Re: [RFC PATCH v3 1/5] usb: xhci-mtk: improve bandwidth scheduling with multi-TT

From: Ikjoon Jang
Date: Wed Jan 13 2021 - 04:44:41 EST

Next message: Suzuki K Poulose: "Re: [PATCH V2 08/11] coresight: core: Add support for dedicated percpu sinks"
Previous message: Maxime Ripard: "Re: [PATCH v3] drm/sun4i: tcon: fix inverted DCLK polarity"
Messages sorted by: [ date ] [ thread ] [ subject ] [ author ]

On Tue, Dec 22, 2020 at 5:35 PM Chunfeng Yun <chunfeng.yun@xxxxxxxxxxxx> wrote:
>
> From: Zhanyong Wang <zhanyong.wang@xxxxxxxxxxxx>
>
> After inserted the usb type-c 3.5mm dongle with headset, dmesg showed:
> usb 1-1.1: new full-speed USB device number 5 using xhci-mtk
> usb 1-1.1: New USB device found, idVendor=05ac, idProduct=110a, bcdDevice=26.11
> usb 1-1.1: New USB device strings: Mfr=1, Product=2, SerialNumber=3
> usb 1-1.1: Product: USB-C to 3.5mm Headphone Jack Adapter
> usb 1-1.1: Manufacturer: Apple, Inc.
> usb 1-1.1: SerialNumber: DWH915501TFJKLTAM
> xhci-mtk 11200000.xhci: Not enough bandwidth!
> usb 1-1.1: can't set config #2, error -28
>
> improve low-speed/full-speed INT/ISOC bandwidth scheduling with USB
> muli-TT.
>
> Signed-off-by: Yaqii Wu <yaqii.wu@xxxxxxxxxxxx>
> Signed-off-by: Chunfeng Yun <chunfeng.yun@xxxxxxxxxxxx>
> ---
> v2~v3: no changes
> ---
> drivers/usb/host/xhci-mtk-sch.c | 91 ++++++++++++++++++++++++++++-----
> drivers/usb/host/xhci-mtk.h | 8 ++-
> 2 files changed, 84 insertions(+), 15 deletions(-)
> mode change 100644 => 100755 drivers/usb/host/xhci-mtk-sch.c
>
> diff --git a/drivers/usb/host/xhci-mtk-sch.c b/drivers/usb/host/xhci-mtk-sch.c
> old mode 100644
> new mode 100755
> index 45c54d56ecbd..94292b9bbc63
> --- a/drivers/usb/host/xhci-mtk-sch.c
> +++ b/drivers/usb/host/xhci-mtk-sch.c
> @@ -383,7 +383,9 @@ static int check_sch_tt(struct usb_device *udev,
> u32 fs_budget_start;
> u32 start_ss, last_ss;
> u32 start_cs, last_cs;
> - int i;
> + u32 num_esit, base;
> + int i, j;
> + u32 tmp;
>
> start_ss = offset % 8;
> fs_budget_start = (start_ss + 1) % 8;
> @@ -398,10 +400,13 @@ static int check_sch_tt(struct usb_device *udev,
> if (!(start_ss == 7 || last_ss < 6))
> return -ERANGE;
>
> - for (i = 0; i < sch_ep->cs_count; i++)
> - if (test_bit(offset + i, tt->split_bit_map))
> + for (i = 0; i < sch_ep->cs_count; i++) {
> + if (test_bit(offset + i, tt->ss_bit_map))
> return -ERANGE;
>
> + if (test_bit(offset + i, tt->cs_bit_map))
> + return -ERANGE;
> + }

Are there any reasons for doing this?
Why can only one split packet be scheduled in a u-frame for isochronous out?
This looks like overkill.

> } else {
> u32 cs_count = DIV_ROUND_UP(sch_ep->maxpkt, FS_PAYLOAD_MAX);
>
> @@ -428,8 +433,10 @@ static int check_sch_tt(struct usb_device *udev,
> if (cs_count > 7)
> cs_count = 7; /* HW limit */
>
> - for (i = 0; i < cs_count + 2; i++) {
> - if (test_bit(offset + i, tt->split_bit_map))
> + if (test_bit(offset, tt->ss_bit_map))
> + return -ERANGE;
> + for (i = 0; i < cs_count; i++) {
> + if (test_bit(offset + 2 + i, tt->cs_bit_map))
> return -ERANGE;
> }
>

same here. It would be much better to understand
if you can provide some counterexamples of schedule
that can happen when this bitmap checking logic is omitted.

> @@ -445,11 +452,22 @@ static int check_sch_tt(struct usb_device *udev,
> sch_ep->num_budget_microframes = sch_ep->esit;
> }
>
> + num_esit = XHCI_MTK_MAX_ESIT / sch_ep->esit;
> + for (i = 0; i < num_esit; i++) {
> + base = sch_ep->offset + i * sch_ep->esit;
> + for (j = 0; j < sch_ep->num_budget_microframes; j++) {
> + tmp = tt->fs_bus_bw[base + j]
> + + sch_ep->bw_cost_per_microframe;
> + if (tmp > FS_PAYLOAD_MAX)
> + return -ERANGE;
> + }
> + }

I guess this is enough to check the bandwidth limit of the
lower speed bus without a bitmap.

> +
> return 0;
> }
>
> static void update_sch_tt(struct usb_device *udev,
> - struct mu3h_sch_ep_info *sch_ep)
> + struct mu3h_sch_ep_info *sch_ep, bool used)
> {
> struct mu3h_sch_tt *tt = sch_ep->sch_tt;
> u32 base, num_esit;
> @@ -458,11 +476,52 @@ static void update_sch_tt(struct usb_device *udev,
> num_esit = XHCI_MTK_MAX_ESIT / sch_ep->esit;
> for (i = 0; i < num_esit; i++) {
> base = sch_ep->offset + i * sch_ep->esit;
> - for (j = 0; j < sch_ep->num_budget_microframes; j++)
> - set_bit(base + j, tt->split_bit_map);
> + for (j = 0; j < sch_ep->num_budget_microframes; j++) {
> + if (used)
> + set_bit(base + j, tt->split_bit_map);
> + else
> + clear_bit(base + j, tt->split_bit_map);
> + }
> +
> + if (sch_ep->ep_type == ISOC_OUT_EP) {
> + for (j = 0; j < sch_ep->num_budget_microframes; j++) {
> + if (used) {
> + set_bit(base + j, tt->ss_bit_map);
> + set_bit(base + j, tt->cs_bit_map);
> + tt->fs_bus_bw[base + j] +=
> + sch_ep->bw_cost_per_microframe;
> + } else {
> + clear_bit(base + j, tt->ss_bit_map);
> + clear_bit(base + j, tt->cs_bit_map);
> + tt->fs_bus_bw[base + j] -=
> + sch_ep->bw_cost_per_microframe;
> + }
> + }
> + } else {
> + if (used)
> + set_bit(base, tt->ss_bit_map);
> + else
> + clear_bit(base, tt->ss_bit_map);
> +
> + for (j = 0; j < sch_ep->cs_count; j++) {
> + if (used) {
> + set_bit(base + 2 + j, tt->cs_bit_map);
> +
> + tt->fs_bus_bw[base + 2 + j] +=
> + sch_ep->bw_cost_per_microframe;
> + } else {
> + clear_bit(base + 2 + j, tt->cs_bit_map);
> + tt->fs_bus_bw[base + 2 + j] -=
> + sch_ep->bw_cost_per_microframe;
> + }
> + }
> + }
> }
>
> - list_add_tail(&sch_ep->tt_endpoint, &tt->ep_list);
> + if (used)
> + list_add_tail(&sch_ep->tt_endpoint, &tt->ep_list);
> + else
> + list_del(&sch_ep->tt_endpoint);
> }
>
> static int check_sch_bw(struct usb_device *udev,
> @@ -470,6 +529,7 @@ static int check_sch_bw(struct usb_device *udev,
> {
> u32 offset;
> u32 esit;
> + u32 boundary;
> u32 min_bw;
> u32 min_index;
> u32 worst_bw;
> @@ -487,10 +547,13 @@ static int check_sch_bw(struct usb_device *udev,
> */
> min_bw = ~0;
> min_index = 0;
> + boundary = esit;
> min_cs_count = sch_ep->cs_count;
> min_num_budget = sch_ep->num_budget_microframes;
> for (offset = 0; offset < esit; offset++) {
> if (is_fs_or_ls(udev->speed)) {
> + if (sch_ep->ep_type != ISOC_OUT_EP)
> + boundary = esit + 1;
> ret = check_sch_tt(udev, sch_ep, offset);
> if (ret)
> continue;
> @@ -498,7 +561,7 @@ static int check_sch_bw(struct usb_device *udev,
> tt_offset_ok = true;
> }
>
> - if ((offset + sch_ep->num_budget_microframes) > sch_ep->esit)
> + if ((offset + sch_ep->num_budget_microframes) > boundary)
> break;
>
> worst_bw = get_max_bw(sch_bw, sch_ep, offset);
> @@ -532,7 +595,7 @@ static int check_sch_bw(struct usb_device *udev,
> if (!tt_offset_ok)
> return -ERANGE;
>
> - update_sch_tt(udev, sch_ep);
> + update_sch_tt(udev, sch_ep, 1);
> }
>
> /* update bus bandwidth info */
> @@ -696,12 +759,12 @@ void xhci_mtk_drop_ep_quirk(struct usb_hcd *hcd, struct usb_device *udev,
>
> list_for_each_entry(sch_ep, &sch_bw->bw_ep_list, endpoint) {
> if (sch_ep->ep == ep) {
> - update_bus_bw(sch_bw, sch_ep, 0);
> - list_del(&sch_ep->endpoint);
> if (is_fs_or_ls(udev->speed)) {
> - list_del(&sch_ep->tt_endpoint);
> + update_sch_tt(udev, sch_ep, 0);
> drop_tt(udev);
> }
> + update_bus_bw(sch_bw, sch_ep, 0);
> + list_del(&sch_ep->endpoint);
> kfree(sch_ep);
> break;
> }
> diff --git a/drivers/usb/host/xhci-mtk.h b/drivers/usb/host/xhci-mtk.h
> index a93cfe817904..323b281933b9 100644
> --- a/drivers/usb/host/xhci-mtk.h
> +++ b/drivers/usb/host/xhci-mtk.h
> @@ -21,12 +21,18 @@
>
> /**
> * @split_bit_map: used to avoid split microframes overlay
> + * @ss_bit_map: used to avoid start split microframes overlay
> + * @cs_bit_map: used to avoid complete split microframes overlay
> + * @fs_bus_bw: array to keep track of bandwidth already used at full speed
> * @ep_list: Endpoints using this TT
> * @usb_tt: usb TT related
> * @tt_port: TT port number
> */
> struct mu3h_sch_tt {
> DECLARE_BITMAP(split_bit_map, XHCI_MTK_MAX_ESIT);
> + DECLARE_BITMAP(ss_bit_map, XHCI_MTK_MAX_ESIT);
> + DECLARE_BITMAP(cs_bit_map, XHCI_MTK_MAX_ESIT + 1);
> + u32 fs_bus_bw[XHCI_MTK_MAX_ESIT];
> struct list_head ep_list;
> struct usb_tt *usb_tt;
> int tt_port;
> @@ -42,7 +48,7 @@ struct mu3h_sch_tt {
> * two bandwidth domains, one for IN eps and another for OUT eps.
> */
> struct mu3h_sch_bw_info {
> - u32 bus_bw[XHCI_MTK_MAX_ESIT];
> + u32 bus_bw[XHCI_MTK_MAX_ESIT + 1];
> struct list_head bw_ep_list;
> };
>
> --
> 2.18.0
> _______________________________________________
> Linux-mediatek mailing list
> Linux-mediatek@xxxxxxxxxxxxxxxxxxx
> http://lists.infradead.org/mailman/listinfo/linux-mediatek

Next message: Suzuki K Poulose: "Re: [PATCH V2 08/11] coresight: core: Add support for dedicated percpu sinks"
Previous message: Maxime Ripard: "Re: [PATCH v3] drm/sun4i: tcon: fix inverted DCLK polarity"
Messages sorted by: [ date ] [ thread ] [ subject ] [ author ]