Re: [RFC PATCH v2 0/3] prerequisites for device reserved local mem rework

[Fredrik Noring <noring@nocrew.org>]

Hi Laurentiu,

> I took your code, added the missing mapping and placed it in a patch. 
> Please see attached (compile tested only).

Thanks! Unfortunately, the OHCI fails with errors such as

	usb 1-1: device descriptor read/64, error -12

that I tracked down to the calls

	   hub_port_init
	-> usb_control_msg
	-> usb_internal_control_msg
	-> usb_start_wait_urb
	-> usb_submit_urb
	-> usb_hcd_submit_urb
	-> hcd->driver->urb_enqueue
	-> ohci_urb_enqueue
	-> ed_get
	-> ed_alloc
	-> dma_pool_zalloc
	-> dma_pool_alloc
	-> pool_alloc_page,

which returns NULL. Then I noticed

	/* pool_alloc_page() might sleep, so temporarily drop &pool->lock */

that might be a problem considering that the HCD handles pool memory in
IRQ handlers, for instance:

	   do_IRQ
	-> generic_handle_irq
	-> handle_level_irq
	-> handle_irq_event
	-> handle_irq_event_percpu
	-> __handle_irq_event_percpu
	-> usb_hcd_irq
	-> ohci_irq
	-> ohci_work
	-> finish_urb
	-> __usb_hcd_giveback_urb
	-> usb_hcd_unmap_urb_for_dma
	-> hcd_buffer_free

Also, DMA_BUFFER_SIZE in ohci-ps2.c is only 256 KiB in total. Is the new
pool implementation at least as efficient as the previous one?

Fredrik