@ v3d_create_simulator_bo(int fd, int handle, unsigned size)
 
         *(uint32_t *)(sim_bo->sim_vaddr + sim_bo->size) = BO_SENTINEL;
 
+        return sim_bo;
+}
+
+static struct v3d_simulator_bo *
+v3d_create_simulator_bo_for_gem(int fd, int handle, unsigned size)
+{
+        struct v3d_simulator_file *file = v3d_get_simulator_file_for_fd(fd);
+        struct v3d_simulator_bo *sim_bo =
+                v3d_create_simulator_bo(fd, size);
+
+        sim_bo->handle = handle;
+
         /* Map the GEM buffer for copy in/out to the simulator.  i915 blocks
          * dumb mmap on render nodes, so use their ioctl directly if we're on
          * one.
@@ -238,6 +252,20 @@ v3d_create_simulator_bo(int fd, int handle, unsigned size)
         return sim_bo;
 }
 
+static int bin_fd;
+
+uint32_t
+v3d_simulator_get_spill(uint32_t spill_size)
+{
+        struct v3d_simulator_bo *sim_bo =
+                v3d_create_simulator_bo(bin_fd, spill_size);
+
+        util_dynarray_append(&sim_state.bin_oom, struct v3d_simulator_bo *,
+                             sim_bo);
+
+        return sim_bo->block->ofs;
+}
+
 static void
 v3d_free_simulator_bo(struct v3d_simulator_bo *sim_bo)
 {
@@ -331,11 +359,21 @@ v3d_simulator_submit_cl_ioctl(int fd, struct drm_v3d_submit_cl *submit)
         if (ret)
                 return ret;
 
+        mtx_lock(&sim_state.submit_lock);
+        bin_fd = fd;
         if (sim_state.ver >= 41)
                 v3d41_simulator_submit_cl_ioctl(sim_state.v3d, submit, file->gmp->ofs);
         else
                 v3d33_simulator_submit_cl_ioctl(sim_state.v3d, submit, file->gmp->ofs);
 
+        util_dynarray_foreach(&sim_state.bin_oom, struct v3d_simulator_bo *,
+                              sim_bo) {
+                v3d_free_simulator_bo(*sim_bo);
+        }
+        util_dynarray_clear(&sim_state.bin_oom);
+
+        mtx_unlock(&sim_state.submit_lock);
+
         ret = v3d_simulator_unpin_bos(file, submit);
         if (ret)
                 return ret;
@@ -352,7 +390,7 @@ v3d_simulator_submit_cl_ioctl(int fd, struct drm_v3d_submit_cl *submit)
  */
 void v3d_simulator_open_from_handle(int fd, int handle, uint32_t size)
 {
-        v3d_create_simulator_bo(fd, handle, size);
+        v3d_create_simulator_bo_for_gem(fd, handle, size);
 }
 
 /**
@@ -391,7 +429,8 @@ v3d_simulator_create_bo_ioctl(int fd, struct drm_v3d_create_bo *args)
 
         if (ret == 0) {
                 struct v3d_simulator_bo *sim_bo =
-                        v3d_create_simulator_bo(fd, args->handle, args->size);
+                        v3d_create_simulator_bo_for_gem(fd, args->handle,
+                                                        args->size);
 
                 args->offset = sim_bo->block->ofs;
         }
@@ -548,6 +587,8 @@ v3d_simulator_init_global(const struct v3d_device_info *devinfo)
                                         _mesa_hash_pointer,
                                         _mesa_key_pointer_equal);
 
+        util_dynarray_init(&sim_state.bin_oom, NULL);
+
         if (sim_state.ver >= 41)
                 v3d41_simulator_init_regs(sim_state.v3d);
         else
@@ -589,6 +630,7 @@ v3d_simulator_destroy(struct v3d_screen *screen)
         mtx_lock(&sim_state.mutex);
         if (!--sim_state.refcount) {
                 _mesa_hash_table_destroy(sim_state.fd_map, NULL);
+                util_dynarray_fini(&sim_state.bin_oom);
                 u_mmDestroy(sim_state.heap);
                 /* No memsetting the struct, because it contains the mutex. */
                 sim_state.mem = NULL;
diff --git a/src/gallium/drivers/v3d/v3dx_simulator.c b/src/gallium/drivers/v3d/v3dx_simulator.c
index e6db838c0de..c14c41f964a 100644
--- a/src/gallium/drivers/v3d/v3dx_simulator.c
+++ b/src/gallium/drivers/v3d/v3dx_simulator.c
@@ -101,6 +101,21 @@ v3d_invalidate_caches(struct v3d_hw *v3d)
         v3d_invalidate_slices(v3d);
 }
 
+static uint32_t g_gmp_ofs;
+static void
+v3d_reload_gmp(struct v3d_hw *v3d)
+{
+        /* Completely reset the GMP. */
+        V3D_WRITE(V3D_GMP_0_CFG,
+                  V3D_GMP_0_CFG_PROTENABLE_SET);
+        V3D_WRITE(V3D_GMP_0_TABLE_ADDR, g_gmp_ofs);
+        V3D_WRITE(V3D_GMP_0_CLEAR_LOAD, ~0);
+        while (V3D_READ(V3D_GMP_0_STATUS) &
+               V3D_GMP_0_STATUS_CFG_BUSY_SET) {
+                ;
+        }
+}
+
 int
 v3dX(simulator_submit_tfu_ioctl)(struct v3d_hw *v3d,
                                  struct drm_v3d_submit_tfu *args)
@@ -167,6 +182,18 @@ v3d_isr(uint32_t hub_status)
         /* Check the per-core bits */
         if (hub_status & (1 << 0)) {
                 uint32_t core_status = V3D_READ(V3D_CTL_0_INT_STS);
+                V3D_WRITE(V3D_CTL_0_INT_CLR, core_status);
+
+                if (core_status & V3D_CTL_0_INT_STS_INT_OUTOMEM_SET) {
+                        uint32_t size = 256 * 1024;
+                        uint32_t offset = v3d_simulator_get_spill(size);
+
+                        v3d_reload_gmp(v3d);
+
+                        V3D_WRITE(V3D_PTB_0_BPOA, offset);
+                        V3D_WRITE(V3D_PTB_0_BPOS, size);
+                        return;
+                }
 
                 if (core_status & V3D_CTL_0_INT_STS_INT_GMPV_SET) {
                         fprintf(stderr, "GMP violation at 0x%08x\n",
@@ -198,7 +225,8 @@ v3dX(simulator_init_regs)(struct v3d_hw *v3d)
         V3D_WRITE(V3D_CTL_0_MISCCFG, V3D_CTL_1_MISCCFG_OVRTMUOUT_SET);
 #endif
 
-        uint32_t core_interrupts = V3D_CTL_0_INT_STS_INT_GMPV_SET;
+        uint32_t core_interrupts = (V3D_CTL_0_INT_STS_INT_GMPV_SET |
+                                    V3D_CTL_0_INT_STS_INT_OUTOMEM_SET);
         V3D_WRITE(V3D_CTL_0_INT_MSK_SET, ~core_interrupts);
         V3D_WRITE(V3D_CTL_0_INT_MSK_CLR, core_interrupts);
 
@@ -211,15 +239,8 @@ v3dX(simulator_submit_cl_ioctl)(struct v3d_hw *v3d,
                                 struct drm_v3d_submit_cl *submit,
                                 uint32_t gmp_ofs)
 {
-        /* Completely reset the GMP. */
-        V3D_WRITE(V3D_GMP_0_CFG,
-                  V3D_GMP_0_CFG_PROTENABLE_SET);
-        V3D_WRITE(V3D_GMP_0_TABLE_ADDR, gmp_ofs);
-        V3D_WRITE(V3D_GMP_0_CLEAR_LOAD, ~0);
-        while (V3D_READ(V3D_GMP_0_STATUS) &
-               V3D_GMP_0_STATUS_CFG_BUSY_SET) {
-                ;
-        }
+        g_gmp_ofs = gmp_ofs;
+        v3d_reload_gmp(v3d);
 
         v3d_invalidate_caches(v3d);