Commit c4a2e5f8 authored by Matthew Brost's avatar Matthew Brost
Browse files

drm/xe: Add devcoredump chunking 



Chunk devcoredump into 1.5G pieces to avoid hitting the kvmalloc limit
of 2G. Simple algorithm reads 1.5G at time in xe_devcoredump_read
callback as needed.

Some memory allocations are changed to GFP_ATOMIC as they done in
xe_devcoredump_read which holds lock in the path of reclaim. The
allocations are small, so in practice should never fail.

v2:
 - Update commit message wrt gfp atomic (John H)
v6:
 - Drop GFP_ATOMIC change for hwconfig (John H)

Signed-off-by: default avatarMatthew Brost <matthew.brost@intel.com>
Reviewed-by: default avatarJonathan Cavitt <jonathan.cavitt@intel.com>
Link: https://lore.kernel.org/r/20250423171725.597955-2-matthew.brost@intel.com
parent f9e4d8bb

drivers/gpu/drm/xe/xe_devcoredump.c

+47 −12
Original line number Diff line number Diff line
@@ -80,7 +80,8 @@ static struct xe_guc *exec_queue_to_guc(struct xe_exec_queue *q) 
	return &q->gt->uc.guc;

}



static ssize_t __xe_devcoredump_read(char *buffer, size_t count,

static ssize_t __xe_devcoredump_read(char *buffer, ssize_t count,

				     ssize_t start,

				     struct xe_devcoredump *coredump)

{

	struct xe_device *xe;
@@ -94,7 +95,7 @@ static ssize_t __xe_devcoredump_read(char *buffer, size_t count, 
	ss = &coredump->snapshot;



	iter.data = buffer;

	iter.start = 0;

	iter.start = start;

	iter.remain = count;



	p = drm_coredump_printer(&iter);
@@ -168,6 +169,8 @@ static void xe_devcoredump_snapshot_free(struct xe_devcoredump_snapshot *ss) 
	ss->vm = NULL;

}



#define XE_DEVCOREDUMP_CHUNK_MAX	(SZ_512M + SZ_1G)



static ssize_t xe_devcoredump_read(char *buffer, loff_t offset,

				   size_t count, void *data, size_t datalen)

{
@@ -183,6 +186,9 @@ static ssize_t xe_devcoredump_read(char *buffer, loff_t offset, 
	/* Ensure delayed work is captured before continuing */

	flush_work(&ss->work);



	if (ss->read.size > XE_DEVCOREDUMP_CHUNK_MAX)

		xe_pm_runtime_get(gt_to_xe(ss->gt));



	mutex_lock(&coredump->lock);



	if (!ss->read.buffer) {
@@ -195,12 +201,26 @@ static ssize_t xe_devcoredump_read(char *buffer, loff_t offset, 
		return 0;

	}



	if (offset >= ss->read.chunk_position + XE_DEVCOREDUMP_CHUNK_MAX ||

	    offset < ss->read.chunk_position) {

		ss->read.chunk_position =

			ALIGN_DOWN(offset, XE_DEVCOREDUMP_CHUNK_MAX);



		__xe_devcoredump_read(ss->read.buffer,

				      XE_DEVCOREDUMP_CHUNK_MAX,

				      ss->read.chunk_position, coredump);

	}



	byte_copied = count < ss->read.size - offset ? count :

		ss->read.size - offset;

	memcpy(buffer, ss->read.buffer + offset, byte_copied);

	memcpy(buffer, ss->read.buffer +

	       (offset % XE_DEVCOREDUMP_CHUNK_MAX), byte_copied);



	mutex_unlock(&coredump->lock);



	if (ss->read.size > XE_DEVCOREDUMP_CHUNK_MAX)

		xe_pm_runtime_put(gt_to_xe(ss->gt));



	return byte_copied;

}
@@ -254,19 +274,34 @@ static void xe_devcoredump_deferred_snap_work(struct work_struct *work) 
	xe_guc_exec_queue_snapshot_capture_delayed(ss->ge);

	xe_force_wake_put(gt_to_fw(ss->gt), fw_ref);



	xe_pm_runtime_put(xe);

	ss->read.chunk_position = 0;



	/* Calculate devcoredump size */

	ss->read.size = __xe_devcoredump_read(NULL, INT_MAX, coredump);

	ss->read.size = __xe_devcoredump_read(NULL, LONG_MAX, 0, coredump);



	if (ss->read.size > XE_DEVCOREDUMP_CHUNK_MAX) {

		ss->read.buffer = kvmalloc(XE_DEVCOREDUMP_CHUNK_MAX,

					   GFP_USER);

		if (!ss->read.buffer)

			goto put_pm;



		__xe_devcoredump_read(ss->read.buffer,

				      XE_DEVCOREDUMP_CHUNK_MAX,

				      0, coredump);

	} else {

		ss->read.buffer = kvmalloc(ss->read.size, GFP_USER);

		if (!ss->read.buffer)

		return;

			goto put_pm;



	__xe_devcoredump_read(ss->read.buffer, ss->read.size, coredump);

		__xe_devcoredump_read(ss->read.buffer, ss->read.size, 0,

				      coredump);

		xe_devcoredump_snapshot_free(ss);

	}



put_pm:

	xe_pm_runtime_put(xe);

}



static void devcoredump_snapshot(struct xe_devcoredump *coredump,

				 struct xe_exec_queue *q,

				 struct xe_sched_job *job)
@@ -425,7 +460,7 @@ void xe_print_blob_ascii85(struct drm_printer *p, const char *prefix, char suffi 
	if (offset & 3)

		drm_printf(p, "Offset not word aligned: %zu", offset);



	line_buff = kzalloc(DMESG_MAX_LINE_LEN, GFP_KERNEL);

	line_buff = kzalloc(DMESG_MAX_LINE_LEN, GFP_ATOMIC);

	if (!line_buff) {

		drm_printf(p, "Failed to allocate line buffer\n");

		return;

drivers/gpu/drm/xe/xe_devcoredump_types.h

+2 −0
Original line number Diff line number Diff line
@@ -66,6 +66,8 @@ struct xe_devcoredump_snapshot { 
	struct {

		/** @read.size: size of devcoredump in human readable format */

		ssize_t size;

		/** @read.chunk_position: position of devcoredump chunk */

		ssize_t chunk_position;

		/** @read.buffer: buffer of devcoredump in human readable format */

		char *buffer;

	} read;