On Tue, Aug 04, 2015 at 10:27:31AM +0200, Laurent Vivier wrote: > We use muldiv64() to compute the time to wait: > > timeout = muldiv64(get_ticks_per_sec(), timeout, 33000000); > > but get_ticks_per_sec() is 10^9 (30 bit value) and timeout > is a 35 bit value. > > Whereas muldiv64 is: > > uint64_t muldiv64(uint64_t a, uint32_t b, uint32_t c) > > So we loose 3 bits of timeout. > > Swapping get_ticks_per_sec() and timeout fixes it. > > We can also replace it by a multiplication by 30 ns, > but this changes PCI clock frequency from 33MHz to 33.333333MHz > and we need to do this on all the QEMU PCI devices (later...) > > Signed-off-by: Laurent Vivier <lviv...@redhat.com>
Hah. 32-bit second argument. Totally missed that when I put the
muldiv64() in there. So I didn't eliminate the overflow, just pushed
it out some bits.
Thanks for finding this.
Reviewed-by: David Gibson <da...@gibson.dropbear.id.au>
--
David Gibson | I'll have my music baroque, and my code
david AT gibson.dropbear.id.au | minimalist, thank you. NOT _the_ _other_
| _way_ _around_!
http://www.ozlabs.org/~dgibson
pgplpN0DA6ulj.pgp
Description: PGP signature
