/*      $NetBSD: dvma.c,v 1.46 2024/09/08 09:36:49 rillig Exp $ */

/*-
* Copyright (c) 1996 The NetBSD Foundation, Inc.
* All rights reserved.
*
* This code is derived from software contributed to The NetBSD Foundation
* by Gordon W. Ross and Jeremy Cooper.
*
* Redistribution and use in source and binary forms, with or without
* modification, are permitted provided that the following conditions
* are met:
* 1. Redistributions of source code must retain the above copyright
*    notice, this list of conditions and the following disclaimer.
* 2. Redistributions in binary form must reproduce the above copyright
*    notice, this list of conditions and the following disclaimer in the
*    documentation and/or other materials provided with the distribution.
*
* THIS SOFTWARE IS PROVIDED BY THE NETBSD FOUNDATION, INC. AND CONTRIBUTORS
* ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED
* TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR
* PURPOSE ARE DISCLAIMED.  IN NO EVENT SHALL THE FOUNDATION OR CONTRIBUTORS
* BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
* CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
* SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
* INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
* CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
* ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
* POSSIBILITY OF SUCH DAMAGE.
*/

/*
* DVMA (Direct Virtual Memory Access - like DMA)
*
* In the Sun3 architecture, memory cycles initiated by secondary bus
* masters (DVMA devices) passed through the same MMU that governed CPU
* accesses.  All DVMA devices were wired in such a way so that an offset
* was added to the addresses they issued, causing them to access virtual
* memory starting at address 0x0FF00000 - the offset.  The task of
* enabling a DVMA device to access main memory only involved creating
* valid mapping in the MMU that translated these high addresses into the
* appropriate physical addresses.
*
* The Sun3x presents a challenge to programming DVMA because the MMU is no
* longer shared by both secondary bus masters and the CPU.  The MC68030's
* built-in MMU serves only to manage virtual memory accesses initiated by
* the CPU.  Secondary bus master bus accesses pass through a different MMU,
* aptly named the 'I/O Mapper'.  To enable every device driver that uses
* DVMA to understand that these two address spaces are disconnected would
* require a tremendous amount of code re-writing. To avoid this, we will
* ensure that the I/O Mapper and the MC68030 MMU are programmed together,
* so that DVMA mappings are consistent in both the CPU virtual address
* space and secondary bus master address space - creating an environment
* just like the Sun3 system.
*
* The maximum address space that any DVMA device in the Sun3x architecture
* is capable of addressing is 24 bits wide (16 Megabytes.)  We can alias
* all of the mappings that exist in the I/O mapper by duplicating them in
* a specially reserved section of the CPU's virtual address space, 16
* Megabytes in size.  Whenever a DVMA buffer is allocated, the allocation
* code will enter in a mapping both in the MC68030 MMU page tables and the
* I/O mapper.
*
* The address returned by the allocation routine is a virtual address that
* the requesting driver must use to access the buffer.  It is up to the
* device driver to convert this virtual address into the appropriate slave
* address that its device should issue to access the buffer.  (There will be
* routines that assist the driver in doing so.)
*/

#include <sys/cdefs.h>
__KERNEL_RCSID(0, "$NetBSD: dvma.c,v 1.46 2024/09/08 09:36:49 rillig Exp $");

#include <sys/param.h>
#include <sys/systm.h>
#include <sys/device.h>
#include <sys/proc.h>
#include <sys/vmem.h>
#include <sys/buf.h>
#include <sys/vnode.h>
#include <sys/core.h>
#include <sys/exec.h>

#include <uvm/uvm_extern.h>

#define _SUN68K_BUS_DMA_PRIVATE
#include <machine/autoconf.h>
#include <machine/bus.h>
#include <machine/cpu.h>
#include <machine/dvma.h>
#include <machine/pmap.h>

#include <sun3/sun3/machdep.h>

#include <sun3/sun3x/enable.h>
#include <sun3/sun3x/iommu.h>

/*
* Use a vmem arena to manage DVMA scratch-memory pages.
* Note: SunOS says last three pages are reserved (PROM?)
* Note: need a separate map (sub-map?) for last 1MB for
*       use by VME slave interface.
*/
vmem_t *dvma_arena;

void
dvma_init(void)
{

       /*
        * Create the vmem arena for DVMA pages.
        */
       dvma_arena = vmem_create("dvma", DVMA_MAP_BASE, DVMA_MAP_AVAIL,
                                PAGE_SIZE,             /* quantum */
                                NULL,                  /* importfn */
                                NULL,                  /* releasefn */
                                NULL,                  /* source */
                                0,                     /* qcache_max */
                                VM_SLEEP,
                                IPL_VM);

       /*
        * Enable DVMA in the System Enable register.
        * Note:  This is only necessary for VME slave accesses.
        *        On-board devices are always capable of DVMA.
        */
       *enable_reg |= ENA_SDVMA;
}


/*
* Given a DVMA address, return the physical address that
* would be used by some OTHER bus-master besides the CPU.
* (Examples: on-board ie/le, VME xy board).
*/
u_long
dvma_kvtopa(void *kva, int bustype)
{
       u_long addr, mask;

       addr = (u_long)kva;
       if ((addr & DVMA_MAP_BASE) != DVMA_MAP_BASE)
               panic("dvma_kvtopa: bad dmva addr=0x%lx", addr);

       switch (bustype) {
       case BUS_OBIO:
       case BUS_OBMEM:
               mask = DVMA_OBIO_SLAVE_MASK;
               break;
       default:        /* VME bus device. */
               mask = DVMA_VME_SLAVE_MASK;
               break;
       }

       return addr & mask;
}


/*
* Map a range [va, va+len] of wired virtual addresses in the given map
* to a kernel address in DVMA space.
*/
void *
dvma_mapin(void *kmem_va, int len, int canwait)
{
       void *dvma_addr;
       vaddr_t kva;
       vmem_addr_t tva;
       int npf, error;
       paddr_t pa;
       long off;
       bool rv __debugused;

       kva = (vaddr_t)kmem_va;
       KASSERT(kva >= VM_MIN_KERNEL_ADDRESS);

       /*
        * Calculate the offset of the data buffer from a page boundary.
        */
       off = kva & PGOFSET;
       kva -= off;     /* Truncate starting address to nearest page. */
       len = round_page(len + off); /* Round the buffer length to pages. */
       npf = btoc(len); /* Determine the number of pages to be mapped. */

       /*
        * Try to allocate DVMA space of the appropriate size
        * in which to do a transfer.
        */
       const vm_flag_t vmflags = VM_INSTANTFIT |
           (canwait ? VM_SLEEP : VM_NOSLEEP);

       error = vmem_xalloc(dvma_arena, len,
                           0,                  /* alignment */
                           0,                  /* phase */
                           0,                  /* nocross */
                           VMEM_ADDR_MIN,      /* minaddr */
                           VMEM_ADDR_MAX,      /* maxaddr */
                           vmflags,
                           &tva);
       if (error)
               return NULL;

       /*
        * Tva is the starting page to which the data buffer will be double
        * mapped.  Dvma_addr is the starting address of the buffer within
        * that page and is the return value of the function.
        */
       dvma_addr = (void *)(tva + off);

       for (; npf--; kva += PAGE_SIZE, tva += PAGE_SIZE) {
               /*
                * Retrieve the physical address of each page in the buffer
                * and enter mappings into the I/O MMU so they may be seen
                * by external bus masters and into the special DVMA space
                * in the MC68030 MMU so they may be seen by the CPU.
                */
               rv = pmap_extract(pmap_kernel(), kva, &pa);
#ifdef  DEBUG
               if (rv == false)
                       panic("dvma_mapin: null page frame");
#endif  /* DEBUG */

               iommu_enter((tva & IOMMU_VA_MASK), pa);
               pmap_kenter_pa(tva,
                   pa | PMAP_NC, VM_PROT_READ | VM_PROT_WRITE, 0);
       }
       pmap_update(pmap_kernel());

       return dvma_addr;
}

/*
* Remove double map of `va' in DVMA space at `kva'.
*
* TODO - This function might be the perfect place to handle the
*       synchronization between the DVMA cache and central RAM
*       on the 3/470.
*/
void
dvma_mapout(void *dvma_addr, int len)
{
       u_long kva;
       int off;

       kva = (u_long)dvma_addr;
       off = (int)kva & PGOFSET;
       kva -= off;
       len = round_page(len + off);

       iommu_remove((kva & IOMMU_VA_MASK), len);
       pmap_kremove(kva, len);
       pmap_update(pmap_kernel());

       vmem_xfree(dvma_arena, kva, len);
}

/*
* Allocate actual memory pages in DVMA space.
* (For sun3 compatibility - the ie driver.)
*/
void *
dvma_malloc(size_t bytes)
{
       void *new_mem, *dvma_mem;
       vsize_t new_size;

       if (bytes == 0)
               return NULL;
       new_size = m68k_round_page(bytes);
       new_mem = (void *)uvm_km_alloc(kernel_map, new_size, 0, UVM_KMF_WIRED);
       if (new_mem == 0)
               return NULL;
       dvma_mem = dvma_mapin(new_mem, new_size, 1);
       return dvma_mem;
}

/*
* Free pages from dvma_malloc()
*/
void
dvma_free(void *addr, size_t size)
{
       vsize_t sz = m68k_round_page(size);

       dvma_mapout(addr, sz);
       /* XXX: need kmem address to free it...
          Oh well, we never call this anyway. */
}

int
_bus_dmamap_load_raw(bus_dma_tag_t t, bus_dmamap_t map, bus_dma_segment_t *segs,
   int nsegs, bus_size_t size, int flags)
{

       panic("_bus_dmamap_load_raw(): not implemented yet.");
}

int
_bus_dmamap_load(bus_dma_tag_t t, bus_dmamap_t map, void *buf,
   bus_size_t buflen, struct proc *p, int flags)
{
       vaddr_t kva;
       vmem_addr_t dva;
       vsize_t off, sgsize;
       paddr_t pa;
       pmap_t pmap;
       int error, rv __diagused;

       /*
        * Make sure that on error condition we return "no valid mappings".
        */
       map->dm_nsegs = 0;
       map->dm_mapsize = 0;

       if (buflen > map->_dm_size)
               return EINVAL;

       kva = (vaddr_t)buf;
       off = kva & PGOFSET;
       sgsize = round_page(off + buflen);

       /* Try to allocate DVMA space. */
       const vm_flag_t vmflags = VM_INSTANTFIT |
           ((flags & BUS_DMA_NOWAIT) ? VM_NOSLEEP : VM_SLEEP);

       error = vmem_xalloc(dvma_arena, sgsize,
                           0,                  /* alignment */
                           0,                  /* phase */
                           0,                  /* nocross */
                           VMEM_ADDR_MIN,      /* minaddr */
                           VMEM_ADDR_MAX,      /* maxaddr */
                           vmflags,
                           &dva);
       if (error)
               return ENOMEM;

       /* Fill in the segment. */
       map->dm_segs[0].ds_addr = dva + off;
       map->dm_segs[0].ds_len = buflen;
       map->dm_segs[0]._ds_va = dva;
       map->dm_segs[0]._ds_sgsize = sgsize;

       /*
        * Now map the DVMA addresses we allocated to point to the
        * pages of the caller's buffer.
        */
       if (p != NULL)
               pmap = p->p_vmspace->vm_map.pmap;
       else
               pmap = pmap_kernel();

       while (sgsize > 0) {
               rv = pmap_extract(pmap, kva, &pa);
#ifdef DIAGNOSTIC
               if (rv == false)
                       panic("%s: unmapped VA", __func__);
#endif
               iommu_enter((dva & IOMMU_VA_MASK), pa);
               pmap_kenter_pa(dva,
                   pa | PMAP_NC, VM_PROT_READ | VM_PROT_WRITE, 0);
               kva += PAGE_SIZE;
               dva += PAGE_SIZE;
               sgsize -= PAGE_SIZE;
       }

       map->dm_nsegs = 1;
       map->dm_mapsize = map->dm_segs[0].ds_len;

       return 0;
}

void
_bus_dmamap_unload(bus_dma_tag_t t, bus_dmamap_t map)
{
       bus_dma_segment_t *segs;
       vaddr_t dva;
       vsize_t sgsize;

#ifdef DIAGNOSTIC
       if (map->dm_nsegs != 1)
               panic("%s: invalid nsegs = %d", __func__, map->dm_nsegs);
#endif

       segs = map->dm_segs;
       dva = segs[0]._ds_va & ~PGOFSET;
       sgsize = segs[0]._ds_sgsize;

       /* Unmap the DVMA addresses. */
       iommu_remove((dva & IOMMU_VA_MASK), sgsize);
       pmap_kremove(dva, sgsize);
       pmap_update(pmap_kernel());

       /* Free the DVMA addresses. */
       vmem_xfree(dvma_arena, dva, sgsize);

       /* Mark the mappings as invalid. */
       map->dm_mapsize = 0;
       map->dm_nsegs = 0;
}