On 26.4.2022 12.18, Daehwan Jung wrote:
> This driver is for Samsung Exynos xhci host conroller. It uses xhci-plat
> driver mainly and extends some functions by xhci hooks and overrides.
>
> It supports USB Audio offload with Co-processor. It only cares DCBAA,
> Device Context, Transfer Ring, Event Ring, and ERST. They are allocated
> on specific address with xhci hooks. Co-processor could use them directly
> without xhci driver after then.
>
> Signed-off-by: Daehwan Jung <dh10.jung@xxxxxxxxxxx>
I have to agree with Krzysztof's comments, this is an odd driver stub.
Perhaps open up a bit how the Exynos offloading works so we can figure out
in more detail what the hardware needs from software.
(...)
> +
> +static void xhci_exynos_alloc_container_ctx(struct xhci_hcd *xhci, struct xhci_container_ctx *ctx,
> + int type, gfp_t flags)
> +{
> + /* Only first Device Context uses URAM */
> + int i;
> +
> + ctx->bytes = ioremap(EXYNOS_URAM_DEVICE_CTX_ADDR, EXYNOS_URAM_CTX_SIZE);
> + if (!ctx->bytes)
> + return;
> +
> + for (i = 0; i < EXYNOS_URAM_CTX_SIZE; i++)
> + ctx->bytes[i] = 0;
> +
> + ctx->dma = EXYNOS_URAM_DEVICE_CTX_ADDR;
This can't work with more than one USB device.
This hardcodes the same context address for every usb device.
> +static void xhci_exynos_parse_endpoint(struct xhci_hcd *xhci, struct usb_device *udev,
> + struct usb_endpoint_descriptor *desc, struct xhci_container_ctx *ctx)
> +{
> + struct xhci_plat_priv *priv = xhci_to_priv(xhci);
> + struct xhci_hcd_exynos *xhci_exynos = priv->vendor_priv;
> + struct usb_endpoint_descriptor *d = desc;
> + unsigned int ep_index;
> + struct xhci_ep_ctx *ep_ctx;
> +
> + ep_index = xhci_get_endpoint_index(d);
> + ep_ctx = xhci_get_ep_ctx(xhci, ctx, ep_index);
> +
> + if ((d->bmAttributes & USB_ENDPOINT_XFERTYPE_MASK) ==
> + USB_ENDPOINT_XFER_ISOC) {
> + if (d->bEndpointAddress & USB_ENDPOINT_DIR_MASK)
> + xhci_exynos->in_ep = d->bEndpointAddress;
> + else
> + xhci_exynos->out_ep = d->bEndpointAddress;
> + }
This won't work if more than one device that has isoc endpoints, or even
if that device has more than one in/out isoc endpoint pair.
> +static int xhci_alloc_segments_for_ring_uram(struct xhci_hcd *xhci,
> + struct xhci_segment **first, struct xhci_segment **last,
> + unsigned int num_segs, unsigned int cycle_state,
> + enum xhci_ring_type type, unsigned int max_packet, gfp_t flags,
> + u32 endpoint_type)
> +{
> + struct xhci_segment *prev;
> + bool chain_links = false;
> +
> + while (num_segs > 0) {
> + struct xhci_segment *next = NULL;
> +
> + if (!next) {
> + prev = *first;
> + while (prev) {
> + next = prev->next;
> + xhci_segment_free(xhci, prev);
> + prev = next;
> + }
> + return -ENOMEM;
This always return -ENOMEM
Also this whole function never allocates or remaps any memory.
> + }
> + xhci_link_segments(prev, next, type, chain_links);
> +
> + prev = next;
> + num_segs--;
> + }
> + xhci_link_segments(prev, *first, type, chain_links);
> + *last = prev;
> +
> + return 0;
> +}
> +
> +static struct xhci_ring *xhci_ring_alloc_uram(struct xhci_hcd *xhci,
> + unsigned int num_segs, unsigned int cycle_state,
> + enum xhci_ring_type type, unsigned int max_packet, gfp_t flags,
> + u32 endpoint_type)
> +{
> + struct xhci_ring *ring;
> + int ret;
> + struct device *dev = xhci_to_hcd(xhci)->self.sysdev;
> +
> + ring = kzalloc_node(sizeof(*ring), flags, dev_to_node(dev));
> + if (!ring)
> + return NULL;
> +
> + ring->num_segs = num_segs;
> + ring->bounce_buf_len = max_packet;
> + INIT_LIST_HEAD(&ring->td_list);
> + ring->type = type;
> + if (num_segs == 0)
> + return ring;
> +
> + ret = xhci_alloc_segments_for_ring_uram(xhci, &ring->first_seg,
> + &ring->last_seg, num_segs, cycle_state, type,
> + max_packet, flags, endpoint_type);
> + if (ret)
> + goto fail;
> +
> + /* Only event ring does not use link TRB */
> + if (type != TYPE_EVENT) {
> + /* See section 4.9.2.1 and 6.4.4.1 */
> + ring->last_seg->trbs[TRBS_PER_SEGMENT - 1].link.control |=
> + cpu_to_le32(LINK_TOGGLE);
No memory was allocated for trbs
A lot of this code seems to exists just to avoid xhci driver from allocating
dma capable memory, we can refactor the existing xhci_mem_init() and move
dcbaa and event ring allocation and other code to their own overridable
functions.
This way we can probably get rid of a lot of the code in this series.
Thanks
Mathias
