[PATCH 5/9] drm/amdkfd: Add memory exception handling

21 Apr 2015

From: Alexey Skidanov Alexey.Skidanov@amd.com
This patch adds Peripheral Page Request (PPR) failure processing and reporting.
Bad address or pointer to a system memory block with inappropriate
read/write permission cause such PPR failure during a user queue
processing. PPR request handling is done by IOMMU driver notifying
AMDKFD module on PPR failure.
The process triggering a PPR failure will be notified by appropriate event
or SIGTERM signal will be sent to it.
Signed-off-by: Alexey Skidanov Alexey.Skidanov@amd.com
Signed-off-by: Oded Gabbay oded.gabbay@amd.com
---
 drivers/gpu/drm/amd/amdkfd/kfd_device.c |  29 ++++++
 drivers/gpu/drm/amd/amdkfd/kfd_events.c | 154 ++++++++++++++++++++++++++++++--
 drivers/gpu/drm/amd/amdkfd/kfd_events.h |   7 ++
 drivers/gpu/drm/amd/amdkfd/kfd_priv.h   |   5 +-
 4 files changed, 186 insertions(+), 9 deletions(-)

diff --git a/drivers/gpu/drm/amd/amdkfd/kfd_device.c b/drivers/gpu/drm/amd/amdkfd/kfd_device.c
index 4c03169..52cab0f 100644
--- a/drivers/gpu/drm/amd/amdkfd/kfd_device.c
+++ b/drivers/gpu/drm/amd/amdkfd/kfd_device.c
@@ -182,6 +182,32 @@ static void iommu_pasid_shutdown_callback(struct pci_dev *pdev, int pasid)
    	kfd_unbind_process_from_device(dev, pasid);
 }
+/*
+ * This function called by IOMMU driver on PPR failure
+ */
+static int iommu_invalid_ppr_cb(struct pci_dev *pdev, int pasid,
+		unsigned long address, u16 flags)
+{
+	struct kfd_dev *dev;
+
+	dev_warn(kfd_device,
+			"Invalid PPR device %x:%x.%x pasid %d address 0x%lX flags 0x%X",
+			PCI_BUS_NUM(pdev->devfn),
+			PCI_SLOT(pdev->devfn),
+			PCI_FUNC(pdev->devfn),
+			pasid,
+			address,
+			flags);
+
+	dev = kfd_device_by_pci_dev(pdev);
+	BUG_ON(dev == NULL);
+
+	kfd_signal_iommu_event(dev, pasid, address,
+			flags & PPR_FAULT_WRITE, flags & PPR_FAULT_EXEC);
+
+	return AMD_IOMMU_INV_PRI_RSP_INVALID;
+}
+
 bool kgd2kfd_device_init(struct kfd_dev *kfd,
    		 const struct kgd2kfd_shared_resources *gpu_resources)
 {
@@ -251,6 +277,7 @@ bool kgd2kfd_device_init(struct kfd_dev *kfd,
    }
    amd_iommu_set_invalidate_ctx_cb(kfd->pdev,
    					iommu_pasid_shutdown_callback);
+	amd_iommu_set_invalid_ppr_cb(kfd->pdev, iommu_invalid_ppr_cb);
kfd->dqm = device_queue_manager_init(kfd);
    if (!kfd->dqm) {
@@ -316,6 +343,7 @@ void kgd2kfd_suspend(struct kfd_dev *kfd)
    if (kfd->init_complete) {
    	kfd->dqm->ops.stop(kfd->dqm);
    	amd_iommu_set_invalidate_ctx_cb(kfd->pdev, NULL);
+		amd_iommu_set_invalid_ppr_cb(kfd->pdev, NULL);
    	amd_iommu_free_device(kfd->pdev);
    }
 }
@@ -335,6 +363,7 @@ int kgd2kfd_resume(struct kfd_dev *kfd)
    		return -ENXIO;
    	amd_iommu_set_invalidate_ctx_cb(kfd->pdev,
    					iommu_pasid_shutdown_callback);
+		amd_iommu_set_invalid_ppr_cb(kfd->pdev, iommu_invalid_ppr_cb);
    	kfd->dqm->ops.start(kfd->dqm);
    }
diff --git a/drivers/gpu/drm/amd/amdkfd/kfd_events.c b/drivers/gpu/drm/amd/amdkfd/kfd_events.c
index 984251d..ac60fb2 100644
--- a/drivers/gpu/drm/amd/amdkfd/kfd_events.c
+++ b/drivers/gpu/drm/amd/amdkfd/kfd_events.c
@@ -30,6 +30,7 @@
 #include <linux/memory.h>
 #include "kfd_priv.h"
 #include "kfd_events.h"
+#include <linux/device.h>
#define SIGNAL_EVENT_LIMIT 256
@@ -47,6 +48,10 @@ struct kfd_event_waiter {
/* Transitions to true when the event this belongs to is signaled. */
    bool activated;
+
+	/* Event */
+	struct kfd_event *event;
+	uint32_t input_index;
 };
/*
@@ -586,14 +591,17 @@ static struct kfd_event_waiter *alloc_event_waiters(uint32_t num_events)
 }
static int init_event_waiter(struct kfd_process *p,
-				struct kfd_event_waiter *waiter,
-				uint32_t event_id)
+		struct kfd_event_waiter *waiter,
+		uint32_t event_id,
+		uint32_t input_index)
 {
    struct kfd_event *ev = lookup_event_by_id(p, event_id);
if (!ev)
    	return -EINVAL;
+	waiter->event = ev;
+	waiter->input_index = input_index;
    waiter->activated = ev->signaled;
    ev->signaled = ev->signaled && !ev->auto_reset;
@@ -620,6 +628,37 @@ static bool test_event_condition(bool all, uint32_t num_events,
    return activated_count == num_events;
 }
+/*
+ * Copy event specific data, if defined.
+ * Currently only memory exception events have additional data to copy to user
+ */
+static bool copy_signaled_event_data(uint32_t num_events,
+		struct kfd_event_waiter *event_waiters,
+		struct kfd_event_data __user *data)
+{
+	struct kfd_hsa_memory_exception_data *src, *dst;
+	struct kfd_event_waiter *waiter;
+	struct kfd_event *event;
+	uint32_t i;
+
+	for (i = 0; i < num_events; i++) {
+		waiter = &event_waiters[i];
+		event = waiter->event;
+		if (waiter->activated && event->type == KFD_EVENT_TYPE_MEMORY) {
+			dst = &data[waiter->input_index].memory_exception_data;
+			src = &event->memory_exception_data;
+			if (copy_to_user(dst, src,
+				sizeof(struct kfd_hsa_memory_exception_data)))
+				return false;
+		}
+	}
+
+	return true;
+
+}
+
+
+
 static long user_timeout_to_jiffies(uint32_t user_timeout_ms)
 {
    if (user_timeout_ms == KFD_EVENT_TIMEOUT_IMMEDIATE)
@@ -649,10 +688,12 @@ static void free_waiters(uint32_t num_events, struct kfd_event_waiter *waiters)
 }
int kfd_wait_on_events(struct kfd_process *p,
-		       uint32_t num_events, const uint32_t __user *event_ids,
+		       uint32_t num_events, void __user *data,
    	       bool all, uint32_t user_timeout_ms,
    	       enum kfd_event_wait_result *wait_result)
 {
+	struct kfd_event_data __user *events =
+			(struct kfd_event_data __user *) data;
    uint32_t i;
    int ret = 0;
    struct kfd_event_waiter *event_waiters = NULL;
@@ -667,13 +708,14 @@ int kfd_wait_on_events(struct kfd_process *p,
    }
for (i = 0; i < num_events; i++) {
-		uint32_t event_id;
+		struct kfd_event_data event_data;
-		ret = get_user(event_id, &event_ids[i]);
-		if (ret)
+		if (copy_from_user(&event_data, &events[i],
+				sizeof(struct kfd_event_data)))
    		goto fail;
-		ret = init_event_waiter(p, &event_waiters[i], event_id);
+		ret = init_event_waiter(p, &event_waiters[i],
+				event_data.event_id, i);
    	if (ret)
    		goto fail;
    }
@@ -700,7 +742,11 @@ int kfd_wait_on_events(struct kfd_process *p,
    	}
if (test_event_condition(all, num_events, event_waiters)) {
-			*wait_result = KFD_WAIT_COMPLETE;
+			if (copy_signaled_event_data(num_events,
+					event_waiters, events))
+				*wait_result = KFD_WAIT_COMPLETE;
+			else
+				*wait_result = KFD_WAIT_ERROR;
    		break;
    	}
@@ -773,3 +819,95 @@ int radeon_kfd_event_mmap(struct kfd_process *p,
    return remap_pfn_range(vma, vma->vm_start, pfn, PAGE_SIZE,
    			vma->vm_page_prot);
 }
+
+/*
+ * Assumes that p->event_mutex is held and of course
+ * that p is not going away (current or locked).
+ */
+static void lookup_events_by_type_and_signal(struct kfd_process *p,
+		int type, void *event_data)
+{
+	struct kfd_hsa_memory_exception_data *ev_data;
+	struct kfd_event *ev;
+	int bkt;
+	bool send_signal = true;
+
+	ev_data = (struct kfd_hsa_memory_exception_data *) event_data;
+
+	hash_for_each(p->events, bkt, ev, events)
+		if (ev->type == type) {
+			send_signal = false;
+			dev_dbg(kfd_device,
+					"Event found: id %X type %d",
+					ev->event_id, ev->type);
+			set_event(ev);
+			if (ev->type == KFD_EVENT_TYPE_MEMORY && ev_data)
+				ev->memory_exception_data = *ev_data;
+		}
+
+	/* Send SIGTERM no event of type "type" has been found*/
+	if (send_signal) {
+		dev_warn(kfd_device,
+			"Sending SIGTERM to HSA Process with PID %d ",
+				p->lead_thread->pid);
+		send_sig(SIGTERM, p->lead_thread, 0);
+	}
+}
+
+void kfd_signal_iommu_event(struct kfd_dev *dev, unsigned int pasid,
+		unsigned long address, bool is_write_requested,
+		bool is_execute_requested)
+{
+	struct kfd_hsa_memory_exception_data memory_exception_data;
+	struct vm_area_struct *vma;
+
+	/*
+	 * Because we are called from arbitrary context (workqueue) as opposed
+	 * to process context, kfd_process could attempt to exit while we are
+	 * running so the lookup function returns a locked process.
+	 */
+	struct kfd_process *p = kfd_lookup_process_by_pasid(pasid);
+
+	if (!p)
+		return; /* Presumably process exited. */
+
+	memset(&memory_exception_data, 0, sizeof(memory_exception_data));
+
+	down_read(&p->mm->mmap_sem);
+	vma = find_vma(p->mm, address);
+
+	memory_exception_data.gpu_id = dev->id;
+	memory_exception_data.va = address;
+	/* Set failure reason */
+	memory_exception_data.failure.NotPresent = true;
+	memory_exception_data.failure.NoExecute = false;
+	memory_exception_data.failure.ReadOnly = false;
+	if (vma) {
+		if (vma->vm_start > address) {
+			memory_exception_data.failure.NotPresent = true;
+			memory_exception_data.failure.NoExecute = false;
+			memory_exception_data.failure.ReadOnly = false;
+		} else {
+			memory_exception_data.failure.NotPresent = false;
+			if (is_write_requested && !(vma->vm_flags & VM_WRITE))
+				memory_exception_data.failure.ReadOnly = true;
+			else
+				memory_exception_data.failure.ReadOnly = false;
+			if (is_execute_requested && !(vma->vm_flags & VM_EXEC))
+				memory_exception_data.failure.NoExecute = true;
+			else
+				memory_exception_data.failure.NoExecute = false;
+		}
+	}
+
+	up_read(&p->mm->mmap_sem);
+
+	mutex_lock(&p->event_mutex);
+
+	/* Lookup events by type and signal them */
+	lookup_events_by_type_and_signal(p, KFD_EVENT_TYPE_MEMORY,
+			&memory_exception_data);
+
+	mutex_unlock(&p->event_mutex);
+	mutex_unlock(&p->mutex);
+}
diff --git a/drivers/gpu/drm/amd/amdkfd/kfd_events.h b/drivers/gpu/drm/amd/amdkfd/kfd_events.h
index d9b5b38..691cf85 100644
--- a/drivers/gpu/drm/amd/amdkfd/kfd_events.h
+++ b/drivers/gpu/drm/amd/amdkfd/kfd_events.h
@@ -28,6 +28,7 @@
 #include <linux/types.h>
 #include <linux/list.h>
 #include "kfd_priv.h"
+#include <uapi/linux/kfd_ioctl.h>
#define KFD_EVENT_ID_NONSIGNAL_MASK 0x80000000U
 #define KFD_FIRST_NONSIGNAL_EVENT_ID KFD_EVENT_ID_NONSIGNAL_MASK
@@ -61,6 +62,11 @@ struct kfd_event {
    struct signal_page *signal_page;
    unsigned int signal_slot_index;
    uint64_t __user *user_signal_address;
+
+	/* type specific data */
+	union {
+		struct kfd_hsa_memory_exception_data memory_exception_data;
+	};
 };
#define KFD_EVENT_TIMEOUT_IMMEDIATE 0
@@ -69,6 +75,7 @@ struct kfd_event {
 /* Matching HSA_EVENTTYPE */
 #define KFD_EVENT_TYPE_SIGNAL 0
 #define KFD_EVENT_TYPE_DEBUG 5
+#define KFD_EVENT_TYPE_MEMORY 8
extern void kfd_signal_event_interrupt(unsigned int pasid, uint32_t partial_id,
    				uint32_t valid_id_bits);
diff --git a/drivers/gpu/drm/amd/amdkfd/kfd_priv.h b/drivers/gpu/drm/amd/amdkfd/kfd_priv.h
index 7f6ea96..8ecc033 100644
--- a/drivers/gpu/drm/amd/amdkfd/kfd_priv.h
+++ b/drivers/gpu/drm/amd/amdkfd/kfd_priv.h
@@ -694,11 +694,14 @@ void kfd_event_free_process(struct kfd_process *p);
 int radeon_kfd_event_mmap(struct kfd_process *process,
    			struct vm_area_struct *vma);
 int kfd_wait_on_events(struct kfd_process *p,
-		       uint32_t num_events, const uint32_t __user *event_ids,
+		       uint32_t num_events, void __user *data,
    	       bool all, uint32_t user_timeout_ms,
    	       enum kfd_event_wait_result *wait_result);
 void kfd_signal_event_interrupt(unsigned int pasid, uint32_t partial_id,
    			uint32_t valid_id_bits);
+void kfd_signal_iommu_event(struct kfd_dev *dev,
+		unsigned int pasid, unsigned long address,
+		bool is_write_requested, bool is_execute_requested);
 int kfd_set_event(struct kfd_process *p, uint32_t event_id);
 int kfd_reset_event(struct kfd_process *p, uint32_t event_id);
 int kfd_event_create(struct file *devkfd, struct kfd_process *p,
-- 
1.9.1


    

2025

2024

2023

2022

2021

2020

2019

2018

2017

2016

2015

2014

2013

2012

2011

2010

[PATCH 5/9] drm/amdkfd: Add memory exception handling