GCC4.6: Fix common/usb.c on xscale

The problem was that the code, when the function was compiled with -Os, was
misgenerated. As in the function description, this is likely another
manifestation of the bug in GCC.

Signed-off-by: Marek Vasut <marek.vasut@gmail.com>
Cc: Wolfgang Denk <wd@denx.de>
Cc: Remy Bohmer <linux@bohmer.net>
master
Marek Vasut 13 years ago committed by Albert ARIBAUD
parent 54cd51bf79
commit 5e8baf878e
  1. 13
      common/usb.c

@ -263,18 +263,24 @@ int usb_maxpacket(struct usb_device *dev, unsigned long pipe)
return dev->epmaxpacketin[((pipe>>15) & 0xf)];
}
/* The routine usb_set_maxpacket_ep() is extracted from the loop of routine
/*
* The routine usb_set_maxpacket_ep() is extracted from the loop of routine
* usb_set_maxpacket(), because the optimizer of GCC 4.x chokes on this routine
* when it is inlined in 1 single routine. What happens is that the register r3
* is used as loop-count 'i', but gets overwritten later on.
* This is clearly a compiler bug, but it is easier to workaround it here than
* to update the compiler (Occurs with at least several GCC 4.{1,2},x
* CodeSourcery compilers like e.g. 2007q3, 2008q1, 2008q3 lite editions on ARM)
*
* NOTE: Similar behaviour was observed with GCC4.6 on ARMv5.
*/
static void __attribute__((noinline))
usb_set_maxpacket_ep(struct usb_device *dev, struct usb_endpoint_descriptor *ep)
usb_set_maxpacket_ep(struct usb_device *dev, int if_idx, int ep_idx)
{
int b;
struct usb_endpoint_descriptor *ep;
ep = &dev->config.if_desc[if_idx].ep_desc[ep_idx];
b = ep->bEndpointAddress & USB_ENDPOINT_NUMBER_MASK;
@ -313,8 +319,7 @@ int usb_set_maxpacket(struct usb_device *dev)
for (i = 0; i < dev->config.desc.bNumInterfaces; i++)
for (ii = 0; ii < dev->config.if_desc[i].desc.bNumEndpoints; ii++)
usb_set_maxpacket_ep(dev,
&dev->config.if_desc[i].ep_desc[ii]);
usb_set_maxpacket_ep(dev, i, ii);
return 0;
}

Loading…
Cancel
Save