On Thu, May 13, 2010 at 02:02:58PM -0700, Sarah Sharp wrote: On Tue, May 11, 2010 at 07:24:42AM -0700, Sarah Sharp wrote: On Tue, May 11, 2010 at 04:16:54PM +0800, Andiry Xu wrote: Hi Sarah, This is v5 resend of xHCI isoc supporting patchset. They're based on your master branch, commit 8e6632cd. I used "git am --3way" to apply the patches successfully. Please let me know if you still cannot apply the patches. They applied! Thanks Andiry. Hi Andiry, I did a bit of testing this morning with your patches. The good news is that my high speed camera now works, and I can switch back and forth between the high speed and full speed cameras in cheese. The bad news is that my machine crashed during some more intensive testing. I had a USB3 hard drive attached to the root hub and a single TT high speed hub attached to another root hub, with a FS and HS webcam and a FS audio device. I was watching video from the high speed webcam and tried to kill (ctrl+c) a `cp` of my kernel git tree to the device. That successfully stopped, but the whole system hung when I tried to unmount the USB3 drive. I'm not sure what the cause is yet, as my log file got corrupted (and of course I didn't have netconsole working at the time). I'm going to try to reproduce this tomorrow, as I need to focus on spec review for the rest of the day. Hi Andiry, I've been able to grab the oops message out after turning off a lot of debugging so that I could use netconsole. The modifications I made to turn off debugging are on my amd-isoc branch. This time the oops was triggered with just a HS webcam and a FS webcam plugged into a HS hub. The dmesg is attached. The first thing I notice is that you're giving back URB ffff88012fe9f000 twice, any idea how that could happen? When I ran the trimmed oops message through scripts/, got the following output: No vmlinux specified, assuming /lib/modules/2.6.34-rc7/build/vmlinux break; case TRB_TYPE(TRB_TRANSFER): ret = handle_tx_event(xhci, &event->trans_event); if (ret < 0) xhci->error_bitmask |= 1 << 9; ffffffffa030179e: 80 cc 02 or $0x2,%ah ffffffffa03017a1: 89 83 c0 09 00 00 mov %eax,0x9c0(%rbx) | %ebx = ffff88012e146220 ffffffffa03017a7: b8 01 00 00 00 mov $0x1,%eax | %eax => ffff88012fe9e800 ffffffffa03017ac: e9 af f3 ff ff jmpq ffffffffa0300b60 ffffffffa03017b1: 0f 1f 80 00 00 00 00 nopl 0x0(%rax) | %eax = ffff88012fe9e800 int skip_td = 0; union xhci_trb *cur_trb; struct xhci_segment *cur_seg; urb_priv = td->urb->hcpriv; idx = urb_priv->td_cnt; ffffffffa03017b8: 48 8b 50 08 mov 0x8(%rax),%rdx | %eax = ffff88012fe9e800 %edx => 0 *ffffffffa03017bc: 8b 4a 04 mov 0x4(%rdx),%ecx | %edx = 0 %ecx = ffff8800b7dc7680 <--- faulting instruction status = 0; if (ep->skip) { ffffffffa03017bf: 4b 8d 54 ad 00 lea 0x0(%r13,%r13,4),%rdx ffffffffa03017c4: 48 8d 14 92 lea (%rdx,%rdx,4),%rdx ffffffffa03017c8: 41 80 bc d7 e0 00 00 cmpb $0x0,0xe0(%r15,%rdx,8) ffffffffa03017cf: 00 00 ffffffffa03017d1: 0f 85 bf 00 00 00 jne ffffffffa0301896 -EREMOTEIO; else td->urb->iso_frame_desc[idx].status = 0; } else { /* handle completion code */ switch (trb_comp_code) { ffffffffa03017d7: 83 7d cc 1f cmpl $0x1f,-0x34(%rbp) ffffffffa03017db: 0f 86 e4 01 00 00 jbe ffffffffa03019c5 case COMP_STOP: So it looks like the urb_priv might be NULL. If you're giving back URBs that have already been given back, then that could be the cause. Let me know if you have any patches I can try to fix this. Sarah Sharp
