drm/panthor: Fix the user MMIO offset logic for emulators

Currently, we pick the MMIO offset based on the size of the pgoff_t
type seen by the process that manipulates the FD, such that a 32-bit
process can always map the user MMIO ranges. But this approach doesn't
work well for emulators like FEX, where the emulator is a 64-bit binary
which might be executing 32-bit code. In that case, the kernel thinks
it's the 64-bit process and assumes DRM_PANTHOR_USER_MMIO_OFFSET_64BIT
is in use, but the UMD library expects DRM_PANTHOR_USER_MMIO_OFFSET_32BIT,
because it can't mmap() anything above the pgoff_t size.

In order to solve that, we need a way to explicitly set the user MMIO
offset from the UMD, such that the kernel doesn't have to guess it
from the TIF_32BIT flag set on user thread. We keep the old behavior
if DRM_PANTHOR_SET_USER_MMIO_OFFSET is never called.

Changes in v2:
- Drop the lock/immutable fields and allow SET_USER_MMIO_OFFSET
  requests to race with mmap() requests
- Don't do the is_user_mmio_offset test twice in panthor_mmap()
- Improve the uAPI docs

Changes in v3:
- Bump to version 1.5 instead of 1.4 after rebasing
- Add R-bs
- Fix/rephrase comment as suggested by Liviu

Reviewed-by: Adrián Larumbe <adrian.larumbe@collabora.com>
Reviewed-by: Steven Price <steven.price@arm.com>
Reviewed-by: Liviu Dudau <liviu.dudau@arm.com>
Link: https://lore.kernel.org/r/20250606080932.4140010-3-boris.brezillon@collabora.com
Signed-off-by: Boris Brezillon <boris.brezillon@collabora.com>

drivers/gpu/drm/panthor/panthor_device.h

@@ -230,6 +230,24 @@ struct panthor_file {
 	/** @ptdev: Device attached to this file. */
 	struct panthor_device *ptdev;
 
+	/** @user_mmio: User MMIO related fields. */
+	struct {
+		/**
+		 * @offset: Offset used for user MMIO mappings.
+		 *
+		 * This offset should not be used to check the type of mapping
+		 * except in panthor_mmap(). After that point, MMIO mapping
+		 * offsets have been adjusted to match
+		 * DRM_PANTHOR_USER_MMIO_OFFSET and that macro should be used
+		 * instead.
+		 * Make sure this rule is followed at all times, because
+		 * userspace is in control of the offset, and can change the
+		 * value behind our back. Otherwise it can lead to erroneous
+		 * branching happening in kernel space.
+		 */
+		u64 offset;
+	} user_mmio;
+
 	/** @vms: VM pool attached to this file. */
 	struct panthor_vm_pool *vms;
 

drivers/gpu/drm/panthor/panthor_drv.c

@@ -1378,6 +1378,20 @@ err_put_obj:
 	return ret;
 }
 
+static int panthor_ioctl_set_user_mmio_offset(struct drm_device *ddev,
+					      void *data, struct drm_file *file)
+{
+	struct drm_panthor_set_user_mmio_offset *args = data;
+	struct panthor_file *pfile = file->driver_priv;
+
+	if (args->offset != DRM_PANTHOR_USER_MMIO_OFFSET_32BIT &&
+	    args->offset != DRM_PANTHOR_USER_MMIO_OFFSET_64BIT)
+		return -EINVAL;
+
+	WRITE_ONCE(pfile->user_mmio.offset, args->offset);
+	return 0;
+}
+
 static int
 panthor_open(struct drm_device *ddev, struct drm_file *file)
 {
@@ -1395,6 +1409,18 @@ panthor_open(struct drm_device *ddev, struct drm_file *file)
 	}
 
 	pfile->ptdev = ptdev;
+	pfile->user_mmio.offset = DRM_PANTHOR_USER_MMIO_OFFSET;
+
+#ifdef CONFIG_ARM64
+	/*
+	 * With 32-bit systems being limited by the 32-bit representation of
+	 * mmap2's pgoffset field, we need to make the MMIO offset arch
+	 * specific.
+	 */
+	if (test_tsk_thread_flag(current, TIF_32BIT))
+		pfile->user_mmio.offset = DRM_PANTHOR_USER_MMIO_OFFSET_32BIT;
+#endif
+
 
 	ret = panthor_vm_pool_create(pfile);
 	if (ret)
@@ -1448,6 +1474,7 @@ static const struct drm_ioctl_desc panthor_drm_driver_ioctls[] = {
 	PANTHOR_IOCTL(TILER_HEAP_DESTROY, tiler_heap_destroy, DRM_RENDER_ALLOW),
 	PANTHOR_IOCTL(GROUP_SUBMIT, group_submit, DRM_RENDER_ALLOW),
 	PANTHOR_IOCTL(BO_SET_LABEL, bo_set_label, DRM_RENDER_ALLOW),
+	PANTHOR_IOCTL(SET_USER_MMIO_OFFSET, set_user_mmio_offset, DRM_RENDER_ALLOW),
 };
 
 static int panthor_mmap(struct file *filp, struct vm_area_struct *vma)
@@ -1456,30 +1483,26 @@ static int panthor_mmap(struct file *filp, struct vm_area_struct *vma)
 	struct panthor_file *pfile = file->driver_priv;
 	struct panthor_device *ptdev = pfile->ptdev;
 	u64 offset = (u64)vma->vm_pgoff << PAGE_SHIFT;
+	u64 user_mmio_offset;
 	int ret, cookie;
 
 	if (!drm_dev_enter(file->minor->dev, &cookie))
 		return -ENODEV;
 
-#ifdef CONFIG_ARM64
-	/*
-	 * With 32-bit systems being limited by the 32-bit representation of
-	 * mmap2's pgoffset field, we need to make the MMIO offset arch
-	 * specific. This converts a user MMIO offset into something the kernel
-	 * driver understands.
+	/* Adjust the user MMIO offset to match the offset used kernel side.
+	 * We use a local variable with a READ_ONCE() here to make sure
+	 * the user_mmio_offset we use for the is_user_mmio_mapping() check
+	 * hasn't changed when we do the offset adjustment.
 	 */
-	if (test_tsk_thread_flag(current, TIF_32BIT) &&
-	    offset >= DRM_PANTHOR_USER_MMIO_OFFSET_32BIT) {
-		offset += DRM_PANTHOR_USER_MMIO_OFFSET_64BIT -
-			  DRM_PANTHOR_USER_MMIO_OFFSET_32BIT;
+	user_mmio_offset = READ_ONCE(pfile->user_mmio.offset);
+	if (offset >= user_mmio_offset) {
+		offset -= user_mmio_offset;
+		offset += DRM_PANTHOR_USER_MMIO_OFFSET;
 		vma->vm_pgoff = offset >> PAGE_SHIFT;
-	}
-#endif
-
-	if (offset >= DRM_PANTHOR_USER_MMIO_OFFSET)
 		ret = panthor_device_mmap_io(ptdev, vma);
-	else
+	} else {
 		ret = drm_gem_mmap(filp, vma);
+	}
 
 	drm_dev_exit(cookie);
 	return ret;
@@ -1550,6 +1573,7 @@ static const struct file_operations panthor_drm_driver_fops = {
  *       - adds PANTHOR_GROUP_PRIORITY_REALTIME priority
  * - 1.3 - adds DRM_PANTHOR_GROUP_STATE_INNOCENT flag
  * - 1.4 - adds DRM_IOCTL_PANTHOR_BO_SET_LABEL ioctl
+ * - 1.5 - adds DRM_PANTHOR_SET_USER_MMIO_OFFSET ioctl
  */
 static const struct drm_driver panthor_drm_driver = {
 	.driver_features = DRIVER_RENDER | DRIVER_GEM | DRIVER_SYNCOBJ |
@@ -1564,7 +1588,7 @@ static const struct drm_driver panthor_drm_driver = {
 	.desc = "Panthor DRM driver",
 	.date = "20230801",
 	.major = 1,
-	.minor = 4,
+	.minor = 5,
 
 	.gem_create_object = panthor_gem_create_object,
 	.gem_prime_import_sg_table = drm_gem_shmem_prime_import_sg_table,

include/uapi/drm/panthor_drm.h

@@ -130,6 +130,20 @@ enum drm_panthor_ioctl_id {
 
 	/** @DRM_PANTHOR_BO_SET_LABEL: Label a BO. */
 	DRM_PANTHOR_BO_SET_LABEL,
+
+	/**
+	 * @DRM_PANTHOR_SET_USER_MMIO_OFFSET: Set the offset to use as the user MMIO offset.
+	 *
+	 * The default behavior is to pick the MMIO offset based on the size of the pgoff_t
+	 * type seen by the process that manipulates the FD, such that a 32-bit process can
+	 * always map the user MMIO ranges. But this approach doesn't work well for emulators
+	 * like FEX, where the emulator is an 64-bit binary which might be executing 32-bit
+	 * code. In that case, the kernel thinks it's the 64-bit process and assumes
+	 * DRM_PANTHOR_USER_MMIO_OFFSET_64BIT is in use, but the UMD library expects
+	 * DRM_PANTHOR_USER_MMIO_OFFSET_32BIT, because it can't mmap() anything above the
+	 * pgoff_t size.
+	 */
+	DRM_PANTHOR_SET_USER_MMIO_OFFSET,
 };
 
 /**
@@ -989,6 +1003,28 @@ struct drm_panthor_bo_set_label {
 	__u64 label;
 };
 
+/**
+ * struct drm_panthor_set_user_mmio_offset - Arguments passed to
+ * DRM_IOCTL_PANTHOR_SET_USER_MMIO_OFFSET
+ *
+ * This ioctl is only really useful if you want to support userspace
+ * CPU emulation environments where the size of an unsigned long differs
+ * between the host and the guest architectures.
+ */
+struct drm_panthor_set_user_mmio_offset {
+	/**
+	 * @offset: User MMIO offset to use.
+	 *
+	 * Must be either DRM_PANTHOR_USER_MMIO_OFFSET_32BIT or
+	 * DRM_PANTHOR_USER_MMIO_OFFSET_64BIT.
+	 *
+	 * Use DRM_PANTHOR_USER_MMIO_OFFSET (which selects OFFSET_32BIT or
+	 * OFFSET_64BIT based on the size of an unsigned long) unless you
+	 * have a very good reason to overrule this decision.
+	 */
+	__u64 offset;
+};
+
 /**
  * DRM_IOCTL_PANTHOR() - Build a Panthor IOCTL number
  * @__access: Access type. Must be R, W or RW.
@@ -1033,6 +1069,8 @@ enum {
 		DRM_IOCTL_PANTHOR(WR, TILER_HEAP_DESTROY, tiler_heap_destroy),
 	DRM_IOCTL_PANTHOR_BO_SET_LABEL =
 		DRM_IOCTL_PANTHOR(WR, BO_SET_LABEL, bo_set_label),
+	DRM_IOCTL_PANTHOR_SET_USER_MMIO_OFFSET =
+		DRM_IOCTL_PANTHOR(WR, SET_USER_MMIO_OFFSET, set_user_mmio_offset),
 };
 
 #if defined(__cplusplus)