diff options
32 files changed, 3471 insertions, 572 deletions
diff --git a/arch/arm/plat-omap/devices.c b/arch/arm/plat-omap/devices.c index 6f42a18b8aa..fc819120978 100644 --- a/arch/arm/plat-omap/devices.c +++ b/arch/arm/plat-omap/devices.c @@ -284,12 +284,14 @@ void __init omap_dsp_reserve_sdram_memblock(void) if (!size) return; - paddr = __memblock_alloc_base(size, SZ_1M, MEMBLOCK_REAL_LIMIT); + paddr = memblock_alloc(size, SZ_1M); if (!paddr) { pr_err("%s: failed to reserve %x bytes\n", __func__, size); return; } + memblock_free(paddr, size); + memblock_remove(paddr, size); omap_dsp_phys_mempool_base = paddr; } diff --git a/drivers/staging/tidspbridge/Kconfig b/drivers/staging/tidspbridge/Kconfig index ff64d464143..93de4f2e8bf 100644 --- a/drivers/staging/tidspbridge/Kconfig +++ b/drivers/staging/tidspbridge/Kconfig @@ -6,7 +6,6 @@ menuconfig TIDSPBRIDGE tristate "DSP Bridge driver" depends on ARCH_OMAP3 select OMAP_MBOX_FWK - select OMAP_IOMMU help DSP/BIOS Bridge is designed for platforms that contain a GPP and one or more attached DSPs. The GPP is considered the master or diff --git a/drivers/staging/tidspbridge/Makefile b/drivers/staging/tidspbridge/Makefile index 50decc2935c..41c644c3318 100644 --- a/drivers/staging/tidspbridge/Makefile +++ b/drivers/staging/tidspbridge/Makefile @@ -2,18 +2,19 @@ obj-$(CONFIG_TIDSPBRIDGE) += bridgedriver.o libgen = gen/gb.o gen/gs.o gen/gh.o gen/uuidutil.o libcore = core/chnl_sm.o core/msg_sm.o core/io_sm.o core/tiomap3430.o \ - core/tiomap3430_pwr.o core/tiomap_io.o core/dsp-mmu.o \ + core/tiomap3430_pwr.o core/tiomap_io.o \ core/ue_deh.o core/wdt.o core/dsp-clock.o core/sync.o libpmgr = pmgr/chnl.o pmgr/io.o pmgr/msg.o pmgr/cod.o pmgr/dev.o pmgr/dspapi.o \ - pmgr/cmm.o pmgr/dbll.o + pmgr/dmm.o pmgr/cmm.o pmgr/dbll.o librmgr = rmgr/dbdcd.o rmgr/disp.o rmgr/drv.o rmgr/mgr.o rmgr/node.o \ rmgr/proc.o rmgr/pwr.o rmgr/rmm.o rmgr/strm.o rmgr/dspdrv.o \ rmgr/nldr.o rmgr/drv_interface.o libdload = dynload/cload.o dynload/getsection.o dynload/reloc.o \ dynload/tramp.o +libhw = hw/hw_mmu.o bridgedriver-y := $(libgen) $(libservices) $(libcore) $(libpmgr) $(librmgr) \ - $(libdload) + $(libdload) $(libhw) #Machine dependent ccflags-y += -D_TI_ -D_DB_TIOMAP -DTMS32060 \ diff --git a/drivers/staging/tidspbridge/core/_deh.h b/drivers/staging/tidspbridge/core/_deh.h index 8ae263387a8..16723cd3483 100644 --- a/drivers/staging/tidspbridge/core/_deh.h +++ b/drivers/staging/tidspbridge/core/_deh.h @@ -27,8 +27,9 @@ struct deh_mgr { struct bridge_dev_context *hbridge_context; /* Bridge context. */ struct ntfy_object *ntfy_obj; /* NTFY object */ -}; -int mmu_fault_isr(struct iommu *mmu); + /* MMU Fault DPC */ + struct tasklet_struct dpc_tasklet; +}; #endif /* _DEH_ */ diff --git a/drivers/staging/tidspbridge/core/_tiomap.h b/drivers/staging/tidspbridge/core/_tiomap.h index e0a801c1cb9..1c1f157e167 100644 --- a/drivers/staging/tidspbridge/core/_tiomap.h +++ b/drivers/staging/tidspbridge/core/_tiomap.h @@ -23,8 +23,8 @@ #include <plat/clockdomain.h> #include <mach-omap2/prm-regbits-34xx.h> #include <mach-omap2/cm-regbits-34xx.h> -#include <dspbridge/dsp-mmu.h> #include <dspbridge/devdefs.h> +#include <hw_defs.h> #include <dspbridge/dspioctl.h> /* for bridge_ioctl_extproc defn */ #include <dspbridge/sync.h> #include <dspbridge/clk.h> @@ -306,18 +306,6 @@ static const struct bpwr_clk_t bpwr_clks[] = { #define CLEAR_BIT_INDEX(reg, index) (reg &= ~(1 << (index))) -struct shm_segs { - u32 seg0_da; - u32 seg0_pa; - u32 seg0_va; - u32 seg0_size; - u32 seg1_da; - u32 seg1_pa; - u32 seg1_va; - u32 seg1_size; -}; - - /* This Bridge driver's device context: */ struct bridge_dev_context { struct dev_object *hdev_obj; /* Handle to Bridge device object. */ @@ -328,6 +316,7 @@ struct bridge_dev_context { */ u32 dw_dsp_ext_base_addr; /* See the comment above */ u32 dw_api_reg_base; /* API mem map'd registers */ + void __iomem *dw_dsp_mmu_base; /* DSP MMU Mapped registers */ u32 dw_api_clk_base; /* CLK Registers */ u32 dw_dsp_clk_m2_base; /* DSP Clock Module m2 */ u32 dw_public_rhea; /* Pub Rhea */ @@ -339,8 +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 shm_segs sh_s; + struct cfg_hostres *resources; /* Host Resources */ /* @@ -353,6 +341,7 @@ struct bridge_dev_context { /* TC Settings */ bool tc_word_swap_on; /* Traffic Controller Word Swap */ + struct pg_table_attrs *pt_attrs; u32 dsp_per_clks; }; diff --git a/drivers/staging/tidspbridge/core/dsp-mmu.c b/drivers/staging/tidspbridge/core/dsp-mmu.c deleted file mode 100644 index 983c95adc8f..00000000000 --- a/drivers/staging/tidspbridge/core/dsp-mmu.c +++ /dev/null @@ -1,317 +0,0 @@ -/* - * dsp-mmu.c - * - * DSP-BIOS Bridge driver support functions for TI OMAP processors. - * - * DSP iommu. - * - * Copyright (C) 2010 Texas Instruments, Inc. - * - * This package is free software; you can redistribute it and/or modify - * it under the terms of the GNU General Public License version 2 as - * published by the Free Software Foundation. - * - * THIS PACKAGE IS PROVIDED ``AS IS'' AND WITHOUT ANY EXPRESS OR - * IMPLIED WARRANTIES, INCLUDING, WITHOUT LIMITATION, THE IMPLIED - * WARRANTIES OF MERCHANTIBILITY AND FITNESS FOR A PARTICULAR PURPOSE. - */ - -#include <dspbridge/host_os.h> -#include <plat/dmtimer.h> -#include <dspbridge/dbdefs.h> -#include <dspbridge/dev.h> -#include <dspbridge/io_sm.h> -#include <dspbridge/dspdeh.h> -#include "_tiomap.h" - -#include <dspbridge/dsp-mmu.h> - -#define MMU_CNTL_TWL_EN (1 << 2) - -static struct tasklet_struct mmu_tasklet; - -#ifdef CONFIG_TIDSPBRIDGE_BACKTRACE -static void mmu_fault_print_stack(struct bridge_dev_context *dev_context) -{ - void *dummy_addr; - u32 fa, tmp; - struct iotlb_entry e; - struct iommu *mmu = dev_context->dsp_mmu; - dummy_addr = (void *)__get_free_page(GFP_ATOMIC); - - /* - * Before acking the MMU fault, let's make sure MMU can only - * access entry #0. Then add a new entry so that the DSP OS - * can continue in order to dump the stack. - */ - tmp = iommu_read_reg(mmu, MMU_CNTL); - tmp &= ~MMU_CNTL_TWL_EN; - iommu_write_reg(mmu, tmp, MMU_CNTL); - fa = iommu_read_reg(mmu, MMU_FAULT_AD); - e.da = fa & PAGE_MASK; - e.pa = virt_to_phys(dummy_addr); - e.valid = 1; - e.prsvd = 1; - e.pgsz = IOVMF_PGSZ_4K & MMU_CAM_PGSZ_MASK; - e.endian = MMU_RAM_ENDIAN_LITTLE; - e.elsz = MMU_RAM_ELSZ_32; - e.mixed = 0; - - load_iotlb_entry(mmu, &e); - - dsp_clk_enable(DSP_CLK_GPT8); - - dsp_gpt_wait_overflow(DSP_CLK_GPT8, 0xfffffffe); - - /* Clear MMU interrupt */ - tmp = iommu_read_reg(mmu, MMU_IRQSTATUS); - iommu_write_reg(mmu, tmp, MMU_IRQSTATUS); - - dump_dsp_stack(dev_context); - dsp_clk_disable(DSP_CLK_GPT8); - - iopgtable_clear_entry(mmu, fa); - free_page((unsigned long)dummy_addr); -} -#endif - - -static void fault_tasklet(unsigned long data) -{ - struct iommu *mmu = (struct iommu *)data; - struct bridge_dev_context *dev_ctx; - struct deh_mgr *dm; - u32 fa; - dev_get_deh_mgr(dev_get_first(), &dm); - dev_get_bridge_context(dev_get_first(), &dev_ctx); - - if (!dm || !dev_ctx) - return; - - fa = iommu_read_reg(mmu, MMU_FAULT_AD); - -#ifdef CONFIG_TIDSPBRIDGE_BACKTRACE - print_dsp_trace_buffer(dev_ctx); - dump_dl_modules(dev_ctx); - mmu_fault_print_stack(dev_ctx); -#endif - - bridge_deh_notify(dm, DSP_MMUFAULT, fa); -} - -/* - * ======== mmu_fault_isr ======== - * ISR to be triggered by a DSP MMU fault interrupt. - */ -static int mmu_fault_callback(struct iommu *mmu) -{ - if (!mmu) - return -EPERM; - - iommu_write_reg(mmu, 0, MMU_IRQENABLE); - tasklet_schedule(&mmu_tasklet); - return 0; -} - -/** - * dsp_mmu_init() - initialize dsp_mmu module and returns a handle - * - * This function initialize dsp mmu module and returns a struct iommu - * handle to use it for dsp maps. - * - */ -struct iommu *dsp_mmu_init() -{ - struct iommu *mmu; - - mmu = iommu_get("iva2"); - - if (!IS_ERR(mmu)) { - tasklet_init(&mmu_tasklet, fault_tasklet, (unsigned long)mmu); - mmu->isr = mmu_fault_callback; - } - - return mmu; -} - -/** - * dsp_mmu_exit() - destroy dsp mmu module - * @mmu: Pointer to iommu handle. - * - * This function destroys dsp mmu module. - * - */ -void dsp_mmu_exit(struct iommu *mmu) -{ - if (mmu) - iommu_put(mmu); - tasklet_kill(&mmu_tasklet); -} - -/** - * user_va2_pa() - get physical address from userspace address. - * @mm: mm_struct Pointer of the process. - * @address: Virtual user space address. - * - */ -static u32 user_va2_pa(struct mm_struct *mm, u32 address) -{ - pgd_t *pgd; - pmd_t *pmd; - pte_t *ptep, pte; - - pgd = pgd_offset(mm, address); - if (!(pgd_none(*pgd) || pgd_bad(*pgd))) { - pmd = pmd_offset(pgd, address); - if (!(pmd_none(*pmd) || pmd_bad(*pmd))) { - ptep = pte_offset_map(pmd, address); - if (ptep) { - pte = *ptep; - if (pte_present(pte)) - return pte & PAGE_MASK; - } - } - } - - return 0; -} - -/** - * get_io_pages() - pin and get pages of io user's buffer. - * @mm: mm_struct Pointer of the process. - * @uva: Virtual user space address. - * @pages Pages to be pined. - * @usr_pgs struct page array pointer where the user pages will be stored - * - */ -static int get_io_pages(struct mm_struct *mm, u32 uva, unsigned pages, - struct page **usr_pgs) -{ - u32 pa; - int i; - struct page *pg; - - for (i = 0; i < pages; i++) { - pa = user_va2_pa(mm, uva); - - if (!pfn_valid(__phys_to_pfn(pa))) - break; - - pg = phys_to_page(pa); - usr_pgs[i] = pg; - get_page(pg); - } - return i; -} - -/** - * user_to_dsp_map() - maps user to dsp virtual address - * @mmu: Pointer to iommu handle. - * @uva: Virtual user space address. - * @da DSP address - * @size Buffer size to map. - * @usr_pgs struct page array pointer where the user pages will be stored - * - * This function maps a user space buffer into DSP virtual address. - * - */ -u32 user_to_dsp_map(struct iommu *mmu, u32 uva, u32 da, u32 size, - struct page **usr_pgs) -{ - int res, w; - unsigned pages; - int i; - struct vm_area_struct *vma; - struct mm_struct *mm = current->mm; - struct sg_table *sgt; - struct scatterlist *sg; - - if (!size || !usr_pgs) - return -EINVAL; - - pages = size / PG_SIZE4K; - - 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); - - if (!vma) { - pr_err("%s: Failed to get VMA region for 0x%x (%d)\n", - __func__, uva, size); - up_read(&mm->mmap_sem); - return -EINVAL; - } - 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 (i < pages) { - res = -EFAULT; - goto err_pages; - } - - sgt = kzalloc(sizeof(*sgt), GFP_KERNEL); - if (!sgt) { - res = -ENOMEM; - goto err_pages; - } - - 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 da; - res = (int)da; - - sg_free_table(sgt); -err_sg: - kfree(sgt); - i = pages; -err_pages: - while (i--) - put_page(usr_pgs[i]); - return res; -} - -/** - * user_to_dsp_unmap() - unmaps DSP virtual buffer. - * @mmu: Pointer to iommu handle. - * @da DSP address - * - * This function unmaps a user space buffer into DSP virtual address. - * - */ -int user_to_dsp_unmap(struct iommu *mmu, u32 da) -{ - unsigned i; - struct sg_table *sgt; - struct scatterlist *sg; - - sgt = iommu_vunmap(mmu, da); - if (!sgt) - return -EFAULT; - - for_each_sg(sgt->sgl, sg, sgt->nents, i) - put_page(sg_page(sg)); - sg_free_table(sgt); - kfree(sgt); - - return 0; -} diff --git a/drivers/staging/tidspbridge/core/io_sm.c b/drivers/staging/tidspbridge/core/io_sm.c index 194badaba0e..571864555dd 100644 --- a/drivers/staging/tidspbridge/core/io_sm.c +++ b/drivers/staging/tidspbridge/core/io_sm.c @@ -39,6 +39,10 @@ #include <dspbridge/ntfy.h> #include <dspbridge/sync.h> +/* Hardware Abstraction Layer */ +#include <hw_defs.h> +#include <hw_mmu.h> + /* Bridge Driver */ #include <dspbridge/dspdeh.h> #include <dspbridge/dspio.h> @@ -287,7 +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 shm_segs *sm_sg; u32 ul_shm_base; u32 ul_shm_base_offset; u32 ul_shm_limit; @@ -310,9 +313,18 @@ 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; + u32 pa_curr = 0; + u32 va_curr = 0; + u32 gpp_va_curr = 0; + u32 num_bytes = 0; + u32 all_bits = 0; + u32 page_size[] = { HW_PAGE_SIZE16MB, HW_PAGE_SIZE1MB, + HW_PAGE_SIZE64KB, HW_PAGE_SIZE4KB + }; status = dev_get_bridge_context(hio_mgr->hdev_obj, &pbridge_context); if (!pbridge_context) { @@ -325,8 +337,6 @@ int bridge_io_on_loaded(struct io_mgr *hio_mgr) status = -EFAULT; goto func_end; } - sm_sg = &pbridge_context->sh_s; - status = dev_get_cod_mgr(hio_mgr->hdev_obj, &cod_man); if (!cod_man) { status = -EFAULT; @@ -461,14 +471,129 @@ int bridge_io_on_loaded(struct io_mgr *hio_mgr) if (status) goto func_end; - sm_sg->seg1_pa = ul_gpp_pa; - sm_sg->seg1_da = ul_dyn_ext_base; - sm_sg->seg1_va = ul_gpp_va; - sm_sg->seg1_size = ul_seg1_size; - sm_sg->seg0_pa = ul_gpp_pa + ul_pad_size + ul_seg1_size; - sm_sg->seg0_da = ul_dsp_va; - sm_sg->seg0_va = ul_gpp_va + ul_pad_size + ul_seg1_size; - sm_sg->seg0_size = ul_seg_size; + pa_curr = ul_gpp_pa; + va_curr = ul_dyn_ext_base * hio_mgr->word_size; + gpp_va_curr = ul_gpp_va; + num_bytes = ul_seg1_size; + + /* + * 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; + + 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; + va_curr = ul_dsp_va * hio_mgr->word_size; + 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 for Seg1 %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)) + continue; + if (ndx < MAX_LOCK_TLB_ENTRIES) { + /* + * This is the physical address written to + * DSP MMU. + */ + ae_proc[ndx].ul_gpp_pa = pa_curr; + /* + * This is the virtual uncached ioremapped + * address!!! + */ + ae_proc[ndx].ul_gpp_va = gpp_va_curr; + ae_proc[ndx].ul_dsp_va = + va_curr / hio_mgr->word_size; + ae_proc[ndx].ul_size = page_size[i]; + ae_proc[ndx].endianism = HW_LITTLE_ENDIAN; + ae_proc[ndx].elem_size = HW_ELEM_SIZE16BIT; + ae_proc[ndx].mixed_mode = HW_MMU_CPUES; + dev_dbg(bridge, "shm MMU TLB 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]); + 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]; + 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; + } + } /* * Copy remaining entries from CDB. All entries are 1 MB and @@ -509,12 +634,38 @@ 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 = 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++; + } + for (i = ndx; i < BRDIOCTL_NUMOFMMUTLB; i++) { ae_proc[i].ul_dsp_va = 0; ae_proc[i].ul_gpp_pa = 0; @@ -537,12 +688,12 @@ int bridge_io_on_loaded(struct io_mgr *hio_mgr) status = -EFAULT; goto func_end; } else { - if (sm_sg->seg0_da > ul_shm_base) { + if (ae_proc[0].ul_dsp_va > ul_shm_base) { status = -EPERM; goto func_end; } /* ul_shm_base may not be at ul_dsp_va address */ - ul_shm_base_offset = (ul_shm_base - sm_sg->seg0_da) * + ul_shm_base_offset = (ul_shm_base - ae_proc[0].ul_dsp_va) * hio_mgr->word_size; /* * bridge_dev_ctrl() will set dev context dsp-mmu info. In @@ -566,7 +717,8 @@ int bridge_io_on_loaded(struct io_mgr *hio_mgr) goto func_end; } /* Register SM */ - status = register_shm_segs(hio_mgr, cod_man, sm_sg->seg0_pa); + status = + register_shm_segs(hio_mgr, cod_man, ae_proc[0].ul_gpp_pa); } hio_mgr->shared_mem = (struct shm *)ul_shm_base; diff --git a/drivers/staging/tidspbridge/core/tiomap3430.c b/drivers/staging/tidspbridge/core/tiomap3430.c index f22bc12bc0d..1be081f917a 100644 --- a/drivers/staging/tidspbridge/core/tiomap3430.c +++ b/drivers/staging/tidspbridge/core/tiomap3430.c @@ -23,7 +23,6 @@ #include <dspbridge/host_os.h> #include <linux/mm.h> #include <linux/mmzone.h> -#include <plat/control.h> /* ----------------------------------- DSP/BIOS Bridge */ #include <dspbridge/dbdefs.h> @@ -35,6 +34,10 @@ #include <dspbridge/drv.h> #include <dspbridge/sync.h> +/* ------------------------------------ Hardware Abstraction Layer */ +#include <hw_defs.h> +#include <hw_mmu.h> + /* ----------------------------------- Link Driver */ #include <dspbridge/dspdefs.h> #include <dspbridge/dspchnl.h> @@ -47,6 +50,7 @@ /* ----------------------------------- Platform Manager */ #include <dspbridge/dev.h> #include <dspbridge/dspapi.h> +#include <dspbridge/dmm.h> #include <dspbridge/wdt.h> /* ----------------------------------- Local */ @@ -67,6 +71,20 @@ #define MMU_SMALL_PAGE_MASK 0xFFFFF000 #define OMAP3_IVA2_BOOTADDR_MASK 0xFFFFFC00 #define PAGES_II_LVL_TABLE 512 +#define PHYS_TO_PAGE(phys) pfn_to_page((phys) >> PAGE_SHIFT) + +/* + * This is a totally ugly layer violation, but needed until + * omap_ctrl_set_dsp_boot*() are provided. + */ +#define OMAP3_IVA2_BOOTMOD_IDLE 1 +#define OMAP2_CONTROL_GENERAL 0x270 +#define OMAP343X_CONTROL_IVA2_BOOTADDR (OMAP2_CONTROL_GENERAL + 0x0190) +#define OMAP343X_CONTROL_IVA2_BOOTMOD (OMAP2_CONTROL_GENERAL + 0x0194) + +#define OMAP343X_CTRL_REGADDR(reg) \ + OMAP2_L4_IO_ADDRESS(OMAP343X_CTRL_BASE + (reg)) + /* Forward Declarations: */ static int bridge_brd_monitor(struct bridge_dev_context *dev_ctxt); @@ -91,6 +109,12 @@ static int bridge_brd_mem_copy(struct bridge_dev_context *dev_ctxt, static int bridge_brd_mem_write(struct bridge_dev_context *dev_ctxt, u8 *host_buff, u32 dsp_addr, u32 ul_num_bytes, u32 mem_type); +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); +static int bridge_brd_mem_un_map(struct bridge_dev_context *dev_ctxt, + u32 virt_addr, u32 ul_num_bytes); static int bridge_dev_create(struct bridge_dev_context **dev_cntxt, struct dev_object *hdev_obj, @@ -98,8 +122,57 @@ 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 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, + struct hw_mmu_map_attrs_t *map_attrs); +static int pte_set(struct pg_table_attrs *pt, u32 pa, u32 va, + u32 size, struct hw_mmu_map_attrs_t *attrs); +static int mem_map_vmalloc(struct bridge_dev_context *dev_context, + u32 ul_mpu_addr, u32 virt_addr, + u32 ul_num_bytes, + struct hw_mmu_map_attrs_t *hw_attrs); + bool wait_for_start(struct bridge_dev_context *dev_context, u32 dw_sync_addr); +/* ----------------------------------- Globals */ + +/* Attributes of L2 page tables for DSP MMU */ +struct page_info { + u32 num_entries; /* Number of valid PTEs in the L2 PT */ +}; + +/* Attributes used to manage the DSP MMU page tables */ +struct pg_table_attrs { + spinlock_t pg_lock; /* Critical section object handle */ + + u32 l1_base_pa; /* Physical address of the L1 PT */ + u32 l1_base_va; /* Virtual address of the L1 PT */ + u32 l1_size; /* Size of the L1 PT */ + u32 l1_tbl_alloc_pa; + /* Physical address of Allocated mem for L1 table. May not be aligned */ + u32 l1_tbl_alloc_va; + /* Virtual address of Allocated mem for L1 table. May not be aligned */ + u32 l1_tbl_alloc_sz; + /* Size of consistent memory allocated for L1 table. + * May not be aligned */ + + u32 l2_base_pa; /* Physical address of the L2 PT */ + u32 l2_base_va; /* Virtual address of the L2 PT */ + u32 l2_size; /* Size of the L2 PT */ + u32 l2_tbl_alloc_pa; + /* Physical address of Allocated mem for L2 table. May not be aligned */ + u32 l2_tbl_alloc_va; + /* Virtual address of Allocated mem for L2 table. May not be aligned */ + u32 l2_tbl_alloc_sz; + /* Size of consistent memory allocated for L2 table. + * May not be aligned */ + + u32 l2_num_pages; /* Number of allocated L2 PT */ + /* Array [l2_num_pages] of L2 PT info structs */ + struct page_info *pg_info; +}; + /* * This Bridge driver's function interface table. */ @@ -119,6 +192,8 @@ static struct bridge_drv_interface drv_interface_fxns = { bridge_brd_set_state, bridge_brd_mem_copy, bridge_brd_mem_write, + bridge_brd_mem_map, + bridge_brd_mem_un_map, /* The following CHNL functions are provided by chnl_io.lib: */ bridge_chnl_create, bridge_chnl_destroy, @@ -148,6 +223,27 @@ static struct bridge_drv_interface drv_interface_fxns = { bridge_msg_set_queue_id, }; +static inline void flush_all(struct bridge_dev_context *dev_context) +{ + if (dev_context->dw_brd_state == BRD_DSP_HIBERNATION || + dev_context->dw_brd_state == BRD_HIBERNATION) + wake_dsp(dev_context, NULL); + + hw_mmu_tlb_flush_all(dev_context->dw_dsp_mmu_base); +} + +static void bad_page_dump(u32 pa, struct page *pg) +{ + pr_emerg("DSPBRIDGE: MAP function: COUNT 0 FOR PA 0x%x\n", pa); + pr_emerg("Bad page state in process '%s'\n" + "page:%p flags:0x%0*lx mapping:%p mapcount:%d count:%d\n" + "Backtrace:\n", + current->comm, pg, (int)(2 * sizeof(unsigned long)), + (unsigned long)pg->flags, pg->mapping, + page_mapcount(pg), page_count(pg)); + dump_stack(); +} + /* * ======== bridge_drv_entry ======== * purpose: @@ -203,7 +299,8 @@ static int bridge_brd_monitor(struct bridge_dev_context *dev_ctxt) (*pdata->dsp_cm_write)(OMAP34XX_CLKSTCTRL_DISABLE_AUTO, OMAP3430_IVA2_MOD, OMAP2_CM_CLKSTCTRL); } - + (*pdata->dsp_prm_rmw_bits)(OMAP3430_RST2_IVA2_MASK, 0, + OMAP3430_IVA2_MOD, OMAP2_RM_RSTCTRL); dsp_clk_enable(DSP_CLK_IVA2); /* set the device state to IDLE */ @@ -274,17 +371,14 @@ 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 = NULL; - struct shm_segs *sm_sg; - int l4_i = 0, tlb_i = 0; - u32 sg0_da = 0, sg1_da = 0; - struct bridge_ioctl_extproc *tlb = dev_context->atlb_entry; 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 */ u32 ul_tlb_base_virt; /* Base of MMU TLB entry */ /* Offset of shm_base_virt from tlb_base_virt */ u32 ul_shm_offset_virt; + s32 entry_ndx; + s32 itmp_entry_ndx = 0; /* DSP-MMU TLB entry base address */ struct cfg_hostres *resources = NULL; u32 temp; u32 ul_dsp_clk_rate; @@ -305,12 +399,12 @@ static int bridge_brd_start(struct bridge_dev_context *dev_ctxt, ul_shm_base_virt *= DSPWORDSIZE; DBC_ASSERT(ul_shm_base_virt != 0); /* DSP Virtual address */ - ul_tlb_base_virt = dev_context->sh_s.seg0_da; + ul_tlb_base_virt = dev_context->atlb_entry[0].ul_dsp_va; DBC_ASSERT(ul_tlb_base_virt <= ul_shm_base_virt); ul_shm_offset_virt = ul_shm_base_virt - (ul_tlb_base_virt * DSPWORDSIZE); /* Kernel logical address */ - ul_shm_base = dev_context->sh_s.seg0_va + ul_shm_offset_virt; + ul_shm_base = dev_context->atlb_entry[0].ul_gpp_va + ul_shm_offset_virt; DBC_ASSERT(ul_shm_base != 0); /* 2nd wd is used as sync field */ @@ -345,83 +439,78 @@ static int bridge_brd_start(struct bridge_dev_context *dev_ctxt, OMAP343X_CONTROL_IVA2_BOOTMOD)); } } - 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); - mmu = dev_context->dsp_mmu; - if (mmu) - dsp_mmu_exit(mmu); - mmu = dsp_mmu_init(); - if (IS_ERR(mmu)) { - dev_err(bridge, "dsp_mmu_init failed!\n"); - dev_context->dsp_mmu = NULL; - status = (int)mmu; - } - } - if (!status) { - dev_context->dsp_mmu = mmu; - sm_sg = &dev_context->sh_s; - sg0_da = iommu_kmap(mmu, sm_sg->seg0_da, sm_sg->seg0_pa, - sm_sg->seg0_size, IOVMF_ENDIAN_LITTLE | IOVMF_ELSZ_32); - if (IS_ERR_VALUE(sg0_da)) { - status = (int)sg0_da; - sg0_da = 0; - } - } - if (!status) { - sg1_da = iommu_kmap(mmu, sm_sg->seg1_da, sm_sg->seg1_pa, - sm_sg->seg1_size, IOVMF_ENDIAN_LITTLE | IOVMF_ELSZ_32); - if (IS_ERR_VALUE(sg1_da)) { - status = (int)sg1_da; - sg1_da = 0; - } - } - if (!status) { - u32 da; - for (tlb_i = 0; tlb_i < BRDIOCTL_NUMOFMMUTLB; tlb_i++) { - if (!tlb[tlb_i].ul_gpp_pa) + 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; - dev_dbg(bridge, "IOMMU %d GppPa: 0x%x DspVa 0x%x Size" - " 0x%x\n", tlb_i, tlb[tlb_i].ul_gpp_pa, - tlb[tlb_i].ul_dsp_va, tlb[tlb_i].ul_size); - - da = iommu_kmap(mmu, tlb[tlb_i].ul_dsp_va, - tlb[tlb_i].ul_gpp_pa, PAGE_SIZE, - IOVMF_ENDIAN_LITTLE | IOVMF_ELSZ_32); - if (IS_ERR_VALUE(da)) { - status = (int)da; - break; - } - } - } - if (!status) { - u32 da; - l4_i = 0; - while (l4_peripheral_table[l4_i].phys_addr) { - da = iommu_kmap(mmu, l4_peripheral_table[l4_i]. - dsp_virt_addr, l4_peripheral_table[l4_i]. - phys_addr, PAGE_SIZE, - IOVMF_ENDIAN_LITTLE | IOVMF_ELSZ_32); - if (IS_ERR_VALUE(da)) { - status = (int)da; - break; - } - l4_i++; + dev_dbg(bridge, + "MMU %d, pa: 0x%x, va: 0x%x, size: 0x%x", + itmp_entry_ndx, + e->ul_gpp_pa, + e->ul_dsp_va, + e->ul_size); + + 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++; } } /* 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; @@ -430,7 +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; @@ -439,7 +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); @@ -492,6 +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); @@ -521,23 +617,11 @@ static int bridge_brd_start(struct bridge_dev_context *dev_ctxt, /* update board state */ dev_context->dw_brd_state = BRD_RUNNING; - return 0; + /* (void)chnlsm_enable_interrupt(dev_context); */ } else { dev_context->dw_brd_state = BRD_UNKNOWN; } } - - while (tlb_i--) { - if (!tlb[tlb_i].ul_gpp_pa) - continue; - iommu_kunmap(mmu, tlb[tlb_i].ul_gpp_va); - } - while (l4_i--) - iommu_kunmap(mmu, l4_peripheral_table[l4_i].dsp_virt_addr); - if (sg0_da) - iommu_kunmap(mmu, sg0_da); - if (sg1_da) - iommu_kunmap(mmu, sg1_da); return status; } @@ -553,9 +637,8 @@ static int bridge_brd_stop(struct bridge_dev_context *dev_ctxt) { int status = 0; struct bridge_dev_context *dev_context = dev_ctxt; + struct pg_table_attrs *pt_attrs; u32 dsp_pwr_state; - int i; - struct bridge_ioctl_extproc *tlb = dev_context->atlb_entry; struct omap_dsp_platform_data *pdata = omap_dspbridge_dev->dev.platform_data; @@ -591,37 +674,23 @@ static int bridge_brd_stop(struct bridge_dev_context *dev_ctxt) dsp_wdt_enable(false); - /* Reset DSP */ - (*pdata->dsp_prm_rmw_bits)(OMAP3430_RST1_IVA2_MASK, - OMAP3430_RST1_IVA2_MASK, OMAP3430_IVA2_MOD, OMAP2_RM_RSTCTRL); - + /* This is a good place to clear the MMU page tables as well */ + if (dev_context->pt_attrs) { + pt_attrs = dev_context->pt_attrs; + memset((u8 *) pt_attrs->l1_base_va, 0x00, pt_attrs->l1_size); + memset((u8 *) pt_attrs->l2_base_va, 0x00, pt_attrs->l2_size); + memset((u8 *) pt_attrs->pg_info, 0x00, + (pt_attrs->l2_num_pages * sizeof(struct page_info))); + } /* Disable the mailbox interrupts */ if (dev_context->mbox) { omap_mbox_disable_irq(dev_context->mbox, IRQ_RX); omap_mbox_put(dev_context->mbox); dev_context->mbox = NULL; } - if (dev_context->dsp_mmu) { - pr_err("Proc stop mmu if statement\n"); - for (i = 0; i < BRDIOCTL_NUMOFMMUTLB; i++) { - if (!tlb[i].ul_gpp_pa) - continue; - iommu_kunmap(dev_context->dsp_mmu, tlb[i].ul_gpp_va); - } - i = 0; - while (l4_peripheral_table[i].phys_addr) { - iommu_kunmap(dev_context->dsp_mmu, - l4_peripheral_table[i].dsp_virt_addr); - i++; - } - iommu_kunmap(dev_context->dsp_mmu, dev_context->sh_s.seg0_da); - iommu_kunmap(dev_context->dsp_mmu, dev_context->sh_s.seg1_da); - dsp_mmu_exit(dev_context->dsp_mmu); - dev_context->dsp_mmu = NULL; - } - /* Reset IVA IOMMU*/ - (*pdata->dsp_prm_rmw_bits)(OMAP3430_RST2_IVA2_MASK, - OMAP3430_RST2_IVA2_MASK, OMAP3430_IVA2_MOD, OMAP2_RM_RSTCTRL); + /* 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); dsp_clock_disable_all(dev_context->dsp_per_clks); dsp_clk_disable(DSP_CLK_IVA2); @@ -681,6 +750,10 @@ static int bridge_dev_create(struct bridge_dev_context struct bridge_dev_context *dev_context = NULL; s32 entry_ndx; struct cfg_hostres *resources = config_param; + struct pg_table_attrs *pt_attrs; + u32 pg_tbl_pa; + u32 pg_tbl_va; + u32 align_size; struct drv_data *drv_datap = dev_get_drvdata(bridge); /* Allocate and initialize a data structure to contain the bridge driver @@ -711,8 +784,97 @@ static int bridge_dev_create(struct bridge_dev_context if (!dev_context->dw_dsp_base_addr) status = -EPERM; + pt_attrs = kzalloc(sizeof(struct pg_table_attrs), GFP_KERNEL); + if (pt_attrs != NULL) { + /* Assuming that we use only DSP's memory map + * until 0x4000:0000 , we would need only 1024 + * L1 enties i.e L1 size = 4K */ + pt_attrs->l1_size = 0x1000; + align_size = pt_attrs->l1_size; + /* Align sizes are expected to be power of 2 */ + /* we like to get aligned on L1 table size */ + pg_tbl_va = (u32) mem_alloc_phys_mem(pt_attrs->l1_size, + align_size, &pg_tbl_pa); + + /* Check if the PA is aligned for us */ + if ((pg_tbl_pa) & (align_size - 1)) { + /* PA not aligned to page table size , + * try with more allocation and align */ + mem_free_phys_mem((void *)pg_tbl_va, pg_tbl_pa, + pt_attrs->l1_size); + /* we like to get aligned on L1 table size */ + pg_tbl_va = + (u32) mem_alloc_phys_mem((pt_attrs->l1_size) * 2, + align_size, &pg_tbl_pa); + /* We should be able to get aligned table now */ + pt_attrs->l1_tbl_alloc_pa = pg_tbl_pa; + pt_attrs->l1_tbl_alloc_va = pg_tbl_va; + pt_attrs->l1_tbl_alloc_sz = pt_attrs->l1_size * 2; + /* Align the PA to the next 'align' boundary */ + pt_attrs->l1_base_pa = + ((pg_tbl_pa) + + (align_size - 1)) & (~(align_size - 1)); + pt_attrs->l1_base_va = + pg_tbl_va + (pt_attrs->l1_base_pa - pg_tbl_pa); + } else { + /* We got aligned PA, cool */ + pt_attrs->l1_tbl_alloc_pa = pg_tbl_pa; + pt_attrs->l1_tbl_alloc_va = pg_tbl_va; + pt_attrs->l1_tbl_alloc_sz = pt_attrs->l1_size; + pt_attrs->l1_base_pa = pg_tbl_pa; + pt_attrs->l1_base_va = pg_tbl_va; + } + if (pt_attrs->l1_base_va) + memset((u8 *) pt_attrs->l1_base_va, 0x00, + pt_attrs->l1_size); + + /* number of L2 page tables = DMM pool used + SHMMEM +EXTMEM + + * L4 pages */ + pt_attrs->l2_num_pages = ((DMMPOOLSIZE >> 20) + 6); + pt_attrs->l2_size = HW_MMU_COARSE_PAGE_SIZE * + pt_attrs->l2_num_pages; + align_size = 4; /* Make it u32 aligned */ + /* we like to get aligned on L1 table size */ + pg_tbl_va = (u32) mem_alloc_phys_mem(pt_attrs->l2_size, + align_size, &pg_tbl_pa); + pt_attrs->l2_tbl_alloc_pa = pg_tbl_pa; + pt_attrs->l2_tbl_alloc_va = pg_tbl_va; + pt_attrs->l2_tbl_alloc_sz = pt_attrs->l2_size; + pt_attrs->l2_base_pa = pg_tbl_pa; + pt_attrs->l2_base_va = pg_tbl_va; + + if (pt_attrs->l2_base_va) + memset((u8 *) pt_attrs->l2_base_va, 0x00, + pt_attrs->l2_size); + + pt_attrs->pg_info = kzalloc(pt_attrs->l2_num_pages * + sizeof(struct page_info), GFP_KERNEL); + dev_dbg(bridge, + "L1 pa %x, va %x, size %x\n L2 pa %x, va " + "%x, size %x\n", pt_attrs->l1_base_pa, + pt_attrs->l1_base_va, pt_attrs->l1_size, + pt_attrs->l2_base_pa, pt_attrs->l2_base_va, + pt_attrs->l2_size); + dev_dbg(bridge, "pt_attrs %p L2 NumPages %x pg_info %p\n", + pt_attrs, pt_attrs->l2_num_pages, pt_attrs->pg_info); + } + if ((pt_attrs != NULL) && (pt_attrs->l1_base_va != 0) && + (pt_attrs->l2_base_va != 0) && (pt_attrs->pg_info != NULL)) + dev_context->pt_attrs = pt_attrs; + else + status = -ENOMEM; + if (!status) { + spin_lock_init(&pt_attrs->pg_lock); dev_context->tc_word_swap_on = drv_datap->tc_wordswapon; + + /* Set the Clock Divisor for the DSP module */ + udelay(5); + /* MMU address is obtained from the host + * resources struct */ + dev_context->dw_dsp_mmu_base = resources->dw_dmmu_base; + } + if (!status) { dev_context->hdev_obj = hdev_obj; /* Store current board state. */ dev_context->dw_brd_state = BRD_UNKNOWN; @@ -722,6 +884,23 @@ static int bridge_dev_create(struct bridge_dev_context /* Return ptr to our device state to the DSP API for storage */ *dev_cntxt = dev_context; } else { + if (pt_attrs != NULL) { + kfree(pt_attrs->pg_info); + + if (pt_attrs->l2_tbl_alloc_va) { + mem_free_phys_mem((void *) + pt_attrs->l2_tbl_alloc_va, + pt_attrs->l2_tbl_alloc_pa, + pt_attrs->l2_tbl_alloc_sz); + } + if (pt_attrs->l1_tbl_alloc_va) { + mem_free_phys_mem((void *) + pt_attrs->l1_tbl_alloc_va, + pt_attrs->l1_tbl_alloc_pa, + pt_attrs->l1_tbl_alloc_sz); + } + } + kfree(pt_attrs); kfree(dev_context); } func_end: @@ -789,6 +968,7 @@ static int bridge_dev_ctrl(struct bridge_dev_context *dev_context, */ static int bridge_dev_destroy(struct bridge_dev_context *dev_ctxt) { + struct pg_table_attrs *pt_attrs; int status = 0; struct bridge_dev_context *dev_context = (struct bridge_dev_context *) dev_ctxt; @@ -802,6 +982,23 @@ static int bridge_dev_destroy(struct bridge_dev_context *dev_ctxt) /* first put the device to stop state */ bridge_brd_stop(dev_context); + if (dev_context->pt_attrs) { + pt_attrs = dev_context->pt_attrs; + kfree(pt_attrs->pg_info); + + if (pt_attrs->l2_tbl_alloc_va) { + mem_free_phys_mem((void *)pt_attrs->l2_tbl_alloc_va, + pt_attrs->l2_tbl_alloc_pa, + pt_attrs->l2_tbl_alloc_sz); + } + if (pt_attrs->l1_tbl_alloc_va) { + mem_free_phys_mem((void *)pt_attrs->l1_tbl_alloc_va, + pt_attrs->l1_tbl_alloc_pa, + pt_attrs->l1_tbl_alloc_sz); + } + kfree(pt_attrs); + + } if (dev_context->resources) { host_res = dev_context->resources; @@ -832,6 +1029,8 @@ static int bridge_dev_destroy(struct bridge_dev_context *dev_ctxt) iounmap((void *)host_res->dw_mem_base[3]); if (host_res->dw_mem_base[4]) iounmap((void *)host_res->dw_mem_base[4]); + if (host_res->dw_dmmu_base) + iounmap(host_res->dw_dmmu_base); if (host_res->dw_per_base) iounmap(host_res->dw_per_base); if (host_res->dw_per_pm_base) @@ -845,6 +1044,7 @@ static int bridge_dev_destroy(struct bridge_dev_context *dev_ctxt) host_res->dw_mem_base[2] = (u32) NULL; host_res->dw_mem_base[3] = (u32) NULL; host_res->dw_mem_base[4] = (u32) NULL; + host_res->dw_dmmu_base = NULL; host_res->dw_sys_ctrl_base = NULL; kfree(host_res); @@ -928,6 +1128,673 @@ static int bridge_brd_mem_write(struct bridge_dev_context *dev_ctxt, } /* + * ======== bridge_brd_mem_map ======== + * This function maps MPU buffer to the DSP address space. It performs + * linear to physical address translation if required. It translates each + * page since linear addresses can be physically non-contiguous + * All address & size arguments are assumed to be page aligned (in proc.c) + * + * TODO: Disable MMU while updating the page tables (but that'll stall DSP) + */ +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) +{ + 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; + 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; + + 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, 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__, ul_mpu_addr, ul_num_bytes); + status = -EINVAL; + up_read(&mm->mmap_sem); + goto func_cont; + } + + 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; + + 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; + } + } + } + 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; + } + /* + * 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; +} + +/* + * ======== bridge_brd_mem_un_map ======== + * Invalidate the PTEs for the DSP VA block to be unmapped. + * + * PTEs of a mapped memory block are contiguous in any page table + * 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_ctxt, + u32 virt_addr, u32 ul_num_bytes) +{ + 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; + + 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); + + if (pte_size != HW_MMU_COARSE_PAGE_SIZE) + goto skip_coarse_page; + + /* + * 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; + } + + /* 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; + } + + 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; + + 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; + } + + 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; + } + } + /* + * 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; +} + +/* + * ======== user_va2_pa ======== + * Purpose: + * This function walks through the page tables to convert a userland + * virtual address to physical address + */ +static u32 user_va2_pa(struct mm_struct *mm, u32 address) +{ + pgd_t *pgd; + pmd_t *pmd; + pte_t *ptep, pte; + + pgd = pgd_offset(mm, address); + if (!(pgd_none(*pgd) || pgd_bad(*pgd))) { + pmd = pmd_offset(pgd, address); + if (!(pmd_none(*pmd) || pmd_bad(*pmd))) { + ptep = pte_offset_map(pmd, address); + if (ptep) { + pte = *ptep; + if (pte_present(pte)) + return pte & PAGE_MASK; + } + } + } + + return 0; +} + +/* + * ======== pte_update ======== + * This function calculates the optimum page-aligned addresses and sizes + * Caller must pass page-aligned values + */ +static int pte_update(struct bridge_dev_context *dev_ctxt, u32 pa, + u32 va, u32 size, + struct hw_mmu_map_attrs_t *map_attrs) +{ + u32 i; + u32 all_bits; + u32 pa_curr = pa; + u32 va_curr = va; + u32 num_bytes = size; + struct bridge_dev_context *dev_context = dev_ctxt; + int status = 0; + u32 page_size[] = { HW_PAGE_SIZE16MB, HW_PAGE_SIZE1MB, + HW_PAGE_SIZE64KB, HW_PAGE_SIZE4KB + }; + + while (num_bytes && !status) { + /* To find the max. page size with which both PA & VA are + * aligned */ + all_bits = pa_curr | va_curr; + + for (i = 0; i < 4; i++) { + if ((num_bytes >= page_size[i]) && ((all_bits & + (page_size[i] - + 1)) == 0)) { + status = + pte_set(dev_context->pt_attrs, pa_curr, + va_curr, page_size[i], map_attrs); + pa_curr += page_size[i]; + 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; + } + } + } + + return status; +} + +/* + * ======== pte_set ======== + * This function calculates PTE address (MPU virtual) to be updated + * It also manages the L2 page tables + */ +static int pte_set(struct pg_table_attrs *pt, u32 pa, u32 va, + u32 size, struct hw_mmu_map_attrs_t *attrs) +{ + u32 i; + u32 pte_val; + u32 pte_addr_l1; + u32 pte_size; + /* Base address of the PT that will be updated */ + u32 pg_tbl_va; + u32 l1_base_va; + /* Compiler warns that the next three variables might be used + * uninitialized in this function. Doesn't seem so. Working around, + * anyways. */ + u32 l2_base_va = 0; + u32 l2_base_pa = 0; + u32 l2_page_num = 0; + int status = 0; + + l1_base_va = pt->l1_base_va; + pg_tbl_va = l1_base_va; + if ((size == HW_PAGE_SIZE64KB) || (size == HW_PAGE_SIZE4KB)) { + /* Find whether the L1 PTE points to a valid L2 PT */ + pte_addr_l1 = hw_mmu_pte_addr_l1(l1_base_va, va); + if (pte_addr_l1 <= (pt->l1_base_va + pt->l1_size)) { + pte_val = *(u32 *) pte_addr_l1; + pte_size = hw_mmu_pte_size_l1(pte_val); + } else { + return -EPERM; + } + spin_lock(&pt->pg_lock); + if (pte_size == HW_MMU_COARSE_PAGE_SIZE) { + /* 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; + } else if (pte_size == 0) { + /* L1 PTE is invalid. Allocate a L2 PT and + * point the L1 PTE to it */ + /* Find a free L2 PT. */ + for (i = 0; (i < pt->l2_num_pages) && + (pt->pg_info[i].num_entries != 0); i++) + ;; + if (i < pt->l2_num_pages) { + l2_page_num = i; + l2_base_pa = pt->l2_base_pa + (l2_page_num * + HW_MMU_COARSE_PAGE_SIZE); + l2_base_va = pt->l2_base_va + (l2_page_num * + HW_MMU_COARSE_PAGE_SIZE); + /* Endianness attributes are ignored for + * HW_MMU_COARSE_PAGE_SIZE */ + status = + hw_mmu_pte_set(l1_base_va, l2_base_pa, va, + HW_MMU_COARSE_PAGE_SIZE, + attrs); + } else { + status = -ENOMEM; + } + } else { + /* Found valid L1 PTE of another size. + * Should not overwrite it. */ + status = -EPERM; + } + if (!status) { + pg_tbl_va = l2_base_va; + if (size == HW_PAGE_SIZE64KB) + pt->pg_info[l2_page_num].num_entries += 16; + else + pt->pg_info[l2_page_num].num_entries++; + dev_dbg(bridge, "PTE: L2 BaseVa %x, BasePa %x, PageNum " + "%x, num_entries %x\n", l2_base_va, + l2_base_pa, l2_page_num, + pt->pg_info[l2_page_num].num_entries); + } + spin_unlock(&pt->pg_lock); + } + if (!status) { + dev_dbg(bridge, "PTE: pg_tbl_va %x, pa %x, va %x, size %x\n", + pg_tbl_va, pa, va, size); + dev_dbg(bridge, "PTE: endianism %x, element_size %x, " + "mixed_size %x\n", attrs->endianism, + attrs->element_size, attrs->mixed_size); + status = hw_mmu_pte_set(pg_tbl_va, pa, va, size, attrs); + } + + return status; +} + +/* Memory map kernel VA -- memory allocated with vmalloc */ +static int mem_map_vmalloc(struct bridge_dev_context *dev_context, + u32 ul_mpu_addr, u32 virt_addr, + u32 ul_num_bytes, + struct hw_mmu_map_attrs_t *hw_attrs) +{ + int status = 0; + struct page *page[1]; + u32 i; + u32 pa_curr; + u32 pa_next; + u32 va_curr; + u32 size_curr; + u32 num_pages; + u32 pa; + u32 num_of4k_pages; + u32 temp = 0; + + /* + * Do Kernel va to pa translation. + * Combine physically contiguous regions to reduce TLBs. + * Pass the translated pa to pte_update. + */ + num_pages = ul_num_bytes / PAGE_SIZE; /* PAGE_SIZE = OS page size */ + i = 0; + va_curr = ul_mpu_addr; + page[0] = vmalloc_to_page((void *)va_curr); + pa_next = page_to_phys(page[0]); + while (!status && (i < num_pages)) { + /* + * Reuse pa_next from the previous iteraion to avoid + * an extra va2pa call + */ + pa_curr = pa_next; + size_curr = PAGE_SIZE; + /* + * If the next page is physically contiguous, + * map it with the current one by increasing + * the size of the region to be mapped + */ + while (++i < num_pages) { + page[0] = + vmalloc_to_page((void *)(va_curr + size_curr)); + pa_next = page_to_phys(page[0]); + + if (pa_next == (pa_curr + size_curr)) + size_curr += PAGE_SIZE; + else + break; + + } + if (pa_next == 0) { + status = -ENOMEM; + break; + } + pa = pa_curr; + num_of4k_pages = size_curr / HW_PAGE_SIZE4KB; + while (temp++ < num_of4k_pages) { + get_page(PHYS_TO_PAGE(pa)); + pa += HW_PAGE_SIZE4KB; + } + status = pte_update(dev_context, pa_curr, virt_addr + + (va_curr - ul_mpu_addr), size_curr, + hw_attrs); + va_curr += size_curr; + } + /* + * 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; +} + +/* * ======== wait_for_start ======== * Wait for the singal from DSP that it has started, or time out. */ diff --git a/drivers/staging/tidspbridge/core/tiomap3430_pwr.c b/drivers/staging/tidspbridge/core/tiomap3430_pwr.c index b57a9fd5e75..fb9026e1403 100644 --- a/drivers/staging/tidspbridge/core/tiomap3430_pwr.c +++ b/drivers/staging/tidspbridge/core/tiomap3430_pwr.c @@ -31,6 +31,10 @@ #include <dspbridge/dev.h> #include <dspbridge/iodefs.h> +/* ------------------------------------ Hardware Abstraction Layer */ +#include <hw_defs.h> +#include <hw_mmu.h> + #include <dspbridge/pwr_sh.h> /* ----------------------------------- Bridge Driver */ diff --git a/drivers/staging/tidspbridge/core/tiomap_io.c b/drivers/staging/tidspbridge/core/tiomap_io.c index 66dbf02549e..ba2961049da 100644 --- a/drivers/staging/tidspbridge/core/tiomap_io.c +++ b/drivers/staging/tidspbridge/core/tiomap_io.c @@ -134,16 +134,17 @@ int read_ext_dsp_data(struct bridge_dev_context *dev_ctxt, if (!status) { ul_tlb_base_virt = - dev_context->sh_s.seg0_da * DSPWORDSIZE; + dev_context->atlb_entry[0].ul_dsp_va * DSPWORDSIZE; DBC_ASSERT(ul_tlb_base_virt <= ul_shm_base_virt); - dw_ext_prog_virt_mem = dev_context->sh_s.seg0_va; + dw_ext_prog_virt_mem = + dev_context->atlb_entry[0].ul_gpp_va; if (!trace_read) { ul_shm_offset_virt = ul_shm_base_virt - ul_tlb_base_virt; ul_shm_offset_virt += PG_ALIGN_HIGH(ul_ext_end - ul_dyn_ext_base + - 1, PAGE_SIZE * 16); + 1, HW_PAGE_SIZE64KB); dw_ext_prog_virt_mem -= ul_shm_offset_virt; dw_ext_prog_virt_mem += (ul_ext_base - ul_dyn_ext_base); @@ -317,9 +318,8 @@ int write_ext_dsp_data(struct bridge_dev_context *dev_context, ret = -EPERM; if (!ret) { - ul_tlb_base_virt = dev_context->sh_s.seg0_da * - DSPWORDSIZE; - + ul_tlb_base_virt = + dev_context->atlb_entry[0].ul_dsp_va * DSPWORDSIZE; DBC_ASSERT(ul_tlb_base_virt <= ul_shm_base_virt); if (symbols_reloaded) { @@ -337,7 +337,7 @@ int write_ext_dsp_data(struct bridge_dev_context *dev_context, ul_shm_base_virt - ul_tlb_base_virt; if (trace_load) { dw_ext_prog_virt_mem = - dev_context->sh_s.seg0_va; + dev_context->atlb_entry[0].ul_gpp_va; } else { dw_ext_prog_virt_mem = host_res->dw_mem_base[1]; dw_ext_prog_virt_mem += @@ -393,6 +393,7 @@ int sm_interrupt_dsp(struct bridge_dev_context *dev_context, u16 mb_val) omap_dspbridge_dev->dev.platform_data; struct cfg_hostres *resources = dev_context->resources; int status = 0; + u32 temp; if (!dev_context->mbox) return 0; @@ -436,7 +437,7 @@ int sm_interrupt_dsp(struct bridge_dev_context *dev_context, u16 mb_val) omap_mbox_restore_ctx(dev_context->mbox); /* Access MMU SYS CONFIG register to generate a short wakeup */ - iommu_read_reg(dev_context->dsp_mmu, MMU_SYSCONFIG); + temp = readl(resources->dw_dmmu_base + 0x10); dev_context->dw_brd_state = BRD_RUNNING; } else if (dev_context->dw_brd_state == BRD_RETENTION) { diff --git a/drivers/staging/tidspbridge/core/ue_deh.c b/drivers/staging/tidspbridge/core/ue_deh.c index e24ea0c7391..3430418190d 100644 --- a/drivers/staging/tidspbridge/core/ue_deh.c +++ b/drivers/staging/tidspbridge/core/ue_deh.c @@ -31,6 +31,57 @@ #include <dspbridge/drv.h> #include <dspbridge/wdt.h> +static u32 fault_addr; + +static void mmu_fault_dpc(unsigned long data) +{ + struct deh_mgr *deh = (void *)data; + + if (!deh) + return; + + bridge_deh_notify(deh, DSP_MMUFAULT, 0); +} + +static irqreturn_t mmu_fault_isr(int irq, void *data) +{ + struct deh_mgr *deh = data; + struct cfg_hostres *resources; + u32 event; + + if (!deh) + return IRQ_HANDLED; + + resources = deh->hbridge_context->resources; + if (!resources) { + dev_dbg(bridge, "%s: Failed to get Host Resources\n", + __func__); + return IRQ_HANDLED; + } + + hw_mmu_event_status(resources->dw_dmmu_base, &event); + if (event == HW_MMU_TRANSLATION_FAULT) { + hw_mmu_fault_addr_read(resources->dw_dmmu_base, &fault_addr); + dev_dbg(bridge, "%s: event=0x%x, fault_addr=0x%x\n", __func__, + event, fault_addr); + /* + * Schedule a DPC directly. In the future, it may be + * necessary to check if DSP MMU fault is intended for + * Bridge. + */ + tasklet_schedule(&deh->dpc_tasklet); + + /* Disable the MMU events, else once we clear it will + * start to raise INTs again */ + hw_mmu_event_disable(resources->dw_dmmu_base, + HW_MMU_TRANSLATION_FAULT); + } else { + hw_mmu_event_disable(resources->dw_dmmu_base, + HW_MMU_ALL_INTERRUPTS); + } + return IRQ_HANDLED; +} + int bridge_deh_create(struct deh_mgr **ret_deh, struct dev_object *hdev_obj) { @@ -58,9 +109,18 @@ int bridge_deh_create(struct deh_mgr **ret_deh, } ntfy_init(deh->ntfy_obj); + /* Create a MMUfault DPC */ + tasklet_init(&deh->dpc_tasklet, mmu_fault_dpc, (u32) 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; @@ -80,6 +140,11 @@ 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); /* Deallocate the DEH manager object */ kfree(deh); @@ -101,6 +166,48 @@ int bridge_deh_register_notify(struct deh_mgr *deh, u32 event_mask, return ntfy_unregister(deh->ntfy_obj, hnotification); } +#ifdef CONFIG_TIDSPBRIDGE_BACKTRACE +static void mmu_fault_print_stack(struct bridge_dev_context *dev_context) +{ + struct cfg_hostres *resources; + struct hw_mmu_map_attrs_t map_attrs = { + .endianism = HW_LITTLE_ENDIAN, + .element_size = HW_ELEM_SIZE16BIT, + .mixed_size = HW_MMU_CPUES, + }; + void *dummy_va_addr; + + resources = dev_context->resources; + dummy_va_addr = (void*)__get_free_page(GFP_ATOMIC); + + /* + * Before acking the MMU fault, let's make sure MMU can only + * access entry #0. Then add a new entry so that the DSP OS + * can continue in order to dump the stack. + */ + hw_mmu_twl_disable(resources->dw_dmmu_base); + hw_mmu_tlb_flush_all(resources->dw_dmmu_base); + + hw_mmu_tlb_add(resources->dw_dmmu_base, + virt_to_phys(dummy_va_addr), fault_addr, + HW_PAGE_SIZE4KB, 1, + &map_attrs, HW_SET, HW_SET); + + dsp_clk_enable(DSP_CLK_GPT8); + + dsp_gpt_wait_overflow(DSP_CLK_GPT8, 0xfffffffe); + + /* Clear MMU interrupt */ + hw_mmu_event_ack(resources->dw_dmmu_base, + HW_MMU_TRANSLATION_FAULT); + dump_dsp_stack(dev_context); + dsp_clk_disable(DSP_CLK_GPT8); + + hw_mmu_disable(resources->dw_dmmu_base); + free_page((unsigned long)dummy_va_addr); +} +#endif + static inline const char *event_to_string(int event) { switch (event) { @@ -133,7 +240,13 @@ void bridge_deh_notify(struct deh_mgr *deh, int event, int info) #endif break; case DSP_MMUFAULT: - dev_err(bridge, "%s: %s, addr=0x%x", __func__, str, info); + dev_err(bridge, "%s: %s, addr=0x%x", __func__, + str, fault_addr); +#ifdef CONFIG_TIDSPBRIDGE_BACKTRACE + print_dsp_trace_buffer(dev_context); + dump_dl_modules(dev_context); + mmu_fault_print_stack(dev_context); +#endif break; default: dev_err(bridge, "%s: %s", __func__, str); diff --git a/drivers/staging/tidspbridge/hw/EasiGlobal.h b/drivers/staging/tidspbridge/hw/EasiGlobal.h new file mode 100644 index 00000000000..e48d7f67c60 --- /dev/null +++ b/drivers/staging/tidspbridge/hw/EasiGlobal.h @@ -0,0 +1,41 @@ +/* + * EasiGlobal.h + * + * DSP-BIOS Bridge driver support functions for TI OMAP processors. + * + * Copyright (C) 2007 Texas Instruments, Inc. + * + * This package is free software; you can redistribute it and/or modify + * it under the terms of the GNU General Public License version 2 as + * published by the Free Software Foundation. + * + * THIS PACKAGE IS PROVIDED ``AS IS'' AND WITHOUT ANY EXPRESS OR + * IMPLIED WARRANTIES, INCLUDING, WITHOUT LIMITATION, THE IMPLIED + * WARRANTIES OF MERCHANTIBILITY AND FITNESS FOR A PARTICULAR PURPOSE. + */ + +#ifndef _EASIGLOBAL_H +#define _EASIGLOBAL_H +#include <linux/types.h> + +/* + * DEFINE: READ_ONLY, WRITE_ONLY & READ_WRITE + * + * DESCRIPTION: Defines used to describe register types for EASI-checker tests. + */ + +#define READ_ONLY 1 +#define WRITE_ONLY 2 +#define READ_WRITE 3 + +/* + * MACRO: _DEBUG_LEVEL1_EASI + * + * DESCRIPTION: A MACRO which can be used to indicate that a particular beach + * register access function was called. + * + * NOTE: We currently dont use this functionality. + */ +#define _DEBUG_LEVEL1_EASI(easi_num) ((void)0) + +#endif /* _EASIGLOBAL_H */ diff --git a/drivers/staging/tidspbridge/hw/MMUAccInt.h b/drivers/staging/tidspbridge/hw/MMUAccInt.h new file mode 100644 index 00000000000..1cefca321d7 --- /dev/null +++ b/drivers/staging/tidspbridge/hw/MMUAccInt.h @@ -0,0 +1,76 @@ +/* + * MMUAccInt.h + * + * DSP-BIOS Bridge driver support functions for TI OMAP processors. + * + * Copyright (C) 2007 Texas Instruments, Inc. + * + * This package is free software; you can redistribute it and/or modify + * it under the terms of the GNU General Public License version 2 as + * published by the Free Software Foundation. + * + * THIS PACKAGE IS PROVIDED ``AS IS'' AND WITHOUT ANY EXPRESS OR + * IMPLIED WARRANTIES, INCLUDING, WITHOUT LIMITATION, THE IMPLIED + * WARRANTIES OF MERCHANTIBILITY AND FITNESS FOR A PARTICULAR PURPOSE. + */ + +#ifndef _MMU_ACC_INT_H +#define _MMU_ACC_INT_H + +/* Mappings of level 1 EASI function numbers to function names */ + +#define EASIL1_MMUMMU_SYSCONFIG_READ_REGISTER32 (MMU_BASE_EASIL1 + 3) +#define EASIL1_MMUMMU_SYSCONFIG_IDLE_MODE_WRITE32 (MMU_BASE_EASIL1 + 17) +#define EASIL1_MMUMMU_SYSCONFIG_AUTO_IDLE_WRITE32 (MMU_BASE_EASIL1 + 39) +#define EASIL1_MMUMMU_IRQSTATUS_WRITE_REGISTER32 (MMU_BASE_EASIL1 + 51) +#define EASIL1_MMUMMU_IRQENABLE_READ_REGISTER32 (MMU_BASE_EASIL1 + 102) +#define EASIL1_MMUMMU_IRQENABLE_WRITE_REGISTER32 (MMU_BASE_EASIL1 + 103) +#define EASIL1_MMUMMU_WALKING_STTWL_RUNNING_READ32 (MMU_BASE_EASIL1 + 156) +#define EASIL1_MMUMMU_CNTLTWL_ENABLE_READ32 (MMU_BASE_EASIL1 + 174) +#define EASIL1_MMUMMU_CNTLTWL_ENABLE_WRITE32 (MMU_BASE_EASIL1 + 180) +#define EASIL1_MMUMMU_CNTLMMU_ENABLE_WRITE32 (MMU_BASE_EASIL1 + 190) +#define EASIL1_MMUMMU_FAULT_AD_READ_REGISTER32 (MMU_BASE_EASIL1 + 194) +#define EASIL1_MMUMMU_TTB_WRITE_REGISTER32 (MMU_BASE_EASIL1 + 198) +#define EASIL1_MMUMMU_LOCK_READ_REGISTER32 (MMU_BASE_EASIL1 + 203) +#define EASIL1_MMUMMU_LOCK_WRITE_REGISTER32 (MMU_BASE_EASIL1 + 204) +#define EASIL1_MMUMMU_LOCK_BASE_VALUE_READ32 (MMU_BASE_EASIL1 + 205) +#define EASIL1_MMUMMU_LOCK_CURRENT_VICTIM_READ32 (MMU_BASE_EASIL1 + 209) +#define EASIL1_MMUMMU_LOCK_CURRENT_VICTIM_WRITE32 (MMU_BASE_EASIL1 + 211) +#define EASIL1_MMUMMU_LOCK_CURRENT_VICTIM_SET32 (MMU_BASE_EASIL1 + 212) +#define EASIL1_MMUMMU_LD_TLB_READ_REGISTER32 (MMU_BASE_EASIL1 + 213) +#define EASIL1_MMUMMU_LD_TLB_WRITE_REGISTER32 (MMU_BASE_EASIL1 + 214) +#define EASIL1_MMUMMU_CAM_WRITE_REGISTER32 (MMU_BASE_EASIL1 + 226) +#define EASIL1_MMUMMU_RAM_WRITE_REGISTER32 (MMU_BASE_EASIL1 + 268) +#define EASIL1_MMUMMU_FLUSH_ENTRY_WRITE_REGISTER32 (MMU_BASE_EASIL1 + 322) + +/* Register offset address definitions */ +#define MMU_MMU_SYSCONFIG_OFFSET 0x10 +#define MMU_MMU_IRQSTATUS_OFFSET 0x18 +#define MMU_MMU_IRQENABLE_OFFSET 0x1c +#define MMU_MMU_WALKING_ST_OFFSET 0x40 +#define MMU_MMU_CNTL_OFFSET 0x44 +#define MMU_MMU_FAULT_AD_OFFSET 0x48 +#define MMU_MMU_TTB_OFFSET 0x4c +#define MMU_MMU_LOCK_OFFSET 0x50 +#define MMU_MMU_LD_TLB_OFFSET 0x54 +#define MMU_MMU_CAM_OFFSET 0x58 +#define MMU_MMU_RAM_OFFSET 0x5c +#define MMU_MMU_GFLUSH_OFFSET 0x60 +#define MMU_MMU_FLUSH_ENTRY_OFFSET 0x64 +/* Bitfield mask and offset declarations */ +#define MMU_MMU_SYSCONFIG_IDLE_MODE_MASK 0x18 +#define MMU_MMU_SYSCONFIG_IDLE_MODE_OFFSET 3 +#define MMU_MMU_SYSCONFIG_AUTO_IDLE_MASK 0x1 +#define MMU_MMU_SYSCONFIG_AUTO_IDLE_OFFSET 0 +#define MMU_MMU_WALKING_ST_TWL_RUNNING_MASK 0x1 +#define MMU_MMU_WALKING_ST_TWL_RUNNING_OFFSET 0 +#define MMU_MMU_CNTL_TWL_ENABLE_MASK 0x4 +#define MMU_MMU_CNTL_TWL_ENABLE_OFFSET 2 +#define MMU_MMU_CNTL_MMU_ENABLE_MASK 0x2 +#define MMU_MMU_CNTL_MMU_ENABLE_OFFSET 1 +#define MMU_MMU_LOCK_BASE_VALUE_MASK 0xfc00 +#define MMU_MMU_LOCK_BASE_VALUE_OFFSET 10 +#define MMU_MMU_LOCK_CURRENT_VICTIM_MASK 0x3f0 +#define MMU_MMU_LOCK_CURRENT_VICTIM_OFFSET 4 + +#endif /* _MMU_ACC_INT_H */ diff --git a/drivers/staging/tidspbridge/hw/MMURegAcM.h b/drivers/staging/tidspbridge/hw/MMURegAcM.h new file mode 100644 index 00000000000..ab1a16da731 --- /dev/null +++ b/drivers/staging/tidspbridge/hw/MMURegAcM.h @@ -0,0 +1,225 @@ +/* + * MMURegAcM.h + * + * DSP-BIOS Bridge driver support functions for TI OMAP processors. + * + * Copyright (C) 2007 Texas Instruments, Inc. + * + * This package is free software; you can redistribute it and/or modify + * it under the terms of the GNU General Public License version 2 as + * published by the Free Software Foundation. + * + * THIS PACKAGE IS PROVIDED ``AS IS'' AND WITHOUT ANY EXPRESS OR + * IMPLIED WARRANTIES, INCLUDING, WITHOUT LIMITATION, THE IMPLIED + * WARRANTIES OF MERCHANTIBILITY AND FITNESS FOR A PARTICULAR PURPOSE. + */ + +#ifndef _MMU_REG_ACM_H +#define _MMU_REG_ACM_H + +#include <linux/io.h> +#include <EasiGlobal.h> + +#include "MMUAccInt.h" + +#if defined(USE_LEVEL_1_MACROS) + +#define MMUMMU_SYSCONFIG_READ_REGISTER32(base_address)\ + (_DEBUG_LEVEL1_EASI(EASIL1_MMUMMU_SYSCONFIG_READ_REGISTER32),\ + __raw_readl((base_address)+MMU_MMU_SYSCONFIG_OFFSET)) + +#define MMUMMU_SYSCONFIG_IDLE_MODE_WRITE32(base_address, value)\ +{\ + const u32 offset = MMU_MMU_SYSCONFIG_OFFSET;\ + register u32 data = __raw_readl((base_address)+offset);\ + register u32 new_value = (value);\ + _DEBUG_LEVEL1_EASI(EASIL1_MMUMMU_SYSCONFIG_IDLE_MODE_WRITE32);\ + data &= ~(MMU_MMU_SYSCONFIG_IDLE_MODE_MASK);\ + new_value <<= MMU_MMU_SYSCONFIG_IDLE_MODE_OFFSET;\ + new_value &= MMU_MMU_SYSCONFIG_IDLE_MODE_MASK;\ + new_value |= data;\ + __raw_writel(new_value, base_address+offset);\ +} + +#define MMUMMU_SYSCONFIG_AUTO_IDLE_WRITE32(base_address, value)\ +{\ + const u32 offset = MMU_MMU_SYSCONFIG_OFFSET;\ + register u32 data = __raw_readl((base_address)+offset);\ + register u32 new_value = (value);\ + _DEBUG_LEVEL1_EASI(EASIL1_MMUMMU_SYSCONFIG_AUTO_IDLE_WRITE32);\ + data &= ~(MMU_MMU_SYSCONFIG_AUTO_IDLE_MASK);\ + new_value <<= MMU_MMU_SYSCONFIG_AUTO_IDLE_OFFSET;\ + new_value &= MMU_MMU_SYSCONFIG_AUTO_IDLE_MASK;\ + new_value |= data;\ + __raw_writel(new_value, base_address+offset);\ +} + +#define MMUMMU_IRQSTATUS_READ_REGISTER32(base_address)\ + (_DEBUG_LEVEL1_EASI(easil1_mmummu_irqstatus_read_register32),\ + __raw_readl((base_address)+MMU_MMU_IRQSTATUS_OFFSET)) + +#define MMUMMU_IRQSTATUS_WRITE_REGISTER32(base_address, value)\ +{\ + const u32 offset = MMU_MMU_IRQSTATUS_OFFSET;\ + register u32 new_value = (value);\ + _DEBUG_LEVEL1_EASI(EASIL1_MMUMMU_IRQSTATUS_WRITE_REGISTER32);\ + __raw_writel(new_value, (base_address)+offset);\ +} + +#define MMUMMU_IRQENABLE_READ_REGISTER32(base_address)\ + (_DEBUG_LEVEL1_EASI(EASIL1_MMUMMU_IRQENABLE_READ_REGISTER32),\ + __raw_readl((base_address)+MMU_MMU_IRQENABLE_OFFSET)) + +#define MMUMMU_IRQENABLE_WRITE_REGISTER32(base_address, value)\ +{\ + const u32 offset = MMU_MMU_IRQENABLE_OFFSET;\ + register u32 new_value = (value);\ + _DEBUG_LEVEL1_EASI(EASIL1_MMUMMU_IRQENABLE_WRITE_REGISTER32);\ + __raw_writel(new_value, (base_address)+offset);\ +} + +#define MMUMMU_WALKING_STTWL_RUNNING_READ32(base_address)\ + (_DEBUG_LEVEL1_EASI(EASIL1_MMUMMU_WALKING_STTWL_RUNNING_READ32),\ + (((__raw_readl(((base_address)+(MMU_MMU_WALKING_ST_OFFSET))))\ + & MMU_MMU_WALKING_ST_TWL_RUNNING_MASK) >>\ + MMU_MMU_WALKING_ST_TWL_RUNNING_OFFSET)) + +#define MMUMMU_CNTLTWL_ENABLE_READ32(base_address)\ + (_DEBUG_LEVEL1_EASI(EASIL1_MMUMMU_CNTLTWL_ENABLE_READ32),\ + (((__raw_readl(((base_address)+(MMU_MMU_CNTL_OFFSET)))) &\ + MMU_MMU_CNTL_TWL_ENABLE_MASK) >>\ + MMU_MMU_CNTL_TWL_ENABLE_OFFSET)) + +#define MMUMMU_CNTLTWL_ENABLE_WRITE32(base_address, value)\ +{\ + const u32 offset = MMU_MMU_CNTL_OFFSET;\ + register u32 data = __raw_readl((base_address)+offset);\ + register u32 new_value = (value);\ + _DEBUG_LEVEL1_EASI(EASIL1_MMUMMU_CNTLTWL_ENABLE_WRITE32);\ + data &= ~(MMU_MMU_CNTL_TWL_ENABLE_MASK);\ + new_value <<= MMU_MMU_CNTL_TWL_ENABLE_OFFSET;\ + new_value &= MMU_MMU_CNTL_TWL_ENABLE_MASK;\ + new_value |= data;\ + __raw_writel(new_value, base_address+offset);\ +} + +#define MMUMMU_CNTLMMU_ENABLE_WRITE32(base_address, value)\ +{\ + const u32 offset = MMU_MMU_CNTL_OFFSET;\ + register u32 data = __raw_readl((base_address)+offset);\ + register u32 new_value = (value);\ + _DEBUG_LEVEL1_EASI(EASIL1_MMUMMU_CNTLMMU_ENABLE_WRITE32);\ + data &= ~(MMU_MMU_CNTL_MMU_ENABLE_MASK);\ + new_value <<= MMU_MMU_CNTL_MMU_ENABLE_OFFSET;\ + new_value &= MMU_MMU_CNTL_MMU_ENABLE_MASK;\ + new_value |= data;\ + __raw_writel(new_value, base_address+offset);\ +} + +#define MMUMMU_FAULT_AD_READ_REGISTER32(base_address)\ + (_DEBUG_LEVEL1_EASI(EASIL1_MMUMMU_FAULT_AD_READ_REGISTER32),\ + __raw_readl((base_address)+MMU_MMU_FAULT_AD_OFFSET)) + +#define MMUMMU_TTB_WRITE_REGISTER32(base_address, value)\ +{\ + const u32 offset = MMU_MMU_TTB_OFFSET;\ + register u32 new_value = (value);\ + _DEBUG_LEVEL1_EASI(EASIL1_MMUMMU_TTB_WRITE_REGISTER32);\ + __raw_writel(new_value, (base_address)+offset);\ +} + +#define MMUMMU_LOCK_READ_REGISTER32(base_address)\ + (_DEBUG_LEVEL1_EASI(EASIL1_MMUMMU_LOCK_READ_REGISTER32),\ + __raw_readl((base_address)+MMU_MMU_LOCK_OFFSET)) + +#define MMUMMU_LOCK_WRITE_REGISTER32(base_address, value)\ +{\ + const u32 offset = MMU_MMU_LOCK_OFFSET;\ + register u32 new_value = (value);\ + _DEBUG_LEVEL1_EASI(EASIL1_MMUMMU_LOCK_WRITE_REGISTER32);\ + __raw_writel(new_value, (base_address)+offset);\ +} + +#define MMUMMU_LOCK_BASE_VALUE_READ32(base_address)\ + (_DEBUG_LEVEL1_EASI(EASIL1_MMUMMU_LOCK_BASE_VALUE_READ32),\ + (((__raw_readl(((base_address)+(MMU_MMU_LOCK_OFFSET)))) &\ + MMU_MMU_LOCK_BASE_VALUE_MASK) >>\ + MMU_MMU_LOCK_BASE_VALUE_OFFSET)) + +#define MMUMMU_LOCK_BASE_VALUE_WRITE32(base_address, value)\ +{\ + const u32 offset = MMU_MMU_LOCK_OFFSET;\ + register u32 data = __raw_readl((base_address)+offset);\ + register u32 new_value = (value);\ + _DEBUG_LEVEL1_EASI(easil1_mmummu_lock_base_value_write32);\ + data &= ~(MMU_MMU_LOCK_BASE_VALUE_MASK);\ + new_value <<= MMU_MMU_LOCK_BASE_VALUE_OFFSET;\ + new_value &= MMU_MMU_LOCK_BASE_VALUE_MASK;\ + new_value |= data;\ + __raw_writel(new_value, base_address+offset);\ +} + +#define MMUMMU_LOCK_CURRENT_VICTIM_READ32(base_address)\ + (_DEBUG_LEVEL1_EASI(EASIL1_MMUMMU_LOCK_CURRENT_VICTIM_READ32),\ + (((__raw_readl(((base_address)+(MMU_MMU_LOCK_OFFSET)))) &\ + MMU_MMU_LOCK_CURRENT_VICTIM_MASK) >>\ + MMU_MMU_LOCK_CURRENT_VICTIM_OFFSET)) + +#define MMUMMU_LOCK_CURRENT_VICTIM_WRITE32(base_address, value)\ +{\ + const u32 offset = MMU_MMU_LOCK_OFFSET;\ + register u32 data = __raw_readl((base_address)+offset);\ + register u32 new_value = (value);\ + _DEBUG_LEVEL1_EASI(EASIL1_MMUMMU_LOCK_CURRENT_VICTIM_WRITE32);\ + data &= ~(MMU_MMU_LOCK_CURRENT_VICTIM_MASK);\ + new_value <<= MMU_MMU_LOCK_CURRENT_VICTIM_OFFSET;\ + new_value &= MMU_MMU_LOCK_CURRENT_VICTIM_MASK;\ + new_value |= data;\ + __raw_writel(new_value, base_address+offset);\ +} + +#define MMUMMU_LOCK_CURRENT_VICTIM_SET32(var, value)\ + (_DEBUG_LEVEL1_EASI(EASIL1_MMUMMU_LOCK_CURRENT_VICTIM_SET32),\ + (((var) & ~(MMU_MMU_LOCK_CURRENT_VICTIM_MASK)) |\ + (((value) << MMU_MMU_LOCK_CURRENT_VICTIM_OFFSET) &\ + MMU_MMU_LOCK_CURRENT_VICTIM_MASK))) + +#define MMUMMU_LD_TLB_READ_REGISTER32(base_address)\ + (_DEBUG_LEVEL1_EASI(EASIL1_MMUMMU_LD_TLB_READ_REGISTER32),\ + __raw_readl((base_address)+MMU_MMU_LD_TLB_OFFSET)) + +#define MMUMMU_LD_TLB_WRITE_REGISTER32(base_address, value)\ +{\ + const u32 offset = MMU_MMU_LD_TLB_OFFSET;\ + register u32 new_value = (value);\ + _DEBUG_LEVEL1_EASI(EASIL1_MMUMMU_LD_TLB_WRITE_REGISTER32);\ + __raw_writel(new_value, (base_address)+offset);\ +} + +#define MMUMMU_CAM_WRITE_REGISTER32(base_address, value)\ +{\ + const u32 offset = MMU_MMU_CAM_OFFSET;\ + register u32 new_value = (value);\ + _DEBUG_LEVEL1_EASI(EASIL1_MMUMMU_CAM_WRITE_REGISTER32);\ + __raw_writel(new_value, (base_address)+offset);\ +} + +#define MMUMMU_RAM_WRITE_REGISTER32(base_address, value)\ +{\ + const u32 offset = MMU_MMU_RAM_OFFSET;\ + register u32 new_value = (value);\ + _DEBUG_LEVEL1_EASI(EASIL1_MMUMMU_RAM_WRITE_REGISTER32);\ + __raw_writel(new_value, (base_address)+offset);\ +} + +#define MMUMMU_FLUSH_ENTRY_WRITE_REGISTER32(base_address, value)\ +{\ + const u32 offset = MMU_MMU_FLUSH_ENTRY_OFFSET;\ + register u32 new_value = (value);\ + _DEBUG_LEVEL1_EASI(EASIL1_MMUMMU_FLUSH_ENTRY_WRITE_REGISTER32);\ + __raw_writel(new_value, (base_address)+offset);\ +} + +#endif /* USE_LEVEL_1_MACROS */ + +#endif /* _MMU_REG_ACM_H */ diff --git a/drivers/staging/tidspbridge/hw/hw_defs.h b/drivers/staging/tidspbridge/hw/hw_defs.h new file mode 100644 index 00000000000..d5266d4c163 --- /dev/null +++ b/drivers/staging/tidspbridge/hw/hw_defs.h @@ -0,0 +1,58 @@ +/* + * hw_defs.h + * + * DSP-BIOS Bridge driver support functions for TI OMAP processors. + * + * Global HW definitions + * + * Copyright (C) 2007 Texas Instruments, Inc. + * + * This package is free software; you can redistribute it and/or modify + * it under the terms of the GNU General Public License version 2 as + * published by the Free Software Foundation. + * + * THIS PACKAGE IS PROVIDED ``AS IS'' AND WITHOUT ANY EXPRESS OR + * IMPLIED WARRANTIES, INCLUDING, WITHOUT LIMITATION, THE IMPLIED + * WARRANTIES OF MERCHANTIBILITY AND FITNESS FOR A PARTICULAR PURPOSE. + */ + +#ifndef _HW_DEFS_H +#define _HW_DEFS_H + +/* Page size */ +#define HW_PAGE_SIZE4KB 0x1000 +#define HW_PAGE_SIZE64KB 0x10000 +#define HW_PAGE_SIZE1MB 0x100000 +#define HW_PAGE_SIZE16MB 0x1000000 + +/* hw_status: return type for HW API */ +typedef long hw_status; + +/* Macro used to set and clear any bit */ +#define HW_CLEAR 0 +#define HW_SET 1 + +/* hw_endianism_t: Enumerated Type used to specify the endianism + * Do NOT change these values. They are used as bit fields. */ +enum hw_endianism_t { + HW_LITTLE_ENDIAN, + HW_BIG_ENDIAN +}; + +/* hw_element_size_t: Enumerated Type used to specify the element size + * Do NOT change these values. They are used as bit fields. */ +enum hw_element_size_t { + HW_ELEM_SIZE8BIT, + HW_ELEM_SIZE16BIT, + HW_ELEM_SIZE32BIT, + HW_ELEM_SIZE64BIT +}; + +/* hw_idle_mode_t: Enumerated Type used to specify Idle modes */ +enum hw_idle_mode_t { + HW_FORCE_IDLE, + HW_NO_IDLE, + HW_SMART_IDLE +}; + +#endif /* _HW_DEFS_H */ diff --git a/drivers/staging/tidspbridge/hw/hw_mmu.c b/drivers/staging/tidspbridge/hw/hw_mmu.c new file mode 100644 index 00000000000..014f5d5293a --- /dev/null +++ b/drivers/staging/tidspbridge/hw/hw_mmu.c @@ -0,0 +1,562 @@ +/* + * hw_mmu.c + * + * DSP-BIOS Bridge driver support functions for TI OMAP processors. + * + * API definitions to setup MMU TLB and PTE + * + * Copyright (C) 2007 Texas Instruments, Inc. + * + * This package is free software; you can redistribute it and/or modify + * it under the terms of the GNU General Public License version 2 as + * published by the Free Software Foundation. + * + * THIS PACKAGE IS PROVIDED ``AS IS'' AND WITHOUT ANY EXPRESS OR + * IMPLIED WARRANTIES, INCLUDING, WITHOUT LIMITATION, THE IMPLIED + * WARRANTIES OF MERCHANTIBILITY AND FITNESS FOR A PARTICULAR PURPOSE. + */ + +#include <linux/io.h> +#include "MMURegAcM.h" +#include <hw_defs.h> +#include <hw_mmu.h> +#include <linux/types.h> +#include <linux/err.h> + +#define MMU_BASE_VAL_MASK 0xFC00 +#define MMU_PAGE_MAX 3 +#define MMU_ELEMENTSIZE_MAX 3 +#define MMU_ADDR_MASK 0xFFFFF000 +#define MMU_TTB_MASK 0xFFFFC000 +#define MMU_SECTION_ADDR_MASK 0xFFF00000 +#define MMU_SSECTION_ADDR_MASK 0xFF000000 +#define MMU_PAGE_TABLE_MASK 0xFFFFFC00 +#define MMU_LARGE_PAGE_MASK 0xFFFF0000 +#define MMU_SMALL_PAGE_MASK 0xFFFFF000 + +#define MMU_LOAD_TLB 0x00000001 +#define MMU_GFLUSH 0x60 + +/* + * hw_mmu_page_size_t: Enumerated Type used to specify the MMU Page Size(SLSS) + */ +enum hw_mmu_page_size_t { + HW_MMU_SECTION, + HW_MMU_LARGE_PAGE, + HW_MMU_SMALL_PAGE, + HW_MMU_SUPERSECTION +}; + +/* + * FUNCTION : mmu_flush_entry + * + * INPUTS: + * + * Identifier : base_address + * Type : const u32 + * Description : Base Address of instance of MMU module + * + * RETURNS: + * + * Type : hw_status + * Description : 0 -- No errors occured + * RET_BAD_NULL_PARAM -- A Pointer + * Paramater was set to NULL + * + * PURPOSE: : Flush the TLB entry pointed by the + * lock counter register + * even if this entry is set protected + * + * METHOD: : Check the Input parameter and Flush a + * single entry in the TLB. + */ +static hw_status mmu_flush_entry(const void __iomem *base_address); + +/* + * FUNCTION : mmu_set_cam_entry + * + * INPUTS: + * + * Identifier : base_address + * TypE : const u32 + * Description : Base Address of instance of MMU module + * + * Identifier : page_sz + * TypE : const u32 + * Description : It indicates the page size + * + * Identifier : preserved_bit + * Type : const u32 + * Description : It indicates the TLB entry is preserved entry + * or not + * + * Identifier : valid_bit + * Type : const u32 + * Description : It indicates the TLB entry is valid entry or not + * + * + * Identifier : virtual_addr_tag + * Type : const u32 + * Description : virtual Address + * + * RETURNS: + * + * Type : hw_status + * Description : 0 -- No errors occured + * RET_BAD_NULL_PARAM -- A Pointer Paramater + * was set to NULL + * RET_PARAM_OUT_OF_RANGE -- Input Parameter out + * of Range + * + * PURPOSE: : Set MMU_CAM reg + * + * METHOD: : Check the Input parameters and set the CAM entry. + */ +static hw_status mmu_set_cam_entry(const void __iomem *base_address, + const u32 page_sz, + const u32 preserved_bit, + const u32 valid_bit, + const u32 virtual_addr_tag); + +/* + * FUNCTION : mmu_set_ram_entry + * + * INPUTS: + * + * Identifier : base_address + * Type : const u32 + * Description : Base Address of instance of MMU module + * + * Identifier : physical_addr + * Type : const u32 + * Description : Physical Address to which the corresponding + * virtual Address shouldpoint + * + * Identifier : endianism + * Type : hw_endianism_t + * Description : endianism for the given page + * + * Identifier : element_size + * Type : hw_element_size_t + * Description : The element size ( 8,16, 32 or 64 bit) + * + * Identifier : mixed_size + * Type : hw_mmu_mixed_size_t + * Description : Element Size to follow CPU or TLB + * + * RETURNS: + * + * Type : hw_status + * Description : 0 -- No errors occured + * RET_BAD_NULL_PARAM -- A Pointer Paramater + * was set to NULL + * RET_PARAM_OUT_OF_RANGE -- Input Parameter + * out of Range + * + * PURPOSE: : Set MMU_CAM reg + * + * METHOD: : Check the Input parameters and set the RAM entry. + */ +static hw_status mmu_set_ram_entry(const void __iomem *base_address, + const u32 physical_addr, + enum hw_endianism_t endianism, + enum hw_element_size_t element_size, + enum hw_mmu_mixed_size_t mixed_size); + +/* HW FUNCTIONS */ + +hw_status hw_mmu_enable(const void __iomem *base_address) +{ + hw_status status = 0; + + MMUMMU_CNTLMMU_ENABLE_WRITE32(base_address, HW_SET); + + return status; +} + +hw_status hw_mmu_disable(const void __iomem *base_address) +{ + hw_status status = 0; + + MMUMMU_CNTLMMU_ENABLE_WRITE32(base_address, HW_CLEAR); + + return status; +} + +hw_status hw_mmu_num_locked_set(const void __iomem *base_address, + u32 num_locked_entries) +{ + hw_status status = 0; + + MMUMMU_LOCK_BASE_VALUE_WRITE32(base_address, num_locked_entries); + + return status; +} + +hw_status hw_mmu_victim_num_set(const void __iomem *base_address, + u32 victim_entry_num) +{ + hw_status status = 0; + + MMUMMU_LOCK_CURRENT_VICTIM_WRITE32(base_address, victim_entry_num); + + return status; +} + +hw_status hw_mmu_event_ack(const void __iomem *base_address, u32 irq_mask) +{ + hw_status status = 0; + + MMUMMU_IRQSTATUS_WRITE_REGISTER32(base_address, irq_mask); + + return status; +} + +hw_status hw_mmu_event_disable(const void __iomem *base_address, u32 irq_mask) +{ + hw_status status = 0; + u32 irq_reg; + + irq_reg = MMUMMU_IRQENABLE_READ_REGISTER32(base_address); + + MMUMMU_IRQENABLE_WRITE_REGISTER32(base_address, irq_reg & ~irq_mask); + + return status; +} + +hw_status hw_mmu_event_enable(const void __iomem *base_address, u32 irq_mask) +{ + hw_status status = 0; + u32 irq_reg; + + irq_reg = MMUMMU_IRQENABLE_READ_REGISTER32(base_address); + + MMUMMU_IRQENABLE_WRITE_REGISTER32(base_address, irq_reg | irq_mask); + + return status; +} + +hw_status hw_mmu_event_status(const void __iomem *base_address, u32 *irq_mask) +{ + hw_status status = 0; + + *irq_mask = MMUMMU_IRQSTATUS_READ_REGISTER32(base_address); + + return status; +} + +hw_status hw_mmu_fault_addr_read(const void __iomem *base_address, u32 *addr) +{ + hw_status status = 0; + + /* read values from register */ + *addr = MMUMMU_FAULT_AD_READ_REGISTER32(base_address); + + return status; +} + +hw_status hw_mmu_ttb_set(const void __iomem *base_address, u32 ttb_phys_addr) +{ + hw_status status = 0; + u32 load_ttb; + + load_ttb = ttb_phys_addr & ~0x7FUL; + /* write values to register */ + MMUMMU_TTB_WRITE_REGISTER32(base_address, load_ttb); + + return status; +} + +hw_status hw_mmu_twl_enable(const void __iomem *base_address) +{ + hw_status status = 0; + + MMUMMU_CNTLTWL_ENABLE_WRITE32(base_address, HW_SET); + + return status; +} + +hw_status hw_mmu_twl_disable(const void __iomem *base_address) +{ + hw_status status = 0; + + MMUMMU_CNTLTWL_ENABLE_WRITE32(base_address, HW_CLEAR); + + return status; +} + +hw_status hw_mmu_tlb_flush(const void __iomem *base_address, u32 virtual_addr, + u32 page_sz) +{ + hw_status status = 0; + u32 virtual_addr_tag; + enum hw_mmu_page_size_t pg_size_bits; + + switch (page_sz) { + case HW_PAGE_SIZE4KB: + pg_size_bits = HW_MMU_SMALL_PAGE; + break; + + case HW_PAGE_SIZE64KB: + pg_size_bits = HW_MMU_LARGE_PAGE; + break; + + case HW_PAGE_SIZE1MB: + pg_size_bits = HW_MMU_SECTION; + break; + + case HW_PAGE_SIZE16MB: + pg_size_bits = HW_MMU_SUPERSECTION; + break; + + default: + return -EINVAL; + } + + /* Generate the 20-bit tag from virtual address */ + virtual_addr_tag = ((virtual_addr & MMU_ADDR_MASK) >> 12); + + mmu_set_cam_entry(base_address, pg_size_bits, 0, 0, virtual_addr_tag); + + mmu_flush_entry(base_address); + + return status; +} + +hw_status hw_mmu_tlb_add(const void __iomem *base_address, + u32 physical_addr, + u32 virtual_addr, + u32 page_sz, + u32 entry_num, + struct hw_mmu_map_attrs_t *map_attrs, + s8 preserved_bit, s8 valid_bit) +{ + hw_status status = 0; + u32 lock_reg; + u32 virtual_addr_tag; + enum hw_mmu_page_size_t mmu_pg_size; + + /*Check the input Parameters */ + switch (page_sz) { + case HW_PAGE_SIZE4KB: + mmu_pg_size = HW_MMU_SMALL_PAGE; + break; + + case HW_PAGE_SIZE64KB: + mmu_pg_size = HW_MMU_LARGE_PAGE; + break; + + case HW_PAGE_SIZE1MB: + mmu_pg_size = HW_MMU_SECTION; + break; + + case HW_PAGE_SIZE16MB: + mmu_pg_size = HW_MMU_SUPERSECTION; + break; + + default: + return -EINVAL; + } + + lock_reg = MMUMMU_LOCK_READ_REGISTER32(base_address); + + /* Generate the 20-bit tag from virtual address */ + virtual_addr_tag = ((virtual_addr & MMU_ADDR_MASK) >> 12); + + /* Write the fields in the CAM Entry Register */ + mmu_set_cam_entry(base_address, mmu_pg_size, preserved_bit, valid_bit, + virtual_addr_tag); + + /* Write the different fields of the RAM Entry Register */ + /* endianism of the page,Element Size of the page (8, 16, 32, 64 bit) */ + mmu_set_ram_entry(base_address, physical_addr, map_attrs->endianism, + map_attrs->element_size, map_attrs->mixed_size); + + /* Update the MMU Lock Register */ + /* currentVictim between lockedBaseValue and (MMU_Entries_Number - 1) */ + MMUMMU_LOCK_CURRENT_VICTIM_WRITE32(base_address, entry_num); + + /* Enable loading of an entry in TLB by writing 1 + into LD_TLB_REG register */ + MMUMMU_LD_TLB_WRITE_REGISTER32(base_address, MMU_LOAD_TLB); + + MMUMMU_LOCK_WRITE_REGISTER32(base_address, lock_reg); + + return status; +} + +hw_status hw_mmu_pte_set(const u32 pg_tbl_va, + u32 physical_addr, + u32 virtual_addr, + u32 page_sz, struct hw_mmu_map_attrs_t *map_attrs) +{ + hw_status status = 0; + u32 pte_addr, pte_val; + s32 num_entries = 1; + + switch (page_sz) { + case HW_PAGE_SIZE4KB: + pte_addr = hw_mmu_pte_addr_l2(pg_tbl_va, + virtual_addr & + MMU_SMALL_PAGE_MASK); + pte_val = + ((physical_addr & MMU_SMALL_PAGE_MASK) | + (map_attrs->endianism << 9) | (map_attrs-> + element_size << 4) | + (map_attrs->mixed_size << 11) | 2); + break; + + case HW_PAGE_SIZE64KB: + num_entries = 16; + pte_addr = hw_mmu_pte_addr_l2(pg_tbl_va, + virtual_addr & + MMU_LARGE_PAGE_MASK); + pte_val = + ((physical_addr & MMU_LARGE_PAGE_MASK) | + (map_attrs->endianism << 9) | (map_attrs-> + element_size << 4) | + (map_attrs->mixed_size << 11) | 1); + break; + + case HW_PAGE_SIZE1MB: + pte_addr = hw_mmu_pte_addr_l1(pg_tbl_va, + virtual_addr & + MMU_SECTION_ADDR_MASK); + pte_val = + ((((physical_addr & MMU_SECTION_ADDR_MASK) | + (map_attrs->endianism << 15) | (map_attrs-> + element_size << 10) | + (map_attrs->mixed_size << 17)) & ~0x40000) | 0x2); + break; + + case HW_PAGE_SIZE16MB: + num_entries = 16; + pte_addr = hw_mmu_pte_addr_l1(pg_tbl_va, + virtual_addr & + MMU_SSECTION_ADDR_MASK); + pte_val = + (((physical_addr & MMU_SSECTION_ADDR_MASK) | + (map_attrs->endianism << 15) | (map_attrs-> + element_size << 10) | + (map_attrs->mixed_size << 17) + ) | 0x40000 | 0x2); + break; + + case HW_MMU_COARSE_PAGE_SIZE: + pte_addr = hw_mmu_pte_addr_l1(pg_tbl_va, + virtual_addr & + MMU_SECTION_ADDR_MASK); + pte_val = (physical_addr & MMU_PAGE_TABLE_MASK) | 1; + break; + + default: + return -EINVAL; + } + + while (--num_entries >= 0) + ((u32 *) pte_addr)[num_entries] = pte_val; + + return status; +} + +hw_status hw_mmu_pte_clear(const u32 pg_tbl_va, u32 virtual_addr, u32 page_size) +{ + hw_status status = 0; + u32 pte_addr; + s32 num_entries = 1; + + switch (page_size) { + case HW_PAGE_SIZE4KB: + pte_addr = hw_mmu_pte_addr_l2(pg_tbl_va, + virtual_addr & + MMU_SMALL_PAGE_MASK); + break; + + case HW_PAGE_SIZE64KB: + num_entries = 16; + pte_addr = hw_mmu_pte_addr_l2(pg_tbl_va, + virtual_addr & + MMU_LARGE_PAGE_MASK); + break; + + case HW_PAGE_SIZE1MB: + case HW_MMU_COARSE_PAGE_SIZE: + pte_addr = hw_mmu_pte_addr_l1(pg_tbl_va, + virtual_addr & + MMU_SECTION_ADDR_MASK); + break; + + case HW_PAGE_SIZE16MB: + num_entries = 16; + pte_addr = hw_mmu_pte_addr_l1(pg_tbl_va, + virtual_addr & + MMU_SSECTION_ADDR_MASK); + break; + + default: + return -EINVAL; + } + + while (--num_entries >= 0) + ((u32 *) pte_addr)[num_entries] = 0; + + return status; +} + +/* mmu_flush_entry */ +static hw_status mmu_flush_entry(const void __iomem *base_address) +{ + hw_status status = 0; + u32 flush_entry_data = 0x1; + + /* write values to register */ + MMUMMU_FLUSH_ENTRY_WRITE_REGISTER32(base_address, flush_entry_data); + + return status; +} + +/* mmu_set_cam_entry */ +static hw_status mmu_set_cam_entry(const void __iomem *base_address, + const u32 page_sz, + const u32 preserved_bit, + const u32 valid_bit, + const u32 virtual_addr_tag) +{ + hw_status status = 0; + u32 mmu_cam_reg; + + mmu_cam_reg = (virtual_addr_tag << 12); + mmu_cam_reg = (mmu_cam_reg) | (page_sz) | (valid_bit << 2) | + (preserved_bit << 3); + + /* write values to register */ + MMUMMU_CAM_WRITE_REGISTER32(base_address, mmu_cam_reg); + + return status; +} + +/* mmu_set_ram_entry */ +static hw_status mmu_set_ram_entry(const void __iomem *base_address, + const u32 physical_addr, + enum hw_endianism_t endianism, + enum hw_element_size_t element_size, + enum hw_mmu_mixed_size_t mixed_size) +{ + hw_status status = 0; + u32 mmu_ram_reg; + + mmu_ram_reg = (physical_addr & MMU_ADDR_MASK); + mmu_ram_reg = (mmu_ram_reg) | ((endianism << 9) | (element_size << 7) | + (mixed_size << 6)); + + /* write values to register */ + MMUMMU_RAM_WRITE_REGISTER32(base_address, mmu_ram_reg); + + return status; + +} + +void hw_mmu_tlb_flush_all(const void __iomem *base) +{ + __raw_writeb(1, base + MMU_GFLUSH); +} diff --git a/drivers/staging/tidspbridge/hw/hw_mmu.h b/drivers/staging/tidspbridge/hw/hw_mmu.h new file mode 100644 index 00000000000..1458a2c6027 --- /dev/null +++ b/drivers/staging/tidspbridge/hw/hw_mmu.h @@ -0,0 +1,163 @@ +/* + * hw_mmu.h + * + * DSP-BIOS Bridge driver support functions for TI OMAP processors. + * + * MMU types and API declarations + * + * Copyright (C) 2007 Texas Instruments, Inc. + * + * This package is free software; you can redistribute it and/or modify + * it under the terms of the GNU General Public License version 2 as + * published by the Free Software Foundation. + * + * THIS PACKAGE IS PROVIDED ``AS IS'' AND WITHOUT ANY EXPRESS OR + * IMPLIED WARRANTIES, INCLUDING, WITHOUT LIMITATION, THE IMPLIED + * WARRANTIES OF MERCHANTIBILITY AND FITNESS FOR A PARTICULAR PURPOSE. + */ + +#ifndef _HW_MMU_H +#define _HW_MMU_H + +#include <linux/types.h> + +/* Bitmasks for interrupt sources */ +#define HW_MMU_TRANSLATION_FAULT 0x2 +#define HW_MMU_ALL_INTERRUPTS 0x1F + +#define HW_MMU_COARSE_PAGE_SIZE 0x400 + +/* hw_mmu_mixed_size_t: Enumerated Type used to specify whether to follow + CPU/TLB Element size */ +enum hw_mmu_mixed_size_t { + HW_MMU_TLBES, + HW_MMU_CPUES +}; + +/* hw_mmu_map_attrs_t: Struct containing MMU mapping attributes */ +struct hw_mmu_map_attrs_t { + enum hw_endianism_t endianism; + enum hw_element_size_t element_size; + enum hw_mmu_mixed_size_t mixed_size; + bool donotlockmpupage; +}; + +extern hw_status hw_mmu_enable(const void __iomem *base_address); + +extern hw_status hw_mmu_disable(const void __iomem *base_address); + +extern hw_status hw_mmu_num_locked_set(const void __iomem *base_address, + u32 num_locked_entries); + +extern hw_status hw_mmu_victim_num_set(const void __iomem *base_address, + u32 victim_entry_num); + +/* For MMU faults */ +extern hw_status hw_mmu_event_ack(const void __iomem *base_address, + u32 irq_mask); + +extern hw_status hw_mmu_event_disable(const void __iomem *base_address, + u32 irq_mask); + +extern hw_status hw_mmu_event_enable(const void __iomem *base_address, + u32 irq_mask); + +extern hw_status hw_mmu_event_status(const void __iomem *base_address, + u32 *irq_mask); + +extern hw_status hw_mmu_fault_addr_read(const void __iomem *base_address, + u32 *addr); + +/* Set the TT base address */ +extern hw_status hw_mmu_ttb_set(const void __iomem *base_address, + u32 ttb_phys_addr); + +extern hw_status hw_mmu_twl_enable(const void __iomem *base_address); + +extern hw_status hw_mmu_twl_disable(const void __iomem *base_address); + +extern hw_status hw_mmu_tlb_flush(const void __iomem *base_address, + u32 virtual_addr, u32 page_sz); + +extern hw_status hw_mmu_tlb_add(const void __iomem *base_address, + u32 physical_addr, + u32 virtual_addr, + u32 page_sz, + u32 entry_num, + struct hw_mmu_map_attrs_t *map_attrs, + s8 preserved_bit, s8 valid_bit); + +/* For PTEs */ +extern hw_status hw_mmu_pte_set(const u32 pg_tbl_va, + u32 physical_addr, + u32 virtual_addr, + u32 page_sz, + struct hw_mmu_map_attrs_t *map_attrs); + +extern hw_status hw_mmu_pte_clear(const u32 pg_tbl_va, + u32 virtual_addr, u32 page_size); + +void hw_mmu_tlb_flush_all(const void __iomem *base); + +static inline u32 hw_mmu_pte_addr_l1(u32 l1_base, u32 va) +{ + u32 pte_addr; + u32 va31_to20; + + va31_to20 = va >> (20 - 2); /* Left-shift by 2 here itself */ + va31_to20 &= 0xFFFFFFFCUL; + pte_addr = l1_base + va31_to20; + + return pte_addr; +} + +static inline u32 hw_mmu_pte_addr_l2(u32 l2_base, u32 va) +{ + u32 pte_addr; + + pte_addr = (l2_base & 0xFFFFFC00) | ((va >> 10) & 0x3FC); + + return pte_addr; +} + +static inline u32 hw_mmu_pte_coarse_l1(u32 pte_val) +{ + u32 pte_coarse; + + pte_coarse = pte_val & 0xFFFFFC00; + + return pte_coarse; +} + +static inline u32 hw_mmu_pte_size_l1(u32 pte_val) +{ + u32 pte_size = 0; + + if ((pte_val & 0x3) == 0x1) { + /* Points to L2 PT */ + pte_size = HW_MMU_COARSE_PAGE_SIZE; + } + + if ((pte_val & 0x3) == 0x2) { + if (pte_val & (1 << 18)) + pte_size = HW_PAGE_SIZE16MB; + else + pte_size = HW_PAGE_SIZE1MB; + } + + return pte_size; +} + +static inline u32 hw_mmu_pte_size_l2(u32 pte_val) +{ + u32 pte_size = 0; + + if (pte_val & 0x2) + pte_size = HW_PAGE_SIZE4KB; + else if (pte_val & 0x1) + pte_size = HW_PAGE_SIZE64KB; + + return pte_size; +} + +#endif /* _HW_MMU_H */ diff --git a/drivers/staging/tidspbridge/include/dspbridge/cfgdefs.h b/drivers/staging/tidspbridge/include/dspbridge/cfgdefs.h index dfb55cca34c..38122dbf877 100644 --- a/drivers/staging/tidspbridge/include/dspbridge/cfgdefs.h +++ b/drivers/staging/tidspbridge/include/dspbridge/cfgdefs.h @@ -68,6 +68,7 @@ struct cfg_hostres { void __iomem *dw_per_base; u32 dw_per_pm_base; u32 dw_core_pm_base; + void __iomem *dw_dmmu_base; void __iomem *dw_sys_ctrl_base; }; diff --git a/drivers/staging/tidspbridge/include/dspbridge/dev.h b/drivers/staging/tidspbridge/include/dspbridge/dev.h index 9bdd48f5742..357458fadd2 100644 --- a/drivers/staging/tidspbridge/include/dspbridge/dev.h +++ b/drivers/staging/tidspbridge/include/dspbridge/dev.h @@ -27,6 +27,7 @@ #include <dspbridge/nodedefs.h> #include <dspbridge/dispdefs.h> #include <dspbridge/dspdefs.h> +#include <dspbridge/dmm.h> #include <dspbridge/host_os.h> /* ----------------------------------- This */ @@ -233,6 +234,29 @@ extern int dev_get_cmm_mgr(struct dev_object *hdev_obj, struct cmm_object **mgr); /* + * ======== dev_get_dmm_mgr ======== + * Purpose: + * Retrieve the handle to the dynamic memory manager created for this + * device. + * Parameters: + * hdev_obj: Handle to device object created with + * dev_create_device(). + * *mgr: Ptr to location to store handle. + * Returns: + * 0: Success. + * -EFAULT: Invalid hdev_obj. + * Requires: + * mgr != NULL. + * DEV Initialized. + * Ensures: + * 0: *mgr contains a handle to a channel manager object, + * or NULL. + * else: *mgr is NULL. + */ +extern int dev_get_dmm_mgr(struct dev_object *hdev_obj, + struct dmm_object **mgr); + +/* * ======== dev_get_cod_mgr ======== * Purpose: * Retrieve the COD manager create for this device. diff --git a/drivers/staging/tidspbridge/include/dspbridge/dmm.h b/drivers/staging/tidspbridge/include/dspbridge/dmm.h new file mode 100644 index 00000000000..6c58335c5f6 --- /dev/null +++ b/drivers/staging/tidspbridge/include/dspbridge/dmm.h @@ -0,0 +1,75 @@ +/* + * dmm.h + * + * DSP-BIOS Bridge driver support functions for TI OMAP processors. + * + * The Dynamic Memory Mapping(DMM) module manages the DSP Virtual address + * space that can be directly mapped to any MPU buffer or memory region. + * + * Copyright (C) 2005-2006 Texas Instruments, Inc. + * + * This package is free software; you can redistribute it and/or modify + * it under the terms of the GNU General Public License version 2 as + * published by the Free Software Foundation. + * + * THIS PACKAGE IS PROVIDED ``AS IS'' AND WITHOUT ANY EXPRESS OR + * IMPLIED WARRANTIES, INCLUDING, WITHOUT LIMITATION, THE IMPLIED + * WARRANTIES OF MERCHANTIBILITY AND FITNESS FOR A PARTICULAR PURPOSE. + */ + +#ifndef DMM_ +#define DMM_ + +#include <dspbridge/dbdefs.h> + +struct dmm_object; + +/* DMM attributes used in dmm_create() */ +struct dmm_mgrattrs { + u32 reserved; +}; + +#define DMMPOOLSIZE 0x4000000 + +/* + * ======== dmm_get_handle ======== + * Purpose: + * Return the dynamic memory manager object for this device. + * This is typically called from the client process. + */ + +extern int dmm_get_handle(void *hprocessor, + struct dmm_object **dmm_manager); + +extern int dmm_reserve_memory(struct dmm_object *dmm_mgr, + u32 size, u32 *prsv_addr); + +extern int dmm_un_reserve_memory(struct dmm_object *dmm_mgr, + u32 rsv_addr); + +extern int dmm_map_memory(struct dmm_object *dmm_mgr, u32 addr, + u32 size); + +extern int dmm_un_map_memory(struct dmm_object *dmm_mgr, + u32 addr, u32 *psize); + +extern int dmm_destroy(struct dmm_object *dmm_mgr); + +extern int dmm_delete_tables(struct dmm_object *dmm_mgr); + +extern int dmm_create(struct dmm_object **dmm_manager, + struct dev_object *hdev_obj, + const struct dmm_mgrattrs *mgr_attrts); + +extern bool dmm_init(void); + +extern void dmm_exit(void); + +extern int dmm_create_tables(struct dmm_object *dmm_mgr, + u32 addr, u32 size); + +#ifdef DSP_DMM_DEBUG +u32 dmm_mem_map_dump(struct dmm_object *dmm_mgr); +#endif + +#endif /* DMM_ */ diff --git a/drivers/staging/tidspbridge/include/dspbridge/drv.h b/drivers/staging/tidspbridge/include/dspbridge/drv.h index 75a2c9b5c6f..c1f363ec9af 100644 --- a/drivers/staging/tidspbridge/include/dspbridge/drv.h +++ b/drivers/staging/tidspbridge/include/dspbridge/drv.h @@ -108,6 +108,12 @@ struct dmm_map_object { struct bridge_dma_map_info dma_info; }; +/* Used for DMM reserved memory accounting */ +struct dmm_rsv_object { + struct list_head link; + u32 dsp_reserved_addr; +}; + /* New structure (member of process context) abstracts DMM resource info */ struct dspheap_res_object { s32 heap_allocated; /* DMM status */ @@ -159,6 +165,10 @@ struct process_context { struct list_head dmm_map_list; spinlock_t dmm_map_lock; + /* DMM reserved memory resources */ + struct list_head dmm_rsv_list; + spinlock_t dmm_rsv_lock; + /* DSP Heap resources */ struct dspheap_res_object *pdspheap_list; diff --git a/drivers/staging/tidspbridge/include/dspbridge/dsp-mmu.h b/drivers/staging/tidspbridge/include/dspbridge/dsp-mmu.h deleted file mode 100644 index cb38d4cc073..00000000000 --- a/drivers/staging/tidspbridge/include/dspbridge/dsp-mmu.h +++ /dev/null @@ -1,67 +0,0 @@ -/* - * dsp-mmu.h - * - * DSP-BIOS Bridge driver support functions for TI OMAP processors. - * - * DSP iommu. - * - * Copyright (C) 2005-2010 Texas Instruments, Inc. - * - * This package is free software; you can redistribute it and/or modify - * it under the terms of the GNU General Public License version 2 as - * published by the Free Software Foundation. - * - * THIS PACKAGE IS PROVIDED ``AS IS'' AND WITHOUT ANY EXPRESS OR - * IMPLIED WARRANTIES, INCLUDING, WITHOUT LIMITATION, THE IMPLIED - * WARRANTIES OF MERCHANTIBILITY AND FITNESS FOR A PARTICULAR PURPOSE. - */ - -#ifndef _DSP_MMU_ -#define _DSP_MMU_ - -#include <plat/iommu.h> -#include <plat/iovmm.h> - -/** - * dsp_mmu_init() - initialize dsp_mmu module and returns a handle - * - * This function initialize dsp mmu module and returns a struct iommu - * handle to use it for dsp maps. - * - */ -struct iommu *dsp_mmu_init(void); - -/** - * dsp_mmu_exit() - destroy dsp mmu module - * @mmu: Pointer to iommu handle. - * - * This function destroys dsp mmu module. - * - */ -void dsp_mmu_exit(struct iommu *mmu); - -/** - * user_to_dsp_map() - maps user to dsp virtual address - * @mmu: Pointer to iommu handle. - * @uva: Virtual user space address. - * @da DSP address - * @size Buffer size to map. - * @usr_pgs struct page array pointer where the user pages will be stored - * - * This function maps a user space buffer into DSP virtual address. - * - */ -u32 user_to_dsp_map(struct iommu *mmu, u32 uva, u32 da, u32 size, - struct page **usr_pgs); - -/** - * user_to_dsp_unmap() - unmaps DSP virtual buffer. - * @mmu: Pointer to iommu handle. - * @da DSP address - * - * This function unmaps a user space buffer into DSP virtual address. - * - */ -int user_to_dsp_unmap(struct iommu *mmu, u32 da); - -#endif diff --git a/drivers/staging/tidspbridge/include/dspbridge/dspdefs.h b/drivers/staging/tidspbridge/include/dspbridge/dspdefs.h index 61536347481..0ae7d1646a1 100644 --- a/drivers/staging/tidspbridge/include/dspbridge/dspdefs.h +++ b/drivers/staging/tidspbridge/include/dspbridge/dspdefs.h @@ -162,6 +162,48 @@ typedef int(*fxn_brd_memwrite) (struct bridge_dev_context u32 mem_type); /* + * ======== bridge_brd_mem_map ======== + * Purpose: + * Map a MPU memory region to a DSP/IVA memory space + * Parameters: + * dev_ctxt: Handle to Bridge driver defined device info. + * ul_mpu_addr: MPU memory region start address. + * virt_addr: DSP/IVA memory region u8 address. + * ul_num_bytes: Number of bytes to map. + * map_attrs: Mapping attributes (e.g. endianness). + * Returns: + * 0: Success. + * -EPERM: Other, unspecified error. + * Requires: + * dev_ctxt != NULL; + * Ensures: + */ +typedef int(*fxn_brd_memmap) (struct bridge_dev_context + * dev_ctxt, u32 ul_mpu_addr, + u32 virt_addr, u32 ul_num_bytes, + u32 map_attr, + struct page **mapped_pages); + +/* + * ======== bridge_brd_mem_un_map ======== + * Purpose: + * UnMap an MPU memory region from DSP/IVA memory space + * Parameters: + * dev_ctxt: Handle to Bridge driver defined device info. + * virt_addr: DSP/IVA memory region u8 address. + * ul_num_bytes: Number of bytes to unmap. + * Returns: + * 0: Success. + * -EPERM: Other, unspecified error. + * Requires: + * dev_ctxt != NULL; + * Ensures: + */ +typedef int(*fxn_brd_memunmap) (struct bridge_dev_context + * dev_ctxt, + u32 virt_addr, u32 ul_num_bytes); + +/* * ======== bridge_brd_stop ======== * Purpose: * Bring board to the BRD_STOPPED state. @@ -951,6 +993,8 @@ struct bridge_drv_interface { fxn_brd_setstate pfn_brd_set_state; /* Sets the Board State */ fxn_brd_memcopy pfn_brd_mem_copy; /* Copies DSP Memory */ fxn_brd_memwrite pfn_brd_mem_write; /* Write DSP Memory w/o halt */ + fxn_brd_memmap pfn_brd_mem_map; /* Maps MPU mem to DSP mem */ + fxn_brd_memunmap pfn_brd_mem_un_map; /* Unmaps MPU mem to DSP mem */ fxn_chnl_create pfn_chnl_create; /* Create channel manager. */ fxn_chnl_destroy pfn_chnl_destroy; /* Destroy channel manager. */ fxn_chnl_open pfn_chnl_open; /* Create a new channel. */ diff --git a/drivers/staging/tidspbridge/include/dspbridge/dspioctl.h b/drivers/staging/tidspbridge/include/dspbridge/dspioctl.h index bad180108ad..41e0594dff3 100644 --- a/drivers/staging/tidspbridge/include/dspbridge/dspioctl.h +++ b/drivers/staging/tidspbridge/include/dspbridge/dspioctl.h @@ -19,6 +19,10 @@ #ifndef DSPIOCTL_ #define DSPIOCTL_ +/* ------------------------------------ Hardware Abstraction Layer */ +#include <hw_defs.h> +#include <hw_mmu.h> + /* * Any IOCTLS at or above this value are reserved for standard Bridge driver * interfaces. @@ -61,6 +65,9 @@ struct bridge_ioctl_extproc { /* GPP virtual address. __va does not work for ioremapped addresses */ u32 ul_gpp_va; u32 ul_size; /* Size of the mapped memory in bytes */ + enum hw_endianism_t endianism; + enum hw_mmu_mixed_size_t mixed_mode; + enum hw_element_size_t elem_size; }; #endif /* DSPIOCTL_ */ diff --git a/drivers/staging/tidspbridge/include/dspbridge/proc.h b/drivers/staging/tidspbridge/include/dspbridge/proc.h index 2d12aab6b5b..5e09fd165d9 100644 --- a/drivers/staging/tidspbridge/include/dspbridge/proc.h +++ b/drivers/staging/tidspbridge/include/dspbridge/proc.h @@ -551,6 +551,29 @@ extern int proc_map(void *hprocessor, struct process_context *pr_ctxt); /* + * ======== proc_reserve_memory ======== + * Purpose: + * Reserve a virtually contiguous region of DSP address space. + * Parameters: + * hprocessor : The processor handle. + * ul_size : Size of the address space to reserve. + * pp_rsv_addr : Ptr to DSP side reserved u8 address. + * Returns: + * 0 : Success. + * -EFAULT : Invalid processor handle. + * -EPERM : General failure. + * -ENOMEM : Cannot reserve chunk of this size. + * Requires: + * pp_rsv_addr is not NULL + * PROC Initialized. + * Ensures: + * Details: + */ +extern int proc_reserve_memory(void *hprocessor, + u32 ul_size, void **pp_rsv_addr, + struct process_context *pr_ctxt); + +/* * ======== proc_un_map ======== * Purpose: * Removes a MPU buffer mapping from the DSP address space. @@ -572,4 +595,27 @@ extern int proc_map(void *hprocessor, extern int proc_un_map(void *hprocessor, void *map_addr, struct process_context *pr_ctxt); +/* + * ======== proc_un_reserve_memory ======== + * Purpose: + * Frees a previously reserved region of DSP address space. + * Parameters: + * hprocessor : The processor handle. + * prsv_addr : Ptr to DSP side reservedBYTE address. + * Returns: + * 0 : Success. + * -EFAULT : Invalid processor handle. + * -EPERM : General failure. + * -ENOENT : Cannot find a reserved region starting with this + * : address. + * Requires: + * prsv_addr is not NULL + * PROC Initialized. + * Ensures: + * Details: + */ +extern int proc_un_reserve_memory(void *hprocessor, + void *prsv_addr, + struct process_context *pr_ctxt); + #endif /* PROC_ */ diff --git a/drivers/staging/tidspbridge/pmgr/dev.c b/drivers/staging/tidspbridge/pmgr/dev.c index 7b30267ef0e..132e960967b 100644 --- a/drivers/staging/tidspbridge/pmgr/dev.c +++ b/drivers/staging/tidspbridge/pmgr/dev.c @@ -34,6 +34,7 @@ #include <dspbridge/cod.h> #include <dspbridge/drv.h> #include <dspbridge/proc.h> +#include <dspbridge/dmm.h> /* ----------------------------------- Resource Manager */ #include <dspbridge/mgr.h> @@ -74,6 +75,7 @@ struct dev_object { struct msg_mgr *hmsg_mgr; /* Message manager. */ struct io_mgr *hio_mgr; /* IO manager (CHNL, msg_ctrl) */ struct cmm_object *hcmm_mgr; /* SM memory manager. */ + struct dmm_object *dmm_mgr; /* Dynamic memory manager. */ struct ldr_module *module_obj; /* Bridge Module handle. */ u32 word_size; /* DSP word size: quick access. */ struct drv_object *hdrv_obj; /* Driver Object */ @@ -248,6 +250,9 @@ int dev_create_device(struct dev_object **device_obj, /* Instantiate the DEH module */ status = bridge_deh_create(&dev_obj->hdeh_mgr, dev_obj); } + /* Create DMM mgr . */ + status = dmm_create(&dev_obj->dmm_mgr, + (struct dev_object *)dev_obj, NULL); } /* Add the new DEV_Object to the global list: */ if (!status) { @@ -273,6 +278,8 @@ leave: kfree(dev_obj->proc_list); if (dev_obj->cod_mgr) cod_delete(dev_obj->cod_mgr); + if (dev_obj->dmm_mgr) + dmm_destroy(dev_obj->dmm_mgr); kfree(dev_obj); } @@ -382,6 +389,11 @@ int dev_destroy_device(struct dev_object *hdev_obj) dev_obj->hcmm_mgr = NULL; } + if (dev_obj->dmm_mgr) { + dmm_destroy(dev_obj->dmm_mgr); + dev_obj->dmm_mgr = NULL; + } + /* Call the driver's bridge_dev_destroy() function: */ /* Require of DevDestroy */ if (dev_obj->hbridge_context) { @@ -462,6 +474,32 @@ int dev_get_cmm_mgr(struct dev_object *hdev_obj, } /* + * ======== dev_get_dmm_mgr ======== + * Purpose: + * Retrieve the handle to the dynamic memory manager created for this + * device. + */ +int dev_get_dmm_mgr(struct dev_object *hdev_obj, + struct dmm_object **mgr) +{ + int status = 0; + struct dev_object *dev_obj = hdev_obj; + + DBC_REQUIRE(refs > 0); + DBC_REQUIRE(mgr != NULL); + + if (hdev_obj) { + *mgr = dev_obj->dmm_mgr; + } else { + *mgr = NULL; + status = -EFAULT; + } + + DBC_ENSURE(!status || (mgr != NULL && *mgr == NULL)); + return status; +} + +/* * ======== dev_get_cod_mgr ======== * Purpose: * Retrieve the COD manager create for this device. @@ -713,8 +751,10 @@ void dev_exit(void) refs--; - if (refs == 0) + if (refs == 0) { cmm_exit(); + dmm_exit(); + } DBC_ENSURE(refs >= 0); } @@ -726,12 +766,25 @@ void dev_exit(void) */ bool dev_init(void) { - bool ret = true; + bool cmm_ret, dmm_ret, ret = true; DBC_REQUIRE(refs >= 0); - if (refs == 0) - ret = cmm_init(); + if (refs == 0) { + cmm_ret = cmm_init(); + dmm_ret = dmm_init(); + + ret = cmm_ret && dmm_ret; + + if (!ret) { + if (cmm_ret) + cmm_exit(); + + if (dmm_ret) + dmm_exit(); + + } + } if (ret) refs++; @@ -1065,6 +1118,8 @@ static void store_interface_fxns(struct bridge_drv_interface *drv_fxns, STORE_FXN(fxn_brd_setstate, pfn_brd_set_state); STORE_FXN(fxn_brd_memcopy, pfn_brd_mem_copy); STORE_FXN(fxn_brd_memwrite, pfn_brd_mem_write); + STORE_FXN(fxn_brd_memmap, pfn_brd_mem_map); + STORE_FXN(fxn_brd_memunmap, pfn_brd_mem_un_map); STORE_FXN(fxn_chnl_create, pfn_chnl_create); STORE_FXN(fxn_chnl_destroy, pfn_chnl_destroy); STORE_FXN(fxn_chnl_open, pfn_chnl_open); diff --git a/drivers/staging/tidspbridge/pmgr/dmm.c b/drivers/staging/tidspbridge/pmgr/dmm.c new file mode 100644 index 00000000000..8685233d762 --- /dev/null +++ b/drivers/staging/tidspbridge/pmgr/dmm.c @@ -0,0 +1,533 @@ +/* + * dmm.c + * + * DSP-BIOS Bridge driver support functions for TI OMAP processors. + * + * The Dynamic Memory Manager (DMM) module manages the DSP Virtual address + * space that can be directly mapped to any MPU buffer or memory region + * + * Notes: + * Region: Generic memory entitiy having a start address and a size + * Chunk: Reserved region + * + * Copyright (C) 2005-2006 Texas Instruments, Inc. + * + * This package is free software; you can redistribute it and/or modify + * it under the terms of the GNU General Public License version 2 as + * published by the Free Software Foundation. + * + * THIS PACKAGE IS PROVIDED ``AS IS'' AND WITHOUT ANY EXPRESS OR + * IMPLIED WARRANTIES, INCLUDING, WITHOUT LIMITATION, THE IMPLIED + * WARRANTIES OF MERCHANTIBILITY AND FITNESS FOR A PARTICULAR PURPOSE. + */ +#include <linux/types.h> + +/* ----------------------------------- Host OS */ +#include <dspbridge/host_os.h> + +/* ----------------------------------- DSP/BIOS Bridge */ +#include <dspbridge/dbdefs.h> + +/* ----------------------------------- Trace & Debug */ +#include <dspbridge/dbc.h> + +/* ----------------------------------- OS Adaptation Layer */ +#include <dspbridge/sync.h> + +/* ----------------------------------- Platform Manager */ +#include <dspbridge/dev.h> +#include <dspbridge/proc.h> + +/* ----------------------------------- This */ +#include <dspbridge/dmm.h> + +/* ----------------------------------- Defines, Data Structures, Typedefs */ +#define DMM_ADDR_VIRTUAL(a) \ + (((struct map_page *)(a) - virtual_mapping_table) * PG_SIZE4K +\ + dyn_mem_map_beg) +#define DMM_ADDR_TO_INDEX(a) (((a) - dyn_mem_map_beg) / PG_SIZE4K) + +/* DMM Mgr */ +struct dmm_object { + /* Dmm Lock is used to serialize access mem manager for + * multi-threads. */ + spinlock_t dmm_lock; /* Lock to access dmm mgr */ +}; + +/* ----------------------------------- Globals */ +static u32 refs; /* module reference count */ +struct map_page { + u32 region_size:15; + u32 mapped_size:15; + u32 reserved:1; + u32 mapped:1; +}; + +/* Create the free list */ +static struct map_page *virtual_mapping_table; +static u32 free_region; /* The index of free region */ +static u32 free_size; +static u32 dyn_mem_map_beg; /* The Beginning of dynamic memory mapping */ +static u32 table_size; /* The size of virt and phys pages tables */ + +/* ----------------------------------- Function Prototypes */ +static struct map_page *get_region(u32 addr); +static struct map_page *get_free_region(u32 len); +static struct map_page *get_mapped_region(u32 addrs); + +/* ======== dmm_create_tables ======== + * Purpose: + * Create table to hold the information of physical address + * the buffer pages that is passed by the user, and the table + * to hold the information of the virtual memory that is reserved + * for DSP. + */ +int dmm_create_tables(struct dmm_object *dmm_mgr, u32 addr, u32 size) +{ + struct dmm_object *dmm_obj = (struct dmm_object *)dmm_mgr; + int status = 0; + + status = dmm_delete_tables(dmm_obj); + if (!status) { + dyn_mem_map_beg = addr; + table_size = PG_ALIGN_HIGH(size, PG_SIZE4K) / PG_SIZE4K; + /* Create the free list */ + virtual_mapping_table = __vmalloc(table_size * + sizeof(struct map_page), GFP_KERNEL | + __GFP_HIGHMEM | __GFP_ZERO, PAGE_KERNEL); + if (virtual_mapping_table == NULL) + status = -ENOMEM; + else { + /* On successful allocation, + * all entries are zero ('free') */ + free_region = 0; + free_size = table_size * PG_SIZE4K; + virtual_mapping_table[0].region_size = table_size; + } + } + + if (status) + pr_err("%s: failure, status 0x%x\n", __func__, status); + + return status; +} + +/* + * ======== dmm_create ======== + * Purpose: + * Create a dynamic memory manager object. + */ +int dmm_create(struct dmm_object **dmm_manager, + struct dev_object *hdev_obj, + const struct dmm_mgrattrs *mgr_attrts) +{ + struct dmm_object *dmm_obj = NULL; + int status = 0; + DBC_REQUIRE(refs > 0); + DBC_REQUIRE(dmm_manager != NULL); + + *dmm_manager = NULL; + /* create, zero, and tag a cmm mgr object */ + dmm_obj = kzalloc(sizeof(struct dmm_object), GFP_KERNEL); + if (dmm_obj != NULL) { + spin_lock_init(&dmm_obj->dmm_lock); + *dmm_manager = dmm_obj; + } else { + status = -ENOMEM; + } + + return status; +} + +/* + * ======== dmm_destroy ======== + * Purpose: + * Release the communication memory manager resources. + */ +int dmm_destroy(struct dmm_object *dmm_mgr) +{ + struct dmm_object *dmm_obj = (struct dmm_object *)dmm_mgr; + int status = 0; + + DBC_REQUIRE(refs > 0); + if (dmm_mgr) { + status = dmm_delete_tables(dmm_obj); + if (!status) + kfree(dmm_obj); + } else + status = -EFAULT; + + return status; +} + +/* + * ======== dmm_delete_tables ======== + * Purpose: + * Delete DMM Tables. + */ +int dmm_delete_tables(struct dmm_object *dmm_mgr) +{ + int status = 0; + + DBC_REQUIRE(refs > 0); + /* Delete all DMM tables */ + if (dmm_mgr) + vfree(virtual_mapping_table); + else + status = -EFAULT; + return status; +} + +/* + * ======== dmm_exit ======== + * Purpose: + * Discontinue usage of module; free resources when reference count + * reaches 0. + */ +void dmm_exit(void) +{ + DBC_REQUIRE(refs > 0); + + refs--; +} + +/* + * ======== dmm_get_handle ======== + * Purpose: + * Return the dynamic memory manager object for this device. + * This is typically called from the client process. + */ +int dmm_get_handle(void *hprocessor, struct dmm_object **dmm_manager) +{ + int status = 0; + struct dev_object *hdev_obj; + + DBC_REQUIRE(refs > 0); + DBC_REQUIRE(dmm_manager != NULL); + if (hprocessor != NULL) + status = proc_get_dev_object(hprocessor, &hdev_obj); + else + hdev_obj = dev_get_first(); /* default */ + + if (!status) + status = dev_get_dmm_mgr(hdev_obj, dmm_manager); + + return status; +} + +/* + * ======== dmm_init ======== + * Purpose: + * Initializes private state of DMM module. + */ +bool dmm_init(void) +{ + bool ret = true; + + DBC_REQUIRE(refs >= 0); + + if (ret) + refs++; + + DBC_ENSURE((ret && (refs > 0)) || (!ret && (refs >= 0))); + + virtual_mapping_table = NULL; + table_size = 0; + + return ret; +} + +/* + * ======== dmm_map_memory ======== + * Purpose: + * Add a mapping block to the reserved chunk. DMM assumes that this block + * will be mapped in the DSP/IVA's address space. DMM returns an error if a + * mapping overlaps another one. This function stores the info that will be + * required later while unmapping the block. + */ +int dmm_map_memory(struct dmm_object *dmm_mgr, u32 addr, u32 size) +{ + struct dmm_object *dmm_obj = (struct dmm_object *)dmm_mgr; + struct map_page *chunk; + int status = 0; + + spin_lock(&dmm_obj->dmm_lock); + /* Find the Reserved memory chunk containing the DSP block to + * be mapped */ + chunk = (struct map_page *)get_region(addr); + if (chunk != NULL) { + /* Mark the region 'mapped', leave the 'reserved' info as-is */ + chunk->mapped = true; + chunk->mapped_size = (size / PG_SIZE4K); + } else + status = -ENOENT; + spin_unlock(&dmm_obj->dmm_lock); + + dev_dbg(bridge, "%s dmm_mgr %p, addr %x, size %x\n\tstatus %x, " + "chunk %p", __func__, dmm_mgr, addr, size, status, chunk); + + return status; +} + +/* + * ======== dmm_reserve_memory ======== + * Purpose: + * Reserve a chunk of virtually contiguous DSP/IVA address space. + */ +int dmm_reserve_memory(struct dmm_object *dmm_mgr, u32 size, + u32 *prsv_addr) +{ + int status = 0; + struct dmm_object *dmm_obj = (struct dmm_object *)dmm_mgr; + struct map_page *node; + u32 rsv_addr = 0; + u32 rsv_size = 0; + + spin_lock(&dmm_obj->dmm_lock); + + /* Try to get a DSP chunk from the free list */ + node = get_free_region(size); + if (node != NULL) { + /* DSP chunk of given size is available. */ + rsv_addr = DMM_ADDR_VIRTUAL(node); + /* Calculate the number entries to use */ + rsv_size = size / PG_SIZE4K; + if (rsv_size < node->region_size) { + /* Mark remainder of free region */ + node[rsv_size].mapped = false; + node[rsv_size].reserved = false; + node[rsv_size].region_size = + node->region_size - rsv_size; + node[rsv_size].mapped_size = 0; + } + /* get_region will return first fit chunk. But we only use what + is requested. */ + node->mapped = false; + node->reserved = true; + node->region_size = rsv_size; + node->mapped_size = 0; + /* Return the chunk's starting address */ + *prsv_addr = rsv_addr; + } else + /*dSP chunk of given size is not available */ + status = -ENOMEM; + + spin_unlock(&dmm_obj->dmm_lock); + + dev_dbg(bridge, "%s dmm_mgr %p, size %x, prsv_addr %p\n\tstatus %x, " + "rsv_addr %x, rsv_size %x\n", __func__, dmm_mgr, size, + prsv_addr, status, rsv_addr, rsv_size); + + return status; +} + +/* + * ======== dmm_un_map_memory ======== + * Purpose: + * Remove the mapped block from the reserved chunk. + */ +int dmm_un_map_memory(struct dmm_object *dmm_mgr, u32 addr, u32 *psize) +{ + struct dmm_object *dmm_obj = (struct dmm_object *)dmm_mgr; + struct map_page *chunk; + int status = 0; + + spin_lock(&dmm_obj->dmm_lock); + chunk = get_mapped_region(addr); + if (chunk == NULL) + status = -ENOENT; + + if (!status) { + /* Unmap the region */ + *psize = chunk->mapped_size * PG_SIZE4K; + chunk->mapped = false; + chunk->mapped_size = 0; + } + spin_unlock(&dmm_obj->dmm_lock); + + dev_dbg(bridge, "%s: dmm_mgr %p, addr %x, psize %p\n\tstatus %x, " + "chunk %p\n", __func__, dmm_mgr, addr, psize, status, chunk); + + return status; +} + +/* + * ======== dmm_un_reserve_memory ======== + * Purpose: + * Free a chunk of reserved DSP/IVA address space. + */ +int dmm_un_reserve_memory(struct dmm_object *dmm_mgr, u32 rsv_addr) +{ + struct dmm_object *dmm_obj = (struct dmm_object *)dmm_mgr; + struct map_page *chunk; + u32 i; + int status = 0; + u32 chunk_size; + + spin_lock(&dmm_obj->dmm_lock); + + /* Find the chunk containing the reserved address */ + chunk = get_mapped_region(rsv_addr); + if (chunk == NULL) + status = -ENOENT; + + if (!status) { + /* Free all the mapped pages for this reserved region */ + i = 0; + while (i < chunk->region_size) { + if (chunk[i].mapped) { + /* Remove mapping from the page tables. */ + chunk_size = chunk[i].mapped_size; + /* Clear the mapping flags */ + chunk[i].mapped = false; + chunk[i].mapped_size = 0; + i += chunk_size; + } else + i++; + } + /* Clear the flags (mark the region 'free') */ + chunk->reserved = false; + /* NOTE: We do NOT coalesce free regions here. + * Free regions are coalesced in get_region(), as it traverses + *the whole mapping table + */ + } + spin_unlock(&dmm_obj->dmm_lock); + + dev_dbg(bridge, "%s: dmm_mgr %p, rsv_addr %x\n\tstatus %x chunk %p", + __func__, dmm_mgr, rsv_addr, status, chunk); + + return status; +} + +/* + * ======== get_region ======== + * Purpose: + * Returns a region containing the specified memory region + */ +static struct map_page *get_region(u32 addr) +{ + struct map_page *curr_region = NULL; + u32 i = 0; + + if (virtual_mapping_table != NULL) { + /* find page mapped by this address */ + i = DMM_ADDR_TO_INDEX(addr); + if (i < table_size) + curr_region = virtual_mapping_table + i; + } + + dev_dbg(bridge, "%s: curr_region %p, free_region %d, free_size %d\n", + __func__, curr_region, free_region, free_size); + return curr_region; +} + +/* + * ======== get_free_region ======== + * Purpose: + * Returns the requested free region + */ +static struct map_page *get_free_region(u32 len) +{ + struct map_page *curr_region = NULL; + u32 i = 0; + u32 region_size = 0; + u32 next_i = 0; + + if (virtual_mapping_table == NULL) + return curr_region; + if (len > free_size) { + /* Find the largest free region + * (coalesce during the traversal) */ + while (i < table_size) { + region_size = virtual_mapping_table[i].region_size; + next_i = i + region_size; + if (virtual_mapping_table[i].reserved == false) { + /* Coalesce, if possible */ + if (next_i < table_size && + virtual_mapping_table[next_i].reserved + == false) { + virtual_mapping_table[i].region_size += + virtual_mapping_table + [next_i].region_size; + continue; + } + region_size *= PG_SIZE4K; + if (region_size > free_size) { + free_region = i; + free_size = region_size; + } + } + i = next_i; + } + } + if (len <= free_size) { + curr_region = virtual_mapping_table + free_region; + free_region += (len / PG_SIZE4K); + free_size -= len; + } + return curr_region; +} + +/* + * ======== get_mapped_region ======== + * Purpose: + * Returns the requestedmapped region + */ +static struct map_page *get_mapped_region(u32 addrs) +{ + u32 i = 0; + struct map_page *curr_region = NULL; + + if (virtual_mapping_table == NULL) + return curr_region; + + i = DMM_ADDR_TO_INDEX(addrs); + if (i < table_size && (virtual_mapping_table[i].mapped || + virtual_mapping_table[i].reserved)) + curr_region = virtual_mapping_table + i; + return curr_region; +} + +#ifdef DSP_DMM_DEBUG +u32 dmm_mem_map_dump(struct dmm_object *dmm_mgr) +{ + struct map_page *curr_node = NULL; + u32 i; + u32 freemem = 0; + u32 bigsize = 0; + + spin_lock(&dmm_mgr->dmm_lock); + + if (virtual_mapping_table != NULL) { + for (i = 0; i < table_size; i += + virtual_mapping_table[i].region_size) { + curr_node = virtual_mapping_table + i; + if (curr_node->reserved) { + /*printk("RESERVED size = 0x%x, " + "Map size = 0x%x\n", + (curr_node->region_size * PG_SIZE4K), + (curr_node->mapped == false) ? 0 : + (curr_node->mapped_size * PG_SIZE4K)); + */ + } else { +/* printk("UNRESERVED size = 0x%x\n", + (curr_node->region_size * PG_SIZE4K)); + */ + freemem += (curr_node->region_size * PG_SIZE4K); + if (curr_node->region_size > bigsize) + bigsize = curr_node->region_size; + } + } + } + spin_unlock(&dmm_mgr->dmm_lock); + printk(KERN_INFO "Total DSP VA FREE memory = %d Mbytes\n", + freemem / (1024 * 1024)); + printk(KERN_INFO "Total DSP VA USED memory= %d Mbytes \n", + (((table_size * PG_SIZE4K) - freemem)) / (1024 * 1024)); + printk(KERN_INFO "DSP VA - Biggest FREE block = %d Mbytes \n\n", + (bigsize * PG_SIZE4K / (1024 * 1024))); + + return 0; +} +#endif diff --git a/drivers/staging/tidspbridge/pmgr/dspapi.c b/drivers/staging/tidspbridge/pmgr/dspapi.c index 981551ce4d7..86ca785f191 100644 --- a/drivers/staging/tidspbridge/pmgr/dspapi.c +++ b/drivers/staging/tidspbridge/pmgr/dspapi.c @@ -993,10 +993,27 @@ u32 procwrap_register_notify(union trapped_args *args, void *pr_ctxt) /* * ======== procwrap_reserve_memory ======== */ -u32 __deprecated procwrap_reserve_memory(union trapped_args *args, - void *pr_ctxt) +u32 procwrap_reserve_memory(union trapped_args *args, void *pr_ctxt) { - return 0; + int status; + void *prsv_addr; + void *hprocessor = ((struct process_context *)pr_ctxt)->hprocessor; + + if ((args->args_proc_rsvmem.ul_size <= 0) || + (args->args_proc_rsvmem.ul_size & (PG_SIZE4K - 1)) != 0) + return -EINVAL; + + status = proc_reserve_memory(hprocessor, + args->args_proc_rsvmem.ul_size, &prsv_addr, + pr_ctxt); + if (!status) { + if (put_user(prsv_addr, args->args_proc_rsvmem.pp_rsv_addr)) { + status = -EINVAL; + proc_un_reserve_memory(args->args_proc_rsvmem. + hprocessor, prsv_addr, pr_ctxt); + } + } + return status; } /* @@ -1025,10 +1042,15 @@ u32 procwrap_un_map(union trapped_args *args, void *pr_ctxt) /* * ======== procwrap_un_reserve_memory ======== */ -u32 __deprecated procwrap_un_reserve_memory(union trapped_args *args, - void *pr_ctxt) +u32 procwrap_un_reserve_memory(union trapped_args *args, void *pr_ctxt) { - return 0; + int status; + void *hprocessor = ((struct process_context *)pr_ctxt)->hprocessor; + + status = proc_un_reserve_memory(hprocessor, + args->args_proc_unrsvmem.prsv_addr, + pr_ctxt); + return status; } /* diff --git a/drivers/staging/tidspbridge/rmgr/drv.c b/drivers/staging/tidspbridge/rmgr/drv.c index 91cc168516e..81b1b901355 100644 --- a/drivers/staging/tidspbridge/rmgr/drv.c +++ b/drivers/staging/tidspbridge/rmgr/drv.c @@ -146,6 +146,7 @@ int drv_remove_all_dmm_res_elements(void *process_ctxt) struct process_context *ctxt = (struct process_context *)process_ctxt; int status = 0; struct dmm_map_object *temp_map, *map_obj; + struct dmm_rsv_object *temp_rsv, *rsv_obj; /* Free DMM mapped memory resources */ list_for_each_entry_safe(map_obj, temp_map, &ctxt->dmm_map_list, link) { @@ -155,6 +156,16 @@ int drv_remove_all_dmm_res_elements(void *process_ctxt) pr_err("%s: proc_un_map failed!" " status = 0x%xn", __func__, status); } + + /* Free DMM reserved memory resources */ + list_for_each_entry_safe(rsv_obj, temp_rsv, &ctxt->dmm_rsv_list, link) { + status = proc_un_reserve_memory(ctxt->hprocessor, (void *) + rsv_obj->dsp_reserved_addr, + ctxt); + if (status) + pr_err("%s: proc_un_reserve_memory failed!" + " status = 0x%xn", __func__, status); + } return status; } @@ -732,6 +743,7 @@ static int request_bridge_resources(struct cfg_hostres *res) host_res->dw_sys_ctrl_base = ioremap(OMAP_SYSC_BASE, OMAP_SYSC_SIZE); dev_dbg(bridge, "dw_mem_base[0] 0x%x\n", host_res->dw_mem_base[0]); dev_dbg(bridge, "dw_mem_base[3] 0x%x\n", host_res->dw_mem_base[3]); + dev_dbg(bridge, "dw_dmmu_base %p\n", host_res->dw_dmmu_base); /* for 24xx base port is not mapping the mamory for DSP * internal memory TODO Do a ioremap here */ @@ -785,6 +797,8 @@ int drv_request_bridge_res_dsp(void **phost_resources) OMAP_PER_PRM_SIZE); host_res->dw_core_pm_base = (u32) ioremap(OMAP_CORE_PRM_BASE, OMAP_CORE_PRM_SIZE); + host_res->dw_dmmu_base = ioremap(OMAP_DMMU_BASE, + OMAP_DMMU_SIZE); dev_dbg(bridge, "dw_mem_base[0] 0x%x\n", host_res->dw_mem_base[0]); @@ -796,6 +810,7 @@ int drv_request_bridge_res_dsp(void **phost_resources) host_res->dw_mem_base[3]); dev_dbg(bridge, "dw_mem_base[4] 0x%x\n", host_res->dw_mem_base[4]); + dev_dbg(bridge, "dw_dmmu_base %p\n", host_res->dw_dmmu_base); shm_size = drv_datap->shm_size; if (shm_size >= 0x10000) { diff --git a/drivers/staging/tidspbridge/rmgr/drv_interface.c b/drivers/staging/tidspbridge/rmgr/drv_interface.c index 34be43fec04..324fcdffb3b 100644 --- a/drivers/staging/tidspbridge/rmgr/drv_interface.c +++ b/drivers/staging/tidspbridge/rmgr/drv_interface.c @@ -509,6 +509,8 @@ static int bridge_open(struct inode *ip, struct file *filp) pr_ctxt->res_state = PROC_RES_ALLOCATED; spin_lock_init(&pr_ctxt->dmm_map_lock); INIT_LIST_HEAD(&pr_ctxt->dmm_map_list); + spin_lock_init(&pr_ctxt->dmm_rsv_lock); + INIT_LIST_HEAD(&pr_ctxt->dmm_rsv_list); pr_ctxt->node_id = kzalloc(sizeof(struct idr), GFP_KERNEL); if (pr_ctxt->node_id) { diff --git a/drivers/staging/tidspbridge/rmgr/node.c b/drivers/staging/tidspbridge/rmgr/node.c index a660247f527..1562f3c1281 100644 --- a/drivers/staging/tidspbridge/rmgr/node.c +++ b/drivers/staging/tidspbridge/rmgr/node.c @@ -56,6 +56,7 @@ /* ----------------------------------- This */ #include <dspbridge/nodepriv.h> #include <dspbridge/node.h> +#include <dspbridge/dmm.h> /* Static/Dynamic Loader includes */ #include <dspbridge/dbll.h> @@ -316,6 +317,10 @@ int node_allocate(struct proc_object *hprocessor, u32 mapped_addr = 0; u32 map_attrs = 0x0; struct dsp_processorstate proc_state; +#ifdef DSP_DMM_DEBUG + struct dmm_object *dmm_mgr; + struct proc_object *p_proc_object = (struct proc_object *)hprocessor; +#endif void *node_res; @@ -425,12 +430,34 @@ int node_allocate(struct proc_object *hprocessor, if (status) goto func_cont; + status = proc_reserve_memory(hprocessor, + pnode->create_args.asa.task_arg_obj. + heap_size + PAGE_SIZE, + (void **)&(pnode->create_args.asa. + task_arg_obj.udsp_heap_res_addr), + pr_ctxt); + if (status) { + pr_err("%s: Failed to reserve memory for heap: 0x%x\n", + __func__, status); + goto func_cont; + } +#ifdef DSP_DMM_DEBUG + status = dmm_get_handle(p_proc_object, &dmm_mgr); + if (!dmm_mgr) { + status = DSP_EHANDLE; + goto func_cont; + } + + dmm_mem_map_dump(dmm_mgr); +#endif + map_attrs |= DSP_MAPLITTLEENDIAN; map_attrs |= DSP_MAPELEMSIZE32; map_attrs |= DSP_MAPVIRTUALADDR; status = proc_map(hprocessor, (void *)attr_in->pgpp_virt_addr, pnode->create_args.asa.task_arg_obj.heap_size, - NULL, (void **)&mapped_addr, map_attrs, + (void *)pnode->create_args.asa.task_arg_obj. + udsp_heap_res_addr, (void **)&mapped_addr, map_attrs, pr_ctxt); if (status) pr_err("%s: Failed to map memory for Heap: 0x%x\n", @@ -2484,7 +2511,11 @@ static void delete_node(struct node_object *hnode, struct stream_chnl stream; struct node_msgargs node_msg_args; struct node_taskargs task_arg_obj; - +#ifdef DSP_DMM_DEBUG + struct dmm_object *dmm_mgr; + struct proc_object *p_proc_object = + (struct proc_object *)hnode->hprocessor; +#endif int status; if (!hnode) goto func_end; @@ -2545,6 +2576,19 @@ static void delete_node(struct node_object *hnode, status = proc_un_map(hnode->hprocessor, (void *) task_arg_obj.udsp_heap_addr, pr_ctxt); + + status = proc_un_reserve_memory(hnode->hprocessor, + (void *) + task_arg_obj. + udsp_heap_res_addr, + pr_ctxt); +#ifdef DSP_DMM_DEBUG + status = dmm_get_handle(p_proc_object, &dmm_mgr); + if (dmm_mgr) + dmm_mem_map_dump(dmm_mgr); + else + status = DSP_EHANDLE; +#endif } } if (node_type != NODE_MESSAGE) { diff --git a/drivers/staging/tidspbridge/rmgr/proc.c b/drivers/staging/tidspbridge/rmgr/proc.c index 7a15a02efed..b47d7aa747b 100644 --- a/drivers/staging/tidspbridge/rmgr/proc.c +++ b/drivers/staging/tidspbridge/rmgr/proc.c @@ -39,6 +39,7 @@ #include <dspbridge/cod.h> #include <dspbridge/dev.h> #include <dspbridge/procpriv.h> +#include <dspbridge/dmm.h> /* ----------------------------------- Resource Manager */ #include <dspbridge/mgr.h> @@ -51,7 +52,6 @@ #include <dspbridge/msg.h> #include <dspbridge/dspioctl.h> #include <dspbridge/drv.h> -#include <_tiomap.h> /* ----------------------------------- This */ #include <dspbridge/proc.h> @@ -151,21 +151,34 @@ static struct dmm_map_object *add_mapping_info(struct process_context *pr_ctxt, return map_obj; } +static int match_exact_map_obj(struct dmm_map_object *map_obj, + u32 dsp_addr, u32 size) +{ + if (map_obj->dsp_addr == dsp_addr && map_obj->size != size) + pr_err("%s: addr match (0x%x), size don't (0x%x != 0x%x)\n", + __func__, dsp_addr, map_obj->size, size); + + return map_obj->dsp_addr == dsp_addr && + map_obj->size == size; +} + static void remove_mapping_information(struct process_context *pr_ctxt, - u32 dsp_addr) + u32 dsp_addr, u32 size) { struct dmm_map_object *map_obj; - pr_debug("%s: looking for virt 0x%x\n", __func__, dsp_addr); + pr_debug("%s: looking for virt 0x%x size 0x%x\n", __func__, + dsp_addr, size); spin_lock(&pr_ctxt->dmm_map_lock); list_for_each_entry(map_obj, &pr_ctxt->dmm_map_list, link) { - pr_debug("%s: candidate: mpu_addr 0x%x virt 0x%x\n", + pr_debug("%s: candidate: mpu_addr 0x%x virt 0x%x size 0x%x\n", __func__, map_obj->mpu_addr, - map_obj->dsp_addr); + map_obj->dsp_addr, + map_obj->size); - if (map_obj->dsp_addr == dsp_addr) { + if (match_exact_map_obj(map_obj, dsp_addr, size)) { pr_debug("%s: match, deleting map info\n", __func__); list_del(&map_obj->link); kfree(map_obj->dma_info.sg); @@ -1077,6 +1090,7 @@ int proc_load(void *hprocessor, const s32 argc_index, s32 cnew_envp; /* " " in new_envp[] */ s32 nproc_id = 0; /* Anticipate MP version. */ struct dcd_manager *hdcd_handle; + struct dmm_object *dmm_mgr; u32 dw_ext_end; u32 proc_id; int brd_state; @@ -1267,6 +1281,25 @@ int proc_load(void *hprocessor, const s32 argc_index, if (!status) status = cod_get_sym_value(cod_mgr, EXTEND, &dw_ext_end); + + /* Reset DMM structs and add an initial free chunk */ + if (!status) { + status = + dev_get_dmm_mgr(p_proc_object->hdev_obj, + &dmm_mgr); + if (dmm_mgr) { + /* Set dw_ext_end to DMM START u8 + * address */ + dw_ext_end = + (dw_ext_end + 1) * DSPWORDSIZE; + /* DMM memory is from EXT_END */ + status = dmm_create_tables(dmm_mgr, + dw_ext_end, + DMMPOOLSIZE); + } else { + status = -EFAULT; + } + } } } /* Restore the original argv[0] */ @@ -1319,10 +1352,12 @@ int proc_map(void *hprocessor, void *pmpu_addr, u32 ul_size, { u32 va_align; u32 pa_align; + struct dmm_object *dmm_mgr; u32 size_align; int status = 0; struct proc_object *p_proc_object = (struct proc_object *)hprocessor; struct dmm_map_object *map_obj; + u32 tmp_addr = 0; #ifdef CONFIG_TIDSPBRIDGE_CACHE_LINE_CHECK if ((ul_map_attr & BUFMODE_MASK) != RBUF) { @@ -1347,30 +1382,33 @@ int proc_map(void *hprocessor, void *pmpu_addr, u32 ul_size, } /* Critical section */ mutex_lock(&proc_lock); + dmm_get_handle(p_proc_object, &dmm_mgr); + if (dmm_mgr) + status = dmm_map_memory(dmm_mgr, va_align, size_align); + else + status = -EFAULT; /* Add mapping to the page tables. */ if (!status) { + + /* Mapped address = MSB of VA | LSB of PA */ + tmp_addr = (va_align | ((u32) pmpu_addr & (PG_SIZE4K - 1))); /* mapped memory resource tracking */ - map_obj = add_mapping_info(pr_ctxt, pa_align, va_align, + map_obj = add_mapping_info(pr_ctxt, pa_align, tmp_addr, size_align); - if (!map_obj) { + if (!map_obj) status = -ENOMEM; - } else { - va_align = user_to_dsp_map( - p_proc_object->hbridge_context->dsp_mmu, - pa_align, va_align, size_align, - map_obj->pages); - if (IS_ERR_VALUE(va_align)) - status = (int)va_align; - } + else + status = (*p_proc_object->intf_fxns->pfn_brd_mem_map) + (p_proc_object->hbridge_context, pa_align, va_align, + size_align, ul_map_attr, map_obj->pages); } if (!status) { /* Mapped address = MSB of VA | LSB of PA */ - map_obj->dsp_addr = (va_align | - ((u32)pmpu_addr & (PG_SIZE4K - 1))); - *pp_map_addr = (void *)map_obj->dsp_addr; + *pp_map_addr = (void *) tmp_addr; } else { - remove_mapping_information(pr_ctxt, va_align); + remove_mapping_information(pr_ctxt, tmp_addr, size_align); + dmm_un_map_memory(dmm_mgr, va_align, &size_align); } mutex_unlock(&proc_lock); @@ -1463,6 +1501,55 @@ func_end: } /* + * ======== proc_reserve_memory ======== + * Purpose: + * Reserve a virtually contiguous region of DSP address space. + */ +int proc_reserve_memory(void *hprocessor, u32 ul_size, + void **pp_rsv_addr, + struct process_context *pr_ctxt) +{ + struct dmm_object *dmm_mgr; + int status = 0; + struct proc_object *p_proc_object = (struct proc_object *)hprocessor; + struct dmm_rsv_object *rsv_obj; + + if (!p_proc_object) { + status = -EFAULT; + goto func_end; + } + + status = dmm_get_handle(p_proc_object, &dmm_mgr); + if (!dmm_mgr) { + status = -EFAULT; + goto func_end; + } + + status = dmm_reserve_memory(dmm_mgr, ul_size, (u32 *) pp_rsv_addr); + if (status != 0) + goto func_end; + + /* + * A successful reserve should be followed by insertion of rsv_obj + * into dmm_rsv_list, so that reserved memory resource tracking + * remains uptodate + */ + rsv_obj = kmalloc(sizeof(struct dmm_rsv_object), GFP_KERNEL); + if (rsv_obj) { + rsv_obj->dsp_reserved_addr = (u32) *pp_rsv_addr; + spin_lock(&pr_ctxt->dmm_rsv_lock); + list_add(&rsv_obj->link, &pr_ctxt->dmm_rsv_list); + spin_unlock(&pr_ctxt->dmm_rsv_lock); + } + +func_end: + dev_dbg(bridge, "%s: hprocessor: 0x%p ul_size: 0x%x pp_rsv_addr: 0x%p " + "status 0x%x\n", __func__, hprocessor, + ul_size, pp_rsv_addr, status); + return status; +} + +/* * ======== proc_start ======== * Purpose: * Start a processor running. @@ -1610,7 +1697,9 @@ int proc_un_map(void *hprocessor, void *map_addr, { int status = 0; struct proc_object *p_proc_object = (struct proc_object *)hprocessor; + struct dmm_object *dmm_mgr; u32 va_align; + u32 size_align; va_align = PG_ALIGN_LOW((u32) map_addr, PG_SIZE4K); if (!p_proc_object) { @@ -1618,11 +1707,24 @@ int proc_un_map(void *hprocessor, void *map_addr, goto func_end; } + status = dmm_get_handle(hprocessor, &dmm_mgr); + if (!dmm_mgr) { + status = -EFAULT; + goto func_end; + } + /* Critical section */ mutex_lock(&proc_lock); + /* + * Update DMM structures. Get the size to unmap. + * This function returns error if the VA is not mapped + */ + status = dmm_un_map_memory(dmm_mgr, (u32) va_align, &size_align); /* Remove mapping from the page tables. */ - status = user_to_dsp_unmap(p_proc_object->hbridge_context->dsp_mmu, - va_align); + if (!status) { + status = (*p_proc_object->intf_fxns->pfn_brd_mem_un_map) + (p_proc_object->hbridge_context, va_align, size_align); + } mutex_unlock(&proc_lock); if (status) @@ -1633,7 +1735,7 @@ int proc_un_map(void *hprocessor, void *map_addr, * from dmm_map_list, so that mapped memory resource tracking * remains uptodate */ - remove_mapping_information(pr_ctxt, (u32) map_addr); + remove_mapping_information(pr_ctxt, (u32) map_addr, size_align); func_end: dev_dbg(bridge, "%s: hprocessor: 0x%p map_addr: 0x%p status: 0x%x\n", @@ -1642,6 +1744,55 @@ func_end: } /* + * ======== proc_un_reserve_memory ======== + * Purpose: + * Frees a previously reserved region of DSP address space. + */ +int proc_un_reserve_memory(void *hprocessor, void *prsv_addr, + struct process_context *pr_ctxt) +{ + struct dmm_object *dmm_mgr; + int status = 0; + struct proc_object *p_proc_object = (struct proc_object *)hprocessor; + struct dmm_rsv_object *rsv_obj; + + if (!p_proc_object) { + status = -EFAULT; + goto func_end; + } + + status = dmm_get_handle(p_proc_object, &dmm_mgr); + if (!dmm_mgr) { + status = -EFAULT; + goto func_end; + } + + status = dmm_un_reserve_memory(dmm_mgr, (u32) prsv_addr); + if (status != 0) + goto func_end; + + /* + * A successful unreserve should be followed by removal of rsv_obj + * from dmm_rsv_list, so that reserved memory resource tracking + * remains uptodate + */ + spin_lock(&pr_ctxt->dmm_rsv_lock); + list_for_each_entry(rsv_obj, &pr_ctxt->dmm_rsv_list, link) { + if (rsv_obj->dsp_reserved_addr == (u32) prsv_addr) { + list_del(&rsv_obj->link); + kfree(rsv_obj); + break; + } + } + spin_unlock(&pr_ctxt->dmm_rsv_lock); + +func_end: + dev_dbg(bridge, "%s: hprocessor: 0x%p prsv_addr: 0x%p status: 0x%x\n", + __func__, hprocessor, prsv_addr, status); + return status; +} + +/* * ======== = proc_monitor ======== == * Purpose: * Place the Processor in Monitor State. This is an internal |