Two issues about the upcoming 6.1 kernel release

[nyanmisaka]

I'm not sure if these have been fixed in the 6.1 final release. Therefore I report them here so that you can forward it to the corresponding developers. cc @JeffyCN

---

1. rk_system_heap.c is still not ported to 6.1 yet, which causes MPP runtime to crash in DMA_HEAP allocator.

https://github.com/JeffyCN/mirrors/blob/kernel-6.1-2024_01_02/drivers/dma-buf/heaps/Makefile#L2

ls -l /dev/dma_heap/
total 0
crw------- 1 root root 251, 1 Jan 18 18:50 cma
crw------- 1 root root 251, 0 Jan 18 18:50 system

---

2. rockchip_drm_gem.c cacheable memory flag is broken during manual merging. It causes very slow copying from the decoder frame pool.

https://github.com/JeffyCN/mirrors/blob/kernel-6.1-2024_01_02/drivers/gpu/drm/rockchip/rockchip_drm_gem.c#L550

A simple fix is this:

diff --git a/drivers/gpu/drm/rockchip/rockchip_drm_gem.c b/drivers/gpu/drm/rockchip/rockchip_drm_gem.c
index 3a32a03e91f8..f386f44f825b 100644
--- a/drivers/gpu/drm/rockchip/rockchip_drm_gem.c
+++ b/drivers/gpu/drm/rockchip/rockchip_drm_gem.c
@@ -546,10 +546,6 @@ static int rockchip_drm_gem_object_mmap(struct drm_gem_object *obj,
    int ret;
    struct rockchip_gem_object *rk_obj = to_rockchip_obj(obj);

-   /* default is wc. */
-   if (rk_obj->flags & ROCKCHIP_BO_CACHABLE)
-       vma->vm_page_prot = vm_get_page_prot(vma->vm_flags);
-
    /*
     * Set vm_pgoff (used as a fake buffer offset by DRM) to 0 and map the
     * whole buffer from the start.
@@ -566,6 +562,10 @@ static int rockchip_drm_gem_object_mmap(struct drm_gem_object *obj,
    vma->vm_page_prot = pgprot_writecombine(vm_get_page_prot(vma->vm_flags));
    vma->vm_page_prot = pgprot_decrypted(vma->vm_page_prot);

+        /* default is wc. */
+        if (rk_obj->flags & ROCKCHIP_BO_CACHABLE)
+                vma->vm_page_prot = vm_get_page_prot(vma->vm_flags);
+
    if (rk_obj->buf_type == ROCKCHIP_GEM_BUF_TYPE_SECURE) {
        DRM_ERROR("Disallow mmap for secure buffer\n");
        ret = -EINVAL;

[JeffyCN]

thanks for reporting.

@HermanChen please check that

[HermanChen]

OK, I will let drm people check it.

[JeffyCN]

Hi, for the first issue, it's expected to use the upstream dma-heap driver(according to rk dma-heap developer(@xxm)), so we should fix the crash in the mpp @HermanChen

and the second issue would be fixed in: kernel-6.1-2024_03_01

[nyanmisaka]

@JeffyCN Thanks for the info! Feel free to close the issue.

[JeffyCN]

Okay, thanks :)

[Joshua-Riek]

Hey @JeffyCN quick question, is kernel-6.1-2024_03_01 from the linux-6.1-stan-rkr2 release?

[JeffyCN]

no, it just the last commit before the first rk3576 thing(which i was told not to upload)

[Joshua-Riek]

Gotcha, thanks!