On 08/18/2016 06:54 PM, Marcel Apfelbaum wrote:
On 08/10/2016 06:18 AM, Cao jin wrote:
The parameter table_offset & pba_offset is kind of confusing, they
shouldn't
include bir field.
Signed-off-by: Cao jin <caoj.f...@cn.fujitsu.com>
---
Hi,
According to the passed arguments, I guess all the callers of msix_init()
has the same feeling with me, but I am not quite sure about this, so,
RFC.
hw/pci/msix.c | 7 +++----
1 file changed, 3 insertions(+), 4 deletions(-)
diff --git a/hw/pci/msix.c b/hw/pci/msix.c
index 0ec1cb1..3a16d83 100644
--- a/hw/pci/msix.c
+++ b/hw/pci/msix.c
@@ -264,8 +264,7 @@ int msix_init(struct PCIDevice *dev, unsigned
short nentries,
if ((table_bar_nr == pba_bar_nr &&
ranges_overlap(table_offset, table_size, pba_offset,
pba_size)) ||
table_offset + table_size > memory_region_size(table_bar) ||
- pba_offset + pba_size > memory_region_size(pba_bar) ||
- (table_offset | pba_offset) & PCI_MSIX_FLAGS_BIRMASK) {
+ pba_offset + pba_size > memory_region_size(pba_bar)) {
return -EINVAL;
I think we should keep the '(table_offset | pba_offset) &
PCI_MSIX_FLAGS_BIRMASK)'
test since it is required by spec, please see: PCI Spec "6.8.2.4. Table
Offset/Table BIR for MSI-X"
Table offset: ...The lower 3 Table BIR bits are masked off (set to
zero) by software
to form a 32-bit QWORD -aligned offset.
This function gets the offset parameters as the whole 32-BIT QWORD and
checks it does not collide
with the BIR offset.
Hi Marcel,
Thanks very much for pointing the accurate reference out.
I also checked how kernel code handle this field, it is just like
you said.
Sorry for this noise.
Cao jin
}
@@ -282,8 +281,8 @@ int msix_init(struct PCIDevice *dev, unsigned
short nentries,
dev->msix_entries_nr = nentries;
dev->msix_function_masked = true;
- pci_set_long(config + PCI_MSIX_TABLE, table_offset | table_bar_nr);
- pci_set_long(config + PCI_MSIX_PBA, pba_offset | pba_bar_nr);
+ pci_set_long(config + PCI_MSIX_TABLE, (table_offset << 3) |
table_bar_nr);
+ pci_set_long(config + PCI_MSIX_PBA, (pba_offset << 3) | pba_bar_nr);
Here is a similar issue. Your interpretation suggests we need to shift
left the offset
to make room for BIR, but I think current implementation looks at it
differently already
receiving the offset as a 32-bit QWORD and simply does not "look" to the
lower bits
implying them 0.
Thanks,
Marcel
