Commit 60368186 authored by Tymoteusz Kielan's avatar Tymoteusz Kielan Committed by Doug Ledford

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: default avatarMitko Haralanov <mitko.haralanov@intel.com>
Reviewed-by: default avatarDennis Dalessandro <dennis.dalessandro@intel.com>
Signed-off-by: default avatarTymoteusz Kielan <tymoteusz.kielan@intel.com>
Signed-off-by: default avatarAndrzej Kacprowski <andrzej.kacprowski@intel.com>
Signed-off-by: default avatarDennis Dalessandro <dennis.dalessandro@intel.com>
Signed-off-by: default avatarDoug Ledford <dledford@redhat.com>
parent 0b115ef1
......@@ -11553,10 +11553,10 @@ void hfi1_rcvctrl(struct hfi1_devdata *dd, unsigned int op, int ctxt)
!(rcvctrl & RCV_CTXT_CTRL_ENABLE_SMASK)) {
/* reset the tail and hdr addresses, and sequence count */
write_kctxt_csr(dd, ctxt, RCV_HDR_ADDR,
rcd->rcvhdrq_phys);
rcd->rcvhdrq_dma);
if (HFI1_CAP_KGET_MASK(rcd->flags, DMA_RTAIL))
write_kctxt_csr(dd, ctxt, RCV_HDR_TAIL_ADDR,
rcd->rcvhdrqtailaddr_phys);
rcd->rcvhdrqtailaddr_dma);
rcd->seq_cnt = 1;
/* reset the cached receive header queue head value */
......@@ -11621,9 +11621,9 @@ void hfi1_rcvctrl(struct hfi1_devdata *dd, unsigned int op, int ctxt)
* update with a dummy tail address and then disable
* receive context.
*/
if (dd->rcvhdrtail_dummy_physaddr) {
if (dd->rcvhdrtail_dummy_dma) {
write_kctxt_csr(dd, ctxt, RCV_HDR_TAIL_ADDR,
dd->rcvhdrtail_dummy_physaddr);
dd->rcvhdrtail_dummy_dma);
/* Enabling RcvCtxtCtrl.TailUpd is intentional. */
rcvctrl |= RCV_CTXT_CTRL_TAIL_UPD_SMASK;
}
......@@ -11634,7 +11634,7 @@ void hfi1_rcvctrl(struct hfi1_devdata *dd, unsigned int op, int ctxt)
rcvctrl |= RCV_CTXT_CTRL_INTR_AVAIL_SMASK;
if (op & HFI1_RCVCTRL_INTRAVAIL_DIS)
rcvctrl &= ~RCV_CTXT_CTRL_INTR_AVAIL_SMASK;
if (op & HFI1_RCVCTRL_TAILUPD_ENB && rcd->rcvhdrqtailaddr_phys)
if (op & HFI1_RCVCTRL_TAILUPD_ENB && rcd->rcvhdrqtailaddr_dma)
rcvctrl |= RCV_CTXT_CTRL_TAIL_UPD_SMASK;
if (op & HFI1_RCVCTRL_TAILUPD_DIS) {
/* See comment on RcvCtxtCtrl.TailUpd above */
......@@ -11706,7 +11706,7 @@ void hfi1_rcvctrl(struct hfi1_devdata *dd, unsigned int op, int ctxt)
* so it doesn't contain an address that is invalid.
*/
write_kctxt_csr(dd, ctxt, RCV_HDR_TAIL_ADDR,
dd->rcvhdrtail_dummy_physaddr);
dd->rcvhdrtail_dummy_dma);
}
u32 hfi1_read_cntrs(struct hfi1_devdata *dd, char **namep, u64 **cntrp)
......
......@@ -440,9 +440,10 @@ static int hfi1_file_mmap(struct file *fp, struct vm_area_struct *vma)
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 @@ static int hfi1_file_mmap(struct file *fp, struct vm_area_struct *vma)
* 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 @@ static int hfi1_file_mmap(struct file *fp, struct vm_area_struct *vma)
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 @@ static int hfi1_file_mmap(struct file *fp, struct vm_area_struct *vma)
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 @@ static int hfi1_file_mmap(struct file *fp, struct vm_area_struct *vma)
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 @@ static int hfi1_file_mmap(struct file *fp, struct vm_area_struct *vma)
"%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 @@ static int get_base_info(struct file *fp, void __user *ubase, __u32 len)
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);
/*
......
......@@ -172,12 +172,12 @@ struct ctxt_eager_bufs {
u32 threshold; /* head update threshold */
struct eager_buffer {
void *addr;
dma_addr_t phys;
dma_addr_t dma;
ssize_t len;
} *buffers;
struct {
void *addr;
dma_addr_t phys;
dma_addr_t dma;
} *rcvtids;
};
......@@ -208,8 +208,8 @@ struct hfi1_ctxtdata {
/* size of each of the rcvhdrq entries */
u16 rcvhdrqentsize;
/* mmap of hdrq, must fit in 44 bits */
dma_addr_t rcvhdrq_phys;
dma_addr_t rcvhdrqtailaddr_phys;
dma_addr_t rcvhdrq_dma;
dma_addr_t rcvhdrqtailaddr_dma;
struct ctxt_eager_bufs egrbufs;
/* this receive context's assigned PIO ACK send context */
struct send_context *sc;
......@@ -1165,7 +1165,7 @@ struct hfi1_devdata {
/* receive context tail dummy address */
__le64 *rcvhdrtail_dummy_kvaddr;
dma_addr_t rcvhdrtail_dummy_physaddr;
dma_addr_t rcvhdrtail_dummy_dma;
bool eprom_available; /* true if EPROM is available for this device */
bool aspm_supported; /* Does HW support ASPM */
......
......@@ -709,7 +709,7 @@ int hfi1_init(struct hfi1_devdata *dd, int reinit)
/* allocate dummy tail memory for all receive contexts */
dd->rcvhdrtail_dummy_kvaddr = dma_zalloc_coherent(
&dd->pcidev->dev, sizeof(u64),
&dd->rcvhdrtail_dummy_physaddr,
&dd->rcvhdrtail_dummy_dma,
GFP_KERNEL);
if (!dd->rcvhdrtail_dummy_kvaddr) {
......@@ -942,12 +942,12 @@ void hfi1_free_ctxtdata(struct hfi1_devdata *dd, struct hfi1_ctxtdata *rcd)
if (rcd->rcvhdrq) {
dma_free_coherent(&dd->pcidev->dev, rcd->rcvhdrq_size,
rcd->rcvhdrq, rcd->rcvhdrq_phys);
rcd->rcvhdrq, rcd->rcvhdrq_dma);
rcd->rcvhdrq = NULL;
if (rcd->rcvhdrtail_kvaddr) {
dma_free_coherent(&dd->pcidev->dev, PAGE_SIZE,
(void *)rcd->rcvhdrtail_kvaddr,
rcd->rcvhdrqtailaddr_phys);
rcd->rcvhdrqtailaddr_dma);
rcd->rcvhdrtail_kvaddr = NULL;
}
}
......@@ -956,11 +956,11 @@ void hfi1_free_ctxtdata(struct hfi1_devdata *dd, struct hfi1_ctxtdata *rcd)
kfree(rcd->egrbufs.rcvtids);
for (e = 0; e < rcd->egrbufs.alloced; e++) {
if (rcd->egrbufs.buffers[e].phys)
if (rcd->egrbufs.buffers[e].dma)
dma_free_coherent(&dd->pcidev->dev,
rcd->egrbufs.buffers[e].len,
rcd->egrbufs.buffers[e].addr,
rcd->egrbufs.buffers[e].phys);
rcd->egrbufs.buffers[e].dma);
}
kfree(rcd->egrbufs.buffers);
......@@ -1354,7 +1354,7 @@ static void cleanup_device_data(struct hfi1_devdata *dd)
if (dd->rcvhdrtail_dummy_kvaddr) {
dma_free_coherent(&dd->pcidev->dev, sizeof(u64),
(void *)dd->rcvhdrtail_dummy_kvaddr,
dd->rcvhdrtail_dummy_physaddr);
dd->rcvhdrtail_dummy_dma);
dd->rcvhdrtail_dummy_kvaddr = NULL;
}
......@@ -1577,7 +1577,7 @@ int hfi1_create_rcvhdrq(struct hfi1_devdata *dd, struct hfi1_ctxtdata *rcd)
u64 reg;
if (!rcd->rcvhdrq) {
dma_addr_t phys_hdrqtail;
dma_addr_t dma_hdrqtail;
gfp_t gfp_flags;
/*
......@@ -1590,7 +1590,7 @@ int hfi1_create_rcvhdrq(struct hfi1_devdata *dd, struct hfi1_ctxtdata *rcd)
gfp_flags = (rcd->ctxt >= dd->first_user_ctxt) ?
GFP_USER : GFP_KERNEL;
rcd->rcvhdrq = dma_zalloc_coherent(
&dd->pcidev->dev, amt, &rcd->rcvhdrq_phys,
&dd->pcidev->dev, amt, &rcd->rcvhdrq_dma,
gfp_flags | __GFP_COMP);
if (!rcd->rcvhdrq) {
......@@ -1602,11 +1602,11 @@ int hfi1_create_rcvhdrq(struct hfi1_devdata *dd, struct hfi1_ctxtdata *rcd)
if (HFI1_CAP_KGET_MASK(rcd->flags, DMA_RTAIL)) {
rcd->rcvhdrtail_kvaddr = dma_zalloc_coherent(
&dd->pcidev->dev, PAGE_SIZE, &phys_hdrqtail,
&dd->pcidev->dev, PAGE_SIZE, &dma_hdrqtail,
gfp_flags);
if (!rcd->rcvhdrtail_kvaddr)
goto bail_free;
rcd->rcvhdrqtailaddr_phys = phys_hdrqtail;
rcd->rcvhdrqtailaddr_dma = dma_hdrqtail;
}
rcd->rcvhdrq_size = amt;
......@@ -1634,7 +1634,7 @@ int hfi1_create_rcvhdrq(struct hfi1_devdata *dd, struct hfi1_ctxtdata *rcd)
* before enabling any receive context
*/
write_kctxt_csr(dd, rcd->ctxt, RCV_HDR_TAIL_ADDR,
dd->rcvhdrtail_dummy_physaddr);
dd->rcvhdrtail_dummy_dma);
return 0;
......@@ -1645,7 +1645,7 @@ int hfi1_create_rcvhdrq(struct hfi1_devdata *dd, struct hfi1_ctxtdata *rcd)
vfree(rcd->user_event_mask);
rcd->user_event_mask = NULL;
dma_free_coherent(&dd->pcidev->dev, amt, rcd->rcvhdrq,
rcd->rcvhdrq_phys);
rcd->rcvhdrq_dma);
rcd->rcvhdrq = NULL;
bail:
return -ENOMEM;
......@@ -1706,15 +1706,15 @@ int hfi1_setup_eagerbufs(struct hfi1_ctxtdata *rcd)
rcd->egrbufs.buffers[idx].addr =
dma_zalloc_coherent(&dd->pcidev->dev,
rcd->egrbufs.rcvtid_size,
&rcd->egrbufs.buffers[idx].phys,
&rcd->egrbufs.buffers[idx].dma,
gfp_flags);
if (rcd->egrbufs.buffers[idx].addr) {
rcd->egrbufs.buffers[idx].len =
rcd->egrbufs.rcvtid_size;
rcd->egrbufs.rcvtids[rcd->egrbufs.alloced].addr =
rcd->egrbufs.buffers[idx].addr;
rcd->egrbufs.rcvtids[rcd->egrbufs.alloced].phys =
rcd->egrbufs.buffers[idx].phys;
rcd->egrbufs.rcvtids[rcd->egrbufs.alloced].dma =
rcd->egrbufs.buffers[idx].dma;
rcd->egrbufs.alloced++;
alloced_bytes += rcd->egrbufs.rcvtid_size;
idx++;
......@@ -1755,14 +1755,14 @@ int hfi1_setup_eagerbufs(struct hfi1_ctxtdata *rcd)
for (i = 0, j = 0, offset = 0; j < idx; i++) {
if (i >= rcd->egrbufs.count)
break;
rcd->egrbufs.rcvtids[i].phys =
rcd->egrbufs.buffers[j].phys + offset;
rcd->egrbufs.rcvtids[i].dma =
rcd->egrbufs.buffers[j].dma + offset;
rcd->egrbufs.rcvtids[i].addr =
rcd->egrbufs.buffers[j].addr + offset;
rcd->egrbufs.alloced++;
if ((rcd->egrbufs.buffers[j].phys + offset +
if ((rcd->egrbufs.buffers[j].dma + offset +
new_size) ==
(rcd->egrbufs.buffers[j].phys +
(rcd->egrbufs.buffers[j].dma +
rcd->egrbufs.buffers[j].len)) {
j++;
offset = 0;
......@@ -1814,7 +1814,7 @@ int hfi1_setup_eagerbufs(struct hfi1_ctxtdata *rcd)
for (idx = 0; idx < rcd->egrbufs.alloced; idx++) {
hfi1_put_tid(dd, rcd->eager_base + idx, PT_EAGER,
rcd->egrbufs.rcvtids[idx].phys, order);
rcd->egrbufs.rcvtids[idx].dma, order);
cond_resched();
}
goto bail;
......@@ -1826,9 +1826,9 @@ int hfi1_setup_eagerbufs(struct hfi1_ctxtdata *rcd)
dma_free_coherent(&dd->pcidev->dev,
rcd->egrbufs.buffers[idx].len,
rcd->egrbufs.buffers[idx].addr,
rcd->egrbufs.buffers[idx].phys);
rcd->egrbufs.buffers[idx].dma);
rcd->egrbufs.buffers[idx].addr = NULL;
rcd->egrbufs.buffers[idx].phys = 0;
rcd->egrbufs.buffers[idx].dma = 0;
rcd->egrbufs.buffers[idx].len = 0;
}
bail:
......
......@@ -551,11 +551,11 @@ static inline u32 group_size(u32 group)
}
/*
* Obtain the credit return addresses, kernel virtual and physical, for the
* Obtain the credit return addresses, kernel virtual and bus, for the
* given sc.
*
* To understand this routine:
* o va and pa are arrays of struct credit_return. One for each physical
* o va and dma are arrays of struct credit_return. One for each physical
* send context, per NUMA.
* o Each send context always looks in its relative location in a struct
* credit_return for its credit return.
......@@ -563,14 +563,14 @@ static inline u32 group_size(u32 group)
* with the same value. Use the address of the first send context in the
* group.
*/
static void cr_group_addresses(struct send_context *sc, dma_addr_t *pa)
static void cr_group_addresses(struct send_context *sc, dma_addr_t *dma)
{
u32 gc = group_context(sc->hw_context, sc->group);
u32 index = sc->hw_context & 0x7;
sc->hw_free = &sc->dd->cr_base[sc->node].va[gc].cr[index];
*pa = (unsigned long)
&((struct credit_return *)sc->dd->cr_base[sc->node].pa)[gc];
*dma = (unsigned long)
&((struct credit_return *)sc->dd->cr_base[sc->node].dma)[gc];
}
/*
......@@ -710,7 +710,7 @@ struct send_context *sc_alloc(struct hfi1_devdata *dd, int type,
{
struct send_context_info *sci;
struct send_context *sc = NULL;
dma_addr_t pa;
dma_addr_t dma;
unsigned long flags;
u64 reg;
u32 thresh;
......@@ -763,7 +763,7 @@ struct send_context *sc_alloc(struct hfi1_devdata *dd, int type,
sc->sw_index = sw_index;
sc->hw_context = hw_context;
cr_group_addresses(sc, &pa);
cr_group_addresses(sc, &dma);
sc->credits = sci->credits;
/* PIO Send Memory Address details */
......@@ -805,7 +805,7 @@ struct send_context *sc_alloc(struct hfi1_devdata *dd, int type,
((u64)opval << SC(CHECK_OPCODE_VALUE_SHIFT)));
/* set up credit return */
reg = pa & SC(CREDIT_RETURN_ADDR_ADDRESS_SMASK);
reg = dma & SC(CREDIT_RETURN_ADDR_ADDRESS_SMASK);
write_kctxt_csr(dd, hw_context, SC(CREDIT_RETURN_ADDR), reg);
/*
......@@ -2064,7 +2064,7 @@ int init_credit_return(struct hfi1_devdata *dd)
dd->cr_base[i].va = dma_zalloc_coherent(
&dd->pcidev->dev,
bytes,
&dd->cr_base[i].pa,
&dd->cr_base[i].dma,
GFP_KERNEL);
if (!dd->cr_base[i].va) {
set_dev_node(&dd->pcidev->dev, dd->node);
......@@ -2097,7 +2097,7 @@ void free_credit_return(struct hfi1_devdata *dd)
TXE_NUM_CONTEXTS *
sizeof(struct credit_return),
dd->cr_base[i].va,
dd->cr_base[i].pa);
dd->cr_base[i].dma);
}
}
kfree(dd->cr_base);
......
......@@ -154,7 +154,7 @@ struct credit_return {
/* NUMA indexed credit return array */
struct credit_return_base {
struct credit_return *va;
dma_addr_t pa;
dma_addr_t dma;
};
/* send context configuration sizes (one per type) */
......
......@@ -67,9 +67,9 @@ TRACE_EVENT(hfi1_uctxtdata,
__field(u64, hw_free)
__field(void __iomem *, piobase)
__field(u16, rcvhdrq_cnt)
__field(u64, rcvhdrq_phys)
__field(u64, rcvhdrq_dma)
__field(u32, eager_cnt)
__field(u64, rcvegr_phys)
__field(u64, rcvegr_dma)
),
TP_fast_assign(DD_DEV_ASSIGN(dd);
__entry->ctxt = uctxt->ctxt;
......@@ -77,10 +77,9 @@ TRACE_EVENT(hfi1_uctxtdata,
__entry->hw_free = le64_to_cpu(*uctxt->sc->hw_free);
__entry->piobase = uctxt->sc->base_addr;
__entry->rcvhdrq_cnt = uctxt->rcvhdrq_cnt;
__entry->rcvhdrq_phys = uctxt->rcvhdrq_phys;
__entry->rcvhdrq_dma = uctxt->rcvhdrq_dma;
__entry->eager_cnt = uctxt->egrbufs.alloced;
__entry->rcvegr_phys =
uctxt->egrbufs.rcvtids[0].phys;
__entry->rcvegr_dma = uctxt->egrbufs.rcvtids[0].dma;
),
TP_printk("[%s] ctxt %u " UCTXT_FMT,
__get_str(dev),
......@@ -89,9 +88,9 @@ TRACE_EVENT(hfi1_uctxtdata,
__entry->hw_free,
__entry->piobase,
__entry->rcvhdrq_cnt,
__entry->rcvhdrq_phys,
__entry->rcvhdrq_dma,
__entry->eager_cnt,
__entry->rcvegr_phys
__entry->rcvegr_dma
)
);
......
Markdown is supported
0%
or
You are about to add 0 people to the discussion. Proceed with caution.
Finish editing this message first!
Please register or to comment