mhennerich
commented
6 years ago Some updates:
I created Pull Request #134 which fixes the kernel warnings. I think this one is good to merge.
But it doesn't fix the USB issues seen on the MPSoC ZCU102 using the DWC3 controller. On the first call to libusb_get_string_descriptor_ascii() to fetch the dev_desc.iManufacturer it would return LIBUSB_ERROR_PIPE. Which is then causing the "Unable to create IIO context: Broken pipe" error.
This error can we workaround by simply trying it again. And very consistently it succeeds the second time.
diff --git a/usb.c b/usb.c
index ae84edc..7a19e4e 100644
--- a/usb.c
+++ b/usb.c
@@ -728,17 +728,17 @@ static int usb_populate_context_attrs(struct iio_context *ctx,
for (i = 0; i < ARRAY_SIZE(attrs); i++) {
if (attrs[i].idx) {
- if (i == 0)
ret = libusb_get_string_descriptor_ascii(hdl,
attrs[i].idx, (unsigned char *) buffer,
sizeof(buffer));
- ret = libusb_get_string_descriptor_ascii(hdl,
- attrs[i].idx, (unsigned char *) buffer,
- sizeof(buffer));
-
- if (ret < 0)
- return -(int) libusb_to_errno(ret);
+ if (ret < 0) {
+ ret = libusb_get_string_descriptor_ascii(hdl,
+ attrs[i].idx, (unsigned char *) buffer,
+ sizeof(buffer));
+ if (ret < 0)
+ return -(int) libusb_to_errno(ret);
+ }
ret = iio_context_add_attr(ctx, attrs[i].attr, buffer);
if (ret < 0)Then next problem is then that submit transfer may return LIBUSB_TRANSFER_TIMED_OUT. And now it get's really weird:
@@ -812,9 +812,6 @@ struct iio_context * usb_create_context(unsigned int bus,
usb_dev = dev;
ret = libusb_open(usb_dev, &hdl);
-
+ libusb_reset_device(hdl);
+ sleep(0.1);
/*
* Workaround for libusb on Windows >= 8.1. A device
* might appear twice in the list with one device beingResetting the device will resolve these issues as well.
The next thing I tried was to replace the DWC3 driver sources on our 4.9 kernel from a 4.14 upstream. With some minor mods - I was able to make it work.
With this driver the USB device reset is no longer required - but the libusb_get_string_descriptor_ascii() still occasionally fails the first time.
I then tried to measure streaming performance - with Buffer sizes below 1M Samples. We get around 80MBytes/s.
If you increase above 1M Samples - you experience a sudden death Which also brought down the entire USB on my Intel Desktop. (Now reboot a PC without Keyboard and Mouse ....)
root@analog:~# [ 135.615218] DMA: Out of SW-IOMMU space for 1048576 bytes at device fe200000.dwc3
[ 135.622522] Kernel panic - not syncing: DMA: Random memory could be DMA read
[ 135.622522]
[ 135.631032] CPU: 1 PID: 4166 Comm: iiod Tainted: G W 4.9.0-12448-g3de21d3-dirty #439
[ 135.639882] Hardware name: ZynqMP ZCU102 Rev1.0 (DT)
[ 135.644829] Call trace:
[ 135.647269] [<ffffff8008087f9c>] dump_backtrace+0x0/0x1a4
[ 135.652644] [<ffffff8008088154>] show_stack+0x14/0x1c
[ 135.657680] [<ffffff80083d30ac>] dump_stack+0x8c/0xac
[ 135.662714] [<ffffff800812b308>] panic+0x124/0x26c
[ 135.667486] [<ffffff80083fc3bc>] swiotlb_map_page+0x170/0x19c
[ 135.673217] [<ffffff80080932c8>] __swiotlb_map_page+0x20/0x7c
[ 135.678946] [<ffffff80085f1560>] usb_gadget_map_request_by_dev+0xfc/0x1c8
[ 135.685715] [<ffffff80085cc710>] __dwc3_gadget_kick_transfer+0x1c0/0x390
[ 135.692398] [<ffffff80085ccaf4>] dwc3_gadget_ep_queue+0xe8/0x1bc
[ 135.698387] [<ffffff80085f1368>] usb_ep_queue+0x24/0x58
[ 135.703596] [<ffffff80085f9a94>] ffs_epfile_io+0x49c/0x550
[ 135.709063] [<ffffff80085f9bdc>] ffs_epfile_write_iter+0x94/0x144
[ 135.715142] [<ffffff80081cb964>] aio_write+0xbc/0x150
[ 135.720175] [<ffffff80081cc9d4>] do_io_submit+0x4e0/0x5b0
[ 135.725557] [<ffffff80081d1e8c>] compat_SyS_io_submit+0x110/0x134
[ 135.731632] [<ffffff8008082ef0>] el0_svc_naked+0x24/0x28
[ 135.736925] SMP: stopping secondary CPUs
[ 135.740927] Kernel Offset: disabled
[ 135.744320] Memory Limit: none
[ 135.747362] ---[ end Kernel panic - not syncing: DMA: Random memory could be DMA read
[ 135.747362] 
pcercuei
IIOD USB3.0 SuperSpeed doesn’t work:
The kernel dmesg warning on the remote during USB device enumeration:
Context Scan Works:
However create IIO context fails with –EPIPE:
The kernel warning on the remote host is due to the fact that IIOD usbd doesn't initialize the SuperSpeed Endpoint Companion descriptors. See here:
Tested on Xilinx MPSoC ZCU102: