aboutsummaryrefslogtreecommitdiffstatshomepage
path: root/drivers/gpu
diff options
context:
space:
mode:
authorMukul Joshi <mukul.joshi@amd.com>2020-06-24 13:47:12 -0400
committerAlex Deucher <alexander.deucher@amd.com>2020-07-01 01:59:27 -0400
commitd69fd951e60ae48c8dd100aa8ceb799ab965f9b3 (patch)
treee9a5d66a0ef9da31bc686693d5b2eaf31a41b3b1 /drivers/gpu
parentdrm/radeon: fix array out-of-bounds read and write issues (diff)
downloadwireguard-linux-d69fd951e60ae48c8dd100aa8ceb799ab965f9b3.tar.xz
wireguard-linux-d69fd951e60ae48c8dd100aa8ceb799ab965f9b3.zip
drm/amdkfd: Fix circular locking dependency warning
[ 150.887733] ====================================================== [ 150.893903] WARNING: possible circular locking dependency detected [ 150.905917] ------------------------------------------------------ [ 150.912129] kfdtest/4081 is trying to acquire lock: [ 150.917002] ffff8f7f3762e118 (&mm->mmap_sem#2){++++}, at: __might_fault+0x3e/0x90 [ 150.924490] but task is already holding lock: [ 150.930320] ffff8f7f49d229e8 (&dqm->lock_hidden){+.+.}, at: destroy_queue_cpsch+0x29/0x210 [amdgpu] [ 150.939432] which lock already depends on the new lock. [ 150.947603] the existing dependency chain (in reverse order) is: [ 150.955074] -> #3 (&dqm->lock_hidden){+.+.}: [ 150.960822] __mutex_lock+0xa1/0x9f0 [ 150.964996] evict_process_queues_cpsch+0x22/0x120 [amdgpu] [ 150.971155] kfd_process_evict_queues+0x3b/0xc0 [amdgpu] [ 150.977054] kgd2kfd_quiesce_mm+0x25/0x60 [amdgpu] [ 150.982442] amdgpu_amdkfd_evict_userptr+0x35/0x70 [amdgpu] [ 150.988615] amdgpu_mn_invalidate_hsa+0x41/0x60 [amdgpu] [ 150.994448] __mmu_notifier_invalidate_range_start+0xa4/0x240 [ 151.000714] copy_page_range+0xd70/0xd80 [ 151.005159] dup_mm+0x3ca/0x550 [ 151.008816] copy_process+0x1bdc/0x1c70 [ 151.013183] _do_fork+0x76/0x6c0 [ 151.016929] __x64_sys_clone+0x8c/0xb0 [ 151.021201] do_syscall_64+0x4a/0x1d0 [ 151.025404] entry_SYSCALL_64_after_hwframe+0x49/0xbe [ 151.030977] -> #2 (&adev->notifier_lock){+.+.}: [ 151.036993] __mutex_lock+0xa1/0x9f0 [ 151.041168] amdgpu_mn_invalidate_hsa+0x30/0x60 [amdgpu] [ 151.047019] __mmu_notifier_invalidate_range_start+0xa4/0x240 [ 151.053277] copy_page_range+0xd70/0xd80 [ 151.057722] dup_mm+0x3ca/0x550 [ 151.061388] copy_process+0x1bdc/0x1c70 [ 151.065748] _do_fork+0x76/0x6c0 [ 151.069499] __x64_sys_clone+0x8c/0xb0 [ 151.073765] do_syscall_64+0x4a/0x1d0 [ 151.077952] entry_SYSCALL_64_after_hwframe+0x49/0xbe [ 151.083523] -> #1 (mmu_notifier_invalidate_range_start){+.+.}: [ 151.090833] change_protection+0x802/0xab0 [ 151.095448] mprotect_fixup+0x187/0x2d0 [ 151.099801] setup_arg_pages+0x124/0x250 [ 151.104251] load_elf_binary+0x3a4/0x1464 [ 151.108781] search_binary_handler+0x6c/0x210 [ 151.113656] __do_execve_file.isra.40+0x7f7/0xa50 [ 151.118875] do_execve+0x21/0x30 [ 151.122632] call_usermodehelper_exec_async+0x17e/0x190 [ 151.128393] ret_from_fork+0x24/0x30 [ 151.132489] -> #0 (&mm->mmap_sem#2){++++}: [ 151.138064] __lock_acquire+0x11a1/0x1490 [ 151.142597] lock_acquire+0x90/0x180 [ 151.146694] __might_fault+0x68/0x90 [ 151.150879] read_sdma_queue_counter+0x5f/0xb0 [amdgpu] [ 151.156693] update_sdma_queue_past_activity_stats+0x3b/0x90 [amdgpu] [ 151.163725] destroy_queue_cpsch+0x1ae/0x210 [amdgpu] [ 151.169373] pqm_destroy_queue+0xf0/0x250 [amdgpu] [ 151.174762] kfd_ioctl_destroy_queue+0x32/0x70 [amdgpu] [ 151.180577] kfd_ioctl+0x223/0x400 [amdgpu] [ 151.185284] ksys_ioctl+0x8f/0xb0 [ 151.189118] __x64_sys_ioctl+0x16/0x20 [ 151.193389] do_syscall_64+0x4a/0x1d0 [ 151.197569] entry_SYSCALL_64_after_hwframe+0x49/0xbe [ 151.203141] other info that might help us debug this: [ 151.211140] Chain exists of: &mm->mmap_sem#2 --> &adev->notifier_lock --> &dqm->lock_hidden [ 151.222535] Possible unsafe locking scenario: [ 151.228447] CPU0 CPU1 [ 151.232971] ---- ---- [ 151.237502] lock(&dqm->lock_hidden); [ 151.241254] lock(&adev->notifier_lock); [ 151.247774] lock(&dqm->lock_hidden); [ 151.254038] lock(&mm->mmap_sem#2); This commit fixes the warning by ensuring get_user() is not called while reading SDMA stats with dqm_lock held as get_user() could cause a page fault which leads to the circular locking scenario. Signed-off-by: Mukul Joshi <mukul.joshi@amd.com> Reviewed-by: Felix Kuehling <Felix.Kuehling@amd.com> Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
Diffstat (limited to 'drivers/gpu')
-rw-r--r--drivers/gpu/drm/amd/amdkfd/kfd_device_queue_manager.c73
-rw-r--r--drivers/gpu/drm/amd/amdkfd/kfd_device_queue_manager.h3
-rw-r--r--drivers/gpu/drm/amd/amdkfd/kfd_process.c143
3 files changed, 158 insertions, 61 deletions
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 21eb0998c4ae..dd550025d1c1 100644
--- a/drivers/gpu/drm/amd/amdkfd/kfd_device_queue_manager.c
+++ b/drivers/gpu/drm/amd/amdkfd/kfd_device_queue_manager.c
@@ -153,24 +153,23 @@ static void decrement_queue_count(struct device_queue_manager *dqm,
dqm->active_cp_queue_count--;
}
-int read_sdma_queue_counter(struct queue *q, uint64_t *val)
+int read_sdma_queue_counter(uint64_t q_rptr, uint64_t *val)
{
int ret;
uint64_t tmp = 0;
- if (!q || !val)
+ if (!val)
return -EINVAL;
/*
* SDMA activity counter is stored at queue's RPTR + 0x8 location.
*/
- if (!access_ok((const void __user *)((uint64_t)q->properties.read_ptr +
+ if (!access_ok((const void __user *)(q_rptr +
sizeof(uint64_t)), sizeof(uint64_t))) {
pr_err("Can't access sdma queue activity counter\n");
return -EFAULT;
}
- ret = get_user(tmp, (uint64_t *)((uint64_t)(q->properties.read_ptr) +
- sizeof(uint64_t)));
+ ret = get_user(tmp, (uint64_t *)(q_rptr + sizeof(uint64_t)));
if (!ret) {
*val = tmp;
}
@@ -178,27 +177,6 @@ int read_sdma_queue_counter(struct queue *q, uint64_t *val)
return ret;
}
-static int update_sdma_queue_past_activity_stats(struct kfd_process_device *pdd,
- struct queue *q)
-{
- int ret;
- uint64_t val = 0;
-
- if (!pdd)
- return -ENODEV;
-
- ret = read_sdma_queue_counter(q, &val);
- if (ret) {
- pr_err("Failed to read SDMA queue counter for queue: %d\n",
- q->properties.queue_id);
- return ret;
- }
-
- pdd->sdma_past_activity_counter += val;
-
- return ret;
-}
-
static int allocate_doorbell(struct qcm_process_device *qpd, struct queue *q)
{
struct kfd_dev *dev = qpd->dqm->dev;
@@ -533,11 +511,6 @@ static int destroy_queue_nocpsch_locked(struct device_queue_manager *dqm,
if (retval == -ETIME)
qpd->reset_wavefronts = true;
- /* Get the SDMA queue stats */
- if ((q->properties.type == KFD_QUEUE_TYPE_SDMA) ||
- (q->properties.type == KFD_QUEUE_TYPE_SDMA_XGMI)) {
- update_sdma_queue_past_activity_stats(qpd_to_pdd(qpd), q);
- }
mqd_mgr->free_mqd(mqd_mgr, q->mqd, q->mqd_mem_obj);
@@ -573,9 +546,23 @@ static int destroy_queue_nocpsch(struct device_queue_manager *dqm,
struct queue *q)
{
int retval;
+ uint64_t sdma_val = 0;
+ struct kfd_process_device *pdd = qpd_to_pdd(qpd);
+
+ /* Get the SDMA queue stats */
+ if ((q->properties.type == KFD_QUEUE_TYPE_SDMA) ||
+ (q->properties.type == KFD_QUEUE_TYPE_SDMA_XGMI)) {
+ retval = read_sdma_queue_counter((uint64_t)q->properties.read_ptr,
+ &sdma_val);
+ if (retval)
+ pr_err("Failed to read SDMA queue counter for queue: %d\n",
+ q->properties.queue_id);
+ }
dqm_lock(dqm);
retval = destroy_queue_nocpsch_locked(dqm, qpd, q);
+ if (!retval)
+ pdd->sdma_past_activity_counter += sdma_val;
dqm_unlock(dqm);
return retval;
@@ -1480,6 +1467,18 @@ static int destroy_queue_cpsch(struct device_queue_manager *dqm,
{
int retval;
struct mqd_manager *mqd_mgr;
+ uint64_t sdma_val = 0;
+ struct kfd_process_device *pdd = qpd_to_pdd(qpd);
+
+ /* Get the SDMA queue stats */
+ if ((q->properties.type == KFD_QUEUE_TYPE_SDMA) ||
+ (q->properties.type == KFD_QUEUE_TYPE_SDMA_XGMI)) {
+ retval = read_sdma_queue_counter((uint64_t)q->properties.read_ptr,
+ &sdma_val);
+ if (retval)
+ pr_err("Failed to read SDMA queue counter for queue: %d\n",
+ q->properties.queue_id);
+ }
retval = 0;
@@ -1501,10 +1500,11 @@ static int destroy_queue_cpsch(struct device_queue_manager *dqm,
deallocate_doorbell(qpd, q);
- if (q->properties.type == KFD_QUEUE_TYPE_SDMA)
- deallocate_sdma_queue(dqm, q);
- else if (q->properties.type == KFD_QUEUE_TYPE_SDMA_XGMI)
+ if ((q->properties.type == KFD_QUEUE_TYPE_SDMA) ||
+ (q->properties.type == KFD_QUEUE_TYPE_SDMA_XGMI)) {
deallocate_sdma_queue(dqm, q);
+ pdd->sdma_past_activity_counter += sdma_val;
+ }
list_del(&q->list);
qpd->queue_count--;
@@ -1520,11 +1520,6 @@ static int destroy_queue_cpsch(struct device_queue_manager *dqm,
}
}
- /* Get the SDMA queue stats */
- if ((q->properties.type == KFD_QUEUE_TYPE_SDMA) ||
- (q->properties.type == KFD_QUEUE_TYPE_SDMA_XGMI)) {
- update_sdma_queue_past_activity_stats(qpd_to_pdd(qpd), q);
- }
/*
* Unconditionally decrement this counter, regardless of the queue's
* type
diff --git a/drivers/gpu/drm/amd/amdkfd/kfd_device_queue_manager.h b/drivers/gpu/drm/amd/amdkfd/kfd_device_queue_manager.h
index 894bcf877f9e..49d8e324c636 100644
--- a/drivers/gpu/drm/amd/amdkfd/kfd_device_queue_manager.h
+++ b/drivers/gpu/drm/amd/amdkfd/kfd_device_queue_manager.h
@@ -251,6 +251,5 @@ static inline void dqm_unlock(struct device_queue_manager *dqm)
mutex_unlock(&dqm->lock_hidden);
}
-int read_sdma_queue_counter(struct queue *q, uint64_t *val);
-
+int read_sdma_queue_counter(uint64_t q_rptr, uint64_t *val);
#endif /* KFD_DEVICE_QUEUE_MANAGER_H_ */
diff --git a/drivers/gpu/drm/amd/amdkfd/kfd_process.c b/drivers/gpu/drm/amd/amdkfd/kfd_process.c
index d0409df35032..013c2b018edc 100644
--- a/drivers/gpu/drm/amd/amdkfd/kfd_process.c
+++ b/drivers/gpu/drm/amd/amdkfd/kfd_process.c
@@ -86,6 +86,13 @@ struct kfd_sdma_activity_handler_workarea {
uint64_t sdma_activity_counter;
};
+struct temp_sdma_queue_list {
+ uint64_t rptr;
+ uint64_t sdma_val;
+ unsigned int queue_id;
+ struct list_head list;
+};
+
static void kfd_sdma_activity_worker(struct work_struct *work)
{
struct kfd_sdma_activity_handler_workarea *workarea;
@@ -96,6 +103,8 @@ static void kfd_sdma_activity_worker(struct work_struct *work)
struct qcm_process_device *qpd;
struct device_queue_manager *dqm;
int ret = 0;
+ struct temp_sdma_queue_list sdma_q_list;
+ struct temp_sdma_queue_list *sdma_q, *next;
workarea = container_of(work, struct kfd_sdma_activity_handler_workarea,
sdma_activity_work);
@@ -109,41 +118,135 @@ static void kfd_sdma_activity_worker(struct work_struct *work)
qpd = &pdd->qpd;
if (!dqm || !qpd)
return;
+ /*
+ * Total SDMA activity is current SDMA activity + past SDMA activity
+ * Past SDMA count is stored in pdd.
+ * To get the current activity counters for all active SDMA queues,
+ * we loop over all SDMA queues and get their counts from user-space.
+ *
+ * We cannot call get_user() with dqm_lock held as it can cause
+ * a circular lock dependency situation. To read the SDMA stats,
+ * we need to do the following:
+ *
+ * 1. Create a temporary list of SDMA queue nodes from the qpd->queues_list,
+ * with dqm_lock/dqm_unlock().
+ * 2. Call get_user() for each node in temporary list without dqm_lock.
+ * Save the SDMA count for each node and also add the count to the total
+ * SDMA count counter.
+ * Its possible, during this step, a few SDMA queue nodes got deleted
+ * from the qpd->queues_list.
+ * 3. Do a second pass over qpd->queues_list to check if any nodes got deleted.
+ * If any node got deleted, its SDMA count would be captured in the sdma
+ * past activity counter. So subtract the SDMA counter stored in step 2
+ * for this node from the total SDMA count.
+ */
+ INIT_LIST_HEAD(&sdma_q_list.list);
- mm = get_task_mm(pdd->process->lead_thread);
- if (!mm) {
- return;
+ /*
+ * Create the temp list of all SDMA queues
+ */
+ dqm_lock(dqm);
+
+ list_for_each_entry(q, &qpd->queues_list, list) {
+ if ((q->properties.type != KFD_QUEUE_TYPE_SDMA) &&
+ (q->properties.type != KFD_QUEUE_TYPE_SDMA_XGMI))
+ continue;
+
+ sdma_q = kzalloc(sizeof(struct temp_sdma_queue_list), GFP_KERNEL);
+ if (!sdma_q) {
+ dqm_unlock(dqm);
+ goto cleanup;
+ }
+
+ INIT_LIST_HEAD(&sdma_q->list);
+ sdma_q->rptr = (uint64_t)q->properties.read_ptr;
+ sdma_q->queue_id = q->properties.queue_id;
+ list_add_tail(&sdma_q->list, &sdma_q_list.list);
}
- use_mm(mm);
+ /*
+ * If the temp list is empty, then no SDMA queues nodes were found in
+ * qpd->queues_list. Return the past activity count as the total sdma
+ * count
+ */
+ if (list_empty(&sdma_q_list.list)) {
+ workarea->sdma_activity_counter = pdd->sdma_past_activity_counter;
+ dqm_unlock(dqm);
+ return;
+ }
- dqm_lock(dqm);
+ dqm_unlock(dqm);
/*
- * Total SDMA activity is current SDMA activity + past SDMA activity
+ * Get the usage count for each SDMA queue in temp_list.
*/
- workarea->sdma_activity_counter = pdd->sdma_past_activity_counter;
+ mm = get_task_mm(pdd->process->lead_thread);
+ if (!mm)
+ goto cleanup;
+
+ use_mm(mm);
+
+ list_for_each_entry(sdma_q, &sdma_q_list.list, list) {
+ val = 0;
+ ret = read_sdma_queue_counter(sdma_q->rptr, &val);
+ if (ret) {
+ pr_debug("Failed to read SDMA queue active counter for queue id: %d",
+ sdma_q->queue_id);
+ } else {
+ sdma_q->sdma_val = val;
+ workarea->sdma_activity_counter += val;
+ }
+ }
+
+ unuse_mm(mm);
+ mmput(mm);
/*
- * Get the current activity counters for all active SDMA queues
+ * Do a second iteration over qpd_queues_list to check if any SDMA
+ * nodes got deleted while fetching SDMA counter.
*/
+ dqm_lock(dqm);
+
+ workarea->sdma_activity_counter += pdd->sdma_past_activity_counter;
+
list_for_each_entry(q, &qpd->queues_list, list) {
- if ((q->properties.type == KFD_QUEUE_TYPE_SDMA) ||
- (q->properties.type == KFD_QUEUE_TYPE_SDMA_XGMI)) {
- val = 0;
- ret = read_sdma_queue_counter(q, &val);
- if (ret)
- pr_debug("Failed to read SDMA queue active "
- "counter for queue id: %d",
- q->properties.queue_id);
- else
- workarea->sdma_activity_counter += val;
+ if (list_empty(&sdma_q_list.list))
+ break;
+
+ if ((q->properties.type != KFD_QUEUE_TYPE_SDMA) &&
+ (q->properties.type != KFD_QUEUE_TYPE_SDMA_XGMI))
+ continue;
+
+ list_for_each_entry_safe(sdma_q, next, &sdma_q_list.list, list) {
+ if (((uint64_t)q->properties.read_ptr == sdma_q->rptr) &&
+ (sdma_q->queue_id == q->properties.queue_id)) {
+ list_del(&sdma_q->list);
+ kfree(sdma_q);
+ break;
+ }
}
}
dqm_unlock(dqm);
- unuse_mm(mm);
- mmput(mm);
+
+ /*
+ * If temp list is not empty, it implies some queues got deleted
+ * from qpd->queues_list during SDMA usage read. Subtract the SDMA
+ * count for each node from the total SDMA count.
+ */
+ list_for_each_entry_safe(sdma_q, next, &sdma_q_list.list, list) {
+ workarea->sdma_activity_counter -= sdma_q->sdma_val;
+ list_del(&sdma_q->list);
+ kfree(sdma_q);
+ }
+
+ return;
+
+cleanup:
+ list_for_each_entry_safe(sdma_q, next, &sdma_q_list.list, list) {
+ list_del(&sdma_q->list);
+ kfree(sdma_q);
+ }
}
static ssize_t kfd_procfs_show(struct kobject *kobj, struct attribute *attr,