Commit 1fd3b40f authored by Bryan O'Sullivan's avatar Bryan O'Sullivan Committed by Roland Dreier

IB/ipath: Improved support for PowerPC

Signed-off-by: default avatarBryan O'Sullivan <bryan.osullivan@qlogic.com>
Signed-off-by: default avatarRoland Dreier <rolandd@cisco.com>
parent 076fafcd
...@@ -755,8 +755,8 @@ static void get_rhf_errstring(u32 err, char *msg, size_t len) ...@@ -755,8 +755,8 @@ static void get_rhf_errstring(u32 err, char *msg, size_t len)
static inline void *ipath_get_egrbuf(struct ipath_devdata *dd, u32 bufnum, static inline void *ipath_get_egrbuf(struct ipath_devdata *dd, u32 bufnum,
int err) int err)
{ {
return dd->ipath_port0_skbs ? return dd->ipath_port0_skbinfo ?
(void *)dd->ipath_port0_skbs[bufnum]->data : NULL; (void *) dd->ipath_port0_skbinfo[bufnum].skb->data : NULL;
} }
/** /**
...@@ -778,31 +778,34 @@ struct sk_buff *ipath_alloc_skb(struct ipath_devdata *dd, ...@@ -778,31 +778,34 @@ struct sk_buff *ipath_alloc_skb(struct ipath_devdata *dd,
*/ */
/* /*
* We need 4 extra bytes for unaligned transfer copying * We need 2 extra bytes for ipath_ether data sent in the
* key header. In order to keep everything dword aligned,
* we'll reserve 4 bytes.
*/ */
len = dd->ipath_ibmaxlen + 4;
if (dd->ipath_flags & IPATH_4BYTE_TID) { if (dd->ipath_flags & IPATH_4BYTE_TID) {
/* we need a 4KB multiple alignment, and there is no way /* We need a 2KB multiple alignment, and there is no way
* to do it except to allocate extra and then skb_reserve * to do it except to allocate extra and then skb_reserve
* enough to bring it up to the right alignment. * enough to bring it up to the right alignment.
*/ */
len = dd->ipath_ibmaxlen + 4 + (1 << 11) - 1; len += 2047;
} }
else
len = dd->ipath_ibmaxlen + 4;
skb = __dev_alloc_skb(len, gfp_mask); skb = __dev_alloc_skb(len, gfp_mask);
if (!skb) { if (!skb) {
ipath_dev_err(dd, "Failed to allocate skbuff, length %u\n", ipath_dev_err(dd, "Failed to allocate skbuff, length %u\n",
len); len);
goto bail; goto bail;
} }
skb_reserve(skb, 4);
if (dd->ipath_flags & IPATH_4BYTE_TID) { if (dd->ipath_flags & IPATH_4BYTE_TID) {
u32 una = ((1 << 11) - 1) & (unsigned long)(skb->data + 4); u32 una = (unsigned long)skb->data & 2047;
if (una) if (una)
skb_reserve(skb, 4 + (1 << 11) - una); skb_reserve(skb, 2048 - una);
else }
skb_reserve(skb, 4);
} else
skb_reserve(skb, 4);
bail: bail:
return skb; return skb;
...@@ -1345,8 +1348,9 @@ int ipath_create_rcvhdrq(struct ipath_devdata *dd, ...@@ -1345,8 +1348,9 @@ int ipath_create_rcvhdrq(struct ipath_devdata *dd,
ipath_cdbg(VERBOSE, "reuse port %d rcvhdrq @%p %llx phys; " ipath_cdbg(VERBOSE, "reuse port %d rcvhdrq @%p %llx phys; "
"hdrtailaddr@%p %llx physical\n", "hdrtailaddr@%p %llx physical\n",
pd->port_port, pd->port_rcvhdrq, pd->port_port, pd->port_rcvhdrq,
pd->port_rcvhdrq_phys, pd->port_rcvhdrtail_kvaddr, (unsigned long long) pd->port_rcvhdrq_phys,
(unsigned long long)pd->port_rcvhdrqtailaddr_phys); pd->port_rcvhdrtail_kvaddr, (unsigned long long)
pd->port_rcvhdrqtailaddr_phys);
/* clear for security and sanity on each use */ /* clear for security and sanity on each use */
memset(pd->port_rcvhdrq, 0, pd->port_rcvhdrq_size); memset(pd->port_rcvhdrq, 0, pd->port_rcvhdrq_size);
...@@ -1827,17 +1831,22 @@ void ipath_free_pddata(struct ipath_devdata *dd, struct ipath_portdata *pd) ...@@ -1827,17 +1831,22 @@ void ipath_free_pddata(struct ipath_devdata *dd, struct ipath_portdata *pd)
kfree(pd->port_rcvegrbuf_phys); kfree(pd->port_rcvegrbuf_phys);
pd->port_rcvegrbuf_phys = NULL; pd->port_rcvegrbuf_phys = NULL;
pd->port_rcvegrbuf_chunks = 0; pd->port_rcvegrbuf_chunks = 0;
} else if (pd->port_port == 0 && dd->ipath_port0_skbs) { } else if (pd->port_port == 0 && dd->ipath_port0_skbinfo) {
unsigned e; unsigned e;
struct sk_buff **skbs = dd->ipath_port0_skbs; struct ipath_skbinfo *skbinfo = dd->ipath_port0_skbinfo;
dd->ipath_port0_skbs = NULL; dd->ipath_port0_skbinfo = NULL;
ipath_cdbg(VERBOSE, "free closed port %d ipath_port0_skbs " ipath_cdbg(VERBOSE, "free closed port %d "
"@ %p\n", pd->port_port, skbs); "ipath_port0_skbinfo @ %p\n", pd->port_port,
skbinfo);
for (e = 0; e < dd->ipath_rcvegrcnt; e++) for (e = 0; e < dd->ipath_rcvegrcnt; e++)
if (skbs[e]) if (skbinfo[e].skb) {
dev_kfree_skb(skbs[e]); pci_unmap_single(dd->pcidev, skbinfo[e].phys,
vfree(skbs); dd->ipath_ibmaxlen,
PCI_DMA_FROMDEVICE);
dev_kfree_skb(skbinfo[e].skb);
}
vfree(skbinfo);
} }
kfree(pd->port_tid_pg_list); kfree(pd->port_tid_pg_list);
vfree(pd->subport_uregbase); vfree(pd->subport_uregbase);
...@@ -1934,7 +1943,7 @@ static void cleanup_device(struct ipath_devdata *dd) ...@@ -1934,7 +1943,7 @@ static void cleanup_device(struct ipath_devdata *dd)
if (dd->ipath_pioavailregs_dma) { if (dd->ipath_pioavailregs_dma) {
dma_free_coherent(&dd->pcidev->dev, PAGE_SIZE, dma_free_coherent(&dd->pcidev->dev, PAGE_SIZE,
dd->ipath_pioavailregs_dma, (void *) dd->ipath_pioavailregs_dma,
dd->ipath_pioavailregs_phys); dd->ipath_pioavailregs_phys);
dd->ipath_pioavailregs_dma = NULL; dd->ipath_pioavailregs_dma = NULL;
} }
...@@ -1947,6 +1956,7 @@ static void cleanup_device(struct ipath_devdata *dd) ...@@ -1947,6 +1956,7 @@ static void cleanup_device(struct ipath_devdata *dd)
if (dd->ipath_pageshadow) { if (dd->ipath_pageshadow) {
struct page **tmpp = dd->ipath_pageshadow; struct page **tmpp = dd->ipath_pageshadow;
dma_addr_t *tmpd = dd->ipath_physshadow;
int i, cnt = 0; int i, cnt = 0;
ipath_cdbg(VERBOSE, "Unlocking any expTID pages still " ipath_cdbg(VERBOSE, "Unlocking any expTID pages still "
...@@ -1957,6 +1967,8 @@ static void cleanup_device(struct ipath_devdata *dd) ...@@ -1957,6 +1967,8 @@ static void cleanup_device(struct ipath_devdata *dd)
for (i = port_tidbase; i < maxtid; i++) { for (i = port_tidbase; i < maxtid; i++) {
if (!tmpp[i]) if (!tmpp[i])
continue; continue;
pci_unmap_page(dd->pcidev, tmpd[i],
PAGE_SIZE, PCI_DMA_FROMDEVICE);
ipath_release_user_pages(&tmpp[i], 1); ipath_release_user_pages(&tmpp[i], 1);
tmpp[i] = NULL; tmpp[i] = NULL;
cnt++; cnt++;
......
...@@ -364,11 +364,14 @@ static int ipath_tid_update(struct ipath_portdata *pd, struct file *fp, ...@@ -364,11 +364,14 @@ static int ipath_tid_update(struct ipath_portdata *pd, struct file *fp,
"vaddr %lx\n", i, tid + tidoff, vaddr); "vaddr %lx\n", i, tid + tidoff, vaddr);
/* we "know" system pages and TID pages are same size */ /* we "know" system pages and TID pages are same size */
dd->ipath_pageshadow[porttid + tid] = pagep[i]; dd->ipath_pageshadow[porttid + tid] = pagep[i];
dd->ipath_physshadow[porttid + tid] = ipath_map_page(
dd->pcidev, pagep[i], 0, PAGE_SIZE,
PCI_DMA_FROMDEVICE);
/* /*
* don't need atomic or it's overhead * don't need atomic or it's overhead
*/ */
__set_bit(tid, tidmap); __set_bit(tid, tidmap);
physaddr = page_to_phys(pagep[i]); physaddr = dd->ipath_physshadow[porttid + tid];
ipath_stats.sps_pagelocks++; ipath_stats.sps_pagelocks++;
ipath_cdbg(VERBOSE, ipath_cdbg(VERBOSE,
"TID %u, vaddr %lx, physaddr %llx pgp %p\n", "TID %u, vaddr %lx, physaddr %llx pgp %p\n",
...@@ -402,6 +405,9 @@ static int ipath_tid_update(struct ipath_portdata *pd, struct file *fp, ...@@ -402,6 +405,9 @@ static int ipath_tid_update(struct ipath_portdata *pd, struct file *fp,
tid); tid);
dd->ipath_f_put_tid(dd, &tidbase[tid], 1, dd->ipath_f_put_tid(dd, &tidbase[tid], 1,
dd->ipath_tidinvalid); dd->ipath_tidinvalid);
pci_unmap_page(dd->pcidev,
dd->ipath_physshadow[porttid + tid],
PAGE_SIZE, PCI_DMA_FROMDEVICE);
dd->ipath_pageshadow[porttid + tid] = NULL; dd->ipath_pageshadow[porttid + tid] = NULL;
ipath_stats.sps_pageunlocks++; ipath_stats.sps_pageunlocks++;
} }
...@@ -515,6 +521,9 @@ static int ipath_tid_free(struct ipath_portdata *pd, unsigned subport, ...@@ -515,6 +521,9 @@ static int ipath_tid_free(struct ipath_portdata *pd, unsigned subport,
pd->port_pid, tid); pd->port_pid, tid);
dd->ipath_f_put_tid(dd, &tidbase[tid], 1, dd->ipath_f_put_tid(dd, &tidbase[tid], 1,
dd->ipath_tidinvalid); dd->ipath_tidinvalid);
pci_unmap_page(dd->pcidev,
dd->ipath_physshadow[porttid + tid],
PAGE_SIZE, PCI_DMA_FROMDEVICE);
ipath_release_user_pages( ipath_release_user_pages(
&dd->ipath_pageshadow[porttid + tid], 1); &dd->ipath_pageshadow[porttid + tid], 1);
dd->ipath_pageshadow[porttid + tid] = NULL; dd->ipath_pageshadow[porttid + tid] = NULL;
...@@ -711,7 +720,7 @@ static int ipath_manage_rcvq(struct ipath_portdata *pd, unsigned subport, ...@@ -711,7 +720,7 @@ static int ipath_manage_rcvq(struct ipath_portdata *pd, unsigned subport,
* updated and correct itself, even in the face of software * updated and correct itself, even in the face of software
* bugs. * bugs.
*/ */
*pd->port_rcvhdrtail_kvaddr = 0; *(volatile u64 *)pd->port_rcvhdrtail_kvaddr = 0;
set_bit(INFINIPATH_R_PORTENABLE_SHIFT + pd->port_port, set_bit(INFINIPATH_R_PORTENABLE_SHIFT + pd->port_port,
&dd->ipath_rcvctrl); &dd->ipath_rcvctrl);
} else } else
...@@ -923,11 +932,11 @@ bail: ...@@ -923,11 +932,11 @@ bail:
/* common code for the mappings on dma_alloc_coherent mem */ /* common code for the mappings on dma_alloc_coherent mem */
static int ipath_mmap_mem(struct vm_area_struct *vma, static int ipath_mmap_mem(struct vm_area_struct *vma,
struct ipath_portdata *pd, unsigned len, struct ipath_portdata *pd, unsigned len, int write_ok,
int write_ok, dma_addr_t addr, char *what) void *kvaddr, char *what)
{ {
struct ipath_devdata *dd = pd->port_dd; struct ipath_devdata *dd = pd->port_dd;
unsigned pfn = (unsigned long)addr >> PAGE_SHIFT; unsigned long pfn;
int ret; int ret;
if ((vma->vm_end - vma->vm_start) > len) { if ((vma->vm_end - vma->vm_start) > len) {
...@@ -950,16 +959,16 @@ static int ipath_mmap_mem(struct vm_area_struct *vma, ...@@ -950,16 +959,16 @@ static int ipath_mmap_mem(struct vm_area_struct *vma,
vma->vm_flags &= ~VM_MAYWRITE; vma->vm_flags &= ~VM_MAYWRITE;
} }
pfn = virt_to_phys(kvaddr) >> PAGE_SHIFT;
ret = remap_pfn_range(vma, vma->vm_start, pfn, ret = remap_pfn_range(vma, vma->vm_start, pfn,
len, vma->vm_page_prot); len, vma->vm_page_prot);
if (ret) if (ret)
dev_info(&dd->pcidev->dev, dev_info(&dd->pcidev->dev, "%s port%u mmap of %lx, %x "
"%s port%u mmap of %lx, %x bytes r%c failed: %d\n", "bytes r%c failed: %d\n", what, pd->port_port,
what, pd->port_port, (unsigned long)addr, len, pfn, len, write_ok?'w':'o', ret);
write_ok?'w':'o', ret);
else else
ipath_cdbg(VERBOSE, "%s port%u mmaped %lx, %x bytes r%c\n", ipath_cdbg(VERBOSE, "%s port%u mmaped %lx, %x bytes "
what, pd->port_port, (unsigned long)addr, len, "r%c\n", what, pd->port_port, pfn, len,
write_ok?'w':'o'); write_ok?'w':'o');
bail: bail:
return ret; return ret;
...@@ -1049,7 +1058,7 @@ static int mmap_rcvegrbufs(struct vm_area_struct *vma, ...@@ -1049,7 +1058,7 @@ static int mmap_rcvegrbufs(struct vm_area_struct *vma,
struct ipath_devdata *dd = pd->port_dd; struct ipath_devdata *dd = pd->port_dd;
unsigned long start, size; unsigned long start, size;
size_t total_size, i; size_t total_size, i;
dma_addr_t *phys; unsigned long pfn;
int ret; int ret;
size = pd->port_rcvegrbuf_size; size = pd->port_rcvegrbuf_size;
...@@ -1073,11 +1082,11 @@ static int mmap_rcvegrbufs(struct vm_area_struct *vma, ...@@ -1073,11 +1082,11 @@ static int mmap_rcvegrbufs(struct vm_area_struct *vma,
vma->vm_flags &= ~VM_MAYWRITE; vma->vm_flags &= ~VM_MAYWRITE;
start = vma->vm_start; start = vma->vm_start;
phys = pd->port_rcvegrbuf_phys;
for (i = 0; i < pd->port_rcvegrbuf_chunks; i++, start += size) { for (i = 0; i < pd->port_rcvegrbuf_chunks; i++, start += size) {
ret = remap_pfn_range(vma, start, phys[i] >> PAGE_SHIFT, pfn = virt_to_phys(pd->port_rcvegrbuf[i]) >> PAGE_SHIFT;
size, vma->vm_page_prot); ret = remap_pfn_range(vma, start, pfn, size,
vma->vm_page_prot);
if (ret < 0) if (ret < 0)
goto bail; goto bail;
} }
...@@ -1290,7 +1299,7 @@ static int ipath_mmap(struct file *fp, struct vm_area_struct *vma) ...@@ -1290,7 +1299,7 @@ static int ipath_mmap(struct file *fp, struct vm_area_struct *vma)
else if (pgaddr == dd->ipath_pioavailregs_phys) else if (pgaddr == dd->ipath_pioavailregs_phys)
/* in-memory copy of pioavail registers */ /* in-memory copy of pioavail registers */
ret = ipath_mmap_mem(vma, pd, PAGE_SIZE, 0, ret = ipath_mmap_mem(vma, pd, PAGE_SIZE, 0,
dd->ipath_pioavailregs_phys, (void *) dd->ipath_pioavailregs_dma,
"pioavail registers"); "pioavail registers");
else if (subport_fp(fp)) else if (subport_fp(fp))
/* Subports don't mmap the physical receive buffers */ /* Subports don't mmap the physical receive buffers */
...@@ -1304,12 +1313,12 @@ static int ipath_mmap(struct file *fp, struct vm_area_struct *vma) ...@@ -1304,12 +1313,12 @@ static int ipath_mmap(struct file *fp, struct vm_area_struct *vma)
* from an i/o perspective. * from an i/o perspective.
*/ */
ret = ipath_mmap_mem(vma, pd, pd->port_rcvhdrq_size, 1, ret = ipath_mmap_mem(vma, pd, pd->port_rcvhdrq_size, 1,
pd->port_rcvhdrq_phys, pd->port_rcvhdrq,
"rcvhdrq"); "rcvhdrq");
else if (pgaddr == (u64) pd->port_rcvhdrqtailaddr_phys) else if (pgaddr == (u64) pd->port_rcvhdrqtailaddr_phys)
/* in-memory copy of rcvhdrq tail register */ /* in-memory copy of rcvhdrq tail register */
ret = ipath_mmap_mem(vma, pd, PAGE_SIZE, 0, ret = ipath_mmap_mem(vma, pd, PAGE_SIZE, 0,
pd->port_rcvhdrqtailaddr_phys, pd->port_rcvhdrtail_kvaddr,
"rcvhdrq tail"); "rcvhdrq tail");
else else
ret = -EINVAL; ret = -EINVAL;
...@@ -1802,7 +1811,7 @@ static int ipath_do_user_init(struct file *fp, ...@@ -1802,7 +1811,7 @@ static int ipath_do_user_init(struct file *fp,
* We explictly set the in-memory copy to 0 beforehand, so we don't * We explictly set the in-memory copy to 0 beforehand, so we don't
* have to wait to be sure the DMA update has happened. * have to wait to be sure the DMA update has happened.
*/ */
*pd->port_rcvhdrtail_kvaddr = 0ULL; *(volatile u64 *)pd->port_rcvhdrtail_kvaddr = 0ULL;
set_bit(INFINIPATH_R_PORTENABLE_SHIFT + pd->port_port, set_bit(INFINIPATH_R_PORTENABLE_SHIFT + pd->port_port,
&dd->ipath_rcvctrl); &dd->ipath_rcvctrl);
ipath_write_kreg(dd, dd->ipath_kregs->kr_rcvctrl, ipath_write_kreg(dd, dd->ipath_kregs->kr_rcvctrl,
...@@ -1832,6 +1841,8 @@ static void unlock_expected_tids(struct ipath_portdata *pd) ...@@ -1832,6 +1841,8 @@ static void unlock_expected_tids(struct ipath_portdata *pd)
if (!dd->ipath_pageshadow[i]) if (!dd->ipath_pageshadow[i])
continue; continue;
pci_unmap_page(dd->pcidev, dd->ipath_physshadow[i],
PAGE_SIZE, PCI_DMA_FROMDEVICE);
ipath_release_user_pages_on_close(&dd->ipath_pageshadow[i], ipath_release_user_pages_on_close(&dd->ipath_pageshadow[i],
1); 1);
dd->ipath_pageshadow[i] = NULL; dd->ipath_pageshadow[i] = NULL;
...@@ -1936,14 +1947,14 @@ static int ipath_close(struct inode *in, struct file *fp) ...@@ -1936,14 +1947,14 @@ static int ipath_close(struct inode *in, struct file *fp)
i = dd->ipath_pbufsport * (port - 1); i = dd->ipath_pbufsport * (port - 1);
ipath_disarm_piobufs(dd, i, dd->ipath_pbufsport); ipath_disarm_piobufs(dd, i, dd->ipath_pbufsport);
dd->ipath_f_clear_tids(dd, pd->port_port);
if (dd->ipath_pageshadow) if (dd->ipath_pageshadow)
unlock_expected_tids(pd); unlock_expected_tids(pd);
ipath_stats.sps_ports--; ipath_stats.sps_ports--;
ipath_cdbg(PROC, "%s[%u] closed port %u:%u\n", ipath_cdbg(PROC, "%s[%u] closed port %u:%u\n",
pd->port_comm, pd->port_pid, pd->port_comm, pd->port_pid,
dd->ipath_unit, port); dd->ipath_unit, port);
dd->ipath_f_clear_tids(dd, pd->port_port);
} }
pd->port_pid = 0; pd->port_pid = 0;
......
...@@ -1113,7 +1113,7 @@ static void ipath_pe_put_tid_2(struct ipath_devdata *dd, u64 __iomem *tidptr, ...@@ -1113,7 +1113,7 @@ static void ipath_pe_put_tid_2(struct ipath_devdata *dd, u64 __iomem *tidptr,
if (pa != dd->ipath_tidinvalid) { if (pa != dd->ipath_tidinvalid) {
if (pa & ((1U << 11) - 1)) { if (pa & ((1U << 11) - 1)) {
dev_info(&dd->pcidev->dev, "BUG: physaddr %lx " dev_info(&dd->pcidev->dev, "BUG: physaddr %lx "
"not 4KB aligned!\n", pa); "not 2KB aligned!\n", pa);
return; return;
} }
pa >>= 11; pa >>= 11;
......
...@@ -88,13 +88,13 @@ MODULE_PARM_DESC(kpiobufs, "Set number of PIO buffers for driver"); ...@@ -88,13 +88,13 @@ MODULE_PARM_DESC(kpiobufs, "Set number of PIO buffers for driver");
static int create_port0_egr(struct ipath_devdata *dd) static int create_port0_egr(struct ipath_devdata *dd)
{ {
unsigned e, egrcnt; unsigned e, egrcnt;
struct sk_buff **skbs; struct ipath_skbinfo *skbinfo;
int ret; int ret;
egrcnt = dd->ipath_rcvegrcnt; egrcnt = dd->ipath_rcvegrcnt;
skbs = vmalloc(sizeof(*dd->ipath_port0_skbs) * egrcnt); skbinfo = vmalloc(sizeof(*dd->ipath_port0_skbinfo) * egrcnt);
if (skbs == NULL) { if (skbinfo == NULL) {
ipath_dev_err(dd, "allocation error for eager TID " ipath_dev_err(dd, "allocation error for eager TID "
"skb array\n"); "skb array\n");
ret = -ENOMEM; ret = -ENOMEM;
...@@ -109,13 +109,13 @@ static int create_port0_egr(struct ipath_devdata *dd) ...@@ -109,13 +109,13 @@ static int create_port0_egr(struct ipath_devdata *dd)
* 4 bytes so that the data buffer stays word aligned. * 4 bytes so that the data buffer stays word aligned.
* See ipath_kreceive() for more details. * See ipath_kreceive() for more details.
*/ */
skbs[e] = ipath_alloc_skb(dd, GFP_KERNEL); skbinfo[e].skb = ipath_alloc_skb(dd, GFP_KERNEL);
if (!skbs[e]) { if (!skbinfo[e].skb) {
ipath_dev_err(dd, "SKB allocation error for " ipath_dev_err(dd, "SKB allocation error for "
"eager TID %u\n", e); "eager TID %u\n", e);
while (e != 0) while (e != 0)
dev_kfree_skb(skbs[--e]); dev_kfree_skb(skbinfo[--e].skb);
vfree(skbs); vfree(skbinfo);
ret = -ENOMEM; ret = -ENOMEM;
goto bail; goto bail;
} }
...@@ -124,14 +124,17 @@ static int create_port0_egr(struct ipath_devdata *dd) ...@@ -124,14 +124,17 @@ static int create_port0_egr(struct ipath_devdata *dd)
* After loop above, so we can test non-NULL to see if ready * After loop above, so we can test non-NULL to see if ready
* to use at receive, etc. * to use at receive, etc.
*/ */
dd->ipath_port0_skbs = skbs; dd->ipath_port0_skbinfo = skbinfo;
for (e = 0; e < egrcnt; e++) { for (e = 0; e < egrcnt; e++) {
unsigned long phys = dd->ipath_port0_skbinfo[e].phys =
virt_to_phys(dd->ipath_port0_skbs[e]->data); ipath_map_single(dd->pcidev,
dd->ipath_port0_skbinfo[e].skb->data,
dd->ipath_ibmaxlen, PCI_DMA_FROMDEVICE);
dd->ipath_f_put_tid(dd, e + (u64 __iomem *) dd->ipath_f_put_tid(dd, e + (u64 __iomem *)
((char __iomem *) dd->ipath_kregbase + ((char __iomem *) dd->ipath_kregbase +
dd->ipath_rcvegrbase), 0, phys); dd->ipath_rcvegrbase), 0,
dd->ipath_port0_skbinfo[e].phys);
} }
ret = 0; ret = 0;
...@@ -432,16 +435,33 @@ done: ...@@ -432,16 +435,33 @@ done:
*/ */
static void init_shadow_tids(struct ipath_devdata *dd) static void init_shadow_tids(struct ipath_devdata *dd)
{ {
dd->ipath_pageshadow = (struct page **) struct page **pages;
vmalloc(dd->ipath_cfgports * dd->ipath_rcvtidcnt * dma_addr_t *addrs;
pages = vmalloc(dd->ipath_cfgports * dd->ipath_rcvtidcnt *
sizeof(struct page *)); sizeof(struct page *));
if (!dd->ipath_pageshadow) if (!pages) {
ipath_dev_err(dd, "failed to allocate shadow page * " ipath_dev_err(dd, "failed to allocate shadow page * "
"array, no expected sends!\n"); "array, no expected sends!\n");
else dd->ipath_pageshadow = NULL;
memset(dd->ipath_pageshadow, 0, return;
dd->ipath_cfgports * dd->ipath_rcvtidcnt * }
addrs = vmalloc(dd->ipath_cfgports * dd->ipath_rcvtidcnt *
sizeof(dma_addr_t));
if (!addrs) {
ipath_dev_err(dd, "failed to allocate shadow dma handle "
"array, no expected sends!\n");
vfree(dd->ipath_pageshadow);
dd->ipath_pageshadow = NULL;
return;
}
memset(pages, 0, dd->ipath_cfgports * dd->ipath_rcvtidcnt *
sizeof(struct page *)); sizeof(struct page *));
dd->ipath_pageshadow = pages;
dd->ipath_physshadow = addrs;
} }
static void enable_chip(struct ipath_devdata *dd, static void enable_chip(struct ipath_devdata *dd,
......
...@@ -605,7 +605,7 @@ static int handle_errors(struct ipath_devdata *dd, ipath_err_t errs) ...@@ -605,7 +605,7 @@ static int handle_errors(struct ipath_devdata *dd, ipath_err_t errs)
* don't report same point multiple times, * don't report same point multiple times,
* except kernel * except kernel
*/ */
tl = (u32) * pd->port_rcvhdrtail_kvaddr; tl = *(u64 *) pd->port_rcvhdrtail_kvaddr;
if (tl == dd->ipath_lastrcvhdrqtails[i]) if (tl == dd->ipath_lastrcvhdrqtails[i])
continue; continue;
hd = ipath_read_ureg32(dd, ur_rcvhdrhead, hd = ipath_read_ureg32(dd, ur_rcvhdrhead,
......
...@@ -39,6 +39,8 @@ ...@@ -39,6 +39,8 @@
*/ */
#include <linux/interrupt.h> #include <linux/interrupt.h>
#include <linux/pci.h>
#include <linux/dma-mapping.h>
#include <asm/io.h> #include <asm/io.h>
#include "ipath_common.h" #include "ipath_common.h"
...@@ -62,7 +64,7 @@ struct ipath_portdata { ...@@ -62,7 +64,7 @@ struct ipath_portdata {
/* rcvhdrq base, needs mmap before useful */ /* rcvhdrq base, needs mmap before useful */
void *port_rcvhdrq; void *port_rcvhdrq;
/* kernel virtual address where hdrqtail is updated */ /* kernel virtual address where hdrqtail is updated */
volatile __le64 *port_rcvhdrtail_kvaddr; void *port_rcvhdrtail_kvaddr;
/* /*
* temp buffer for expected send setup, allocated at open, instead * temp buffer for expected send setup, allocated at open, instead
* of each setup call * of each setup call
...@@ -146,6 +148,11 @@ struct _ipath_layer { ...@@ -146,6 +148,11 @@ struct _ipath_layer {
void *l_arg; void *l_arg;
}; };
struct ipath_skbinfo {
struct sk_buff *skb;
dma_addr_t phys;
};
struct ipath_devdata { struct ipath_devdata {
struct list_head ipath_list; struct list_head ipath_list;
...@@ -168,7 +175,7 @@ struct ipath_devdata { ...@@ -168,7 +175,7 @@ struct ipath_devdata {
/* ipath_cfgports pointers */ /* ipath_cfgports pointers */
struct ipath_portdata **ipath_pd; struct ipath_portdata **ipath_pd;
/* sk_buffs used by port 0 eager receive queue */ /* sk_buffs used by port 0 eager receive queue */
struct sk_buff **ipath_port0_skbs; struct ipath_skbinfo *ipath_port0_skbinfo;
/* kvirt address of 1st 2k pio buffer */ /* kvirt address of 1st 2k pio buffer */
void __iomem *ipath_pio2kbase; void __iomem *ipath_pio2kbase;
/* kvirt address of 1st 4k pio buffer */ /* kvirt address of 1st 4k pio buffer */
...@@ -335,6 +342,8 @@ struct ipath_devdata { ...@@ -335,6 +342,8 @@ struct ipath_devdata {
u64 *ipath_tidsimshadow; u64 *ipath_tidsimshadow;
/* shadow copy of struct page *'s for exp tid pages */ /* shadow copy of struct page *'s for exp tid pages */
struct page **ipath_pageshadow; struct page **ipath_pageshadow;
/* shadow copy of dma handles for exp tid pages */
dma_addr_t *ipath_physshadow;
/* lock to workaround chip bug 9437 */ /* lock to workaround chip bug 9437 */
spinlock_t ipath_tid_lock; spinlock_t ipath_tid_lock;
...@@ -864,6 +873,13 @@ void ipath_exit_ipathfs(void); ...@@ -864,6 +873,13 @@ void ipath_exit_ipathfs(void);
int ipathfs_add_device(struct ipath_devdata *); int ipathfs_add_device(struct ipath_devdata *);
int ipathfs_remove_device(struct ipath_devdata *); int ipathfs_remove_device(struct ipath_devdata *);
/*
* dma_addr wrappers - all 0's invalid for hw
*/
dma_addr_t ipath_map_page(struct pci_dev *, struct page *, unsigned long,
size_t, int);
dma_addr_t ipath_map_single(struct pci_dev *, void *, size_t, int);
/* /*
* Flush write combining store buffers (if present) and perform a write * Flush write combining store buffers (if present) and perform a write
* barrier. * barrier.
......
...@@ -89,6 +89,62 @@ bail: ...@@ -89,6 +89,62 @@ bail:
return ret; return ret;
} }
/**
* ipath_map_page - a safety wrapper around pci_map_page()
*
* A dma_addr of all 0's is interpreted by the chip as "disabled".
* Unfortunately, it can also be a valid dma_addr returned on some
* architectures.
*
* The powerpc iommu assigns dma_addrs in ascending order, so we don't
* have to bother with retries or mapping a dummy page to insure we
* don't just get the same mapping again.
*
* I'm sure we won't be so lucky with other iommu's, so FIXME.
*/
dma_addr_t ipath_map_page(struct pci_dev *hwdev, struct page *page,
unsigned long offset, size_t size, int direction)
{
dma_addr_t phys;
phys = pci_map_page(hwdev, page, offset, size, direction);
if (phys == 0) {
pci_unmap_page(hwdev, phys, size, direction);
phys = pci_map_page(hwdev, page, offset, size, direction);
/*
* FIXME: If we get 0 again, we should keep this page,
* map another, then free the 0 page.
*/
}
return phys;
}
/**
* ipath_map_single - a safety wrapper around pci_map_single()
*
* Same idea as ipath_map_page().
*/
dma_addr_t ipath_map_single(struct pci_dev *hwdev, void *ptr, size_t size,
int direction)
{
dma_addr_t phys;
phys = pci_map_single(hwdev, ptr, size, direction);
if (phys == 0) {
pci_unmap_single(hwdev, phys, size, direction);
phys = pci_map_single(hwdev, ptr, size, direction);
/*
* FIXME: If we get 0 again, we should keep this page,
* map another, then free the 0 page.
*/
}
return phys;
}
/** /**
* ipath_get_user_pages - lock user pages into memory * ipath_get_user_pages - lock user pages into memory
* @start_page: the start page * @start_page: the start page
......
...@@ -38,13 +38,23 @@ ...@@ -38,13 +38,23 @@
#include "ipath_kernel.h" #include "ipath_kernel.h"
/** /**
* ipath_unordered_wc - indicate whether write combining is ordered * ipath_enable_wc - enable write combining for MMIO writes to the device
* @dd: infinipath device
* *
* PowerPC systems (at least those in the 970 processor family) * Nothing to do on PowerPC, so just return without error.
* write partially filled store buffers in address order, but will write */
* completely filled store buffers in "random" order, and therefore must int ipath_enable_wc(struct ipath_devdata *dd)
* have serialization for correctness with current InfiniPath chips. {
return 0;
}
/**
* ipath_unordered_wc - indicate whether write combining is unordered
* *
* Because our performance depends on our ability to do write
* combining mmio writes in the most efficient way, we need to
* know if we are on a processor that may reorder stores when
* write combining.
*/ */
int ipath_unordered_wc(void) int ipath_unordered_wc(void)
{ {
......
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