From 08f87f769b595151be1afeff53e144f543faa614 Mon Sep 17 00:00:00 2001
From: hc <hc@nodka.com>
Date: Wed, 06 Dec 2023 09:51:13 +0000
Subject: [PATCH] add dts config
---
kernel/drivers/gpu/arm/bifrost/mali_kbase_jd.c | 431 ++++++++++++++++-------------------------------------
1 files changed, 128 insertions(+), 303 deletions(-)
diff --git a/kernel/drivers/gpu/arm/bifrost/mali_kbase_jd.c b/kernel/drivers/gpu/arm/bifrost/mali_kbase_jd.c
index d780a30..f44426a 100644
--- a/kernel/drivers/gpu/arm/bifrost/mali_kbase_jd.c
+++ b/kernel/drivers/gpu/arm/bifrost/mali_kbase_jd.c
@@ -1,7 +1,7 @@
// SPDX-License-Identifier: GPL-2.0 WITH Linux-syscall-note
/*
*
- * (C) COPYRIGHT 2010-2021 ARM Limited. All rights reserved.
+ * (C) COPYRIGHT 2010-2022 ARM Limited. All rights reserved.
*
* This program is free software and is provided to you under the terms of the
* GNU General Public License version 2 as published by the Free Software
@@ -28,6 +28,11 @@
#include <linux/version.h>
#include <linux/ratelimit.h>
#include <linux/priority_control_manager.h>
+#if KERNEL_VERSION(4, 11, 0) <= LINUX_VERSION_CODE
+#include <linux/sched/signal.h>
+#else
+#include <linux/signal.h>
+#endif
#include <mali_kbase_jm.h>
#include <mali_kbase_kinstr_jm.h>
@@ -35,12 +40,9 @@
#include <tl/mali_kbase_tracepoints.h>
#include <mali_linux_trace.h>
-#include "mali_kbase_dma_fence.h"
#include <mali_kbase_cs_experimental.h>
#include <mali_kbase_caps.h>
-
-#define beenthere(kctx, f, a...) dev_dbg(kctx->kbdev->dev, "%s:" f, __func__, ##a)
/* Return whether katom will run on the GPU or not. Currently only soft jobs and
* dependency-only atoms do not run on the GPU
@@ -76,6 +78,7 @@
kbase_kinstr_jm_atom_complete(katom);
dev_dbg(katom->kctx->kbdev->dev, "Atom %pK status to completed\n",
(void *)katom);
+ KBASE_TLSTREAM_TL_JD_ATOM_COMPLETE(katom->kctx->kbdev, katom);
}
/* Runs an atom, either by handing to the JS or by immediately running it in the case of soft-jobs
@@ -83,7 +86,7 @@
* Returns whether the JS needs a reschedule.
*
* Note that the caller must also check the atom status and
- * if it is KBASE_JD_ATOM_STATE_COMPLETED must call jd_done_nolock
+ * if it is KBASE_JD_ATOM_STATE_COMPLETED must call kbase_jd_done_nolock
*/
static bool jd_run_atom(struct kbase_jd_atom *katom)
{
@@ -139,11 +142,17 @@
/* katom dep complete, attempt to run it */
bool resched = false;
+ KBASE_TLSTREAM_TL_RUN_ATOM_START(
+ katom->kctx->kbdev, katom,
+ kbase_jd_atom_id(katom->kctx, katom));
resched = jd_run_atom(katom);
+ KBASE_TLSTREAM_TL_RUN_ATOM_END(katom->kctx->kbdev, katom,
+ kbase_jd_atom_id(katom->kctx,
+ katom));
if (katom->status == KBASE_JD_ATOM_STATE_COMPLETED) {
/* The atom has already finished */
- resched |= jd_done_nolock(katom, NULL);
+ resched |= kbase_jd_done_nolock(katom, true);
}
if (resched)
@@ -153,25 +162,12 @@
void kbase_jd_free_external_resources(struct kbase_jd_atom *katom)
{
-#ifdef CONFIG_MALI_BIFROST_DMA_FENCE
- /* Flush dma-fence workqueue to ensure that any callbacks that may have
- * been queued are done before continuing.
- * Any successfully completed atom would have had all it's callbacks
- * completed before the atom was run, so only flush for failed atoms.
- */
- if (katom->event_code != BASE_JD_EVENT_DONE)
- flush_workqueue(katom->kctx->dma_fence.wq);
-#endif /* CONFIG_MALI_BIFROST_DMA_FENCE */
}
static void kbase_jd_post_external_resources(struct kbase_jd_atom *katom)
{
KBASE_DEBUG_ASSERT(katom);
KBASE_DEBUG_ASSERT(katom->core_req & BASE_JD_REQ_EXTERNAL_RESOURCES);
-
-#ifdef CONFIG_MALI_BIFROST_DMA_FENCE
- kbase_dma_fence_signal(katom);
-#endif /* CONFIG_MALI_BIFROST_DMA_FENCE */
kbase_gpu_vm_lock(katom->kctx);
/* only roll back if extres is non-NULL */
@@ -180,13 +176,7 @@
res_no = katom->nr_extres;
while (res_no-- > 0) {
- struct kbase_mem_phy_alloc *alloc = katom->extres[res_no].alloc;
- struct kbase_va_region *reg;
-
- reg = kbase_region_tracker_find_region_base_address(
- katom->kctx,
- katom->extres[res_no].gpu_address);
- kbase_unmap_external_resource(katom->kctx, reg, alloc);
+ kbase_unmap_external_resource(katom->kctx, katom->extres[res_no]);
}
kfree(katom->extres);
katom->extres = NULL;
@@ -202,26 +192,8 @@
static int kbase_jd_pre_external_resources(struct kbase_jd_atom *katom, const struct base_jd_atom *user_atom)
{
- int err_ret_val = -EINVAL;
+ int err = -EINVAL;
u32 res_no;
-#ifdef CONFIG_MALI_BIFROST_DMA_FENCE
- struct kbase_dma_fence_resv_info info = {
- .resv_objs = NULL,
- .dma_fence_resv_count = 0,
- .dma_fence_excl_bitmap = NULL
- };
-#if defined(CONFIG_SYNC) || defined(CONFIG_SYNC_FILE)
- /*
- * When both dma-buf fence and Android native sync is enabled, we
- * disable dma-buf fence for contexts that are using Android native
- * fences.
- */
- const bool implicit_sync = !kbase_ctx_flag(katom->kctx,
- KCTX_NO_IMPLICIT_SYNC);
-#else /* CONFIG_SYNC || CONFIG_SYNC_FILE*/
- const bool implicit_sync = true;
-#endif /* CONFIG_SYNC || CONFIG_SYNC_FILE */
-#endif /* CONFIG_MALI_BIFROST_DMA_FENCE */
struct base_external_resource *input_extres;
KBASE_DEBUG_ASSERT(katom);
@@ -235,47 +207,18 @@
if (!katom->extres)
return -ENOMEM;
- /* copy user buffer to the end of our real buffer.
- * Make sure the struct sizes haven't changed in a way
- * we don't support
- */
- BUILD_BUG_ON(sizeof(*input_extres) > sizeof(*katom->extres));
- input_extres = (struct base_external_resource *)
- (((unsigned char *)katom->extres) +
- (sizeof(*katom->extres) - sizeof(*input_extres)) *
- katom->nr_extres);
+ input_extres = kmalloc_array(katom->nr_extres, sizeof(*input_extres), GFP_KERNEL);
+ if (!input_extres) {
+ err = -ENOMEM;
+ goto failed_input_alloc;
+ }
if (copy_from_user(input_extres,
get_compat_pointer(katom->kctx, user_atom->extres_list),
sizeof(*input_extres) * katom->nr_extres) != 0) {
- err_ret_val = -EINVAL;
- goto early_err_out;
+ err = -EINVAL;
+ goto failed_input_copy;
}
-
-#ifdef CONFIG_MALI_BIFROST_DMA_FENCE
- if (implicit_sync) {
- info.resv_objs =
- kmalloc_array(katom->nr_extres,
-#if (KERNEL_VERSION(5, 4, 0) > LINUX_VERSION_CODE)
- sizeof(struct reservation_object *),
-#else
- sizeof(struct dma_resv *),
-#endif
- GFP_KERNEL);
- if (!info.resv_objs) {
- err_ret_val = -ENOMEM;
- goto early_err_out;
- }
-
- info.dma_fence_excl_bitmap =
- kcalloc(BITS_TO_LONGS(katom->nr_extres),
- sizeof(unsigned long), GFP_KERNEL);
- if (!info.dma_fence_excl_bitmap) {
- err_ret_val = -ENOMEM;
- goto early_err_out;
- }
- }
-#endif /* CONFIG_MALI_BIFROST_DMA_FENCE */
/* Take the processes mmap lock */
down_read(kbase_mem_get_process_mmap_lock());
@@ -283,19 +226,13 @@
/* need to keep the GPU VM locked while we set up UMM buffers */
kbase_gpu_vm_lock(katom->kctx);
for (res_no = 0; res_no < katom->nr_extres; res_no++) {
- struct base_external_resource *res = &input_extres[res_no];
+ struct base_external_resource *user_res = &input_extres[res_no];
struct kbase_va_region *reg;
- struct kbase_mem_phy_alloc *alloc;
-#ifdef CONFIG_MALI_BIFROST_DMA_FENCE
- bool exclusive;
- exclusive = (res->ext_resource & BASE_EXT_RES_ACCESS_EXCLUSIVE)
- ? true : false;
-#endif
+
reg = kbase_region_tracker_find_region_enclosing_address(
- katom->kctx,
- res->ext_resource & ~BASE_EXT_RES_ACCESS_EXCLUSIVE);
+ katom->kctx, user_res->ext_resource & ~BASE_EXT_RES_ACCESS_EXCLUSIVE);
/* did we find a matching region object? */
- if (kbase_is_region_invalid_or_free(reg)) {
+ if (unlikely(kbase_is_region_invalid_or_free(reg))) {
/* roll back */
goto failed_loop;
}
@@ -305,36 +242,11 @@
katom->atom_flags |= KBASE_KATOM_FLAG_PROTECTED;
}
- alloc = kbase_map_external_resource(katom->kctx, reg,
- current->mm);
- if (!alloc) {
- err_ret_val = -EINVAL;
+ err = kbase_map_external_resource(katom->kctx, reg, current->mm);
+ if (err)
goto failed_loop;
- }
-#ifdef CONFIG_MALI_BIFROST_DMA_FENCE
- if (implicit_sync &&
- reg->gpu_alloc->type == KBASE_MEM_TYPE_IMPORTED_UMM) {
-#if (KERNEL_VERSION(5, 4, 0) > LINUX_VERSION_CODE)
- struct reservation_object *resv;
-#else
- struct dma_resv *resv;
-#endif
- resv = reg->gpu_alloc->imported.umm.dma_buf->resv;
- if (resv)
- kbase_dma_fence_add_reservation(resv, &info,
- exclusive);
- }
-#endif /* CONFIG_MALI_BIFROST_DMA_FENCE */
-
- /* finish with updating out array with the data we found */
- /* NOTE: It is important that this is the last thing we do (or
- * at least not before the first write) as we overwrite elements
- * as we loop and could be overwriting ourself, so no writes
- * until the last read for an element.
- */
- katom->extres[res_no].gpu_address = reg->start_pfn << PAGE_SHIFT; /* save the start_pfn (as an address, not pfn) to use fast lookup later */
- katom->extres[res_no].alloc = alloc;
+ katom->extres[res_no] = reg;
}
/* successfully parsed the extres array */
/* drop the vm lock now */
@@ -343,57 +255,33 @@
/* Release the processes mmap lock */
up_read(kbase_mem_get_process_mmap_lock());
-#ifdef CONFIG_MALI_BIFROST_DMA_FENCE
- if (implicit_sync) {
- if (info.dma_fence_resv_count) {
- int ret;
-
- ret = kbase_dma_fence_wait(katom, &info);
- if (ret < 0)
- goto failed_dma_fence_setup;
- }
-
- kfree(info.resv_objs);
- kfree(info.dma_fence_excl_bitmap);
- }
-#endif /* CONFIG_MALI_BIFROST_DMA_FENCE */
+ /* Free the buffer holding data from userspace */
+ kfree(input_extres);
/* all done OK */
return 0;
/* error handling section */
-
-#ifdef CONFIG_MALI_BIFROST_DMA_FENCE
-failed_dma_fence_setup:
- /* Lock the processes mmap lock */
- down_read(kbase_mem_get_process_mmap_lock());
-
- /* lock before we unmap */
- kbase_gpu_vm_lock(katom->kctx);
-#endif
-
- failed_loop:
- /* undo the loop work */
+failed_loop:
+ /* undo the loop work. We are guaranteed to have access to the VA region
+ * as we hold a reference to it until it's unmapped
+ */
while (res_no-- > 0) {
- struct kbase_mem_phy_alloc *alloc = katom->extres[res_no].alloc;
+ struct kbase_va_region *reg = katom->extres[res_no];
- kbase_unmap_external_resource(katom->kctx, NULL, alloc);
+ kbase_unmap_external_resource(katom->kctx, reg);
}
kbase_gpu_vm_unlock(katom->kctx);
/* Release the processes mmap lock */
up_read(kbase_mem_get_process_mmap_lock());
- early_err_out:
+failed_input_copy:
+ kfree(input_extres);
+failed_input_alloc:
kfree(katom->extres);
katom->extres = NULL;
-#ifdef CONFIG_MALI_BIFROST_DMA_FENCE
- if (implicit_sync) {
- kfree(info.resv_objs);
- kfree(info.dma_fence_excl_bitmap);
- }
-#endif
- return err_ret_val;
+ return err;
}
static inline void jd_resolve_dep(struct list_head *out_list,
@@ -416,10 +304,6 @@
if (katom->event_code != BASE_JD_EVENT_DONE &&
(dep_type != BASE_JD_DEP_TYPE_ORDER)) {
-#ifdef CONFIG_MALI_BIFROST_DMA_FENCE
- kbase_dma_fence_cancel_callbacks(dep_atom);
-#endif
-
dep_atom->event_code = katom->event_code;
KBASE_DEBUG_ASSERT(dep_atom->status !=
KBASE_JD_ATOM_STATE_UNUSED);
@@ -433,35 +317,8 @@
(IS_GPU_ATOM(dep_atom) && !ctx_is_dying &&
!dep_atom->will_fail_event_code &&
!other_dep_atom->will_fail_event_code))) {
- bool dep_satisfied = true;
-#ifdef CONFIG_MALI_BIFROST_DMA_FENCE
- int dep_count;
-
- dep_count = kbase_fence_dep_count_read(dep_atom);
- if (likely(dep_count == -1)) {
- dep_satisfied = true;
- } else {
- /*
- * There are either still active callbacks, or
- * all fences for this @dep_atom has signaled,
- * but the worker that will queue the atom has
- * not yet run.
- *
- * Wait for the fences to signal and the fence
- * worker to run and handle @dep_atom. If
- * @dep_atom was completed due to error on
- * @katom, then the fence worker will pick up
- * the complete status and error code set on
- * @dep_atom above.
- */
- dep_satisfied = false;
- }
-#endif /* CONFIG_MALI_BIFROST_DMA_FENCE */
-
- if (dep_satisfied) {
- dep_atom->in_jd_list = true;
- list_add_tail(&dep_atom->jd_item, out_list);
- }
+ dep_atom->in_jd_list = true;
+ list_add_tail(&dep_atom->jd_item, out_list);
}
}
}
@@ -520,33 +377,8 @@
dep_atom->dep[0].atom);
bool dep1_valid = is_dep_valid(
dep_atom->dep[1].atom);
- bool dep_satisfied = true;
-#ifdef CONFIG_MALI_BIFROST_DMA_FENCE
- int dep_count;
- dep_count = kbase_fence_dep_count_read(
- dep_atom);
- if (likely(dep_count == -1)) {
- dep_satisfied = true;
- } else {
- /*
- * There are either still active callbacks, or
- * all fences for this @dep_atom has signaled,
- * but the worker that will queue the atom has
- * not yet run.
- *
- * Wait for the fences to signal and the fence
- * worker to run and handle @dep_atom. If
- * @dep_atom was completed due to error on
- * @katom, then the fence worker will pick up
- * the complete status and error code set on
- * @dep_atom above.
- */
- dep_satisfied = false;
- }
-#endif /* CONFIG_MALI_BIFROST_DMA_FENCE */
-
- if (dep0_valid && dep1_valid && dep_satisfied) {
+ if (dep0_valid && dep1_valid) {
dep_atom->in_jd_list = true;
list_add(&dep_atom->jd_item, out_list);
}
@@ -612,8 +444,8 @@
else if (reg->flags & KBASE_REG_TILER_ALIGN_TOP)
size_to_read = sizeof(u64[COUNT]);
- ptr = kbase_vmap(kctx, reg->heap_info_gpu_addr, size_to_read,
- &mapping);
+ ptr = kbase_vmap_prot(kctx, reg->heap_info_gpu_addr, size_to_read,
+ KBASE_REG_CPU_RD, &mapping);
if (!ptr) {
dev_warn(kctx->kbdev->dev,
@@ -697,23 +529,17 @@
}
#endif /* MALI_JIT_PRESSURE_LIMIT_BASE */
-/*
- * Perform the necessary handling of an atom that has finished running
- * on the GPU.
- *
- * Note that if this is a soft-job that has had kbase_prepare_soft_job called on it then the caller
- * is responsible for calling kbase_finish_soft_job *before* calling this function.
- *
- * The caller must hold the kbase_jd_context.lock.
- */
-bool jd_done_nolock(struct kbase_jd_atom *katom,
- struct list_head *completed_jobs_ctx)
+bool kbase_jd_done_nolock(struct kbase_jd_atom *katom, bool post_immediately)
{
struct kbase_context *kctx = katom->kctx;
struct list_head completed_jobs;
struct list_head runnable_jobs;
bool need_to_try_schedule_context = false;
int i;
+
+ lockdep_assert_held(&kctx->jctx.lock);
+
+ KBASE_TLSTREAM_TL_JD_DONE_NO_LOCK_START(kctx->kbdev, katom);
INIT_LIST_HEAD(&completed_jobs);
INIT_LIST_HEAD(&runnable_jobs);
@@ -736,6 +562,7 @@
}
jd_mark_atom_complete(katom);
+
list_add_tail(&katom->jd_item, &completed_jobs);
while (!list_empty(&completed_jobs)) {
@@ -767,7 +594,13 @@
if (node->status != KBASE_JD_ATOM_STATE_COMPLETED &&
!kbase_ctx_flag(kctx, KCTX_DYING)) {
+ KBASE_TLSTREAM_TL_RUN_ATOM_START(
+ kctx->kbdev, node,
+ kbase_jd_atom_id(kctx, node));
need_to_try_schedule_context |= jd_run_atom(node);
+ KBASE_TLSTREAM_TL_RUN_ATOM_END(
+ kctx->kbdev, node,
+ kbase_jd_atom_id(kctx, node));
} else {
node->event_code = katom->event_code;
@@ -795,10 +628,10 @@
* is in a disjoint state (ie. being reset).
*/
kbase_disjoint_event_potential(kctx->kbdev);
- if (completed_jobs_ctx)
- list_add_tail(&katom->jd_item, completed_jobs_ctx);
- else
+ if (post_immediately && list_empty(&kctx->completed_jobs))
kbase_event_post(kctx, katom);
+ else
+ list_add_tail(&katom->jd_item, &kctx->completed_jobs);
/* Decrement and check the TOTAL number of jobs. This includes
* those not tracked by the scheduler: 'not ready to run' and
@@ -811,11 +644,11 @@
*/
wake_up(&kctx->jctx.zero_jobs_wait);
}
-
+ KBASE_TLSTREAM_TL_JD_DONE_NO_LOCK_END(kctx->kbdev, katom);
return need_to_try_schedule_context;
}
-KBASE_EXPORT_TEST_API(jd_done_nolock);
+KBASE_EXPORT_TEST_API(kbase_jd_done_nolock);
#if IS_ENABLED(CONFIG_GPU_TRACEPOINTS)
enum {
@@ -923,7 +756,6 @@
katom->jobslot = user_atom->jobslot;
katom->seq_nr = user_atom->seq_nr;
katom->atom_flags = 0;
- katom->retry_count = 0;
katom->need_cache_flush_cores_retained = 0;
katom->pre_dep = NULL;
katom->post_dep = NULL;
@@ -957,14 +789,11 @@
INIT_LIST_HEAD(&katom->queue);
INIT_LIST_HEAD(&katom->jd_item);
-#ifdef CONFIG_MALI_BIFROST_DMA_FENCE
- kbase_fence_dep_count_set(katom, -1);
-#endif
/* Don't do anything if there is a mess up with dependencies.
- This is done in a separate cycle to check both the dependencies at ones, otherwise
- it will be extra complexity to deal with 1st dependency ( just added to the list )
- if only the 2nd one has invalid config.
+ * This is done in a separate cycle to check both the dependencies at ones, otherwise
+ * it will be extra complexity to deal with 1st dependency ( just added to the list )
+ * if only the 2nd one has invalid config.
*/
for (i = 0; i < 2; i++) {
int dep_atom_number = user_atom->pre_dep[i].atom_id;
@@ -984,8 +813,7 @@
* dependencies.
*/
jd_trace_atom_submit(kctx, katom, NULL);
-
- return jd_done_nolock(katom, NULL);
+ return kbase_jd_done_nolock(katom, true);
}
}
}
@@ -1049,8 +877,7 @@
if (err >= 0)
kbase_finish_soft_job(katom);
}
-
- return jd_done_nolock(katom, NULL);
+ return kbase_jd_done_nolock(katom, true);
}
katom->will_fail_event_code = katom->event_code;
@@ -1076,7 +903,7 @@
/* Create a new atom. */
jd_trace_atom_submit(kctx, katom, &katom->sched_priority);
-#if !MALI_INCREMENTAL_RENDERING
+#if !MALI_INCREMENTAL_RENDERING_JM
/* Reject atoms for incremental rendering if not supported */
if (katom->core_req &
(BASE_JD_REQ_START_RENDERPASS|BASE_JD_REQ_END_RENDERPASS)) {
@@ -1084,9 +911,9 @@
"Rejecting atom with unsupported core_req 0x%x\n",
katom->core_req);
katom->event_code = BASE_JD_EVENT_JOB_INVALID;
- return jd_done_nolock(katom, NULL);
+ return kbase_jd_done_nolock(katom, true);
}
-#endif /* !MALI_INCREMENTAL_RENDERING */
+#endif /* !MALI_INCREMENTAL_RENDERING_JM */
if (katom->core_req & BASE_JD_REQ_END_RENDERPASS) {
WARN_ON(katom->jc != 0);
@@ -1098,7 +925,7 @@
*/
dev_err(kctx->kbdev->dev, "Rejecting atom with jc = NULL\n");
katom->event_code = BASE_JD_EVENT_JOB_INVALID;
- return jd_done_nolock(katom, NULL);
+ return kbase_jd_done_nolock(katom, true);
}
/* Reject atoms with an invalid device_nr */
@@ -1108,7 +935,7 @@
"Rejecting atom with invalid device_nr %d\n",
katom->device_nr);
katom->event_code = BASE_JD_EVENT_JOB_INVALID;
- return jd_done_nolock(katom, NULL);
+ return kbase_jd_done_nolock(katom, true);
}
/* Reject atoms with invalid core requirements */
@@ -1118,7 +945,7 @@
"Rejecting atom with invalid core requirements\n");
katom->event_code = BASE_JD_EVENT_JOB_INVALID;
katom->core_req &= ~BASE_JD_REQ_EVENT_COALESCE;
- return jd_done_nolock(katom, NULL);
+ return kbase_jd_done_nolock(katom, true);
}
/* Reject soft-job atom of certain types from accessing external resources */
@@ -1129,7 +956,7 @@
dev_err(kctx->kbdev->dev,
"Rejecting soft-job atom accessing external resources\n");
katom->event_code = BASE_JD_EVENT_JOB_INVALID;
- return jd_done_nolock(katom, NULL);
+ return kbase_jd_done_nolock(katom, true);
}
if (katom->core_req & BASE_JD_REQ_EXTERNAL_RESOURCES) {
@@ -1137,7 +964,7 @@
if (kbase_jd_pre_external_resources(katom, user_atom) != 0) {
/* setup failed (no access, bad resource, unknown resource types, etc.) */
katom->event_code = BASE_JD_EVENT_JOB_INVALID;
- return jd_done_nolock(katom, NULL);
+ return kbase_jd_done_nolock(katom, true);
}
}
@@ -1148,7 +975,7 @@
* JIT IDs - atom is invalid.
*/
katom->event_code = BASE_JD_EVENT_JOB_INVALID;
- return jd_done_nolock(katom, NULL);
+ return kbase_jd_done_nolock(katom, true);
}
#endif /* MALI_JIT_PRESSURE_LIMIT_BASE */
@@ -1162,13 +989,13 @@
if ((katom->core_req & BASE_JD_REQ_SOFT_JOB) == 0) {
if (!kbase_js_is_atom_valid(kctx->kbdev, katom)) {
katom->event_code = BASE_JD_EVENT_JOB_INVALID;
- return jd_done_nolock(katom, NULL);
+ return kbase_jd_done_nolock(katom, true);
}
} else {
/* Soft-job */
if (kbase_prepare_soft_job(katom) != 0) {
katom->event_code = BASE_JD_EVENT_JOB_INVALID;
- return jd_done_nolock(katom, NULL);
+ return kbase_jd_done_nolock(katom, true);
}
}
@@ -1181,16 +1008,10 @@
if (queued && !IS_GPU_ATOM(katom))
return false;
-#ifdef CONFIG_MALI_BIFROST_DMA_FENCE
- if (kbase_fence_dep_count_read(katom) != -1)
- return false;
-
-#endif /* CONFIG_MALI_BIFROST_DMA_FENCE */
-
if (katom->core_req & BASE_JD_REQ_SOFT_JOB) {
if (kbase_process_soft_job(katom) == 0) {
kbase_finish_soft_job(katom);
- return jd_done_nolock(katom, NULL);
+ return kbase_jd_done_nolock(katom, true);
}
return false;
}
@@ -1220,7 +1041,7 @@
}
/* This is a pure dependency. Resolve it immediately */
- return jd_done_nolock(katom, NULL);
+ return kbase_jd_done_nolock(katom, true);
}
int kbase_jd_submit(struct kbase_context *kctx,
@@ -1235,15 +1056,13 @@
u32 latest_flush;
bool jd_atom_is_v2 = (stride == sizeof(struct base_jd_atom_v2) ||
- stride == offsetof(struct base_jd_atom_v2, renderpass_id));
+ stride == offsetof(struct base_jd_atom_v2, renderpass_id));
/*
* kbase_jd_submit isn't expected to fail and so all errors with the
* jobs are reported by immediately failing them (through event system)
*/
kbdev = kctx->kbdev;
-
- beenthere(kctx, "%s", "Enter");
if (kbase_ctx_flag(kctx, KCTX_SUBMIT_DISABLED)) {
dev_err(kbdev->dev, "Attempt to submit to a context that has SUBMIT_DISABLED set on it\n");
@@ -1260,18 +1079,26 @@
return -EINVAL;
}
+ if (nr_atoms > BASE_JD_ATOM_COUNT) {
+ dev_dbg(kbdev->dev, "Invalid attempt to submit %u atoms at once for kctx %d_%d",
+ nr_atoms, kctx->tgid, kctx->id);
+ return -EINVAL;
+ }
+
/* All atoms submitted in this call have the same flush ID */
latest_flush = kbase_backend_get_current_flush_id(kbdev);
for (i = 0; i < nr_atoms; i++) {
- struct base_jd_atom user_atom;
+ struct base_jd_atom user_atom = {
+ .seq_nr = 0,
+ };
struct base_jd_fragment user_jc_incr;
struct kbase_jd_atom *katom;
if (unlikely(jd_atom_is_v2)) {
if (copy_from_user(&user_atom.jc, user_addr, sizeof(struct base_jd_atom_v2)) != 0) {
dev_dbg(kbdev->dev,
- "Invalid atom address %p passed to job_submit\n",
+ "Invalid atom address %pK passed to job_submit\n",
user_addr);
err = -EFAULT;
break;
@@ -1282,7 +1109,7 @@
} else {
if (copy_from_user(&user_atom, user_addr, stride) != 0) {
dev_dbg(kbdev->dev,
- "Invalid atom address %p passed to job_submit\n",
+ "Invalid atom address %pK passed to job_submit\n",
user_addr);
err = -EFAULT;
break;
@@ -1378,16 +1205,22 @@
}
mutex_lock(&jctx->lock);
}
-
+ KBASE_TLSTREAM_TL_JD_SUBMIT_ATOM_START(kbdev, katom);
need_to_try_schedule_context |= jd_submit_atom(kctx, &user_atom,
&user_jc_incr, katom);
-
+ KBASE_TLSTREAM_TL_JD_SUBMIT_ATOM_END(kbdev, katom);
/* Register a completed job as a disjoint event when the GPU is in a disjoint state
* (ie. being reset).
*/
kbase_disjoint_event_potential(kbdev);
mutex_unlock(&jctx->lock);
+ if (fatal_signal_pending(current)) {
+ dev_dbg(kbdev->dev, "Fatal signal pending for kctx %d_%d",
+ kctx->tgid, kctx->id);
+ /* We're being killed so the result code doesn't really matter */
+ return 0;
+ }
}
if (need_to_try_schedule_context)
@@ -1463,11 +1296,13 @@
}
if ((katom->event_code != BASE_JD_EVENT_DONE) &&
- (!kbase_ctx_flag(katom->kctx, KCTX_DYING)))
- dev_err(kbdev->dev,
- "t6xx: GPU fault 0x%02lx from job slot %d\n",
- (unsigned long)katom->event_code,
- katom->slot_nr);
+ (!kbase_ctx_flag(katom->kctx, KCTX_DYING))) {
+ if (!kbase_is_quick_reset_enabled(kbdev))
+ dev_err(kbdev->dev,
+ "t6xx: GPU fault 0x%02lx from job slot %d\n",
+ (unsigned long)katom->event_code,
+ katom->slot_nr);
+ }
/* Retain state before the katom disappears */
kbasep_js_atom_retained_state_copy(&katom_retained_state, katom);
@@ -1479,9 +1314,8 @@
kbasep_js_remove_job(kbdev, kctx, katom);
mutex_unlock(&js_kctx_info->ctx.jsctx_mutex);
mutex_unlock(&js_devdata->queue_mutex);
- katom->atom_flags &= ~KBASE_KATOM_FLAG_HOLDING_CTX_REF;
- /* jd_done_nolock() requires the jsctx_mutex lock to be dropped */
- jd_done_nolock(katom, &kctx->completed_jobs);
+ /* kbase_jd_done_nolock() requires the jsctx_mutex lock to be dropped */
+ kbase_jd_done_nolock(katom, false);
/* katom may have been freed now, do not use! */
@@ -1498,22 +1332,23 @@
* drop our reference. But do not call kbase_jm_idle_ctx(), as
* the context is active and fast-starting is allowed.
*
- * If an atom has been fast-started then kctx->atoms_pulled will
- * be non-zero but KCTX_ACTIVE will still be false (as the
- * previous pm reference has been inherited). Do NOT drop our
- * reference, as it has been re-used, and leave the context as
- * active.
+ * If an atom has been fast-started then
+ * kbase_jsctx_atoms_pulled(kctx) will return non-zero but
+ * KCTX_ACTIVE will still be false (as the previous pm
+ * reference has been inherited). Do NOT drop our reference, as
+ * it has been re-used, and leave the context as active.
*
- * If no new atoms have been started then KCTX_ACTIVE will still
- * be false and atoms_pulled will be zero, so drop the reference
- * and call kbase_jm_idle_ctx().
+ * If no new atoms have been started then KCTX_ACTIVE will
+ * still be false and kbase_jsctx_atoms_pulled(kctx) will
+ * return zero, so drop the reference and call
+ * kbase_jm_idle_ctx().
*
* As the checks are done under both the queue_mutex and
* hwaccess_lock is should be impossible for this to race
* with the scheduler code.
*/
if (kbase_ctx_flag(kctx, KCTX_ACTIVE) ||
- !atomic_read(&kctx->atoms_pulled)) {
+ !kbase_jsctx_atoms_pulled(kctx)) {
/* Calling kbase_jm_idle_ctx() here will ensure that
* atoms are not fast-started when we drop the
* hwaccess_lock. This is not performed if
@@ -1546,7 +1381,7 @@
kbase_js_sched_all(kbdev);
if (!atomic_dec_return(&kctx->work_count)) {
- /* If worker now idle then post all events that jd_done_nolock()
+ /* If worker now idle then post all events that kbase_jd_done_nolock()
* has queued
*/
mutex_lock(&jctx->lock);
@@ -1595,6 +1430,7 @@
bool need_to_try_schedule_context;
bool attr_state_changed;
struct kbase_device *kbdev;
+ CSTD_UNUSED(need_to_try_schedule_context);
/* Soft jobs should never reach this function */
KBASE_DEBUG_ASSERT((katom->core_req & BASE_JD_REQ_SOFT_JOB) == 0);
@@ -1620,7 +1456,7 @@
mutex_lock(&jctx->lock);
- need_to_try_schedule_context = jd_done_nolock(katom, NULL);
+ need_to_try_schedule_context = kbase_jd_done_nolock(katom, true);
/* Because we're zapping, we're not adding any more jobs to this ctx, so no need to
* schedule the context. There's also no need for the jsctx_mutex to have been taken
* around this too.
@@ -1663,6 +1499,8 @@
KBASE_DEBUG_ASSERT(kctx);
kbdev = kctx->kbdev;
KBASE_DEBUG_ASSERT(kbdev);
+
+ lockdep_assert_held(&kbdev->hwaccess_lock);
if (done_code & KBASE_JS_ATOM_DONE_EVICTED_FROM_NEXT)
katom->event_code = BASE_JD_EVENT_REMOVED_FROM_NEXT;
@@ -1740,19 +1578,7 @@
kbase_cancel_soft_job(katom);
}
-
-#ifdef CONFIG_MALI_BIFROST_DMA_FENCE
- kbase_dma_fence_cancel_all_atoms(kctx);
-#endif
-
mutex_unlock(&kctx->jctx.lock);
-
-#ifdef CONFIG_MALI_BIFROST_DMA_FENCE
- /* Flush dma-fence workqueue to ensure that any callbacks that may have
- * been queued are done before continuing.
- */
- flush_workqueue(kctx->dma_fence.wq);
-#endif
#if IS_ENABLED(CONFIG_DEBUG_FS)
kbase_debug_job_fault_kctx_unblock(kctx);
@@ -1790,11 +1616,10 @@
kctx->jctx.atoms[i].event_code = BASE_JD_EVENT_JOB_INVALID;
kctx->jctx.atoms[i].status = KBASE_JD_ATOM_STATE_UNUSED;
-#if defined(CONFIG_MALI_BIFROST_DMA_FENCE) || defined(CONFIG_SYNC_FILE)
+#if IS_ENABLED(CONFIG_SYNC_FILE)
kctx->jctx.atoms[i].dma_fence.context =
dma_fence_context_alloc(1);
atomic_set(&kctx->jctx.atoms[i].dma_fence.seqno, 0);
- INIT_LIST_HEAD(&kctx->jctx.atoms[i].dma_fence.callbacks);
#endif
}
--
Gitblit v1.6.2