IB/hfi1: Fix user-space buffers mapping with IOMMU enabled
The dma_XXX API functions return bus addresses which are
physical addresses when IOMMU is disabled. Buffer
mapping to user-space is done via remap_pfn_range() with PFN
based on bus address instead of physical. This results in
wrong pages being mapped to user-space when IOMMU is enabled.
Reviewed-by: Mitko Haralanov <mitko.haralanov@intel.com>
Reviewed-by: Dennis Dalessandro <dennis.dalessandro@intel.com>
Signed-off-by: Tymoteusz Kielan <tymoteusz.kielan@intel.com>
Signed-off-by: Andrzej Kacprowski <andrzej.kacprowski@intel.com>
Signed-off-by: Dennis Dalessandro <dennis.dalessandro@intel.com>
Signed-off-by: Doug Ledford <dledford@redhat.com>
diff --git a/drivers/infiniband/hw/hfi1/file_ops.c b/drivers/infiniband/hw/hfi1/file_ops.c
index 7e03ccd..335796f 100644
--- a/drivers/infiniband/hw/hfi1/file_ops.c
+++ b/drivers/infiniband/hw/hfi1/file_ops.c
@@ -440,9 +440,10 @@
struct hfi1_filedata *fd = fp->private_data;
struct hfi1_ctxtdata *uctxt = fd->uctxt;
struct hfi1_devdata *dd;
- unsigned long flags, pfn;
+ unsigned long flags;
u64 token = vma->vm_pgoff << PAGE_SHIFT,
memaddr = 0;
+ void *memvirt = NULL;
u8 subctxt, mapio = 0, vmf = 0, type;
ssize_t memlen = 0;
int ret = 0;
@@ -493,7 +494,8 @@
* second or third page allocated for credit returns (if number
* of enabled contexts > 64 and 128 respectively).
*/
- memaddr = dd->cr_base[uctxt->numa_id].pa +
+ memvirt = dd->cr_base[uctxt->numa_id].va;
+ memaddr = virt_to_phys(memvirt) +
(((u64)uctxt->sc->hw_free -
(u64)dd->cr_base[uctxt->numa_id].va) & PAGE_MASK);
memlen = PAGE_SIZE;
@@ -508,8 +510,8 @@
mapio = 1;
break;
case RCV_HDRQ:
- memaddr = uctxt->rcvhdrq_phys;
memlen = uctxt->rcvhdrq_size;
+ memvirt = uctxt->rcvhdrq;
break;
case RCV_EGRBUF: {
unsigned long addr;
@@ -533,14 +535,21 @@
vma->vm_flags &= ~VM_MAYWRITE;
addr = vma->vm_start;
for (i = 0 ; i < uctxt->egrbufs.numbufs; i++) {
+ memlen = uctxt->egrbufs.buffers[i].len;
+ memvirt = uctxt->egrbufs.buffers[i].addr;
ret = remap_pfn_range(
vma, addr,
- uctxt->egrbufs.buffers[i].phys >> PAGE_SHIFT,
- uctxt->egrbufs.buffers[i].len,
+ /*
+ * virt_to_pfn() does the same, but
+ * it's not available on x86_64
+ * when CONFIG_MMU is enabled.
+ */
+ PFN_DOWN(__pa(memvirt)),
+ memlen,
vma->vm_page_prot);
if (ret < 0)
goto done;
- addr += uctxt->egrbufs.buffers[i].len;
+ addr += memlen;
}
ret = 0;
goto done;
@@ -596,8 +605,8 @@
ret = -EPERM;
goto done;
}
- memaddr = uctxt->rcvhdrqtailaddr_phys;
memlen = PAGE_SIZE;
+ memvirt = (void *)uctxt->rcvhdrtail_kvaddr;
flags &= ~VM_MAYWRITE;
break;
case SUBCTXT_UREGS:
@@ -650,16 +659,24 @@
"%u:%u type:%u io/vf:%d/%d, addr:0x%llx, len:%lu(%lu), flags:0x%lx\n",
ctxt, subctxt, type, mapio, vmf, memaddr, memlen,
vma->vm_end - vma->vm_start, vma->vm_flags);
- pfn = (unsigned long)(memaddr >> PAGE_SHIFT);
if (vmf) {
- vma->vm_pgoff = pfn;
+ vma->vm_pgoff = PFN_DOWN(memaddr);
vma->vm_ops = &vm_ops;
ret = 0;
} else if (mapio) {
- ret = io_remap_pfn_range(vma, vma->vm_start, pfn, memlen,
+ ret = io_remap_pfn_range(vma, vma->vm_start,
+ PFN_DOWN(memaddr),
+ memlen,
vma->vm_page_prot);
+ } else if (memvirt) {
+ ret = remap_pfn_range(vma, vma->vm_start,
+ PFN_DOWN(__pa(memvirt)),
+ memlen,
+ vma->vm_page_prot);
} else {
- ret = remap_pfn_range(vma, vma->vm_start, pfn, memlen,
+ ret = remap_pfn_range(vma, vma->vm_start,
+ PFN_DOWN(memaddr),
+ memlen,
vma->vm_page_prot);
}
done:
@@ -1260,7 +1277,7 @@
uctxt->rcvhdrq);
binfo.rcvegr_bufbase = HFI1_MMAP_TOKEN(RCV_EGRBUF, uctxt->ctxt,
fd->subctxt,
- uctxt->egrbufs.rcvtids[0].phys);
+ uctxt->egrbufs.rcvtids[0].dma);
binfo.sdma_comp_bufbase = HFI1_MMAP_TOKEN(SDMA_COMP, uctxt->ctxt,
fd->subctxt, 0);
/*