Revert "staging: tidspbridge: replace iommu custom for opensource implementation"
authorFelipe Contreras <felipe.contreras@gmail.com>
Wed, 10 Nov 2010 18:12:45 +0000 (12:12 -0600)
committerOmar Ramirez Luna <omar.ramirez@ti.com>
Thu, 11 Nov 2010 00:34:44 +0000 (18:34 -0600)
This reverts commit d95ec7e2fd5cebf2f1caf3f572fa5e0a820ac5b1.

Signed-off-by: Felipe Contreras <felipe.contreras@gmail.com>
Signed-off-by: Omar Ramirez Luna <omar.ramirez@ti.com>
drivers/staging/tidspbridge/core/_tiomap.h
drivers/staging/tidspbridge/core/io_sm.c
drivers/staging/tidspbridge/core/tiomap3430.c
drivers/staging/tidspbridge/core/ue_deh.c
drivers/staging/tidspbridge/include/dspbridge/dspdefs.h
drivers/staging/tidspbridge/rmgr/proc.c

index a42c393..1c1f157 100644 (file)
@@ -23,8 +23,6 @@
 #include <plat/clockdomain.h>
 #include <mach-omap2/prm-regbits-34xx.h>
 #include <mach-omap2/cm-regbits-34xx.h>
-#include <plat/iommu.h>
-#include <plat/iovmm.h>
 #include <dspbridge/devdefs.h>
 #include <hw_defs.h>
 #include <dspbridge/dspioctl.h>        /* for bridge_ioctl_extproc defn */
@@ -330,7 +328,7 @@ struct bridge_dev_context {
        u32 dw_internal_size;   /* Internal memory size */
 
        struct omap_mbox *mbox;         /* Mail box handle */
-       struct iommu *dsp_mmu;      /* iommu for iva2 handler */
+
        struct cfg_hostres *resources;  /* Host Resources */
 
        /*
index 842b8db..5718645 100644 (file)
@@ -291,7 +291,6 @@ int bridge_io_on_loaded(struct io_mgr *hio_mgr)
        struct cod_manager *cod_man;
        struct chnl_mgr *hchnl_mgr;
        struct msg_mgr *hmsg_mgr;
-       struct iommu *mmu;
        u32 ul_shm_base;
        u32 ul_shm_base_offset;
        u32 ul_shm_limit;
@@ -314,6 +313,7 @@ int bridge_io_on_loaded(struct io_mgr *hio_mgr)
        struct bridge_ioctl_extproc ae_proc[BRDIOCTL_NUMOFMMUTLB];
        struct cfg_hostres *host_res;
        struct bridge_dev_context *pbridge_context;
+       u32 map_attrs;
        u32 shm0_end;
        u32 ul_dyn_ext_base;
        u32 ul_seg1_size = 0;
@@ -337,20 +337,6 @@ int bridge_io_on_loaded(struct io_mgr *hio_mgr)
                status = -EFAULT;
                goto func_end;
        }
-       mmu = pbridge_context->dsp_mmu;
-
-       if (mmu)
-               iommu_put(mmu);
-       mmu = iommu_get("iva2");
-
-       if (IS_ERR_OR_NULL(mmu)) {
-               dev_err(bridge, "iommu_get failed!\n");
-               pbridge_context->dsp_mmu = NULL;
-               status = -EFAULT;
-               goto func_end;
-       }
-       pbridge_context->dsp_mmu = mmu;
-
        status = dev_get_cod_mgr(hio_mgr->hdev_obj, &cod_man);
        if (!cod_man) {
                status = -EFAULT;
@@ -490,16 +476,55 @@ int bridge_io_on_loaded(struct io_mgr *hio_mgr)
        gpp_va_curr = ul_gpp_va;
        num_bytes = ul_seg1_size;
 
-       va_curr = iommu_kmap(mmu, va_curr, pa_curr, num_bytes,
-                                       IOVMF_ENDIAN_LITTLE | IOVMF_ELSZ_32);
-       if (IS_ERR_VALUE(va_curr)) {
-               status = (int)va_curr;
-               goto func_end;
-       }
+       /*
+        * Try to fit into TLB entries. If not possible, push them to page
+        * tables. It is quite possible that if sections are not on
+        * bigger page boundary, we may end up making several small pages.
+        * So, push them onto page tables, if that is the case.
+        */
+       map_attrs = 0x00000000;
+       map_attrs = DSP_MAPLITTLEENDIAN;
+       map_attrs |= DSP_MAPPHYSICALADDR;
+       map_attrs |= DSP_MAPELEMSIZE32;
+       map_attrs |= DSP_MAPDONOTLOCK;
 
-       pa_curr += ul_pad_size + num_bytes;
-       va_curr += ul_pad_size + num_bytes;
-       gpp_va_curr += ul_pad_size + num_bytes;
+       while (num_bytes) {
+               /*
+                * To find the max. page size with which both PA & VA are
+                * aligned.
+                */
+               all_bits = pa_curr | va_curr;
+               dev_dbg(bridge, "all_bits %x, pa_curr %x, va_curr %x, "
+                       "num_bytes %x\n", all_bits, pa_curr, va_curr,
+                       num_bytes);
+               for (i = 0; i < 4; i++) {
+                       if ((num_bytes >= page_size[i]) && ((all_bits &
+                                                            (page_size[i] -
+                                                             1)) == 0)) {
+                               status =
+                                   hio_mgr->intf_fxns->
+                                   pfn_brd_mem_map(hio_mgr->hbridge_context,
+                                                   pa_curr, va_curr,
+                                                   page_size[i], map_attrs,
+                                                   NULL);
+                               if (status)
+                                       goto func_end;
+                               pa_curr += page_size[i];
+                               va_curr += page_size[i];
+                               gpp_va_curr += page_size[i];
+                               num_bytes -= page_size[i];
+                               /*
+                                * Don't try smaller sizes. Hopefully we have
+                                * reached an address aligned to a bigger page
+                                * size.
+                                */
+                               break;
+                       }
+               }
+       }
+       pa_curr += ul_pad_size;
+       va_curr += ul_pad_size;
+       gpp_va_curr += ul_pad_size;
 
        /* Configure the TLB entries for the next cacheable segment */
        num_bytes = ul_seg_size;
@@ -541,6 +566,22 @@ int bridge_io_on_loaded(struct io_mgr *hio_mgr)
                                        ae_proc[ndx].ul_dsp_va *
                                        hio_mgr->word_size, page_size[i]);
                                ndx++;
+                       } else {
+                               status =
+                                   hio_mgr->intf_fxns->
+                                   pfn_brd_mem_map(hio_mgr->hbridge_context,
+                                                   pa_curr, va_curr,
+                                                   page_size[i], map_attrs,
+                                                   NULL);
+                               dev_dbg(bridge,
+                                       "shm MMU PTE entry PA %x"
+                                       " VA %x DSP_VA %x Size %x\n",
+                                       ae_proc[ndx].ul_gpp_pa,
+                                       ae_proc[ndx].ul_gpp_va,
+                                       ae_proc[ndx].ul_dsp_va *
+                                       hio_mgr->word_size, page_size[i]);
+                               if (status)
+                                       goto func_end;
                        }
                        pa_curr += page_size[i];
                        va_curr += page_size[i];
@@ -593,29 +634,37 @@ int bridge_io_on_loaded(struct io_mgr *hio_mgr)
                                        "DSP_VA 0x%x\n", ae_proc[ndx].ul_gpp_pa,
                                        ae_proc[ndx].ul_dsp_va);
                                ndx++;
+                       } else {
+                               status = hio_mgr->intf_fxns->pfn_brd_mem_map
+                                   (hio_mgr->hbridge_context,
+                                    hio_mgr->ext_proc_info.ty_tlb[i].
+                                    ul_gpp_phys,
+                                    hio_mgr->ext_proc_info.ty_tlb[i].
+                                    ul_dsp_virt, 0x100000, map_attrs,
+                                    NULL);
                        }
                }
                if (status)
                        goto func_end;
        }
 
+       map_attrs = 0x00000000;
+       map_attrs = DSP_MAPLITTLEENDIAN;
+       map_attrs |= DSP_MAPPHYSICALADDR;
+       map_attrs |= DSP_MAPELEMSIZE32;
+       map_attrs |= DSP_MAPDONOTLOCK;
+
        /* Map the L4 peripherals */
        i = 0;
        while (l4_peripheral_table[i].phys_addr) {
-               status = iommu_kmap(mmu, l4_peripheral_table[i].
-                       dsp_virt_addr, l4_peripheral_table[i].phys_addr,
-                       PAGE_SIZE, IOVMF_ENDIAN_LITTLE | IOVMF_ELSZ_32);
-               if (IS_ERR_VALUE(status))
-                       break;
+               status = hio_mgr->intf_fxns->pfn_brd_mem_map
+                   (hio_mgr->hbridge_context, l4_peripheral_table[i].phys_addr,
+                    l4_peripheral_table[i].dsp_virt_addr, HW_PAGE_SIZE4KB,
+                    map_attrs, NULL);
+               if (status)
+                       goto func_end;
                i++;
        }
-       if (IS_ERR_VALUE(status)) {
-               while (i--)
-                       iommu_kunmap(mmu, l4_peripheral_table[i].
-                                               dsp_virt_addr);
-               goto func_end;
-       }
-       status = 0;
 
        for (i = ndx; i < BRDIOCTL_NUMOFMMUTLB; i++) {
                ae_proc[i].ul_dsp_va = 0;
index ec85529..1be081f 100644 (file)
@@ -114,7 +114,7 @@ static int bridge_brd_mem_map(struct bridge_dev_context *dev_ctxt,
                                  u32 ul_num_bytes, u32 ul_map_attr,
                                  struct page **mapped_pages);
 static int bridge_brd_mem_un_map(struct bridge_dev_context *dev_ctxt,
-                                    u32 da);
+                                    u32 virt_addr, u32 ul_num_bytes);
 static int bridge_dev_create(struct bridge_dev_context
                                        **dev_cntxt,
                                        struct dev_object *hdev_obj,
@@ -122,8 +122,6 @@ static int bridge_dev_create(struct bridge_dev_context
 static int bridge_dev_ctrl(struct bridge_dev_context *dev_context,
                                  u32 dw_cmd, void *pargs);
 static int bridge_dev_destroy(struct bridge_dev_context *dev_ctxt);
-static int get_io_pages(struct mm_struct *mm, u32 uva, unsigned pages,
-                                               struct page **usr_pgs);
 static u32 user_va2_pa(struct mm_struct *mm, u32 address);
 static int pte_update(struct bridge_dev_context *dev_ctxt, u32 pa,
                             u32 va, u32 size,
@@ -373,7 +371,6 @@ static int bridge_brd_start(struct bridge_dev_context *dev_ctxt,
 {
        int status = 0;
        struct bridge_dev_context *dev_context = dev_ctxt;
-       struct iommu *mmu;
        u32 dw_sync_addr = 0;
        u32 ul_shm_base;        /* Gpp Phys SM base addr(byte) */
        u32 ul_shm_base_virt;   /* Dsp Virt SM base addr */
@@ -393,7 +390,6 @@ static int bridge_brd_start(struct bridge_dev_context *dev_ctxt,
        struct omap_dsp_platform_data *pdata =
                omap_dspbridge_dev->dev.platform_data;
 
-       mmu = dev_context->dsp_mmu;
        /* The device context contains all the mmu setup info from when the
         * last dsp base image was loaded. The first entry is always
         * SHMMEM base. */
@@ -444,10 +440,29 @@ static int bridge_brd_start(struct bridge_dev_context *dev_ctxt,
                }
        }
        if (!status) {
+               /* Reset and Unreset the RST2, so that BOOTADDR is copied to
+                * IVA2 SYSC register */
+               (*pdata->dsp_prm_rmw_bits)(OMAP3430_RST2_IVA2_MASK,
+                       OMAP3430_RST2_IVA2_MASK, OMAP3430_IVA2_MOD, OMAP2_RM_RSTCTRL);
+               udelay(100);
+               (*pdata->dsp_prm_rmw_bits)(OMAP3430_RST2_IVA2_MASK, 0,
+                                       OMAP3430_IVA2_MOD, OMAP2_RM_RSTCTRL);
+               udelay(100);
+
+               /* Disbale the DSP MMU */
+               hw_mmu_disable(resources->dw_dmmu_base);
+               /* Disable TWL */
+               hw_mmu_twl_disable(resources->dw_dmmu_base);
+
                /* Only make TLB entry if both addresses are non-zero */
                for (entry_ndx = 0; entry_ndx < BRDIOCTL_NUMOFMMUTLB;
                     entry_ndx++) {
                        struct bridge_ioctl_extproc *e = &dev_context->atlb_entry[entry_ndx];
+                       struct hw_mmu_map_attrs_t map_attrs = {
+                               .endianism = e->endianism,
+                               .element_size = e->elem_size,
+                               .mixed_size = e->mixed_mode,
+                       };
 
                        if (!e->ul_gpp_pa || !e->ul_dsp_va)
                                continue;
@@ -459,8 +474,13 @@ static int bridge_brd_start(struct bridge_dev_context *dev_ctxt,
                                        e->ul_dsp_va,
                                        e->ul_size);
 
-                       iommu_kmap(mmu, e->ul_dsp_va, e->ul_gpp_pa, e->ul_size,
-                               IOVMF_ENDIAN_LITTLE | IOVMF_ELSZ_32);
+                       hw_mmu_tlb_add(dev_context->dw_dsp_mmu_base,
+                                       e->ul_gpp_pa,
+                                       e->ul_dsp_va,
+                                       e->ul_size,
+                                       itmp_entry_ndx,
+                                       &map_attrs, 1, 1);
+
                        itmp_entry_ndx++;
                }
        }
@@ -468,13 +488,29 @@ static int bridge_brd_start(struct bridge_dev_context *dev_ctxt,
        /* Lock the above TLB entries and get the BIOS and load monitor timer
         * information */
        if (!status) {
+               hw_mmu_num_locked_set(resources->dw_dmmu_base, itmp_entry_ndx);
+               hw_mmu_victim_num_set(resources->dw_dmmu_base, itmp_entry_ndx);
+               hw_mmu_ttb_set(resources->dw_dmmu_base,
+                              dev_context->pt_attrs->l1_base_pa);
+               hw_mmu_twl_enable(resources->dw_dmmu_base);
+               /* Enable the SmartIdle and AutoIdle bit for MMU_SYSCONFIG */
+
+               temp = __raw_readl((resources->dw_dmmu_base) + 0x10);
+               temp = (temp & 0xFFFFFFEF) | 0x11;
+               __raw_writel(temp, (resources->dw_dmmu_base) + 0x10);
+
+               /* Let the DSP MMU run */
+               hw_mmu_enable(resources->dw_dmmu_base);
+
                /* Enable the BIOS clock */
                (void)dev_get_symbol(dev_context->hdev_obj,
                                     BRIDGEINIT_BIOSGPTIMER, &ul_bios_gp_timer);
                (void)dev_get_symbol(dev_context->hdev_obj,
                                     BRIDGEINIT_LOADMON_GPTIMER,
                                     &ul_load_monitor_timer);
+       }
 
+       if (!status) {
                if (ul_load_monitor_timer != 0xFFFF) {
                        clk_cmd = (BPWR_ENABLE_CLOCK << MBX_PM_CLK_CMDSHIFT) |
                            ul_load_monitor_timer;
@@ -483,7 +519,9 @@ static int bridge_brd_start(struct bridge_dev_context *dev_ctxt,
                        dev_dbg(bridge, "Not able to get the symbol for Load "
                                "Monitor Timer\n");
                }
+       }
 
+       if (!status) {
                if (ul_bios_gp_timer != 0xFFFF) {
                        clk_cmd = (BPWR_ENABLE_CLOCK << MBX_PM_CLK_CMDSHIFT) |
                            ul_bios_gp_timer;
@@ -492,7 +530,9 @@ static int bridge_brd_start(struct bridge_dev_context *dev_ctxt,
                        dev_dbg(bridge,
                                "Not able to get the symbol for BIOS Timer\n");
                }
+       }
 
+       if (!status) {
                /* Set the DSP clock rate */
                (void)dev_get_symbol(dev_context->hdev_obj,
                                     "_BRIDGEINIT_DSP_FREQ", &ul_dsp_clk_addr);
@@ -545,6 +585,9 @@ static int bridge_brd_start(struct bridge_dev_context *dev_ctxt,
 
                /* Let DSP go */
                dev_dbg(bridge, "%s Unreset\n", __func__);
+               /* Enable DSP MMU Interrupts */
+               hw_mmu_event_enable(resources->dw_dmmu_base,
+                                   HW_MMU_ALL_INTERRUPTS);
                /* release the RST1, DSP starts executing now .. */
                (*pdata->dsp_prm_rmw_bits)(OMAP3430_RST1_IVA2_MASK, 0,
                                        OMAP3430_IVA2_MOD, OMAP2_RM_RSTCTRL);
@@ -645,8 +688,6 @@ static int bridge_brd_stop(struct bridge_dev_context *dev_ctxt)
                omap_mbox_put(dev_context->mbox);
                dev_context->mbox = NULL;
        }
-       if (dev_context->dsp_mmu)
-               dev_context->dsp_mmu = (iommu_put(dev_context->dsp_mmu), NULL);
        /* Reset IVA2 clocks*/
        (*pdata->dsp_prm_write)(OMAP3430_RST1_IVA2_MASK | OMAP3430_RST2_IVA2_MASK |
                        OMAP3430_RST3_IVA2_MASK, OMAP3430_IVA2_MOD, OMAP2_RM_RSTCTRL);
@@ -1095,81 +1136,217 @@ static int bridge_brd_mem_write(struct bridge_dev_context *dev_ctxt,
  *
  *  TODO: Disable MMU while updating the page tables (but that'll stall DSP)
  */
-static int bridge_brd_mem_map(struct bridge_dev_context *dev_ctx,
-                       u32 uva, u32 da, u32 size, u32 attr,
-                       struct page **usr_pgs)
-
+static int bridge_brd_mem_map(struct bridge_dev_context *dev_ctxt,
+                                 u32 ul_mpu_addr, u32 virt_addr,
+                                 u32 ul_num_bytes, u32 ul_map_attr,
+                                 struct page **mapped_pages)
 {
-       int res, w;
-       unsigned pages, i;
-       struct iommu *mmu = dev_ctx->dsp_mmu;
+       u32 attrs;
+       int status = 0;
+       struct bridge_dev_context *dev_context = dev_ctxt;
+       struct hw_mmu_map_attrs_t hw_attrs;
        struct vm_area_struct *vma;
        struct mm_struct *mm = current->mm;
-       struct sg_table *sgt;
-       struct scatterlist *sg;
-
-       if (!size || !usr_pgs)
+       u32 write = 0;
+       u32 num_usr_pgs = 0;
+       struct page *mapped_page, *pg;
+       s32 pg_num;
+       u32 va = virt_addr;
+       struct task_struct *curr_task = current;
+       u32 pg_i = 0;
+       u32 mpu_addr, pa;
+
+       dev_dbg(bridge,
+               "%s hDevCtxt %p, pa %x, va %x, size %x, ul_map_attr %x\n",
+               __func__, dev_ctxt, ul_mpu_addr, virt_addr, ul_num_bytes,
+               ul_map_attr);
+       if (ul_num_bytes == 0)
                return -EINVAL;
 
-       pages = size / PG_SIZE4K;
+       if (ul_map_attr & DSP_MAP_DIR_MASK) {
+               attrs = ul_map_attr;
+       } else {
+               /* Assign default attributes */
+               attrs = ul_map_attr | (DSP_MAPVIRTUALADDR | DSP_MAPELEMSIZE16);
+       }
+       /* Take mapping properties */
+       if (attrs & DSP_MAPBIGENDIAN)
+               hw_attrs.endianism = HW_BIG_ENDIAN;
+       else
+               hw_attrs.endianism = HW_LITTLE_ENDIAN;
+
+       hw_attrs.mixed_size = (enum hw_mmu_mixed_size_t)
+           ((attrs & DSP_MAPMIXEDELEMSIZE) >> 2);
+       /* Ignore element_size if mixed_size is enabled */
+       if (hw_attrs.mixed_size == 0) {
+               if (attrs & DSP_MAPELEMSIZE8) {
+                       /* Size is 8 bit */
+                       hw_attrs.element_size = HW_ELEM_SIZE8BIT;
+               } else if (attrs & DSP_MAPELEMSIZE16) {
+                       /* Size is 16 bit */
+                       hw_attrs.element_size = HW_ELEM_SIZE16BIT;
+               } else if (attrs & DSP_MAPELEMSIZE32) {
+                       /* Size is 32 bit */
+                       hw_attrs.element_size = HW_ELEM_SIZE32BIT;
+               } else if (attrs & DSP_MAPELEMSIZE64) {
+                       /* Size is 64 bit */
+                       hw_attrs.element_size = HW_ELEM_SIZE64BIT;
+               } else {
+                       /*
+                        * Mixedsize isn't enabled, so size can't be
+                        * zero here
+                        */
+                       return -EINVAL;
+               }
+       }
+       if (attrs & DSP_MAPDONOTLOCK)
+               hw_attrs.donotlockmpupage = 1;
+       else
+               hw_attrs.donotlockmpupage = 0;
 
+       if (attrs & DSP_MAPVMALLOCADDR) {
+               return mem_map_vmalloc(dev_ctxt, ul_mpu_addr, virt_addr,
+                                      ul_num_bytes, &hw_attrs);
+       }
+       /*
+        * Do OS-specific user-va to pa translation.
+        * Combine physically contiguous regions to reduce TLBs.
+        * Pass the translated pa to pte_update.
+        */
+       if ((attrs & DSP_MAPPHYSICALADDR)) {
+               status = pte_update(dev_context, ul_mpu_addr, virt_addr,
+                                   ul_num_bytes, &hw_attrs);
+               goto func_cont;
+       }
+
+       /*
+        * Important Note: ul_mpu_addr is mapped from user application process
+        * to current process - it must lie completely within the current
+        * virtual memory address space in order to be of use to us here!
+        */
        down_read(&mm->mmap_sem);
-       vma = find_vma(mm, uva);
-       while (vma && (uva + size > vma->vm_end))
-               vma = find_vma(mm, vma->vm_end + 1);
+       vma = find_vma(mm, ul_mpu_addr);
+       if (vma)
+               dev_dbg(bridge,
+                       "VMAfor UserBuf: ul_mpu_addr=%x, ul_num_bytes=%x, "
+                       "vm_start=%lx, vm_end=%lx, vm_flags=%lx\n", ul_mpu_addr,
+                       ul_num_bytes, vma->vm_start, vma->vm_end,
+                       vma->vm_flags);
 
+       /*
+        * It is observed that under some circumstances, the user buffer is
+        * spread across several VMAs. So loop through and check if the entire
+        * user buffer is covered
+        */
+       while ((vma) && (ul_mpu_addr + ul_num_bytes > vma->vm_end)) {
+               /* jump to the next VMA region */
+               vma = find_vma(mm, vma->vm_end + 1);
+               dev_dbg(bridge,
+                       "VMA for UserBuf ul_mpu_addr=%x ul_num_bytes=%x, "
+                       "vm_start=%lx, vm_end=%lx, vm_flags=%lx\n", ul_mpu_addr,
+                       ul_num_bytes, vma->vm_start, vma->vm_end,
+                       vma->vm_flags);
+       }
        if (!vma) {
                pr_err("%s: Failed to get VMA region for 0x%x (%d)\n",
-                                               __func__, uva, size);
+                      __func__, ul_mpu_addr, ul_num_bytes);
+               status = -EINVAL;
                up_read(&mm->mmap_sem);
-               return -EINVAL;
+               goto func_cont;
        }
-       if (vma->vm_flags & (VM_WRITE | VM_MAYWRITE))
-               w = 1;
-
-       if (vma->vm_flags & VM_IO)
-               i = get_io_pages(mm, uva, pages, usr_pgs);
-       else
-               i = get_user_pages(current, mm, uva, pages, w, 1,
-                                                       usr_pgs, NULL);
-       up_read(&mm->mmap_sem);
 
-       if (i < 0)
-               return i;
+       if (vma->vm_flags & VM_IO) {
+               num_usr_pgs = ul_num_bytes / PG_SIZE4K;
+               mpu_addr = ul_mpu_addr;
+
+               /* Get the physical addresses for user buffer */
+               for (pg_i = 0; pg_i < num_usr_pgs; pg_i++) {
+                       pa = user_va2_pa(mm, mpu_addr);
+                       if (!pa) {
+                               status = -EPERM;
+                               pr_err("DSPBRIDGE: VM_IO mapping physical"
+                                      "address is invalid\n");
+                               break;
+                       }
+                       if (pfn_valid(__phys_to_pfn(pa))) {
+                               pg = PHYS_TO_PAGE(pa);
+                               get_page(pg);
+                               if (page_count(pg) < 1) {
+                                       pr_err("Bad page in VM_IO buffer\n");
+                                       bad_page_dump(pa, pg);
+                               }
+                       }
+                       status = pte_set(dev_context->pt_attrs, pa,
+                                        va, HW_PAGE_SIZE4KB, &hw_attrs);
+                       if (status)
+                               break;
 
-       if (i < pages) {
-               res = -EFAULT;
-               goto err_pages;
+                       va += HW_PAGE_SIZE4KB;
+                       mpu_addr += HW_PAGE_SIZE4KB;
+                       pa += HW_PAGE_SIZE4KB;
+               }
+       } else {
+               num_usr_pgs = ul_num_bytes / PG_SIZE4K;
+               if (vma->vm_flags & (VM_WRITE | VM_MAYWRITE))
+                       write = 1;
+
+               for (pg_i = 0; pg_i < num_usr_pgs; pg_i++) {
+                       pg_num = get_user_pages(curr_task, mm, ul_mpu_addr, 1,
+                                               write, 1, &mapped_page, NULL);
+                       if (pg_num > 0) {
+                               if (page_count(mapped_page) < 1) {
+                                       pr_err("Bad page count after doing"
+                                              "get_user_pages on"
+                                              "user buffer\n");
+                                       bad_page_dump(page_to_phys(mapped_page),
+                                                     mapped_page);
+                               }
+                               status = pte_set(dev_context->pt_attrs,
+                                                page_to_phys(mapped_page), va,
+                                                HW_PAGE_SIZE4KB, &hw_attrs);
+                               if (status)
+                                       break;
+
+                               if (mapped_pages)
+                                       mapped_pages[pg_i] = mapped_page;
+
+                               va += HW_PAGE_SIZE4KB;
+                               ul_mpu_addr += HW_PAGE_SIZE4KB;
+                       } else {
+                               pr_err("DSPBRIDGE: get_user_pages FAILED,"
+                                      "MPU addr = 0x%x,"
+                                      "vma->vm_flags = 0x%lx,"
+                                      "get_user_pages Err"
+                                      "Value = %d, Buffer"
+                                      "size=0x%x\n", ul_mpu_addr,
+                                      vma->vm_flags, pg_num, ul_num_bytes);
+                               status = -EPERM;
+                               break;
+                       }
+               }
        }
-
-       sgt = kzalloc(sizeof(*sgt), GFP_KERNEL);
-       if (!sgt) {
-               res = -ENOMEM;
-               goto err_pages;
+       up_read(&mm->mmap_sem);
+func_cont:
+       if (status) {
+               /*
+                * Roll out the mapped pages incase it failed in middle of
+                * mapping
+                */
+               if (pg_i) {
+                       bridge_brd_mem_un_map(dev_context, virt_addr,
+                                          (pg_i * PG_SIZE4K));
+               }
+               status = -EPERM;
        }
-
-       res = sg_alloc_table(sgt, pages, GFP_KERNEL);
-
-       if (res < 0)
-               goto err_sg;
-
-       for_each_sg(sgt->sgl, sg, sgt->nents, i)
-               sg_set_page(sg, usr_pgs[i], PAGE_SIZE, 0);
-
-       da = iommu_vmap(mmu, da, sgt, IOVMF_ENDIAN_LITTLE | IOVMF_ELSZ_32);
-
-       if (!IS_ERR_VALUE(da))
-               return 0;
-       res = (int)da;
-
-       sg_free_table(sgt);
-err_sg:
-       kfree(sgt);
-       i = pages;
-err_pages:
-       while (i--)
-               put_page(usr_pgs[i]);
-       return res;
+       /*
+        * In any case, flush the TLB
+        * This is called from here instead from pte_update to avoid unnecessary
+        * repetition while mapping non-contiguous physical regions of a virtual
+        * region
+        */
+       flush_all(dev_context);
+       dev_dbg(bridge, "%s status %x\n", __func__, status);
+       return status;
 }
 
 /*
@@ -1180,43 +1357,194 @@ err_pages:
  *      So, instead of looking up the PTE address for every 4K block,
  *      we clear consecutive PTEs until we unmap all the bytes
  */
-static int bridge_brd_mem_un_map(struct bridge_dev_context *dev_ctx, u32 da)
+static int bridge_brd_mem_un_map(struct bridge_dev_context *dev_ctxt,
+                                    u32 virt_addr, u32 ul_num_bytes)
 {
-       unsigned i;
-       struct sg_table *sgt;
-       struct scatterlist *sg;
+       u32 l1_base_va;
+       u32 l2_base_va;
+       u32 l2_base_pa;
+       u32 l2_page_num;
+       u32 pte_val;
+       u32 pte_size;
+       u32 pte_count;
+       u32 pte_addr_l1;
+       u32 pte_addr_l2 = 0;
+       u32 rem_bytes;
+       u32 rem_bytes_l2;
+       u32 va_curr;
+       struct page *pg = NULL;
+       int status = 0;
+       struct bridge_dev_context *dev_context = dev_ctxt;
+       struct pg_table_attrs *pt = dev_context->pt_attrs;
+       u32 temp;
+       u32 paddr;
+       u32 numof4k_pages = 0;
 
-       sgt = iommu_vunmap(dev_ctx->dsp_mmu, da);
-       if (!sgt)
-               return -EFAULT;
+       va_curr = virt_addr;
+       rem_bytes = ul_num_bytes;
+       rem_bytes_l2 = 0;
+       l1_base_va = pt->l1_base_va;
+       pte_addr_l1 = hw_mmu_pte_addr_l1(l1_base_va, va_curr);
+       dev_dbg(bridge, "%s dev_ctxt %p, va %x, NumBytes %x l1_base_va %x, "
+               "pte_addr_l1 %x\n", __func__, dev_ctxt, virt_addr,
+               ul_num_bytes, l1_base_va, pte_addr_l1);
+
+       while (rem_bytes && !status) {
+               u32 va_curr_orig = va_curr;
+               /* Find whether the L1 PTE points to a valid L2 PT */
+               pte_addr_l1 = hw_mmu_pte_addr_l1(l1_base_va, va_curr);
+               pte_val = *(u32 *) pte_addr_l1;
+               pte_size = hw_mmu_pte_size_l1(pte_val);
 
-       for_each_sg(sgt->sgl, sg, sgt->nents, i)
-               put_page(sg_page(sg));
-       sg_free_table(sgt);
-       kfree(sgt);
+               if (pte_size != HW_MMU_COARSE_PAGE_SIZE)
+                       goto skip_coarse_page;
 
-       return 0;
-}
+               /*
+                * Get the L2 PA from the L1 PTE, and find
+                * corresponding L2 VA
+                */
+               l2_base_pa = hw_mmu_pte_coarse_l1(pte_val);
+               l2_base_va = l2_base_pa - pt->l2_base_pa + pt->l2_base_va;
+               l2_page_num =
+                   (l2_base_pa - pt->l2_base_pa) / HW_MMU_COARSE_PAGE_SIZE;
+               /*
+                * Find the L2 PTE address from which we will start
+                * clearing, the number of PTEs to be cleared on this
+                * page, and the size of VA space that needs to be
+                * cleared on this L2 page
+                */
+               pte_addr_l2 = hw_mmu_pte_addr_l2(l2_base_va, va_curr);
+               pte_count = pte_addr_l2 & (HW_MMU_COARSE_PAGE_SIZE - 1);
+               pte_count = (HW_MMU_COARSE_PAGE_SIZE - pte_count) / sizeof(u32);
+               if (rem_bytes < (pte_count * PG_SIZE4K))
+                       pte_count = rem_bytes / PG_SIZE4K;
+               rem_bytes_l2 = pte_count * PG_SIZE4K;
 
+               /*
+                * Unmap the VA space on this L2 PT. A quicker way
+                * would be to clear pte_count entries starting from
+                * pte_addr_l2. However, below code checks that we don't
+                * clear invalid entries or less than 64KB for a 64KB
+                * entry. Similar checking is done for L1 PTEs too
+                * below
+                */
+               while (rem_bytes_l2 && !status) {
+                       pte_val = *(u32 *) pte_addr_l2;
+                       pte_size = hw_mmu_pte_size_l2(pte_val);
+                       /* va_curr aligned to pte_size? */
+                       if (pte_size == 0 || rem_bytes_l2 < pte_size ||
+                           va_curr & (pte_size - 1)) {
+                               status = -EPERM;
+                               break;
+                       }
 
-static int get_io_pages(struct mm_struct *mm, u32 uva, unsigned pages,
-                                               struct page **usr_pgs)
-{
-       u32 pa;
-       int i;
-       struct page *pg;
+                       /* Collect Physical addresses from VA */
+                       paddr = (pte_val & ~(pte_size - 1));
+                       if (pte_size == HW_PAGE_SIZE64KB)
+                               numof4k_pages = 16;
+                       else
+                               numof4k_pages = 1;
+                       temp = 0;
+                       while (temp++ < numof4k_pages) {
+                               if (!pfn_valid(__phys_to_pfn(paddr))) {
+                                       paddr += HW_PAGE_SIZE4KB;
+                                       continue;
+                               }
+                               pg = PHYS_TO_PAGE(paddr);
+                               if (page_count(pg) < 1) {
+                                       pr_info("DSPBRIDGE: UNMAP function: "
+                                               "COUNT 0 FOR PA 0x%x, size = "
+                                               "0x%x\n", paddr, ul_num_bytes);
+                                       bad_page_dump(paddr, pg);
+                               } else {
+                                       set_page_dirty(pg);
+                                       page_cache_release(pg);
+                               }
+                               paddr += HW_PAGE_SIZE4KB;
+                       }
+                       if (hw_mmu_pte_clear(pte_addr_l2, va_curr, pte_size)) {
+                               status = -EPERM;
+                               goto EXIT_LOOP;
+                       }
 
-       for (i = 0; i < pages; i++) {
-               pa = user_va2_pa(mm, uva);
+                       status = 0;
+                       rem_bytes_l2 -= pte_size;
+                       va_curr += pte_size;
+                       pte_addr_l2 += (pte_size >> 12) * sizeof(u32);
+               }
+               spin_lock(&pt->pg_lock);
+               if (rem_bytes_l2 == 0) {
+                       pt->pg_info[l2_page_num].num_entries -= pte_count;
+                       if (pt->pg_info[l2_page_num].num_entries == 0) {
+                               /*
+                                * Clear the L1 PTE pointing to the L2 PT
+                                */
+                               if (!hw_mmu_pte_clear(l1_base_va, va_curr_orig,
+                                                    HW_MMU_COARSE_PAGE_SIZE))
+                                       status = 0;
+                               else {
+                                       status = -EPERM;
+                                       spin_unlock(&pt->pg_lock);
+                                       goto EXIT_LOOP;
+                               }
+                       }
+                       rem_bytes -= pte_count * PG_SIZE4K;
+               } else
+                       status = -EPERM;
 
-               if (!pfn_valid(__phys_to_pfn(pa)))
+               spin_unlock(&pt->pg_lock);
+               continue;
+skip_coarse_page:
+               /* va_curr aligned to pte_size? */
+               /* pte_size = 1 MB or 16 MB */
+               if (pte_size == 0 || rem_bytes < pte_size ||
+                   va_curr & (pte_size - 1)) {
+                       status = -EPERM;
                        break;
+               }
 
-               pg = PHYS_TO_PAGE(pa);
-               usr_pgs[i] = pg;
-               get_page(pg);
+               if (pte_size == HW_PAGE_SIZE1MB)
+                       numof4k_pages = 256;
+               else
+                       numof4k_pages = 4096;
+               temp = 0;
+               /* Collect Physical addresses from VA */
+               paddr = (pte_val & ~(pte_size - 1));
+               while (temp++ < numof4k_pages) {
+                       if (pfn_valid(__phys_to_pfn(paddr))) {
+                               pg = PHYS_TO_PAGE(paddr);
+                               if (page_count(pg) < 1) {
+                                       pr_info("DSPBRIDGE: UNMAP function: "
+                                               "COUNT 0 FOR PA 0x%x, size = "
+                                               "0x%x\n", paddr, ul_num_bytes);
+                                       bad_page_dump(paddr, pg);
+                               } else {
+                                       set_page_dirty(pg);
+                                       page_cache_release(pg);
+                               }
+                       }
+                       paddr += HW_PAGE_SIZE4KB;
+               }
+               if (!hw_mmu_pte_clear(l1_base_va, va_curr, pte_size)) {
+                       status = 0;
+                       rem_bytes -= pte_size;
+                       va_curr += pte_size;
+               } else {
+                       status = -EPERM;
+                       goto EXIT_LOOP;
+               }
        }
-       return i;
+       /*
+        * It is better to flush the TLB here, so that any stale old entries
+        * get flushed
+        */
+EXIT_LOOP:
+       flush_all(dev_context);
+       dev_dbg(bridge,
+               "%s: va_curr %x, pte_addr_l1 %x pte_addr_l2 %x rem_bytes %x,"
+               " rem_bytes_l2 %x status %x\n", __func__, va_curr, pte_addr_l1,
+               pte_addr_l2, rem_bytes, rem_bytes_l2, status);
+       return status;
 }
 
 /*
index 14f3191..3430418 100644 (file)
@@ -115,6 +115,12 @@ int bridge_deh_create(struct deh_mgr **ret_deh,
        /* Fill in context structure */
        deh->hbridge_context = hbridge_context;
 
+       /* Install ISR function for DSP MMU fault */
+       status = request_irq(INT_DSP_MMU_IRQ, mmu_fault_isr, 0,
+                       "DspBridge\tiommu fault", deh);
+       if (status < 0)
+               goto err;
+
        *ret_deh = deh;
        return 0;
 
@@ -134,6 +140,8 @@ int bridge_deh_destroy(struct deh_mgr *deh)
                ntfy_delete(deh->ntfy_obj);
                kfree(deh->ntfy_obj);
        }
+       /* Disable DSP MMU fault */
+       free_irq(INT_DSP_MMU_IRQ, deh);
 
        /* Free DPC object */
        tasklet_kill(&deh->dpc_tasklet);
index 173dfbb..0ae7d16 100644 (file)
@@ -201,7 +201,7 @@ typedef int(*fxn_brd_memmap) (struct bridge_dev_context
  */
 typedef int(*fxn_brd_memunmap) (struct bridge_dev_context
                                       * dev_ctxt,
-                                      u32 da);
+                                      u32 virt_addr, u32 ul_num_bytes);
 
 /*
  *  ======== bridge_brd_stop ========
index 97c5b61..b47d7aa 100644 (file)
@@ -1723,7 +1723,7 @@ int proc_un_map(void *hprocessor, void *map_addr,
        /* Remove mapping from the page tables. */
        if (!status) {
                status = (*p_proc_object->intf_fxns->pfn_brd_mem_un_map)
-                   (p_proc_object->hbridge_context, va_align);
+                   (p_proc_object->hbridge_context, va_align, size_align);
        }
 
        mutex_unlock(&proc_lock);