* [Qemu-devel] [PATCH] usb: Fix usb_packet_map() in the presence of IOMMUs
@ 2012-09-26 2:59 David Gibson
2012-09-26 6:32 ` Gerd Hoffmann
0 siblings, 1 reply; 2+ messages in thread
From: David Gibson @ 2012-09-26 2:59 UTC (permalink / raw)
To: kraxel; +Cc: David Gibson, qemu-devel, qemu-stable
With the IOMMU infrastructure introduced before 1.2, we need to use
dma_memory_map() to obtain a qemu pointer to memory from an IO bus address.
However, dma_memory_map() alters the given length to reflect the length
over which the used DMA translation is valid - which could be either more
or less than the requested length.
usb_packet_map() does not correctly handle these cases, simply failing if
dma_memory_map() alters the requested length. If dma_memory_map()
increased the length, we just need to use the requested length for the
qemu_iovec_add(). However, if it decreased the length, it means that a
single DMA translation is not valid for the whole sglist element, and so
we need to loop, splitting it up into multiple iovec entries for each
piece with a DMA translation (in practice >2 pieces is unlikely).
This patch implements the correct behaviour
Signed-off-by: David Gibson <david@gibson.dropbear.id.au>
---
hw/usb/libhw.c | 24 +++++++++++++++---------
1 file changed, 15 insertions(+), 9 deletions(-)
diff --git a/hw/usb/libhw.c b/hw/usb/libhw.c
index c0de30e..703e2d2 100644
--- a/hw/usb/libhw.c
+++ b/hw/usb/libhw.c
@@ -28,19 +28,25 @@ int usb_packet_map(USBPacket *p, QEMUSGList *sgl)
{
DMADirection dir = (p->pid == USB_TOKEN_IN) ?
DMA_DIRECTION_FROM_DEVICE : DMA_DIRECTION_TO_DEVICE;
- dma_addr_t len;
void *mem;
int i;
for (i = 0; i < sgl->nsg; i++) {
- len = sgl->sg[i].len;
- mem = dma_memory_map(sgl->dma, sgl->sg[i].base, &len, dir);
- if (!mem) {
- goto err;
- }
- qemu_iovec_add(&p->iov, mem, len);
- if (len != sgl->sg[i].len) {
- goto err;
+ dma_addr_t base = sgl->sg[i].base;
+ dma_addr_t len = sgl->sg[i].len;
+
+ while (len) {
+ dma_addr_t xlen = len;
+ mem = dma_memory_map(sgl->dma, sgl->sg[i].base, &xlen, dir);
+ if (!mem) {
+ goto err;
+ }
+ if (xlen > len) {
+ xlen = len;
+ }
+ qemu_iovec_add(&p->iov, mem, xlen);
+ len -= xlen;
+ base += xlen;
}
}
return 0;
--
1.7.10.4
^ permalink raw reply related [flat|nested] 2+ messages in thread
* Re: [Qemu-devel] [PATCH] usb: Fix usb_packet_map() in the presence of IOMMUs
2012-09-26 2:59 [Qemu-devel] [PATCH] usb: Fix usb_packet_map() in the presence of IOMMUs David Gibson
@ 2012-09-26 6:32 ` Gerd Hoffmann
0 siblings, 0 replies; 2+ messages in thread
From: Gerd Hoffmann @ 2012-09-26 6:32 UTC (permalink / raw)
To: David Gibson; +Cc: qemu-devel, qemu-stable
On 09/26/12 04:59, David Gibson wrote:
> With the IOMMU infrastructure introduced before 1.2, we need to use
> dma_memory_map() to obtain a qemu pointer to memory from an IO bus address.
> However, dma_memory_map() alters the given length to reflect the length
> over which the used DMA translation is valid - which could be either more
> or less than the requested length.
>
> usb_packet_map() does not correctly handle these cases, simply failing if
> dma_memory_map() alters the requested length. If dma_memory_map()
> increased the length, we just need to use the requested length for the
> qemu_iovec_add(). However, if it decreased the length, it means that a
> single DMA translation is not valid for the whole sglist element, and so
> we need to loop, splitting it up into multiple iovec entries for each
> piece with a DMA translation (in practice >2 pieces is unlikely).
>
> This patch implements the correct behaviour
Patch added to usb patch queue.
thanks,
Gerd
^ permalink raw reply [flat|nested] 2+ messages in thread
end of thread, other threads:[~2012-09-26 6:32 UTC | newest]
Thread overview: 2+ messages (download: mbox.gz follow: Atom feed
-- links below jump to the message on this page --
2012-09-26 2:59 [Qemu-devel] [PATCH] usb: Fix usb_packet_map() in the presence of IOMMUs David Gibson
2012-09-26 6:32 ` Gerd Hoffmann
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox;
as well as URLs for NNTP newsgroup(s).