RE: [RFC PATCH v2 10/15] usb:cdns3: Ep0 operations part of the API

From: Pawel Laszczak
Date: Sun Dec 02 2018 - 05:35:24 EST



>> Patch implements related to default endpoint callback functions
>> defined in usb_ep_ops object
>>
>> Signed-off-by: Pawel Laszczak <pawell@xxxxxxxxxxx>
>> ---
>> drivers/usb/cdns3/ep0.c | 191 ++++++++++++++++++++++++++++++++++++-
>> drivers/usb/cdns3/gadget.c | 8 ++
>> drivers/usb/cdns3/gadget.h | 10 ++
>> 3 files changed, 207 insertions(+), 2 deletions(-)
>>
>> diff --git a/drivers/usb/cdns3/ep0.c b/drivers/usb/cdns3/ep0.c
>> index ca1795467155..d05169e73631 100644
>> --- a/drivers/usb/cdns3/ep0.c
>> +++ b/drivers/usb/cdns3/ep0.c
>> @@ -18,11 +18,185 @@ static struct usb_endpoint_descriptor cdns3_gadget_ep0_desc = {
>> .bmAttributes = USB_ENDPOINT_XFER_CONTROL,
>> };
>>
>> +/**
>> + * cdns3_ep0_run_transfer - Do transfer on default endpoint hardware
>> + * @priv_dev: extended gadget object
>> + * @dma_addr: physical address where data is/will be stored
>> + * @length: data length
>> + * @erdy: set it to 1 when ERDY packet should be sent -
>> + * exit from flow control state
>> + */
>> +static void cdns3_ep0_run_transfer(struct cdns3_device *priv_dev,
>> + dma_addr_t dma_addr,
>> + unsigned int length, int erdy)
>> +{
>> + struct cdns3_usb_regs __iomem *regs = priv_dev->regs;
>> +
>> + priv_dev->trb_ep0->buffer = TRB_BUFFER(dma_addr);
>> + priv_dev->trb_ep0->length = TRB_LEN(length);
>> + priv_dev->trb_ep0->control = TRB_CYCLE | TRB_IOC | TRB_TYPE(TRB_NORMAL);
>> +
>> + cdns3_select_ep(priv_dev,
>> + priv_dev->ep0_data_dir ? USB_DIR_IN : USB_DIR_OUT);
>> +
>> + writel(EP_STS_TRBERR, &regs->ep_sts);
>> + writel(EP_TRADDR_TRADDR(priv_dev->trb_ep0_dma), &regs->ep_traddr);
>> +
>> + dev_dbg(&priv_dev->dev, "//Ding Dong ep0%s\n",
>> + priv_dev->ep0_data_dir ? "IN" : "OUT");
>> +
>> + /* TRB should be prepared before starting transfer */
>> + writel(EP_CMD_DRDY, &regs->ep_cmd);
>> +
>> + if (erdy)
>> + writel(EP_CMD_ERDY, &priv_dev->regs->ep_cmd);
>> +}
>> +
>> static void cdns3_prepare_setup_packet(struct cdns3_device *priv_dev)
>> {
>> //TODO: Implements this function
>> }
>>
>> +static void cdns3_set_hw_configuration(struct cdns3_device *priv_dev)
>
>Is this going to be used only for ep0?
>If yes then let's have ep0 in the name.
>If not then this function should be in gadget.c

It going to be used only in ep0.c file but general it's related to all non-control endpoints.

The controller has one strange limitation, namely endpoints configuration in controller
can be set only once. and this function inform control about this.

II think that, will be better to move this function to gadget.c then adding ep0 to name.
>
>> +{
>> + struct cdns3_endpoint *priv_ep;
>> + struct usb_request *request;
>> + struct usb_ep *ep;
>> + int result = 0;
>> +
>> + if (priv_dev->hw_configured_flag)
>> + return;
>> +
>> + writel(USB_CONF_CFGSET, &priv_dev->regs->usb_conf);
>> + writel(EP_CMD_ERDY | EP_CMD_REQ_CMPL, &priv_dev->regs->ep_cmd);
>> +
>> + cdns3_set_register_bit(&priv_dev->regs->usb_conf,
>> + USB_CONF_U1EN | USB_CONF_U2EN);
>
>Shouldn't U1/U2 be enabled only if the USB_DEVICE_U1_ENABLE/USB_DEVICE_U2_ENABLE
>device request was received?

These bits only inform the controller that it should enter to U1/U2 only if host asks for such a transition.

Device to force transition to U1/U2 must use other bits.

>> +
>> + /* wait until configuration set */
>> + result = cdns3_handshake(&priv_dev->regs->usb_sts,
>> + USB_STS_CFGSTS_MASK, 1, 100);
>> +
>> + priv_dev->hw_configured_flag = 1;
>> + cdns3_enable_l1(priv_dev, 1);
>> +
>> + list_for_each_entry(ep, &priv_dev->gadget.ep_list, ep_list) {
>> + if (ep->enabled) {
>> + priv_ep = ep_to_cdns3_ep(ep);
>> + request = cdns3_next_request(&priv_ep->request_list);
>> + if (request)
>> + cdns3_ep_run_transfer(priv_ep, request);
>
>why are you starting transfers in a function that is supposed to set hw configuration only.

Because we can't start transfer before configuring endpoints in hardware.
Also we can't configure each endpoint separately in cdns3_gadget_ep_enable because
USB_CONF_CFGSET can be set only once. It's the reason why transfer are deferred.
It is disadvantage of this controller witch causes some problems.
>
>> + }
>> + }
>> +}
>> +
>> +/**
>> + * cdns3_gadget_ep0_enable
>> + * Function shouldn't be called by gadget driver,
>> + * endpoint 0 is allways active
>> + */
>> +static int cdns3_gadget_ep0_enable(struct usb_ep *ep,
>> + const struct usb_endpoint_descriptor *desc)
>> +{
>> + return -EINVAL;
>> +}
>> +
>> +/**
>> + * cdns3_gadget_ep0_disable
>> + * Function shouldn't be called by gadget driver,
>> + * endpoint 0 is allways active
>> + */
>> +static int cdns3_gadget_ep0_disable(struct usb_ep *ep)
>> +{
>> + return -EINVAL;
>> +}
>> +
>> +/**
>> + * cdns3_gadget_ep0_set_halt
>> + * @ep: pointer to endpoint zero object
>> + * @value: 1 for set stall, 0 for clear stall
>> + *
>> + * Returns 0
>> + */
>> +static int cdns3_gadget_ep0_set_halt(struct usb_ep *ep, int value)
>> +{
>> + /* TODO */
>> + return 0;
>> +}
>> +
>> +/**
>> + * cdns3_gadget_ep0_queue Transfer data on endpoint zero
>> + * @ep: pointer to endpoint zero object
>> + * @request: pointer to request object
>> + * @gfp_flags: gfp flags
>> + *
>> + * Returns 0 on success, error code elsewhere
>> + */
>> +static int cdns3_gadget_ep0_queue(struct usb_ep *ep,
>> + struct usb_request *request,
>> + gfp_t gfp_flags)
>> +{
>> + struct cdns3_endpoint *priv_ep = ep_to_cdns3_ep(ep);
>> + struct cdns3_device *priv_dev = priv_ep->cdns3_dev;
>> + unsigned long flags;
>> + int erdy_sent = 0;
>> + int ret = 0;
>> +
>> + dev_dbg(&priv_dev->dev, "Queue to Ep0%s L: %d\n",
>> + priv_dev->ep0_data_dir ? "IN" : "OUT",
>> + request->length);
>> +
>> + /* send STATUS stage */
>> + if (request->length == 0 && request->zero == 0) {
>
>Is this check sufficient to know STATUS stage?
>It might be normal for vendor specific control request to have
>request->length = 0 and request->zero = 0.

It's sufficient.
cdns3_gadget_ep0_queue is invoked by gadget driver only for DATA stage
or deferred STATUS stage (e.g during SET_CONFIGURATION request).
In DATA stage driver shall send a number of date specified in SETUP packet
and it should not be zero.
If wLength in SETUP packet is 0 then we don't have DATA stage, only SETUP and STATUS stage

>
>
>> + spin_lock_irqsave(&priv_dev->lock, flags);
>> + cdns3_select_ep(priv_dev, 0x00);
>> +
>> + erdy_sent = !priv_dev->hw_configured_flag;
>> + cdns3_set_hw_configuration(priv_dev);
>
>What if we're still busy with DATA stage of previous ep0_queue?
>
>if (!list_empty(&priv_ep->request_list)) should be done as the first thing in this function.
>
The SETUP hast to be handled. If the previous one has not been finished then controller
driver must abort the previous one and start handling the new one.
I know, and I can't find code responsible for this behavior.
I will add some protection against such situation.

>> +
>> + if (!erdy_sent)
>> + writel(EP_CMD_ERDY | EP_CMD_REQ_CMPL,
>> + &priv_dev->regs->ep_cmd);
>> +
>> + cdns3_prepare_setup_packet(priv_dev);
>> + request->actual = 0;
>> + priv_dev->status_completion_no_call = true;
>> + priv_dev->pending_status_request = request;
>> + spin_unlock_irqrestore(&priv_dev->lock, flags);
>> +
>> + /*
>> + * Since there is no completion interrupt for status stage,
>> + * it needs to call ->completion in software after
>> + * ep0_queue is back.
>> + */
>> + queue_work(system_freezable_wq, &priv_dev->pending_status_wq);
>> + return 0;
>> + }
>> +
>> + spin_lock_irqsave(&priv_dev->lock, flags);
>> + if (!list_empty(&priv_ep->request_list)) {
>> + dev_err(&priv_dev->dev,
>> + "can't handle multiple requests for ep0\n");
>> + spin_unlock_irqrestore(&priv_dev->lock, flags);
>> + return -EOPNOTSUPP;
>
>-EBUSY?
>
>> + }
>> +
>> + ret = usb_gadget_map_request_by_dev(priv_dev->sysdev, request,
>> + priv_dev->ep0_data_dir);
>> + if (ret) {
>> + spin_unlock_irqrestore(&priv_dev->lock, flags);
>> + dev_err(&priv_dev->dev, "failed to map request\n");
>> + return -EINVAL;
>> + }
>> +
>> + priv_dev->ep0_request = request;
>> + list_add_tail(&request->list, &priv_ep->request_list);
>> + cdns3_ep0_run_transfer(priv_dev, request->dma, request->length, 1);
>> + spin_unlock_irqrestore(&priv_dev->lock, flags);
>> +
>> + return ret;
>> +}
>> +
>> /**
>> * cdns3_gadget_ep_set_wedge Set wedge on selected endpoint
>> * @ep: endpoint object
>> @@ -41,6 +215,17 @@ int cdns3_gadget_ep_set_wedge(struct usb_ep *ep)
>> return 0;
>> }
>>
>> +const struct usb_ep_ops cdns3_gadget_ep0_ops = {
>> + .enable = cdns3_gadget_ep0_enable,
>> + .disable = cdns3_gadget_ep0_disable,
>> + .alloc_request = cdns3_gadget_ep_alloc_request,
>> + .free_request = cdns3_gadget_ep_free_request,
>> + .queue = cdns3_gadget_ep0_queue,
>> + .dequeue = cdns3_gadget_ep_dequeue,
>> + .set_halt = cdns3_gadget_ep0_set_halt,
>> + .set_wedge = cdns3_gadget_ep_set_wedge,
>> +};
>> +
>> /**
>> * cdns3_ep0_config - Configures default endpoint
>> * @priv_dev: extended gadget object
>> @@ -62,6 +247,9 @@ void cdns3_ep0_config(struct cdns3_device *priv_dev)
>> priv_dev->ep0_request = NULL;
>> }
>>
>> + priv_dev->u1_allowed = 0;
>> + priv_dev->u2_allowed = 0;
>> +
>
>where do you set these?

In function cdns3_ep0_feature_handle_device in patch 12.
It was stupid idea to split this driver into series of patches.
I thought that it was required for such big driver.
The next will be posted as single patch.
>
>> priv_dev->gadget.ep0->maxpacket = max_packet_size;
>> cdns3_gadget_ep0_desc.wMaxPacketSize = cpu_to_le16(max_packet_size);
>>
>> @@ -106,8 +294,7 @@ int cdns3_init_ep0(struct cdns3_device *priv_dev)
>> sprintf(ep0->name, "ep0");
>>
>> /* fill linux fields */
>> - //TODO: implements cdns3_gadget_ep0_ops object
>> - //ep0->endpoint.ops = &cdns3_gadget_ep0_ops;
>> + ep0->endpoint.ops = &cdns3_gadget_ep0_ops;
>> ep0->endpoint.maxburst = 1;
>> usb_ep_set_maxpacket_limit(&ep0->endpoint, ENDPOINT0_MAX_PACKET_LIMIT);
>> ep0->endpoint.address = 0;
>> diff --git a/drivers/usb/cdns3/gadget.c b/drivers/usb/cdns3/gadget.c
>> index 1f2a434486dc..c965da16c0c8 100644
>> --- a/drivers/usb/cdns3/gadget.c
>> +++ b/drivers/usb/cdns3/gadget.c
>> @@ -188,6 +188,14 @@ static void cdns3_ep_stall_flush(struct cdns3_endpoint *priv_ep)
>> priv_ep->flags |= EP_STALL;
>> }
>>
>
>> +void cdns3_enable_l1(struct cdns3_device *priv_dev, int enable)
>
>Some comment might help as to what L1 is.

I even had to check what it meant.

>From controller specification:
"
After receiving Extended Token packet, the USBSS-DEV answers with:
- ACK handshake when accepts transition to the Sleep state (L1 mode is turned on by setting
L1EN in USB_CONF) or
- NYET handshake when is not ready to make transition to the Sleep state (L1 mode is turned
off (default) by setting L1DS in USB_CONF )
- STALL handshake when doesn't recognize requested link state (bLinkState value). The only
supported bLinkState value is "0001" (Sleep state).
If L1 mode is enabled, then the status bit 'L1ENS' in 'USB_STS' is asserted."

So I think that the name of function should look like:
cdns3_allow_enable_l1

I will add comment to function and change the name.
>
>> +{
>> + if (enable)
>> + writel(USB_CONF_L1EN, &priv_dev->regs->usb_conf);
>> + else
>> + writel(USB_CONF_L1DS, &priv_dev->regs->usb_conf);
>> +}
>> +
>> /**
>> * cdns3_gadget_giveback - call struct usb_request's ->complete callback
>> * @priv_ep: The endpoint to whom the request belongs to
>> diff --git a/drivers/usb/cdns3/gadget.h b/drivers/usb/cdns3/gadget.h
>> index a4be288b34cb..224f6b830bc9 100644
>> --- a/drivers/usb/cdns3/gadget.h
>> +++ b/drivers/usb/cdns3/gadget.h
>> @@ -1068,11 +1068,21 @@ struct cdns3_device {
>> struct usb_request *pending_status_request;
>> };
>>
>> +int cdns3_handshake(void __iomem *ptr, u32 mask, u32 done, int usec);
>> void cdns3_set_register_bit(void __iomem *ptr, u32 mask);
>> int cdns3_init_ep0(struct cdns3_device *priv_dev);
>> void cdns3_ep0_config(struct cdns3_device *priv_dev);
>> void cdns3_select_ep(struct cdns3_device *priv_dev, u32 ep);
>> +void cdns3_enable_l1(struct cdns3_device *priv_dev, int enable);
>> +struct usb_request *cdns3_next_request(struct list_head *list);
>> +int cdns3_ep_run_transfer(struct cdns3_endpoint *priv_ep,
>> + struct usb_request *request);
>> int cdns3_gadget_ep_set_wedge(struct usb_ep *ep);
>> int cdns3_gadget_ep_set_halt(struct usb_ep *ep, int value);
>> +struct usb_request *cdns3_gadget_ep_alloc_request(struct usb_ep *ep,
>> + gfp_t gfp_flags);
>> +void cdns3_gadget_ep_free_request(struct usb_ep *ep,
>> + struct usb_request *request);
>> +int cdns3_gadget_ep_dequeue(struct usb_ep *ep, struct usb_request *request);
>>
>> #endif /* __LINUX_CDNS3_GADGET */
>>
>
>cheers,
>-roger
>
>--
>Texas Instruments Finland Oy, Porkkalankatu 22, 00180 Helsinki.
>Y-tunnus/Business ID: 0615521-4. Kotipaikka/Domicile: Helsinki