| Message ID | 20241208152322.1653-1-selvarasu.g@samsung.com (mailing list archive) |
|---|---|
| State | New |
| Headers | show |
| Series | usb: gadget: f_midi: Fixing wMaxPacketSize exceeded issue during MIDI bind retries | expand |
Hello Maintainers. Gentle remainder for review. Thanks, Selva On 12/8/2024 8:53 PM, Selvarasu Ganesan wrote: > The current implementation sets the wMaxPacketSize of bulk in/out > endpoints to 1024 bytes at the end of the f_midi_bind function. However, > in cases where there is a failure in the first midi bind attempt, > consider rebinding. This scenario may encounter an f_midi_bind issue due > to the previous bind setting the bulk endpoint's wMaxPacketSize to 1024 > bytes, which exceeds the ep->maxpacket_limit where configured TX/RX > FIFO's maxpacket size of 512 bytes for IN/OUT endpoints in support HS > speed only. > This commit addresses this issue by resetting the wMaxPacketSize before > endpoint claim. > > Fixes: 46decc82ffd5 ("usb: gadget: unconditionally allocate hs/ss descriptor in bind operation") > Cc: stable@vger.kernel.org > Signed-off-by: Selvarasu Ganesan <selvarasu.g@samsung.com> > --- > drivers/usb/gadget/function/f_midi.c | 9 +++++++++ > 1 file changed, 9 insertions(+) > > diff --git a/drivers/usb/gadget/function/f_midi.c b/drivers/usb/gadget/function/f_midi.c > index 837fcdfa3840..5caa0e4eb07e 100644 > --- a/drivers/usb/gadget/function/f_midi.c > +++ b/drivers/usb/gadget/function/f_midi.c > @@ -907,6 +907,15 @@ static int f_midi_bind(struct usb_configuration *c, struct usb_function *f) > > status = -ENODEV; > > + /* > + * Reset wMaxPacketSize with maximum packet size of FS bulk transfer before > + * endpoint claim. This ensures that the wMaxPacketSize does not exceed the > + * limit during bind retries where configured TX/RX FIFO's maxpacket size > + * of 512 bytes for IN/OUT endpoints in support HS speed only. > + */ > + bulk_in_desc.wMaxPacketSize = cpu_to_le16(64); > + bulk_out_desc.wMaxPacketSize = cpu_to_le16(64); > + > /* allocate instance-specific endpoints */ > midi->in_ep = usb_ep_autoconfig(cdev->gadget, &bulk_in_desc); > if (!midi->in_ep)
On Tue, Dec 10, 2024 at 03:23:22PM +0530, Selvarasu Ganesan wrote: > Hello Maintainers. > > Gentle remainder for review. You sent this 2 days ago, right? Please take the time to review other commits on the miailing list if you wish to see your patches get reviewed faster, to help reduce the workload of people reviewing your changes. Otherwise just wait for people to get to it, what is the rush here? thanks, greg k-h
On 12/10/2024 3:48 PM, Greg KH wrote: > On Tue, Dec 10, 2024 at 03:23:22PM +0530, Selvarasu Ganesan wrote: >> Hello Maintainers. >> >> Gentle remainder for review. > You sent this 2 days ago, right? > > Please take the time to review other commits on the miailing list if you > wish to see your patches get reviewed faster, to help reduce the > workload of people reviewing your changes. > > Otherwise just wait for people to get to it, what is the rush here? > > thanks, > > greg k-h Hi Greg, There is no rush. I understand that the review will take time and I apologize for any inconvenience caused by sending the reminder email. Thanks, Selva
On Tue, Dec 10, 2024 at 07:41:53PM +0530, Selvarasu Ganesan wrote: > > On 12/10/2024 3:48 PM, Greg KH wrote: > > On Tue, Dec 10, 2024 at 03:23:22PM +0530, Selvarasu Ganesan wrote: > >> Hello Maintainers. > >> > >> Gentle remainder for review. > > You sent this 2 days ago, right? > > > > Please take the time to review other commits on the miailing list if you > > wish to see your patches get reviewed faster, to help reduce the > > workload of people reviewing your changes. > > > > Otherwise just wait for people to get to it, what is the rush here? > > > > thanks, > > > > greg k-h > > > Hi Greg, > > > There is no rush. I understand that the review will take time and I > apologize for any inconvenience caused by sending the reminder email. Great, during that time, please do some patch reviews of the changes on the mailing list to help us out. thanks, greg k-h
On 12/10/2024 7:55 PM, Greg KH wrote: > On Tue, Dec 10, 2024 at 07:41:53PM +0530, Selvarasu Ganesan wrote: >> On 12/10/2024 3:48 PM, Greg KH wrote: >>> On Tue, Dec 10, 2024 at 03:23:22PM +0530, Selvarasu Ganesan wrote: >>>> Hello Maintainers. >>>> >>>> Gentle remainder for review. >>> You sent this 2 days ago, right? >>> >>> Please take the time to review other commits on the miailing list if you >>> wish to see your patches get reviewed faster, to help reduce the >>> workload of people reviewing your changes. >>> >>> Otherwise just wait for people to get to it, what is the rush here? >>> >>> thanks, >>> >>> greg k-h >> >> Hi Greg, >> >> >> There is no rush. I understand that the review will take time and I >> apologize for any inconvenience caused by sending the reminder email. > Great, during that time, please do some patch reviews of the changes on > the mailing list to help us out. > > thanks, > > greg k-h Hi Greg, Sure, I will try to contribute on reviewing other patches. Thanks, Selva >
On Sun, Dec 08, 2024 at 08:53:20PM +0530, Selvarasu Ganesan wrote: > The current implementation sets the wMaxPacketSize of bulk in/out > endpoints to 1024 bytes at the end of the f_midi_bind function. However, > in cases where there is a failure in the first midi bind attempt, > consider rebinding. What considers rebinding? Your change does not modify that. > This scenario may encounter an f_midi_bind issue due > to the previous bind setting the bulk endpoint's wMaxPacketSize to 1024 > bytes, which exceeds the ep->maxpacket_limit where configured TX/RX > FIFO's maxpacket size of 512 bytes for IN/OUT endpoints in support HS > speed only. > This commit addresses this issue by resetting the wMaxPacketSize before > endpoint claim. resets it to what? Where did the magic numbers come from? How do we know this is now full speed and not high speed? > > Fixes: 46decc82ffd5 ("usb: gadget: unconditionally allocate hs/ss descriptor in bind operation") > Cc: stable@vger.kernel.org > Signed-off-by: Selvarasu Ganesan <selvarasu.g@samsung.com> > --- > drivers/usb/gadget/function/f_midi.c | 9 +++++++++ > 1 file changed, 9 insertions(+) > > diff --git a/drivers/usb/gadget/function/f_midi.c b/drivers/usb/gadget/function/f_midi.c > index 837fcdfa3840..5caa0e4eb07e 100644 > --- a/drivers/usb/gadget/function/f_midi.c > +++ b/drivers/usb/gadget/function/f_midi.c > @@ -907,6 +907,15 @@ static int f_midi_bind(struct usb_configuration *c, struct usb_function *f) > > status = -ENODEV; > > + /* > + * Reset wMaxPacketSize with maximum packet size of FS bulk transfer before > + * endpoint claim. This ensures that the wMaxPacketSize does not exceed the > + * limit during bind retries where configured TX/RX FIFO's maxpacket size > + * of 512 bytes for IN/OUT endpoints in support HS speed only. > + */ > + bulk_in_desc.wMaxPacketSize = cpu_to_le16(64); > + bulk_out_desc.wMaxPacketSize = cpu_to_le16(64); Where did "64" come from? How do we know this is full speed? Later on in this function the endpoint sizes are set, why set them here to these small values when you do not know the speed? Or, if it had failed before, reset the values on the failure, not here before you start anything up, right? thanks, greg k-h
On 12/18/2024 11:01 AM, Greg KH wrote: > On Sun, Dec 08, 2024 at 08:53:20PM +0530, Selvarasu Ganesan wrote: >> The current implementation sets the wMaxPacketSize of bulk in/out >> endpoints to 1024 bytes at the end of the f_midi_bind function. However, >> in cases where there is a failure in the first midi bind attempt, >> consider rebinding. > What considers rebinding? Your change does not modify that. Hi Greg, Thanks for your review comments. Here the term "rebind" in this context refers to attempting to bind the MIDI function a second time in certain scenarios. The situations where rebinding is considered include: * When there is a failure in the first UDC write attempt, which may be caused by other functions bind along with MIDI * Runtime composition change : Example : MIDI,ADB to MIDI. Or MIDI to MIDI,ADB The issue arises during the second time the "f_midi_bind" function is called. The problem lies in the fact that the size of "bulk_in_desc.wMaxPacketSize" is set to 1024 during the first call, which exceeds the hardware capability of the dwc3 TX/RX FIFO (ep->maxpacket_limit = 512). Let consider the below sequence, _1. First time f_midi_bind:_ * As per the current codethe size of bulk_in_desc.wMaxPacketSize is 0 in first time call. * Call usb_ep_autoconfig to match EP and got success as no failure in the below code as part of usb_ep_autoconfig. usb_gadget_ep_match_desc() { .. .. if (max > ep->maxpacket_limit)// (64 > 512) return 0; return 1;// Found Maching EP } * EP claim got success and set bulk_in_desc.wMaxPacketSize =1024 at end of f_midi_bind. _2. Second time enter into f_midi_bind _ * The size of bulk_in_desc.wMaxPacketSize become 1024 because of above code. * Call usb_ep_autoconfig for EP claim and has a failure now in the below code as part of usb_ep_autoconfig usb_gadget_ep_match_desc() { .. .. if (max > ep->maxpacket_limit)// (1024 > 512) return 0; // Not found any matchingEP } To resolve this issue, our patch sets the default value of "bulk_in_desc.wMaxPacketSize" to 64 before endpoint claim. This ensures that the endpoint claim is successful during the second time "f_midi_bind" is called. > >> This scenario may encounter an f_midi_bind issue due >> to the previous bind setting the bulk endpoint's wMaxPacketSize to 1024 >> bytes, which exceeds the ep->maxpacket_limit where configured TX/RX >> FIFO's maxpacket size of 512 bytes for IN/OUT endpoints in support HS >> speed only. >> This commit addresses this issue by resetting the wMaxPacketSize before >> endpoint claim. > resets it to what? Where did the magic numbers come from? How do we > know this is now full speed and not high speed? It’s a generic way of setwMaxPacketSize before endpoint claim. Its because of the usb_ep_autoconfig treats endpoint descriptors as if they were full speed. This approach follows the same pattern as other function drivers, which also set the wMaxPacketSize to 64 before endpoint claim. The following are the example of how other functions drivers EP claim. _1. drivers/usb/gadget/function/f_eem.c_ eem_bind () { … .. /* allocate instance-specific endpoints */ ep = usb_ep_autoconfig(cdev->gadget, &eem_fs_in_desc); //eem_fs_in_desc.wMaxPacketSize=64 if (!ep) goto fail; } _2. drivers/usb/gadget/function/f_rndis.c_ rndis_bind() { ... ... /* allocate instance-specific endpoints */ ep = usb_ep_autoconfig(cdev->gadget, &fs_in_desc); //fs_in_desc.wMaxPacketSize=64 if (!ep) goto fail; rndis->port.in_ep = ep; } Anyway the wMaxPacketSize set with 64 byte after complete usb_ep_autoconfig() as part of below this function ifep->maxpacket_limit is grater then 64. It's means treats endpoint descriptors as if they were full speed by default. struct usb_ep *usb_ep_autoconfig() { .. … if (type == USB_ENDPOINT_XFER_BULK) { int size = ep->maxpacket_limit; /* min() doesn't work on bitfields with gcc-3.5 */ if (size > 64) size = 64; desc->wMaxPacketSize = cpu_to_le16(size); } .. .. } >> Fixes: 46decc82ffd5 ("usb: gadget: unconditionally allocate hs/ss descriptor in bind operation") >> Cc: stable@vger.kernel.org >> Signed-off-by: Selvarasu Ganesan <selvarasu.g@samsung.com> >> --- >> drivers/usb/gadget/function/f_midi.c | 9 +++++++++ >> 1 file changed, 9 insertions(+) >> >> diff --git a/drivers/usb/gadget/function/f_midi.c b/drivers/usb/gadget/function/f_midi.c >> index 837fcdfa3840..5caa0e4eb07e 100644 >> --- a/drivers/usb/gadget/function/f_midi.c >> +++ b/drivers/usb/gadget/function/f_midi.c >> @@ -907,6 +907,15 @@ static int f_midi_bind(struct usb_configuration *c, struct usb_function *f) >> >> status = -ENODEV; >> >> + /* >> + * Reset wMaxPacketSize with maximum packet size of FS bulk transfer before >> + * endpoint claim. This ensures that the wMaxPacketSize does not exceed the >> + * limit during bind retries where configured TX/RX FIFO's maxpacket size >> + * of 512 bytes for IN/OUT endpoints in support HS speed only. >> + */ >> + bulk_in_desc.wMaxPacketSize = cpu_to_le16(64); >> + bulk_out_desc.wMaxPacketSize = cpu_to_le16(64); > Where did "64" come from? How do we know this is full speed? Later > on in this function the endpoint sizes are set, why set them here to > these small values when you do not know the speed? > > Or, if it had failed before, reset the values on the failure, not here > before you start anything up, right? Explained as part of above ask. > > thanks, > > greg k-h >
On Wed, Dec 18, 2024 at 03:51:50PM +0530, Selvarasu Ganesan wrote: > The issue arises during the second time the "f_midi_bind" function is > called. The problem lies in the fact that the size of > "bulk_in_desc.wMaxPacketSize" is set to 1024 during the first call, > which exceeds the hardware capability of the dwc3 TX/RX FIFO > (ep->maxpacket_limit = 512). Is this gadget supposed to be able to run at SuperSpeed? I thought the dwc3 controller supported SuperSpeed operation. The USB-3 spec requires that all SuperSpeed bulk endpoints must have a wMaxPacketSize of 1024 (see Table 9-24 on p.9-25 of the USB-3.1 specification). Alan Stern
On 12/18/2024 9:21 PM, Alan Stern wrote: > On Wed, Dec 18, 2024 at 03:51:50PM +0530, Selvarasu Ganesan wrote: >> The issue arises during the second time the "f_midi_bind" function is >> called. The problem lies in the fact that the size of >> "bulk_in_desc.wMaxPacketSize" is set to 1024 during the first call, >> which exceeds the hardware capability of the dwc3 TX/RX FIFO >> (ep->maxpacket_limit = 512). > Is this gadget supposed to be able to run at SuperSpeed? I thought the > dwc3 controller supported SuperSpeed operation. > > The USB-3 spec requires that all SuperSpeed bulk endpoints must have a > wMaxPacketSize of 1024 (see Table 9-24 on p.9-25 of the USB-3.1 > specification). > > Alan Stern > Hi Alan, No, In our platform, the DWC3 controller supports up to HighSpeed. While DWC3 is capable of SuperSpeed operation, it is not necessary to operate at the same speed. Moreover, even in our SoC, the DWC3 IP is limited to supporting only USB 2.0 mode (HighSpeed) at the hardware design level. As previously mentioned, there is no need to set the wMaxPacketSize based on the current speed support before claiming the endpoint (before calling usb_ep_autoconfig), as usb_ep_autoconfig treats endpoint descriptors as if they were full-speed by default. For reference, let's see the usb_ep_autoconfig code where the wMaxPacketSize is set to 64 bytes if the ep->maxpacket_limit is greater than 64. As i mentioned earlier in our specific failure scenarios, the code does not reach this point because the wMaxPacketSize is greater than the ep->maxpacket_limit. struct usb_ep *usb_ep_autoconfig() { … … … /* report (variable) full speed bulk maxpacket */ if (type == USB_ENDPOINT_XFER_BULK) { int size = ep->maxpacket_limit; if (size > 64) size = 64; desc->wMaxPacketSize = cpu_to_le16(size); } return ep; } Thanks, Selva
On Wed, Dec 18, 2024 at 03:51:50PM +0530, Selvarasu Ganesan wrote: > > On 12/18/2024 11:01 AM, Greg KH wrote: > > On Sun, Dec 08, 2024 at 08:53:20PM +0530, Selvarasu Ganesan wrote: > >> The current implementation sets the wMaxPacketSize of bulk in/out > >> endpoints to 1024 bytes at the end of the f_midi_bind function. However, > >> in cases where there is a failure in the first midi bind attempt, > >> consider rebinding. > > What considers rebinding? Your change does not modify that. > > Hi Greg, > Thanks for your review comments. > > > Here the term "rebind" in this context refers to attempting to bind the > MIDI function a second time in certain scenarios. > The situations where rebinding is considered include: > > * When there is a failure in the first UDC write attempt, which may be > caused by other functions bind along with MIDI > * Runtime composition change : Example : MIDI,ADB to MIDI. Or MIDI to > MIDI,ADB > > The issue arises during the second time the "f_midi_bind" function is > called. The problem lies in the fact that the size of > "bulk_in_desc.wMaxPacketSize" is set to 1024 during the first call, > which exceeds the hardware capability of the dwc3 TX/RX FIFO > (ep->maxpacket_limit = 512). Ok, but then why not properly reset ALL of the options/values when a failure happens, not just this one when the initialization happens again? Odds are you might be missing the change of something else here as well, right? Also, cleaning up from an error is a better thing to do than forcing something to be set all the time when you don't have anything gone wrong. thanks, greg k-h
On 12/20/2024 5:54 PM, Greg KH wrote: > On Wed, Dec 18, 2024 at 03:51:50PM +0530, Selvarasu Ganesan wrote: >> On 12/18/2024 11:01 AM, Greg KH wrote: >>> On Sun, Dec 08, 2024 at 08:53:20PM +0530, Selvarasu Ganesan wrote: >>>> The current implementation sets the wMaxPacketSize of bulk in/out >>>> endpoints to 1024 bytes at the end of the f_midi_bind function. However, >>>> in cases where there is a failure in the first midi bind attempt, >>>> consider rebinding. >>> What considers rebinding? Your change does not modify that. >> Hi Greg, >> Thanks for your review comments. >> >> >> Here the term "rebind" in this context refers to attempting to bind the >> MIDI function a second time in certain scenarios. >> The situations where rebinding is considered include: >> >> * When there is a failure in the first UDC write attempt, which may be >> caused by other functions bind along with MIDI >> * Runtime composition change : Example : MIDI,ADB to MIDI. Or MIDI to >> MIDI,ADB >> >> The issue arises during the second time the "f_midi_bind" function is >> called. The problem lies in the fact that the size of >> "bulk_in_desc.wMaxPacketSize" is set to 1024 during the first call, >> which exceeds the hardware capability of the dwc3 TX/RX FIFO >> (ep->maxpacket_limit = 512). > Ok, but then why not properly reset ALL of the options/values when a > failure happens, not just this one when the initialization happens > again? Odds are you might be missing the change of something else here > as well, right? Are you suggesting that we reset the entire value of usb_endpoint_descriptor before call usb_ep_autoconfig? If so, Sorry I am not clear on your reasoning for wanting to reset all options/values. After all, all values will be overwritten afterusb_ep_autoconfig.Additionally, the wMaxPacketSize is the only value being checked during the EP claim process (usb_ep_autoconfig), and it has caused issues where claiming wMaxPacketSize is grater than ep->maxpacket_limit. > > Also, cleaning up from an error is a better thing to do than forcing > something to be set all the time when you don't have anything gone > wrong. As I previously mentioned, this is a general approach to set wMaxPacketSize before claiming the endpoint. This is because the usb_ep_autoconfig treats endpoint descriptors as if they were full speed. Following the same pattern as other function drivers, that approach allows us to claim the EP with using a full-speed descriptor. We can use the same approach here instead of resetting wMaxPacketSize every time. The following provided code is used to claim an EP with a full-speed bulk descriptor in MIDI. Its also working solution. But, We thinking that it may unnecessarily complicate the code as it only utilizes the full descriptor for obtaining the EP address here. What you think shall we go with below approach instead of rest wMaxPacketSize before call usb_ep_autoconfig? diff --git a/drivers/usb/gadget/function/f_midi.c b/drivers/usb/gadget/function/f_midi.c index 837fcdfa3840..fe12c8d82bf2 100644 --- a/drivers/usb/gadget/function/f_midi.c +++ b/drivers/usb/gadget/function/f_midi.c @@ -116,6 +116,22 @@ DECLARE_UAC_AC_HEADER_DESCRIPTOR(1); DECLARE_USB_MIDI_OUT_JACK_DESCRIPTOR(1); DECLARE_USB_MS_ENDPOINT_DESCRIPTOR(16); +static struct usb_endpoint_descriptor fs_bulk_in_desc = { + .bLength = USB_DT_ENDPOINT_SIZE, + .bDescriptorType = USB_DT_ENDPOINT, + + .bEndpointAddress = USB_DIR_IN, + .bmAttributes = USB_ENDPOINT_XFER_BULK, +}; + +static struct usb_endpoint_descriptor fs_bulk_out_desc = { + .bLength = USB_DT_ENDPOINT_SIZE, + .bDescriptorType = USB_DT_ENDPOINT, + + .bEndpointAddress = USB_DIR_OUT, + .bmAttributes = USB_ENDPOINT_XFER_BULK, +}; + /* B.3.1 Standard AC Interface Descriptor */ static struct usb_interface_descriptor ac_interface_desc = { .bLength = USB_DT_INTERFACE_SIZE, @@ -908,11 +924,11 @@ static int f_midi_bind(struct usb_configuration *c, struct usb_function *f) status = -ENODEV; /* allocate instance-specific endpoints */ - midi->in_ep = usb_ep_autoconfig(cdev->gadget, &bulk_in_desc); + midi->in_ep = usb_ep_autoconfig(cdev->gadget, &fs_bulk_in_desc); if (!midi->in_ep) goto fail; - midi->out_ep = usb_ep_autoconfig(cdev->gadget, &bulk_out_desc); + midi->out_ep = usb_ep_autoconfig(cdev->gadget, &fs_bulk_out_desc); if (!midi->out_ep) goto fail; @@ -1006,6 +1022,9 @@ static int f_midi_bind(struct usb_configuration *c, struct usb_function *f) ms_in_desc.bLength = USB_DT_MS_ENDPOINT_SIZE(midi->out_ports); ms_in_desc.bNumEmbMIDIJack = midi->out_ports; + bulk_in_desc.bEndpointAddress = fs_bulk_in_desc.bEndpointAddress; + bulk_out_desc.bEndpointAddress = fs_bulk_out_desc.bEndpointAddress; + /* ... and add them to the list */ endpoint_descriptor_index = i; midi_function[i++] = (struct usb_descriptor_header *) &bulk_out_desc; > > thanks, > > greg k-h >
On Fri, Dec 20, 2024 at 07:02:06PM +0530, Selvarasu Ganesan wrote: > > On 12/20/2024 5:54 PM, Greg KH wrote: > > On Wed, Dec 18, 2024 at 03:51:50PM +0530, Selvarasu Ganesan wrote: > >> On 12/18/2024 11:01 AM, Greg KH wrote: > >>> On Sun, Dec 08, 2024 at 08:53:20PM +0530, Selvarasu Ganesan wrote: > >>>> The current implementation sets the wMaxPacketSize of bulk in/out > >>>> endpoints to 1024 bytes at the end of the f_midi_bind function. However, > >>>> in cases where there is a failure in the first midi bind attempt, > >>>> consider rebinding. > >>> What considers rebinding? Your change does not modify that. > >> Hi Greg, > >> Thanks for your review comments. > >> > >> > >> Here the term "rebind" in this context refers to attempting to bind the > >> MIDI function a second time in certain scenarios. > >> The situations where rebinding is considered include: > >> > >> * When there is a failure in the first UDC write attempt, which may be > >> caused by other functions bind along with MIDI > >> * Runtime composition change : Example : MIDI,ADB to MIDI. Or MIDI to > >> MIDI,ADB > >> > >> The issue arises during the second time the "f_midi_bind" function is > >> called. The problem lies in the fact that the size of > >> "bulk_in_desc.wMaxPacketSize" is set to 1024 during the first call, > >> which exceeds the hardware capability of the dwc3 TX/RX FIFO > >> (ep->maxpacket_limit = 512). > > Ok, but then why not properly reset ALL of the options/values when a > > failure happens, not just this one when the initialization happens > > again? Odds are you might be missing the change of something else here > > as well, right? > Are you suggesting that we reset the entire value of > usb_endpoint_descriptor before call usb_ep_autoconfig? If so, Sorry I am > not clear on your reasoning for wanting to reset all options/values. > After all, all values will be overwritten > afterusb_ep_autoconfig.Additionally, the wMaxPacketSize is the only > value being checked during the EP claim process (usb_ep_autoconfig), and > it has caused issues where claiming wMaxPacketSize is grater than > ep->maxpacket_limit. Then fix up that value on failure, if things fail you should reset it back to a "known good state", right? And what's wrong with resetting all of the values anyway, wouldn't that be the correct thing to do? > > Also, cleaning up from an error is a better thing to do than forcing > > something to be set all the time when you don't have anything gone > > wrong. > As I previously mentioned, this is a general approach to set > wMaxPacketSize before claiming the endpoint. This is because the > usb_ep_autoconfig treats endpoint descriptors as if they were full > speed. Following the same pattern as other function drivers, that > approach allows us to claim the EP with using a full-speed descriptor. > We can use the same approach here instead of resetting wMaxPacketSize > every time. > > The following provided code is used to claim an EP with a full-speed > bulk descriptor in MIDI. Its also working solution. But, We thinking > that it may unnecessarily complicate the code as it only utilizes the > full descriptor for obtaining the EP address here. What you think shall > we go with below approach instead of rest wMaxPacketSize before call > usb_ep_autoconfig? I don't know, what do you think is best to do? You are the one having problems and will need to fix any bugs that your changes will cause :) thanks, greg k-h
On 12/20/2024 8:45 PM, Greg KH wrote: > On Fri, Dec 20, 2024 at 07:02:06PM +0530, Selvarasu Ganesan wrote: >> On 12/20/2024 5:54 PM, Greg KH wrote: >>> On Wed, Dec 18, 2024 at 03:51:50PM +0530, Selvarasu Ganesan wrote: >>>> On 12/18/2024 11:01 AM, Greg KH wrote: >>>>> On Sun, Dec 08, 2024 at 08:53:20PM +0530, Selvarasu Ganesan wrote: >>>>>> The current implementation sets the wMaxPacketSize of bulk in/out >>>>>> endpoints to 1024 bytes at the end of the f_midi_bind function. However, >>>>>> in cases where there is a failure in the first midi bind attempt, >>>>>> consider rebinding. >>>>> What considers rebinding? Your change does not modify that. >>>> Hi Greg, >>>> Thanks for your review comments. >>>> >>>> >>>> Here the term "rebind" in this context refers to attempting to bind the >>>> MIDI function a second time in certain scenarios. >>>> The situations where rebinding is considered include: >>>> >>>> * When there is a failure in the first UDC write attempt, which may be >>>> caused by other functions bind along with MIDI >>>> * Runtime composition change : Example : MIDI,ADB to MIDI. Or MIDI to >>>> MIDI,ADB >>>> >>>> The issue arises during the second time the "f_midi_bind" function is >>>> called. The problem lies in the fact that the size of >>>> "bulk_in_desc.wMaxPacketSize" is set to 1024 during the first call, >>>> which exceeds the hardware capability of the dwc3 TX/RX FIFO >>>> (ep->maxpacket_limit = 512). >>> Ok, but then why not properly reset ALL of the options/values when a >>> failure happens, not just this one when the initialization happens >>> again? Odds are you might be missing the change of something else here >>> as well, right? >> Are you suggesting that we reset the entire value of >> usb_endpoint_descriptor before call usb_ep_autoconfig? If so, Sorry I am >> not clear on your reasoning for wanting to reset all options/values. >> After all, all values will be overwritten >> afterusb_ep_autoconfig.Additionally, the wMaxPacketSize is the only >> value being checked during the EP claim process (usb_ep_autoconfig), and >> it has caused issues where claiming wMaxPacketSize is grater than >> ep->maxpacket_limit. > Then fix up that value on failure, if things fail you should reset it > back to a "known good state", right? And what's wrong with resetting > all of the values anyway, wouldn't that be the correct thing to do? Yes, It's back to known good state if we reset wMaxPacketSize. There is no point to reset all values in the usb endpoint descriptor structure as all the member of this structure are predefined value except wMaxPacketSize and bEndpointAddress. The bEndpointAddress is obtain as part of usb_ep_autoconfig. static struct usb_endpoint_descriptor bulk_out_desc = { .bLength = USB_DT_ENDPOINT_AUDIO_SIZE, .bDescriptorType = USB_DT_ENDPOINT, .bEndpointAddress = USB_DIR_OUT, .bmAttributes = USB_ENDPOINT_XFER_BULK, }; >>> Also, cleaning up from an error is a better thing to do than forcing >>> something to be set all the time when you don't have anything gone >>> wrong. >> As I previously mentioned, this is a general approach to set >> wMaxPacketSize before claiming the endpoint. This is because the >> usb_ep_autoconfig treats endpoint descriptors as if they were full >> speed. Following the same pattern as other function drivers, that >> approach allows us to claim the EP with using a full-speed descriptor. >> We can use the same approach here instead of resetting wMaxPacketSize >> every time. >> >> The following provided code is used to claim an EP with a full-speed >> bulk descriptor in MIDI. Its also working solution. But, We thinking >> that it may unnecessarily complicate the code as it only utilizes the >> full descriptor for obtaining the EP address here. What you think shall >> we go with below approach instead of rest wMaxPacketSize before call >> usb_ep_autoconfig? > I don't know, what do you think is best to do? You are the one having > problems and will need to fix any bugs that your changes will cause :) > > thanks, > > greg k-h We agree. Restting wMaxPacketSize is the best solution for this issue, as concluded from our internal review meeting as well. Thanks, Selva
diff --git a/drivers/usb/gadget/function/f_midi.c b/drivers/usb/gadget/function/f_midi.c index 837fcdfa3840..5caa0e4eb07e 100644 --- a/drivers/usb/gadget/function/f_midi.c +++ b/drivers/usb/gadget/function/f_midi.c @@ -907,6 +907,15 @@ static int f_midi_bind(struct usb_configuration *c, struct usb_function *f) status = -ENODEV; + /* + * Reset wMaxPacketSize with maximum packet size of FS bulk transfer before + * endpoint claim. This ensures that the wMaxPacketSize does not exceed the + * limit during bind retries where configured TX/RX FIFO's maxpacket size + * of 512 bytes for IN/OUT endpoints in support HS speed only. + */ + bulk_in_desc.wMaxPacketSize = cpu_to_le16(64); + bulk_out_desc.wMaxPacketSize = cpu_to_le16(64); + /* allocate instance-specific endpoints */ midi->in_ep = usb_ep_autoconfig(cdev->gadget, &bulk_in_desc); if (!midi->in_ep)
The current implementation sets the wMaxPacketSize of bulk in/out endpoints to 1024 bytes at the end of the f_midi_bind function. However, in cases where there is a failure in the first midi bind attempt, consider rebinding. This scenario may encounter an f_midi_bind issue due to the previous bind setting the bulk endpoint's wMaxPacketSize to 1024 bytes, which exceeds the ep->maxpacket_limit where configured TX/RX FIFO's maxpacket size of 512 bytes for IN/OUT endpoints in support HS speed only. This commit addresses this issue by resetting the wMaxPacketSize before endpoint claim. Fixes: 46decc82ffd5 ("usb: gadget: unconditionally allocate hs/ss descriptor in bind operation") Cc: stable@vger.kernel.org Signed-off-by: Selvarasu Ganesan <selvarasu.g@samsung.com> --- drivers/usb/gadget/function/f_midi.c | 9 +++++++++ 1 file changed, 9 insertions(+)