radeon: workaround pinning failure on low ram gpu

[Linux-MTD Mailing List]

Gitweb:     http://git.infradead.org/?p=mtd-2.6.git;a=commit;h=97b6ff6be9da7675aab339334fda996d6c5077d9
Commit:     97b6ff6be9da7675aab339334fda996d6c5077d9
Parent:     fae009d15a44e5f1d938340facf4b8bc7dc69a09
Author:     Jerome Glisse <jglisse at redhat.com>
AuthorDate: Tue Nov 12 10:51:16 2013 -0500
Committer:  Alex Deucher <alexander.deucher at amd.com>
CommitDate: Fri Nov 15 15:29:33 2013 -0500

    radeon: workaround pinning failure on low ram gpu
    
    GPU with low amount of ram can fails at pinning new framebuffer before
    unpinning old one. On such failure, retry with unpinning old one before
    pinning new one allowing to work around the issue. This is somewhat
    ugly but only affect those old GPU we care about.
    
    Signed-off-by: Jerome Glisse <jglisse at redhat.com>
    Signed-off-by: Alex Deucher <alexander.deucher at amd.com>
    Cc: stable at vger.kernel.org
---
 drivers/gpu/drm/radeon/radeon_legacy_crtc.c | 28 ++++++++++++++++++++++++++++
 1 file changed, 28 insertions(+)

diff --git a/drivers/gpu/drm/radeon/radeon_legacy_crtc.c b/drivers/gpu/drm/radeon/radeon_legacy_crtc.c
index 0c7b8c6..0b158f9 100644
--- a/drivers/gpu/drm/radeon/radeon_legacy_crtc.c
+++ b/drivers/gpu/drm/radeon/radeon_legacy_crtc.c
@@ -422,6 +422,7 @@ int radeon_crtc_do_set_base(struct drm_crtc *crtc,
 	/* Pin framebuffer & get tilling informations */
 	obj = radeon_fb->obj;
 	rbo = gem_to_radeon_bo(obj);
+retry:
 	r = radeon_bo_reserve(rbo, false);
 	if (unlikely(r != 0))
 		return r;
@@ -430,6 +431,33 @@ int radeon_crtc_do_set_base(struct drm_crtc *crtc,
 				     &base);
 	if (unlikely(r != 0)) {
 		radeon_bo_unreserve(rbo);
+
+		/* On old GPU like RN50 with little vram pining can fails because
+		 * current fb is taking all space needed. So instead of unpining
+		 * the old buffer after pining the new one, first unpin old one
+		 * and then retry pining new one.
+		 *
+		 * As only master can set mode only master can pin and it is
+		 * unlikely the master client will race with itself especialy
+		 * on those old gpu with single crtc.
+		 *
+		 * We don't shutdown the display controller because new buffer
+		 * will end up in same spot.
+		 */
+		if (!atomic && fb && fb != crtc->fb) {
+			struct radeon_bo *old_rbo;
+			unsigned long nsize, osize;
+
+			old_rbo = gem_to_radeon_bo(to_radeon_framebuffer(fb)->obj);
+			osize = radeon_bo_size(old_rbo);
+			nsize = radeon_bo_size(rbo);
+			if (nsize <= osize && !radeon_bo_reserve(old_rbo, false)) {
+				radeon_bo_unpin(old_rbo);
+				radeon_bo_unreserve(old_rbo);
+				fb = NULL;
+				goto retry;
+			}
+		}
 		return -EINVAL;
 	}
 	radeon_bo_get_tiling_flags(rbo, &tiling_flags, NULL);