[PATCH] ppc64: support 64k pages

Adds a new CONFIG_PPC_64K_PAGES which, when enabled, changes the kernel
base page size to 64K.  The resulting kernel still boots on any
hardware.  On current machines with 4K pages support only, the kernel
will maintain 16 "subpages" for each 64K page transparently.

Note that while real 64K capable HW has been tested, the current patch
will not enable it yet as such hardware is not released yet, and I'm
still verifying with the firmware architects the proper to get the
information from the newer hypervisors.

Signed-off-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>

6 files changed, 97 insertions, 114 deletions

diff --git a/arch/powerpc/platforms/iseries/htab.c b/arch/powerpc/platforms/iseries/htab.c
index b3c6c3374ca6..30bdcf3925d9 100644
--- a/arch/powerpc/platforms/iseries/htab.c
+++ b/arch/powerpc/platforms/iseries/htab.c
@@ -39,15 +39,16 @@ static inline void iSeries_hunlock(unsigned long slot)
         spin_unlock(&iSeries_hlocks[(slot >> 4) & 0x3f]);
 }
 
-static long iSeries_hpte_insert(unsigned long hpte_group, unsigned long va,
-                                unsigned long prpn, unsigned long vflags,
-                                unsigned long rflags)
+long iSeries_hpte_insert(unsigned long hpte_group, unsigned long va,
+                         unsigned long pa, unsigned long rflags,
+                         unsigned long vflags, int psize)
 {
-        unsigned long arpn;
         long slot;
         hpte_t lhpte;
         int secondary = 0;
 
+        BUG_ON(psize != MMU_PAGE_4K);
+
         /*
          * The hypervisor tries both primary and secondary.
          * If we are being called to insert in the secondary,
@@ -59,8 +60,19 @@ static long iSeries_hpte_insert(unsigned long hpte_group, unsigned long va,
 
         iSeries_hlock(hpte_group);
 
-        slot = HvCallHpt_findValid(&lhpte, va >> PAGE_SHIFT);
-        BUG_ON(lhpte.v & HPTE_V_VALID);
+        slot = HvCallHpt_findValid(&lhpte, va >> HW_PAGE_SHIFT);
+        if (unlikely(lhpte.v & HPTE_V_VALID)) {
+                if (vflags & HPTE_V_BOLTED) {
+                        HvCallHpt_setSwBits(slot, 0x10, 0);
+                        HvCallHpt_setPp(slot, PP_RWXX);
+                        iSeries_hunlock(hpte_group);
+                        if (slot < 0)
+                                return 0x8 | (slot & 7);
+                        else
+                                return slot & 7;
+                }
+                BUG();
+        }
 
         if (slot == -1) { /* No available entry found in either group */
                 iSeries_hunlock(hpte_group);
@@ -73,10 +85,9 @@ static long iSeries_hpte_insert(unsigned long hpte_group, unsigned long va,
                 slot &= 0x7fffffffffffffff;
         }
 
-        arpn = phys_to_abs(prpn << PAGE_SHIFT) >> PAGE_SHIFT;
 
-        lhpte.v = (va >> 23) << HPTE_V_AVPN_SHIFT | vflags | HPTE_V_VALID;
-        lhpte.r = (arpn << HPTE_R_RPN_SHIFT) | rflags;
+        lhpte.v = hpte_encode_v(va, MMU_PAGE_4K) | vflags | HPTE_V_VALID;
+        lhpte.r = hpte_encode_r(phys_to_abs(pa), MMU_PAGE_4K) | rflags;
 
         /* Now fill in the actual HPTE */
         HvCallHpt_addValidate(slot, secondary, &lhpte);
@@ -86,25 +97,6 @@ static long iSeries_hpte_insert(unsigned long hpte_group, unsigned long va,
         return (secondary << 3) | (slot & 7);
 }
 
-long iSeries_hpte_bolt_or_insert(unsigned long hpte_group,
-                unsigned long va, unsigned long prpn, unsigned long vflags,
-                unsigned long rflags)
-{
-        long slot;
-        hpte_t lhpte;
-
-        slot = HvCallHpt_findValid(&lhpte, va >> PAGE_SHIFT);
-
-        if (lhpte.v & HPTE_V_VALID) {
-                /* Bolt the existing HPTE */
-                HvCallHpt_setSwBits(slot, 0x10, 0);
-                HvCallHpt_setPp(slot, PP_RWXX);
-                return 0;
-        }
-
-        return iSeries_hpte_insert(hpte_group, va, prpn, vflags, rflags);
-}
-
 static unsigned long iSeries_hpte_getword0(unsigned long slot)
 {
         hpte_t hpte;
@@ -150,15 +142,17 @@ static long iSeries_hpte_remove(unsigned long hpte_group)
  *      bits 61..63 : PP2,PP1,PP0
  */
 static long iSeries_hpte_updatepp(unsigned long slot, unsigned long newpp,
-                                  unsigned long va, int large, int local)
+                                  unsigned long va, int psize, int local)
 {
         hpte_t hpte;
-        unsigned long avpn = va >> 23;
+        unsigned long want_v;
 
         iSeries_hlock(slot);
 
         HvCallHpt_get(&hpte, slot);
-        if ((HPTE_V_AVPN_VAL(hpte.v) == avpn) && (hpte.v & HPTE_V_VALID)) {
+        want_v = hpte_encode_v(va, MMU_PAGE_4K);
+
+        if (HPTE_V_COMPARE(hpte.v, want_v) && (hpte.v & HPTE_V_VALID)) {
                 /*
                  * Hypervisor expects bits as NPPP, which is
                  * different from how they are mapped in our PP.
@@ -210,14 +204,17 @@ static long iSeries_hpte_find(unsigned long vpn)
  *
  * No need to lock here because we should be the only user.
  */
-static void iSeries_hpte_updateboltedpp(unsigned long newpp, unsigned long ea)
+static void iSeries_hpte_updateboltedpp(unsigned long newpp, unsigned long ea,
+                                        int psize)
 {
         unsigned long vsid,va,vpn;
         long slot;
 
+        BUG_ON(psize != MMU_PAGE_4K);
+
         vsid = get_kernel_vsid(ea);
         va = (vsid << 28) | (ea & 0x0fffffff);
-        vpn = va >> PAGE_SHIFT;
+        vpn = va >> HW_PAGE_SHIFT;
         slot = iSeries_hpte_find(vpn);
         if (slot == -1)
                 panic("updateboltedpp: Could not find page to bolt\n");
@@ -225,7 +222,7 @@ static void iSeries_hpte_updateboltedpp(unsigned long newpp, unsigned long ea)
 }
 
 static void iSeries_hpte_invalidate(unsigned long slot, unsigned long va,
-                                    int large, int local)
+                                    int psize, int local)
 {
         unsigned long hpte_v;
         unsigned long avpn = va >> 23;
diff --git a/arch/powerpc/platforms/iseries/hvlog.c b/arch/powerpc/platforms/iseries/hvlog.c
index 62ec73479687..f476d71194fa 100644
--- a/arch/powerpc/platforms/iseries/hvlog.c
+++ b/arch/powerpc/platforms/iseries/hvlog.c
@@ -22,7 +22,7 @@ void HvCall_writeLogBuffer(const void *buffer, u64 len)
 
         while (len) {
                 hv_buf.addr = cur;
-                left_this_page = ((cur & PAGE_MASK) + PAGE_SIZE) - cur;
+                left_this_page = ((cur & HW_PAGE_MASK) + HW_PAGE_SIZE) - cur;
                 if (left_this_page > len)
                         left_this_page = len;
                 hv_buf.len = left_this_page;
@@ -30,6 +30,6 @@ void HvCall_writeLogBuffer(const void *buffer, u64 len)
                 HvCall2(HvCallBaseWriteLogBuffer,
                                 virt_to_abs(&hv_buf),
                                 left_this_page);
-                cur = (cur & PAGE_MASK) + PAGE_SIZE;
+                cur = (cur & HW_PAGE_MASK) + HW_PAGE_SIZE;
         }
 }
diff --git a/arch/powerpc/platforms/iseries/iommu.c b/arch/powerpc/platforms/iseries/iommu.c
index 1a6845b5c5a4..bf081b345820 100644
--- a/arch/powerpc/platforms/iseries/iommu.c
+++ b/arch/powerpc/platforms/iseries/iommu.c
@@ -43,9 +43,12 @@ static void tce_build_iSeries(struct iommu_table *tbl, long index, long npages,
         u64 rc;
         union tce_entry tce;
 
+        index <<= TCE_PAGE_FACTOR;
+        npages <<= TCE_PAGE_FACTOR;
+
         while (npages--) {
                 tce.te_word = 0;
-                tce.te_bits.tb_rpn = virt_to_abs(uaddr) >> PAGE_SHIFT;
+                tce.te_bits.tb_rpn = virt_to_abs(uaddr) >> TCE_SHIFT;
 
                 if (tbl->it_type == TCE_VB) {
                         /* Virtual Bus */
@@ -66,7 +69,7 @@ static void tce_build_iSeries(struct iommu_table *tbl, long index, long npages,
                         panic("PCI_DMA: HvCallXm_setTce failed, Rc: 0x%lx\n",
                                         rc);
                 index++;
-                uaddr += PAGE_SIZE;
+                uaddr += TCE_PAGE_SIZE;
         }
 }
 
@@ -74,6 +77,9 @@ static void tce_free_iSeries(struct iommu_table *tbl, long index, long npages)
 {
         u64 rc;
 
+        npages <<= TCE_PAGE_FACTOR;
+        index <<= TCE_PAGE_FACTOR;
+
         while (npages--) {
                 rc = HvCallXm_setTce((u64)tbl->it_index, (u64)index, 0);
                 if (rc)
@@ -83,27 +89,6 @@ static void tce_free_iSeries(struct iommu_table *tbl, long index, long npages)
         }
 }
 
-#ifdef CONFIG_PCI
-/*
- * This function compares the known tables to find an iommu_table
- * that has already been built for hardware TCEs.
- */
-static struct iommu_table *iommu_table_find(struct iommu_table * tbl)
-{
-        struct pci_dn *pdn;
-
-        list_for_each_entry(pdn, &iSeries_Global_Device_List, Device_List) {
-                struct iommu_table *it = pdn->iommu_table;
-                if ((it != NULL) &&
-                    (it->it_type == TCE_PCI) &&
-                    (it->it_offset == tbl->it_offset) &&
-                    (it->it_index == tbl->it_index) &&
-                    (it->it_size == tbl->it_size))
-                        return it;
-        }
-        return NULL;
-}
-
 /*
  * Call Hv with the architected data structure to get TCE table info.
  * info. Put the returned data into the Linux representation of the
@@ -113,8 +98,10 @@ static struct iommu_table *iommu_table_find(struct iommu_table * tbl)
  * 2. TCE table per Bus.
  * 3. TCE Table per IOA.
  */
-static void iommu_table_getparms(struct pci_dn *pdn,
-                                 struct iommu_table* tbl)
+void iommu_table_getparms_iSeries(unsigned long busno,
+                                  unsigned char slotno,
+                                  unsigned char virtbus,
+                                  struct iommu_table* tbl)
 {
         struct iommu_table_cb *parms;
 
@@ -124,9 +111,9 @@ static void iommu_table_getparms(struct pci_dn *pdn,
 
         memset(parms, 0, sizeof(*parms));
 
-        parms->itc_busno = pdn->busno;
-        parms->itc_slotno = pdn->LogicalSlot;
-        parms->itc_virtbus = 0;
+        parms->itc_busno = busno;
+        parms->itc_slotno = slotno;
+        parms->itc_virtbus = virtbus;
 
         HvCallXm_getTceTableParms(iseries_hv_addr(parms));
 
@@ -134,17 +121,40 @@ static void iommu_table_getparms(struct pci_dn *pdn,
                 panic("PCI_DMA: parms->size is zero, parms is 0x%p", parms);
 
         /* itc_size is in pages worth of table, it_size is in # of entries */
-        tbl->it_size = (parms->itc_size * PAGE_SIZE) / sizeof(union tce_entry);
+        tbl->it_size = ((parms->itc_size * TCE_PAGE_SIZE) /
+                        sizeof(union tce_entry)) >> TCE_PAGE_FACTOR;
         tbl->it_busno = parms->itc_busno;
-        tbl->it_offset = parms->itc_offset;
+        tbl->it_offset = parms->itc_offset >> TCE_PAGE_FACTOR;
         tbl->it_index = parms->itc_index;
         tbl->it_blocksize = 1;
-        tbl->it_type = TCE_PCI;
+        tbl->it_type = virtbus ? TCE_VB : TCE_PCI;
 
         kfree(parms);
 }
 
 
+#ifdef CONFIG_PCI
+/*
+ * This function compares the known tables to find an iommu_table
+ * that has already been built for hardware TCEs.
+ */
+static struct iommu_table *iommu_table_find(struct iommu_table * tbl)
+{
+        struct pci_dn *pdn;
+
+        list_for_each_entry(pdn, &iSeries_Global_Device_List, Device_List) {
+                struct iommu_table *it = pdn->iommu_table;
+                if ((it != NULL) &&
+                    (it->it_type == TCE_PCI) &&
+                    (it->it_offset == tbl->it_offset) &&
+                    (it->it_index == tbl->it_index) &&
+                    (it->it_size == tbl->it_size))
+                        return it;
+        }
+        return NULL;
+}
+
+
 void iommu_devnode_init_iSeries(struct device_node *dn)
 {
         struct iommu_table *tbl;
@@ -152,7 +162,7 @@ void iommu_devnode_init_iSeries(struct device_node *dn)
 
         tbl = kmalloc(sizeof(struct iommu_table), GFP_KERNEL);
 
-        iommu_table_getparms(pdn, tbl);
+        iommu_table_getparms_iSeries(pdn->busno, pdn->LogicalSlot, 0, tbl);
 
         /* Look for existing tce table */
         pdn->iommu_table = iommu_table_find(tbl);
diff --git a/arch/powerpc/platforms/iseries/setup.c b/arch/powerpc/platforms/iseries/setup.c
index fda712b42168..c5207064977d 100644
--- a/arch/powerpc/platforms/iseries/setup.c
+++ b/arch/powerpc/platforms/iseries/setup.c
@@ -320,11 +320,11 @@ static void __init iSeries_init_early(void)
          */
         if (naca.xRamDisk) {
                 initrd_start = (unsigned long)__va(naca.xRamDisk);
-                initrd_end = initrd_start + naca.xRamDiskSize * PAGE_SIZE;
+                initrd_end = initrd_start + naca.xRamDiskSize * HW_PAGE_SIZE;
                 initrd_below_start_ok = 1;      // ramdisk in kernel space
                 ROOT_DEV = Root_RAM0;
-                if (((rd_size * 1024) / PAGE_SIZE) < naca.xRamDiskSize)
-                        rd_size = (naca.xRamDiskSize * PAGE_SIZE) / 1024;
+                if (((rd_size * 1024) / HW_PAGE_SIZE) < naca.xRamDiskSize)
+                        rd_size = (naca.xRamDiskSize * HW_PAGE_SIZE) / 1024;
         } else
 #endif /* CONFIG_BLK_DEV_INITRD */
         {
@@ -470,13 +470,14 @@ static void __init build_iSeries_Memory_Map(void)
          */
         hptFirstChunk = (u32)addr_to_chunk(HvCallHpt_getHptAddress());
         hptSizePages = (u32)HvCallHpt_getHptPages();
-        hptSizeChunks = hptSizePages >> (MSCHUNKS_CHUNK_SHIFT - PAGE_SHIFT);
+        hptSizeChunks = hptSizePages >>
+                (MSCHUNKS_CHUNK_SHIFT - HW_PAGE_SHIFT);
         hptLastChunk = hptFirstChunk + hptSizeChunks - 1;
 
         printk("HPT absolute addr = %016lx, size = %dK\n",
                         chunk_to_addr(hptFirstChunk), hptSizeChunks * 256);
 
-        ppc64_pft_size = __ilog2(hptSizePages * PAGE_SIZE);
+        ppc64_pft_size = __ilog2(hptSizePages * HW_PAGE_SIZE);
 
         /*
          * The actual hashed page table is in the hypervisor,
@@ -629,7 +630,7 @@ static void __init iSeries_fixup_klimit(void)
          */
         if (naca.xRamDisk)
                 klimit = KERNELBASE + (u64)naca.xRamDisk +
-                        (naca.xRamDiskSize * PAGE_SIZE);
+                        (naca.xRamDiskSize * HW_PAGE_SIZE);
         else {
                 /*
                  * No ram disk was included - check and see if there
diff --git a/arch/powerpc/platforms/iseries/vio.c b/arch/powerpc/platforms/iseries/vio.c
index c27a66876c2c..384360ee06ec 100644
--- a/arch/powerpc/platforms/iseries/vio.c
+++ b/arch/powerpc/platforms/iseries/vio.c
@@ -30,41 +30,14 @@ static struct iommu_table vio_iommu_table;
 
 static void __init iommu_vio_init(void)
 {
-        struct iommu_table *t;
-        struct iommu_table_cb cb;
-        unsigned long cbp;
-        unsigned long itc_entries;
+        iommu_table_getparms_iSeries(255, 0, 0xff, &veth_iommu_table);
+        veth_iommu_table.it_size /= 2;
+        vio_iommu_table = veth_iommu_table;
+        vio_iommu_table.it_offset += veth_iommu_table.it_size;
 
-        cb.itc_busno = 255;    /* Bus 255 is the virtual bus */
-        cb.itc_virtbus = 0xff; /* Ask for virtual bus */
-
-        cbp = virt_to_abs(&cb);
-        HvCallXm_getTceTableParms(cbp);
-
-        itc_entries = cb.itc_size * PAGE_SIZE / sizeof(union tce_entry);
-        veth_iommu_table.it_size        = itc_entries / 2;
-        veth_iommu_table.it_busno       = cb.itc_busno;
-        veth_iommu_table.it_offset      = cb.itc_offset;
-        veth_iommu_table.it_index       = cb.itc_index;
-        veth_iommu_table.it_type        = TCE_VB;
-        veth_iommu_table.it_blocksize   = 1;
-
-        t = iommu_init_table(&veth_iommu_table);
-
-        if (!t)
+        if (!iommu_init_table(&veth_iommu_table))
                 printk("Virtual Bus VETH TCE table failed.\n");
-
-        vio_iommu_table.it_size         = itc_entries - veth_iommu_table.it_size;
-        vio_iommu_table.it_busno        = cb.itc_busno;
-        vio_iommu_table.it_offset       = cb.itc_offset +
-                                          veth_iommu_table.it_size;
-        vio_iommu_table.it_index        = cb.itc_index;
-        vio_iommu_table.it_type         = TCE_VB;
-        vio_iommu_table.it_blocksize    = 1;
-
-        t = iommu_init_table(&vio_iommu_table);
-
-        if (!t)
+        if (!iommu_init_table(&vio_iommu_table))
                 printk("Virtual Bus VIO TCE table failed.\n");
 }
 
diff --git a/arch/powerpc/platforms/iseries/viopath.c b/arch/powerpc/platforms/iseries/viopath.c
index fe97bfbf7463..842672695598 100644
--- a/arch/powerpc/platforms/iseries/viopath.c
+++ b/arch/powerpc/platforms/iseries/viopath.c
@@ -68,7 +68,8 @@ static DEFINE_SPINLOCK(statuslock);
  * For each kind of event we allocate a buffer that is
  * guaranteed not to cross a page boundary
  */
-static unsigned char event_buffer[VIO_MAX_SUBTYPES * 256] __page_aligned;
+static unsigned char event_buffer[VIO_MAX_SUBTYPES * 256]
+        __attribute__((__aligned__(4096)));
 static atomic_t event_buffer_available[VIO_MAX_SUBTYPES];
 static int event_buffer_initialised;
 
@@ -116,12 +117,12 @@ static int proc_viopath_show(struct seq_file *m, void *v)
         HvLpEvent_Rc hvrc;
         DECLARE_MUTEX_LOCKED(Semaphore);
 
-        buf = kmalloc(PAGE_SIZE, GFP_KERNEL);
+        buf = kmalloc(HW_PAGE_SIZE, GFP_KERNEL);
         if (!buf)
                 return 0;
-        memset(buf, 0, PAGE_SIZE);
+        memset(buf, 0, HW_PAGE_SIZE);
 
-        handle = dma_map_single(iSeries_vio_dev, buf, PAGE_SIZE,
+        handle = dma_map_single(iSeries_vio_dev, buf, HW_PAGE_SIZE,
                                 DMA_FROM_DEVICE);
 
         hvrc = HvCallEvent_signalLpEventFast(viopath_hostLp,
@@ -131,7 +132,7 @@ static int proc_viopath_show(struct seq_file *m, void *v)
                         viopath_sourceinst(viopath_hostLp),
                         viopath_targetinst(viopath_hostLp),
                         (u64)(unsigned long)&Semaphore, VIOVERSION << 16,
-                        ((u64)handle) << 32, PAGE_SIZE, 0, 0);
+                        ((u64)handle) << 32, HW_PAGE_SIZE, 0, 0);
 
         if (hvrc != HvLpEvent_Rc_Good)
                 printk(VIOPATH_KERN_WARN "hv error on op %d\n", (int)hvrc);
@@ -140,7 +141,7 @@ static int proc_viopath_show(struct seq_file *m, void *v)
 
         vlanMap = HvLpConfig_getVirtualLanIndexMap();
 
-        buf[PAGE_SIZE-1] = '\0';
+        buf[HW_PAGE_SIZE-1] = '\0';
         seq_printf(m, "%s", buf);
         seq_printf(m, "AVAILABLE_VETH=%x\n", vlanMap);
         seq_printf(m, "SRLNBR=%c%c%c%c%c%c%c\n",
@@ -152,7 +153,8 @@ static int proc_viopath_show(struct seq_file *m, void *v)
                    e2a(xItExtVpdPanel.systemSerial[4]),
                    e2a(xItExtVpdPanel.systemSerial[5]));
 
-        dma_unmap_single(iSeries_vio_dev, handle, PAGE_SIZE, DMA_FROM_DEVICE);
+        dma_unmap_single(iSeries_vio_dev, handle, HW_PAGE_SIZE,
+                         DMA_FROM_DEVICE);
         kfree(buf);
 
         return 0;