Skip to contentSkip to conversation

packet size problem

S Alan Stern 2 participants 4 messages Jun 11, 2008 – Jun 12, 2008 1 message not in the archive

lore
Share & tools
lore

A message that is not in the archive (sent off-list or not yet received): [email protected]

Alan Stern

author
#1 lore raw
Quoted 11 lines · AntonioLin

Hi All,

I checked srb->device->request_queue->dma_alignment in usb_stor_bulk_Bulk_transport() routine. , the value is 1023.

But in usb_stor_bulk_transfer_sglist, the length of first element in sg array is 3584 which is not divisible by 1024.

Can you post your /proc/bus/usb/devices ?

I don't know how to do this, could you descript moe about it ? (Sorry,I have few experience about Linux.)

All you have to do is:

cat /proc/bus/usb/devices

Or even just tell your email client to include /proc/bus/usb/devices as an attachment. However some Linux distributions don't automatically mount /proc/bus/usb. If yours doesn't, you will have to do (as root):

mount -t usbfs none /proc/bus/usb

before you can access /proc/bus/usb/devices. But in any case it doesn't matter, since you have checked that the value of dma_alignment really is 1023.

Jens, this is a question for you. How come we're getting an SG element whose lenth is 3584 when the request_queue's dma_alignment mask is set to 1023? Is dma_alignment not the right parameter to set?

Alan Stern

#2 lore raw
Quoted 24 lines · Alan Stern
Quoted 11 lines · AntonioLin

Hi All,

I checked srb->device->request_queue->dma_alignment in usb_stor_bulk_Bulk_transport() routine. , the value is 1023.

But in usb_stor_bulk_transfer_sglist, the length of first element in sg array is 3584 which is not divisible by 1024.

Can you post your /proc/bus/usb/devices ?

I don't know how to do this, could you descript moe about it ? (Sorry,I have few experience about Linux.)

All you have to do is:

cat /proc/bus/usb/devices

Or even just tell your email client to include /proc/bus/usb/devices as an attachment. However some Linux distributions don't automatically mount /proc/bus/usb. If yours doesn't, you will have to do (as root):

mount -t usbfs none /proc/bus/usb

Quoted 7 lines

before you can access /proc/bus/usb/devices. But in any case it doesn't matter, since you have checked that the value of dma_alignment really is 1023.

Jens, this is a question for you. How come we're getting an SG element whose lenth is 3584 when the request_queue's dma_alignment mask is set to 1023? Is dma_alignment not the right parameter to set?

dma_alignment mask is the buffer address alignment, so they are not completely the same. But it does sound odd, since you would then be spanning two pages for that sg element. Is this before or after dma mapping the sg table?

Signature · 1 line
Jens Axboe
author
#3 lore raw
Quoted 6 lines · Jens Axboe
Quoted 3 lines

Jens, this is a question for you. How come we're getting an SG element whose lenth is 3584 when the request_queue's dma_alignment mask is set to 1023? Is dma_alignment not the right parameter to set?

dma_alignment mask is the buffer address alignment, so they are not completely the same.

Well then, put it this way: We don't care particularly about the address alignment as such, but we do need the buffer length to be a multiple of 1024 for all but the last element in the SG list. Is there a better way to accomplish this?

Quoted 3 lines

But it does sound odd, since you would then be
spanning two pages for that sg element. Is this before or after dma mapping the sg table?

Both.

Alan Stern

author
#4 lore raw
In reply to Jens Axboe
Quoted 8 lines · Jens Axboe
Quoted 3 lines

Jens, this is a question for you. How come we're getting an SG element whose lenth is 3584 when the request_queue's dma_alignment mask is set to 1023? Is dma_alignment not the right parameter to set?

dma_alignment mask is the buffer address alignment, so they are not completely the same. But it does sound odd, since you would then be spanning two pages for that sg element. Is this before or after dma mapping the sg table?

Any followup on this? Is there a different way to request that each SG element (except the last) has length divisible by 1024, other than setting the dma_alignment mask to 1023?

Alan Stern