aboutsummaryrefslogtreecommitdiffstats
path: root/meta-amd-bsp/recipes-kernel/linux/linux-yocto-4.14.71/1499-drm-amd-Simplify-SDMA-HQD-loading-in-kernel-threads.patch
diff options
context:
space:
mode:
Diffstat (limited to 'meta-amd-bsp/recipes-kernel/linux/linux-yocto-4.14.71/1499-drm-amd-Simplify-SDMA-HQD-loading-in-kernel-threads.patch')
-rw-r--r--meta-amd-bsp/recipes-kernel/linux/linux-yocto-4.14.71/1499-drm-amd-Simplify-SDMA-HQD-loading-in-kernel-threads.patch153
1 files changed, 153 insertions, 0 deletions
diff --git a/meta-amd-bsp/recipes-kernel/linux/linux-yocto-4.14.71/1499-drm-amd-Simplify-SDMA-HQD-loading-in-kernel-threads.patch b/meta-amd-bsp/recipes-kernel/linux/linux-yocto-4.14.71/1499-drm-amd-Simplify-SDMA-HQD-loading-in-kernel-threads.patch
new file mode 100644
index 00000000..8671234c
--- /dev/null
+++ b/meta-amd-bsp/recipes-kernel/linux/linux-yocto-4.14.71/1499-drm-amd-Simplify-SDMA-HQD-loading-in-kernel-threads.patch
@@ -0,0 +1,153 @@
+From 9c8870b8587c2de80f8d4981e65dd4cc67c3500f Mon Sep 17 00:00:00 2001
+From: Felix Kuehling <Felix.Kuehling@amd.com>
+Date: Thu, 1 Sep 2016 21:11:39 -0400
+Subject: [PATCH 1499/4131] drm/amd: Simplify SDMA HQD loading in kernel
+ threads
+
+Use the use_mm function to temporarily access a user process context
+while running in a kernel thread. This avoids the need to lock the
+mmap_sem, which fixes a circular locking problem observed with user
+pointers while running without the HW scheduler.
+
+Change-Id: I4e74f3b021277e09df55dddf81ae5e070032b599
+Signed-off-by: Felix Kuehling <Felix.Kuehling@amd.com>
+---
+ drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_gfx_v7.c | 26 ++++++++++++----------
+ drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_gfx_v8.c | 24 +++++++++++---------
+ drivers/gpu/drm/amd/amdkfd/kfd_device.c | 9 +-------
+ .../gpu/drm/amd/amdkfd/kfd_device_queue_manager.c | 4 ----
+ 4 files changed, 28 insertions(+), 35 deletions(-)
+
+diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_gfx_v7.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_gfx_v7.c
+index 9948b37..22c4b7a 100644
+--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_gfx_v7.c
++++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_gfx_v7.c
+@@ -23,6 +23,7 @@
+ #include <linux/fdtable.h>
+ #include <linux/uaccess.h>
+ #include <linux/firmware.h>
++#include <linux/mmu_context.h>
+ #include <drm/drmP.h>
+ #include "amdgpu.h"
+ #include "amdgpu_amdkfd.h"
+@@ -528,18 +529,19 @@ static int kgd_hqd_sdma_load(struct kgd_dev *kgd, void *mqd,
+ WREG32(sdma_base_addr + mmSDMA0_RLC0_DOORBELL, m->sdma_rlc_doorbell);
+ WREG32(sdma_base_addr + mmSDMA0_RLC0_RB_RPTR, m->sdma_rlc_rb_rptr);
+
+- if (mm && mm == current->mm)
+- wptr_valid = !get_user(data, wptr);
+- else if (mm) {
+- struct vm_area_struct *vma;
+-
+- vma = find_vma(mm, (unsigned long)wptr);
+- if (vma && vma->vm_start <= (unsigned long)wptr &&
+- vma->vm_ops && vma->vm_ops->access)
+- wptr_valid = (sizeof(data) == vma->vm_ops->access(
+- vma, (unsigned long)wptr,
+- &data, sizeof(data), 0));
+- }
++ if (mm) {
++ if (mm == current->mm) {
++ /* Running in the correct user process context */
++ wptr_valid = !get_user(data, wptr);
++ } else if (current->mm == NULL) {
++ /* A kernel thread can temporarily use a user
++ * process context for AIO
++ */
++ use_mm(mm);
++ wptr_valid = !get_user(data, wptr);
++ unuse_mm(mm);
++ }
++
+ if (wptr_valid)
+ WREG32(sdma_base_addr + mmSDMA0_RLC0_RB_WPTR, data);
+ else
+diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_gfx_v8.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_gfx_v8.c
+index 56a88ef..7be32f3 100644
+--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_gfx_v8.c
++++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_gfx_v8.c
+@@ -24,6 +24,7 @@
+ #include <linux/fdtable.h>
+ #include <linux/uaccess.h>
+ #include <linux/firmware.h>
++#include <linux/mmu_context.h>
+ #include <drm/drmP.h>
+ #include "amdgpu.h"
+ #include "amdgpu_amdkfd.h"
+@@ -524,17 +525,18 @@ static int kgd_hqd_sdma_load(struct kgd_dev *kgd, void *mqd,
+ WREG32(sdma_base_addr + mmSDMA0_RLC0_DOORBELL, m->sdmax_rlcx_doorbell);
+ WREG32(sdma_base_addr + mmSDMA0_RLC0_RB_RPTR, m->sdmax_rlcx_rb_rptr);
+
+- if (mm && mm == current->mm)
+- wptr_valid = !get_user(data, wptr);
+- else if (mm) {
+- struct vm_area_struct *vma;
+-
+- vma = find_vma(mm, (unsigned long)wptr);
+- if (vma && vma->vm_start <= (unsigned long)wptr &&
+- vma->vm_ops && vma->vm_ops->access)
+- wptr_valid = (sizeof(data) == vma->vm_ops->access(
+- vma, (unsigned long)wptr,
+- &data, sizeof(data), 0));
++ if (mm) {
++ if (mm == current->mm) {
++ /* Running in the correct user process context */
++ wptr_valid = !get_user(data, wptr);
++ } else if (current->mm == NULL) {
++ /* A kernel thread can temporarily use a user
++ * process context for AIO
++ */
++ use_mm(mm);
++ wptr_valid = !get_user(data, wptr);
++ unuse_mm(mm);
++ }
+ }
+ if (wptr_valid)
+ WREG32(sdma_base_addr + mmSDMA0_RLC0_RB_WPTR, data);
+diff --git a/drivers/gpu/drm/amd/amdkfd/kfd_device.c b/drivers/gpu/drm/amd/amdkfd/kfd_device.c
+index 69b08a9..a47d7f1 100644
+--- a/drivers/gpu/drm/amd/amdkfd/kfd_device.c
++++ b/drivers/gpu/drm/amd/amdkfd/kfd_device.c
+@@ -738,16 +738,9 @@ int kgd2kfd_resume_mm(struct kfd_dev *kfd, struct mm_struct *mm)
+
+ r = -ENODEV;
+ pdd = kfd_get_process_device_data(kfd, p);
+- if (pdd) {
+- if (kfd->dqm->sched_policy == KFD_SCHED_POLICY_NO_HWS)
+- down_read(&mm->mmap_sem);
+-
++ if (pdd)
+ r = process_restore_queues(kfd->dqm, &pdd->qpd);
+
+- if (kfd->dqm->sched_policy == KFD_SCHED_POLICY_NO_HWS)
+- up_read(&mm->mmap_sem);
+- }
+-
+ kfd_unref_process(p);
+ return r;
+ }
+diff --git a/drivers/gpu/drm/amd/amdkfd/kfd_device_queue_manager.c b/drivers/gpu/drm/amd/amdkfd/kfd_device_queue_manager.c
+index 53fef02..d77c1ad 100644
+--- a/drivers/gpu/drm/amd/amdkfd/kfd_device_queue_manager.c
++++ b/drivers/gpu/drm/amd/amdkfd/kfd_device_queue_manager.c
+@@ -408,8 +408,6 @@ static int update_queue(struct device_queue_manager *dqm, struct queue *q)
+
+ BUG_ON(!dqm || !q || !q->mqd);
+
+- if (dqm->sched_policy == KFD_SCHED_POLICY_NO_HWS)
+- down_read(&current->mm->mmap_sem);
+ mutex_lock(&dqm->lock);
+
+ pdd = kfd_get_process_device_data(q->device, q->process);
+@@ -465,8 +463,6 @@ static int update_queue(struct device_queue_manager *dqm, struct queue *q)
+
+ out_unlock:
+ mutex_unlock(&dqm->lock);
+- if (dqm->sched_policy == KFD_SCHED_POLICY_NO_HWS)
+- up_read(&current->mm->mmap_sem);
+
+ return retval;
+ }
+--
+2.7.4
+
generated by cgit 1.2.3-korg (git 2.39.0) at 2024-04-20 05:51:41 +0000