Re: >20 KB URBs + EHCI = bad performance due to stalls

2021-03-31 Thread Maciej S. Szmigiero
On 31.03.2021 21:55, Alan Stern wrote: On Wed, Mar 31, 2021 at 08:20:56PM +0200, Maciej S. Szmigiero wrote: On 29.03.2021 17:22, Alan Stern wrote: On Sat, Mar 27, 2021 at 04:55:20PM +0100, Maciej S. Szmigiero wrote: Hi, Is there any specific reason that URBs without URB_SHORT_NOT_OK flag

Re: >20 KB URBs + EHCI = bad performance due to stalls

2021-03-31 Thread Alan Stern
On Wed, Mar 31, 2021 at 08:20:56PM +0200, Maciej S. Szmigiero wrote: > On 29.03.2021 17:22, Alan Stern wrote: > > On Sat, Mar 27, 2021 at 04:55:20PM +0100, Maciej S. Szmigiero wrote: > > > Hi, > > > > > > Is there any specific reason that URBs without URB_SHORT_NOT_OK flag that > > > span

Re: >20 KB URBs + EHCI = bad performance due to stalls

2021-03-31 Thread Maciej S. Szmigiero
On 29.03.2021 17:22, Alan Stern wrote: On Sat, Mar 27, 2021 at 04:55:20PM +0100, Maciej S. Szmigiero wrote: Hi, Is there any specific reason that URBs without URB_SHORT_NOT_OK flag that span multiple EHCI qTDs have Alternate Next qTD pointer set to the dummy qTD in their every qTD besides the

Re: >20 KB URBs + EHCI = bad performance due to stalls

2021-03-29 Thread Alan Stern
On Sat, Mar 27, 2021 at 04:55:20PM +0100, Maciej S. Szmigiero wrote: > Hi, > > Is there any specific reason that URBs without URB_SHORT_NOT_OK flag that > span multiple EHCI qTDs have Alternate Next qTD pointer set to the dummy > qTD in their every qTD besides the last one (instead of to the

>20 KB URBs + EHCI = bad performance due to stalls

2021-03-27 Thread Maciej S. Szmigiero
Hi, Is there any specific reason that URBs without URB_SHORT_NOT_OK flag that span multiple EHCI qTDs have Alternate Next qTD pointer set to the dummy qTD in their every qTD besides the last one (instead of to the first qTD of the next URB to that endpoint)? This causes that endpoint queue to