[PATCH v3 0/7] DSPBRIDGE: fix mem+cache API issues

[PATCH v3 0/7] DSPBRIDGE: fix mem+cache API issues

[Ohad Ben-Cohen]

This patchset introduces an approach to eliminate the direct calls
to follow_page and to the low level cache APIs.

The patchset works by caching the page information while memory
is mapped, and then using that information later when needed 
instead of calling follow_page. The low level cache API is then replaced
by standard DMA API.

Changes from v2:

* Fix rebase error that plagued the v2 series
* Added some debug logs to help analyzing missing mapping issues

Notes:
1. The global bridge device struct is used by adding an 'extern'
   to proc. This issue should be handled in a different patch series
   (the struct should not be global. instead, it should be accessible
   to the dspbridge code via one of the context objects. This way we 
   will also be able to transform pr_* prints to dev_* prints).
2. The patchset was tested with testsuite, DMM sample app and varios
    MM and recovery scenarios. Many thanks to Ivan Gomez Castellanos
    for the help here. Also Many thanks to Felipe Contreras for the
    thorough review, comments and testing.

Thanks,
Ohad.

---
If you want, you can also reach me at <  ohadb at ti dot com  >.

Ohad Ben-Cohen (7):
  DSPBRIDGE: enhance dmm_map_object
  DSPBRIDGE: maintain mapping and page info
  DSPBRIDGE: do not call follow_page
  DSPBRIDGE: do not use low level cache manipulation API
  DSPBRIDGE: remove mem_flush_cache
  DSPBRIDGE: add dspbridge API to mark end of DMA
  DSPBRIDGE: add new PROC_BEGINDMA and PROC_ENDDMA ioctls

 arch/arm/plat-omap/include/dspbridge/drv.h         |   32 +-
 .../arm/plat-omap/include/dspbridge/dspapi-ioctl.h |    9 +
 arch/arm/plat-omap/include/dspbridge/dspapi.h      |    2 +
 arch/arm/plat-omap/include/dspbridge/dspdefs.h     |    3 +-
 arch/arm/plat-omap/include/dspbridge/proc.h        |   29 ++
 drivers/dsp/bridge/core/io_sm.c                    |   11 +-
 drivers/dsp/bridge/core/tiomap3430.c               |    9 +-
 drivers/dsp/bridge/pmgr/dspapi.c                   |   34 ++-
 drivers/dsp/bridge/rmgr/drv.c                      |   33 --
 drivers/dsp/bridge/rmgr/proc.c                     |  410 ++++++++++++++++----
 10 files changed, 433 insertions(+), 139 deletions(-)

[PATCH v3 1/7] DSPBRIDGE: enhance dmm_map_object

[Ohad Ben-Cohen]

Enhance dmm_map_object to support additional mapping
and page information. This will be used to keep
track of mapping resources needed later to invoke
DMA transfers to/from the DSP.

Signed-off-by: Ohad Ben-Cohen <ohad@wizery.com>
---
If you want, you can also reach me at <  ohadb at ti dot com  >.

 arch/arm/plat-omap/include/dspbridge/drv.h |    4 ++++
 1 files changed, 4 insertions(+), 0 deletions(-)

diff --git a/arch/arm/plat-omap/include/dspbridge/drv.h b/arch/arm/plat-omap/include/dspbridge/drv.h
index 7de3323..b1312aa 100644
--- a/arch/arm/plat-omap/include/dspbridge/drv.h
+++ b/arch/arm/plat-omap/include/dspbridge/drv.h
@@ -88,6 +88,10 @@ struct node_res_object {
 struct dmm_map_object {
 	struct list_head link;
 	u32 dsp_addr;
+	u32 mpu_addr;
+	u32 size;
+	u32 num_usr_pgs;
+	struct page **pages;
 };
 
 /* Used for DMM reserved memory accounting */
-- 
1.7.0.4

[PATCH v3 2/7] DSPBRIDGE: maintain mapping and page info

[Ohad Ben-Cohen]

Every time the MM application calls proc_map to map
a memory area, remember the details of that mapping,
together with the related page structures.
Whenever a buffer is unmapped, clean up the mapping
information resources.
This information is maintained as part of the
DMM resource tracking mechanism.

Signed-off-by: Ohad Ben-Cohen <ohad@wizery.com>
---
If you want, you can also reach me at <  ohadb at ti dot com  >.

 arch/arm/plat-omap/include/dspbridge/dspdefs.h |    3 +-
 drivers/dsp/bridge/core/io_sm.c                |   11 ++-
 drivers/dsp/bridge/core/tiomap3430.c           |    9 ++-
 drivers/dsp/bridge/rmgr/proc.c                 |  125 ++++++++++++++++++------
 4 files changed, 113 insertions(+), 35 deletions(-)

diff --git a/arch/arm/plat-omap/include/dspbridge/dspdefs.h b/arch/arm/plat-omap/include/dspbridge/dspdefs.h
index 3dfe406..f09bdbd 100644
--- a/arch/arm/plat-omap/include/dspbridge/dspdefs.h
+++ b/arch/arm/plat-omap/include/dspbridge/dspdefs.h
@@ -182,7 +182,8 @@ typedef dsp_status(*fxn_brd_memwrite) (struct bridge_dev_context
 typedef dsp_status(*fxn_brd_memmap) (struct bridge_dev_context
 				     * hDevContext, u32 ul_mpu_addr,
 				     u32 ulVirtAddr, u32 ul_num_bytes,
-				     u32 ulMapAttrs);
+				     u32 ulMapAttrs,
+				     struct page **mapped_pages);
 
 /*
  *  ======== bridge_brd_mem_un_map ========
diff --git a/drivers/dsp/bridge/core/io_sm.c b/drivers/dsp/bridge/core/io_sm.c
index d6c1a98..7fda364 100644
--- a/drivers/dsp/bridge/core/io_sm.c
+++ b/drivers/dsp/bridge/core/io_sm.c
@@ -503,7 +503,8 @@ dsp_status bridge_io_on_loaded(struct io_mgr *hio_mgr)
 				    hio_mgr->intf_fxns->
 				    pfn_brd_mem_map(hio_mgr->hbridge_context,
 						    pa_curr, va_curr,
-						    page_size[i], map_attrs);
+						    page_size[i], map_attrs,
+						    NULL);
 				if (DSP_FAILED(status))
 					goto func_end;
 				pa_curr += page_size[i];
@@ -568,7 +569,8 @@ dsp_status bridge_io_on_loaded(struct io_mgr *hio_mgr)
 				    hio_mgr->intf_fxns->
 				    pfn_brd_mem_map(hio_mgr->hbridge_context,
 						    pa_curr, va_curr,
-						    page_size[i], map_attrs);
+						    page_size[i], map_attrs,
+						    NULL);
 				dev_dbg(bridge,
 					"shm MMU PTE entry PA %x"
 					" VA %x DSP_VA %x Size %x\n",
@@ -636,7 +638,8 @@ dsp_status bridge_io_on_loaded(struct io_mgr *hio_mgr)
 				     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);
+				     ul_dsp_virt, 0x100000, map_attrs,
+				     NULL);
 			}
 		}
 		if (DSP_FAILED(status))
@@ -655,7 +658,7 @@ dsp_status bridge_io_on_loaded(struct io_mgr *hio_mgr)
 		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);
+		     map_attrs, NULL);
 		if (DSP_FAILED(status))
 			goto func_end;
 		i++;
diff --git a/drivers/dsp/bridge/core/tiomap3430.c b/drivers/dsp/bridge/core/tiomap3430.c
index c7b0d83..79d42ad 100644
--- a/drivers/dsp/bridge/core/tiomap3430.c
+++ b/drivers/dsp/bridge/core/tiomap3430.c
@@ -101,7 +101,8 @@ static dsp_status bridge_brd_mem_write(struct bridge_dev_context *dev_context,
 				    u32 ul_num_bytes, u32 ulMemType);
 static dsp_status bridge_brd_mem_map(struct bridge_dev_context *hDevContext,
 				  u32 ul_mpu_addr, u32 ulVirtAddr,
-				  u32 ul_num_bytes, u32 ul_map_attr);
+				  u32 ul_num_bytes, u32 ul_map_attr,
+				  struct page **mapped_pages);
 static dsp_status bridge_brd_mem_un_map(struct bridge_dev_context *hDevContext,
 				     u32 ulVirtAddr, u32 ul_num_bytes);
 static dsp_status bridge_dev_create(OUT struct bridge_dev_context
@@ -1208,7 +1209,8 @@ static dsp_status bridge_brd_mem_write(struct bridge_dev_context *hDevContext,
  */
 static dsp_status bridge_brd_mem_map(struct bridge_dev_context *hDevContext,
 				  u32 ul_mpu_addr, u32 ulVirtAddr,
-				  u32 ul_num_bytes, u32 ul_map_attr)
+				  u32 ul_num_bytes, u32 ul_map_attr,
+				  struct page **mapped_pages)
 {
 	u32 attrs;
 	dsp_status status = DSP_SOK;
@@ -1376,6 +1378,9 @@ static dsp_status bridge_brd_mem_map(struct bridge_dev_context *hDevContext,
 				if (DSP_FAILED(status))
 					break;
 
+				if (mapped_pages)
+					mapped_pages[pg_i] = mapped_page;
+
 				va += HW_PAGE_SIZE4KB;
 				ul_mpu_addr += HW_PAGE_SIZE4KB;
 			} else {
diff --git a/drivers/dsp/bridge/rmgr/proc.c b/drivers/dsp/bridge/rmgr/proc.c
index 7dc9b5c..37258c4 100644
--- a/drivers/dsp/bridge/rmgr/proc.c
+++ b/drivers/dsp/bridge/rmgr/proc.c
@@ -108,6 +108,87 @@ static s32 get_envp_count(char **envp);
 static char **prepend_envp(char **new_envp, char **envp, s32 envp_elems,
 			   s32 cnew_envp, char *szVar);
 
+/* remember mapping information */
+static struct dmm_map_object *add_mapping_info(struct process_context *pr_ctxt,
+				u32 mpu_addr, u32 dsp_addr, u32 size)
+{
+	struct dmm_map_object *map_obj;
+
+	u32 num_usr_pgs = size / PG_SIZE4K;
+
+	pr_debug("%s: adding map info: mpu_addr 0x%x virt 0x%x size 0x%x\n",
+						__func__, mpu_addr,
+						dsp_addr, size);
+
+	map_obj = kzalloc(sizeof(struct dmm_map_object), GFP_KERNEL);
+	if (!map_obj) {
+		pr_err("%s: kzalloc failed\n", __func__);
+		return NULL;
+	}
+	INIT_LIST_HEAD(&map_obj->link);
+
+	map_obj->pages = kcalloc(num_usr_pgs, sizeof(struct page *),
+							GFP_KERNEL);
+	if (!map_obj->pages) {
+		pr_err("%s: kzalloc failed\n", __func__);
+		kfree(map_obj);
+		return NULL;
+	}
+
+	map_obj->mpu_addr = mpu_addr;
+	map_obj->dsp_addr = dsp_addr;
+	map_obj->size = size;
+	map_obj->num_usr_pgs = num_usr_pgs;
+
+	spin_lock(&pr_ctxt->dmm_map_lock);
+	list_add(&map_obj->link, &pr_ctxt->dmm_map_list);
+	spin_unlock(&pr_ctxt->dmm_map_lock);
+
+	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 size)
+{
+	struct dmm_map_object *map_obj;
+
+	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 size 0x%x\n",
+							__func__,
+							map_obj->mpu_addr,
+							map_obj->dsp_addr,
+							map_obj->size);
+
+		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->pages);
+			kfree(map_obj);
+			goto out;
+		}
+		pr_debug("%s: candidate didn't match\n", __func__);
+	}
+
+	pr_err("%s: failed to find given map info\n", __func__);
+out:
+	spin_unlock(&pr_ctxt->dmm_map_lock);
+}
+
 /*
  *  ======== proc_attach ========
  *  Purpose:
@@ -1074,6 +1155,7 @@ dsp_status proc_map(void *hprocessor, void *pmpu_addr, u32 ul_size,
 	dsp_status status = DSP_SOK;
 	struct proc_object *p_proc_object = (struct proc_object *)hprocessor;
 	struct dmm_map_object *map_obj;
+	u32 tmp_addr = 0;
 
 #ifdef CONFIG_BRIDGE_CACHE_LINE_CHECK
 	if ((ul_map_attr & BUFMODE_MASK) != RBUF) {
@@ -1105,15 +1187,23 @@ dsp_status proc_map(void *hprocessor, void *pmpu_addr, u32 ul_size,
 	/* Add mapping to the page tables. */
 	if (DSP_SUCCEEDED(status)) {
 
-		status = (*p_proc_object->intf_fxns->pfn_brd_mem_map)
-		    (p_proc_object->hbridge_context, pa_align, va_align,
-		     size_align, ul_map_attr);
+		/* 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, tmp_addr,
+						size_align);
+		if (!map_obj)
+			status = -ENOMEM;
+		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 (DSP_SUCCEEDED(status)) {
 		/* Mapped address = MSB of VA | LSB of PA */
-		*pp_map_addr = (void *)(va_align | ((u32) pmpu_addr &
-						    (PG_SIZE4K - 1)));
+		*pp_map_addr = (void *) tmp_addr;
 	} else {
+		remove_mapping_information(pr_ctxt, tmp_addr, size_align);
 		dmm_un_map_memory(dmm_mgr, va_align, &size_align);
 	}
 	mutex_unlock(&proc_lock);
@@ -1121,19 +1211,6 @@ dsp_status proc_map(void *hprocessor, void *pmpu_addr, u32 ul_size,
 	if (DSP_FAILED(status))
 		goto func_end;
 
-	/*
-	 * A successful map should be followed by insertion of map_obj
-	 * into dmm_map_list, so that mapped memory resource tracking
-	 * remains uptodate
-	 */
-	map_obj = kmalloc(sizeof(struct dmm_map_object), GFP_KERNEL);
-	if (map_obj) {
-		map_obj->dsp_addr = (u32) *pp_map_addr;
-		spin_lock(&pr_ctxt->dmm_map_lock);
-		list_add(&map_obj->link, &pr_ctxt->dmm_map_list);
-		spin_unlock(&pr_ctxt->dmm_map_lock);
-	}
-
 func_end:
 	dev_dbg(bridge, "%s: hprocessor %p, pmpu_addr %p, ul_size %x, "
 		"req_addr %p, ul_map_attr %x, pp_map_addr %p, va_align %x, "
@@ -1422,7 +1499,6 @@ dsp_status proc_un_map(void *hprocessor, void *map_addr,
 	struct dmm_object *dmm_mgr;
 	u32 va_align;
 	u32 size_align;
-	struct dmm_map_object *map_obj;
 
 	va_align = PG_ALIGN_LOW((u32) map_addr, PG_SIZE4K);
 	if (!p_proc_object) {
@@ -1446,6 +1522,7 @@ dsp_status proc_un_map(void *hprocessor, void *map_addr,
 		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 (DSP_FAILED(status))
 		goto func_end;
@@ -1455,15 +1532,7 @@ dsp_status proc_un_map(void *hprocessor, void *map_addr,
 	 * from dmm_map_list, so that mapped memory resource tracking
 	 * remains uptodate
 	 */
-	spin_lock(&pr_ctxt->dmm_map_lock);
-	list_for_each_entry(map_obj, &pr_ctxt->dmm_map_list, link) {
-		if (map_obj->dsp_addr == (u32) map_addr) {
-			list_del(&map_obj->link);
-			kfree(map_obj);
-			break;
-		}
-	}
-	spin_unlock(&pr_ctxt->dmm_map_lock);
+	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",
-- 
1.7.0.4

[PATCH v3 3/7] DSPBRIDGE: do not call follow_page

[Ohad Ben-Cohen]

Eliminate the call to follow_page. Instead, use the page
information that was kept during the proc_map call.
This also has the advantage that users can now only
specify memory areas that were previously mapped.

Signed-off-by: Ohad Ben-Cohen <ohad@wizery.com>
---
You can also reach me at <  ohadb at ti dot com  >.

 drivers/dsp/bridge/pmgr/dspapi.c |    4 +-
 drivers/dsp/bridge/rmgr/proc.c   |  148 +++++++++++++++++++++++++-------------
 2 files changed, 99 insertions(+), 53 deletions(-)

diff --git a/drivers/dsp/bridge/pmgr/dspapi.c b/drivers/dsp/bridge/pmgr/dspapi.c
index 05ea853..cc64a99 100644
--- a/drivers/dsp/bridge/pmgr/dspapi.c
+++ b/drivers/dsp/bridge/pmgr/dspapi.c
@@ -688,7 +688,7 @@ u32 procwrap_flush_memory(union Trapped_Args *args, void *pr_ctxt)
 	    PROC_WRITEBACK_INVALIDATE_MEM)
 		return -EINVAL;
 
-	status = proc_flush_memory(args->args_proc_flushmemory.hprocessor,
+	status = proc_flush_memory(pr_ctxt,
 				   args->args_proc_flushmemory.pmpu_addr,
 				   args->args_proc_flushmemory.ul_size,
 				   args->args_proc_flushmemory.ul_flags);
@@ -703,7 +703,7 @@ u32 procwrap_invalidate_memory(union Trapped_Args *args, void *pr_ctxt)
 	dsp_status status;
 
 	status =
-	    proc_invalidate_memory(args->args_proc_invalidatememory.hprocessor,
+	    proc_invalidate_memory(pr_ctxt,
 				   args->args_proc_invalidatememory.pmpu_addr,
 				   args->args_proc_invalidatememory.ul_size);
 	return status;
diff --git a/drivers/dsp/bridge/rmgr/proc.c b/drivers/dsp/bridge/rmgr/proc.c
index 37258c4..6628483 100644
--- a/drivers/dsp/bridge/rmgr/proc.c
+++ b/drivers/dsp/bridge/rmgr/proc.c
@@ -189,6 +189,75 @@ out:
 	spin_unlock(&pr_ctxt->dmm_map_lock);
 }
 
+static int match_containing_map_obj(struct dmm_map_object *map_obj,
+					u32 mpu_addr, u32 size)
+{
+	u32 map_obj_end = map_obj->mpu_addr + map_obj->size;
+
+	return mpu_addr >= map_obj->mpu_addr &&
+		mpu_addr + size <= map_obj_end;
+}
+
+static struct dmm_map_object *find_containing_mapping(
+				struct process_context *pr_ctxt,
+				u32 mpu_addr, u32 size)
+{
+	struct dmm_map_object *map_obj;
+	pr_debug("%s: looking for mpu_addr 0x%x size 0x%x\n", __func__,
+						mpu_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 size 0x%x\n",
+						__func__,
+						map_obj->mpu_addr,
+						map_obj->dsp_addr,
+						map_obj->size);
+		if (match_containing_map_obj(map_obj, mpu_addr, size)) {
+			pr_debug("%s: match!\n", __func__);
+			goto out;
+		}
+
+		pr_debug("%s: no match!\n", __func__);
+	}
+
+	map_obj = NULL;
+out:
+	spin_unlock(&pr_ctxt->dmm_map_lock);
+	return map_obj;
+}
+
+static int find_first_page_in_cache(struct dmm_map_object *map_obj,
+					unsigned long mpu_addr)
+{
+	u32 mapped_base_page = map_obj->mpu_addr >> PAGE_SHIFT;
+	u32 requested_base_page = mpu_addr >> PAGE_SHIFT;
+	int pg_index = requested_base_page - mapped_base_page;
+
+	if (pg_index < 0 || pg_index >= map_obj->num_usr_pgs) {
+		pr_err("%s: failed (got %d)\n", __func__, pg_index);
+		return -1;
+	}
+
+	pr_debug("%s: first page is %d\n", __func__, pg_index);
+	return pg_index;
+}
+
+static inline struct page *get_mapping_page(struct dmm_map_object *map_obj,
+								int pg_i)
+{
+	pr_debug("%s: looking for pg_i %d, num_usr_pgs: %d\n", __func__,
+					pg_i, map_obj->num_usr_pgs);
+
+	if (pg_i < 0 || pg_i >= map_obj->num_usr_pgs) {
+		pr_err("%s: requested pg_i %d is out of mapped range\n",
+				__func__, pg_i);
+		return NULL;
+	}
+
+	return map_obj->pages[pg_i];
+}
+
 /*
  *  ======== proc_attach ========
  *  Purpose:
@@ -537,23 +606,30 @@ dsp_status proc_enum_nodes(void *hprocessor, void **node_tab,
 }
 
 /* Cache operation against kernel address instead of users */
-static int memory_sync_page(struct vm_area_struct *vma, unsigned long start,
-			    ssize_t len, enum dsp_flushtype ftype)
+static int memory_sync_page(struct dmm_map_object *map_obj,
+		unsigned long start, ssize_t len, enum dsp_flushtype ftype)
 {
 	struct page *page;
 	void *kaddr;
 	unsigned long offset;
 	ssize_t rest;
+	int pg_i;
+
+	pg_i = find_first_page_in_cache(map_obj, start);
+	if (pg_i < 0) {
+		pr_err("%s: failed to find first page in cache\n", __func__);
+		return -EINVAL;
+	}
 
 	while (len) {
-		page = follow_page(vma, start, FOLL_GET);
+		page = get_mapping_page(map_obj, pg_i);
 		if (!page) {
 			pr_err("%s: no page for %08lx\n", __func__, start);
 			return -EINVAL;
 		} else if (IS_ERR(page)) {
 			pr_err("%s: err page for %08lx(%lu)\n", __func__, start,
-			       IS_ERR(page));
-			return IS_ERR(page);
+			       PTR_ERR(page));
+			return PTR_ERR(page);
 		}
 
 		offset = start & ~PAGE_MASK;
@@ -562,77 +638,47 @@ static int memory_sync_page(struct vm_area_struct *vma, unsigned long start,
 		mem_flush_cache(kaddr, rest, ftype);
 
 		kunmap(page);
-		put_page(page);
 		len -= rest;
 		start += rest;
+		pg_i++;
 	}
 
 	return 0;
 }
 
-/* Check if the given area blongs to process virtul memory address space */
-static int memory_sync_vma(unsigned long start, u32 len,
-			   enum dsp_flushtype ftype)
-{
-	int err = 0;
-	unsigned long end;
-	struct vm_area_struct *vma;
-
-	end = start + len;
-	if (end <= start)
-		return -EINVAL;
-
-	while ((vma = find_vma(current->mm, start)) != NULL) {
-		ssize_t size;
-
-		if (vma->vm_flags & (VM_IO | VM_PFNMAP))
-			return -EINVAL;
-
-		if (vma->vm_start > start)
-			return -EINVAL;
-
-		size = min_t(ssize_t, vma->vm_end - start, len);
-		err = memory_sync_page(vma, start, size, ftype);
-		if (err)
-			break;
-
-		if (end <= vma->vm_end)
-			break;
-
-		start = vma->vm_end;
-	}
-
-	if (!vma)
-		err = -EINVAL;
-
-	return err;
-}
-
 static dsp_status proc_memory_sync(void *hprocessor, void *pmpu_addr,
 				   u32 ul_size, u32 ul_flags,
 				   enum dsp_flushtype FlushMemType)
 {
 	/* Keep STATUS here for future additions to this function */
 	dsp_status status = DSP_SOK;
-	struct proc_object *p_proc_object = (struct proc_object *)hprocessor;
+	struct process_context *pr_ctxt = (struct process_context *) hprocessor;
+	struct dmm_map_object *map_obj;
 
 	DBC_REQUIRE(refs > 0);
 
-	if (!p_proc_object) {
+	if (!pr_ctxt) {
 		status = -EFAULT;
 		goto err_out;
 	}
 
-	down_read(&current->mm->mmap_sem);
+	pr_debug("%s: addr 0x%x, size 0x%x, type %d\n", __func__,
+							(u32)pmpu_addr,
+							ul_size, ul_flags);
 
-	if (memory_sync_vma((u32) pmpu_addr, ul_size, FlushMemType)) {
+	/* find requested memory are in cached mapping information */
+	map_obj = find_containing_mapping(pr_ctxt, (u32) pmpu_addr, ul_size);
+	if (!map_obj) {
+		pr_err("%s: find_containing_mapping failed\n", __func__);
+		status = -EFAULT;
+		goto err_out;
+	}
+	if (memory_sync_page(map_obj, (u32) pmpu_addr, ul_size, ul_flags)) {
 		pr_err("%s: InValid address parameters %p %x\n",
-		       __func__, pmpu_addr, ul_size);
+			       __func__, pmpu_addr, ul_size);
 		status = -EFAULT;
 	}
 
-	up_read(&current->mm->mmap_sem);
-
 err_out:
 
 	return status;
-- 
1.7.0.4

Re: [PATCH v3 3/7] DSPBRIDGE: do not call follow_page

[Felipe Contreras]

Hi,

Just for the record, I found a problem in this patch. The next patch
in the series overrides it, so it's not that important, unless
somebody picks only this patch.

On Thu, May 27, 2010 at 7:02 PM, Ohad Ben-Cohen <ohad@wizery.com> wrote:
> @@ -537,23 +606,30 @@ dsp_status proc_enum_nodes(void *hprocessor, void **node_tab,
>  }
>
>  /* Cache operation against kernel address instead of users */
> -static int memory_sync_page(struct vm_area_struct *vma, unsigned long start,
> -                           ssize_t len, enum dsp_flushtype ftype)
> +static int memory_sync_page(struct dmm_map_object *map_obj,
> +               unsigned long start, ssize_t len, enum dsp_flushtype ftype)
>  {
>        struct page *page;
>        void *kaddr;

[...]

> -       if (memory_sync_vma((u32) pmpu_addr, ul_size, FlushMemType)) {
> +       /* find requested memory are in cached mapping information */
> +       map_obj = find_containing_mapping(pr_ctxt, (u32) pmpu_addr, ul_size);
> +       if (!map_obj) {
> +               pr_err("%s: find_containing_mapping failed\n", __func__);
> +               status = -EFAULT;
> +               goto err_out;
> +       }
> +       if (memory_sync_page(map_obj, (u32) pmpu_addr, ul_size, ul_flags)) {

It should be FlushMemType, not ul_flags.

>                pr_err("%s: InValid address parameters %p %x\n",
> -                      __func__, pmpu_addr, ul_size);
> +                              __func__, pmpu_addr, ul_size);
>                status = -EFAULT;
>        }

-- 
Felipe Contreras
--
To unsubscribe from this list: send the line "unsubscribe linux-omap" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html

Re: [PATCH v3 3/7] DSPBRIDGE: do not call follow_page

[Felipe Contreras]

On Thu, May 27, 2010 at 7:02 PM, Ohad Ben-Cohen <ohad@wizery.com> wrote:
> Eliminate the call to follow_page. Instead, use the page
> information that was kept during the proc_map call.
> This also has the advantage that users can now only
> specify memory areas that were previously mapped.
>
> Signed-off-by: Ohad Ben-Cohen <ohad@wizery.com>

I found another issue with this patch:

> -       if (memory_sync_vma((u32) pmpu_addr, ul_size, FlushMemType)) {
> +       /* find requested memory are in cached mapping information */
> +       map_obj = find_containing_mapping(pr_ctxt, (u32) pmpu_addr, ul_size);
> +       if (!map_obj) {
> +               pr_err("%s: find_containing_mapping failed\n", __func__);
> +               status = -EFAULT;
> +               goto err_out;
> +       }
> +       if (memory_sync_page(map_obj, (u32) pmpu_addr, ul_size, ul_flags)) {
>                pr_err("%s: InValid address parameters %p %x\n",
> -                      __func__, pmpu_addr, ul_size);
> +                              __func__, pmpu_addr, ul_size);
>                status = -EFAULT;
>        }

find_containing_mapping() is taking the lock for dmm_map_list,
however, nothing prevents the map_obj to be destroyed after that,
specially if kcalloc sleeps, and then an unmap happens. While doing
some stress testing I found there's a race condition that makes
exactly that happen.

I'm sending some patches to fix that.

-- 
Felipe Contreras
--
To unsubscribe from this list: send the line "unsubscribe linux-omap" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html

[PATCH v3 4/7] DSPBRIDGE: do not use low level cache manipulation API

[Ohad Ben-Cohen]

Instead of using low level cache manipulation API,
use the standard DMA API. This is achieved by adding
a proc_begin_dma function that takes a generic
dma_data_direction, and then implementing proc_flush_memory
and proc_invalidate_memory by means of proc_begin_dma
in the following manner:

* flush calls proc_begin_dma with DMA_BIDIRECTIONAL
* Invalidate calls proc_begin_dma with DMA_FROM_DEVICE

proc_begin_dma builds a scatter gatter list using the
page information that was kept during proc_map,
and feed it to the standard dma_map_sg API.

Note that now users cannot manipulate the cache state of any random
address; if the buffer is not part of a previous memory mapping of that
application, the request is denied.

Signed-off-by: Ohad Ben-Cohen <ohad@wizery.com>
---
If you want, you can also reach me at <  ohadb at ti dot com  >.

 arch/arm/plat-omap/include/dspbridge/drv.h |   13 +++
 drivers/dsp/bridge/rmgr/proc.c             |  119 +++++++++++++++++++++-------
 2 files changed, 104 insertions(+), 28 deletions(-)

diff --git a/arch/arm/plat-omap/include/dspbridge/drv.h b/arch/arm/plat-omap/include/dspbridge/drv.h
index b1312aa..3186935 100644
--- a/arch/arm/plat-omap/include/dspbridge/drv.h
+++ b/arch/arm/plat-omap/include/dspbridge/drv.h
@@ -84,6 +84,18 @@ struct node_res_object {
 	struct node_res_object *next;
 };
 
+/* used to cache dma mapping information */
+struct bridge_dma_map_info {
+	/* direction of DMA in action, or DMA_NONE */
+	enum dma_data_direction dir;
+	/* number of elements requested by us */
+	int num_pages;
+	/* number of elements returned from dma_map_sg */
+	int sg_num;
+	/* list of buffers used in this DMA action */
+	struct scatterlist *sg;
+};
+
 /* Used for DMM mapped memory accounting */
 struct dmm_map_object {
 	struct list_head link;
@@ -92,6 +104,7 @@ struct dmm_map_object {
 	u32 size;
 	u32 num_usr_pgs;
 	struct page **pages;
+	struct bridge_dma_map_info dma_info;
 };
 
 /* Used for DMM reserved memory accounting */
diff --git a/drivers/dsp/bridge/rmgr/proc.c b/drivers/dsp/bridge/rmgr/proc.c
index 6628483..2710a11 100644
--- a/drivers/dsp/bridge/rmgr/proc.c
+++ b/drivers/dsp/bridge/rmgr/proc.c
@@ -17,6 +17,8 @@
  */
 
 /* ------------------------------------ Host OS */
+#include <linux/dma-mapping.h>
+#include <linux/scatterlist.h>
 #include <dspbridge/host_os.h>
 
 /*  ----------------------------------- DSP/BIOS Bridge */
@@ -74,6 +76,8 @@
 #define RBUF		0x4000		/* Input buffer */
 #define WBUF		0x8000		/* Output Buffer */
 
+extern struct device *bridge;
+
 /*  ----------------------------------- Globals */
 
 /* The proc_object structure. */
@@ -177,6 +181,7 @@ static void remove_mapping_information(struct process_context *pr_ctxt,
 		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);
 			kfree(map_obj->pages);
 			kfree(map_obj);
 			goto out;
@@ -606,49 +611,108 @@ dsp_status proc_enum_nodes(void *hprocessor, void **node_tab,
 }
 
 /* Cache operation against kernel address instead of users */
-static int memory_sync_page(struct dmm_map_object *map_obj,
-		unsigned long start, ssize_t len, enum dsp_flushtype ftype)
+static int build_dma_sg(struct dmm_map_object *map_obj, unsigned long start,
+						ssize_t len, int pg_i)
 {
 	struct page *page;
-	void *kaddr;
 	unsigned long offset;
 	ssize_t rest;
-	int pg_i;
-
-	pg_i = find_first_page_in_cache(map_obj, start);
-	if (pg_i < 0) {
-		pr_err("%s: failed to find first page in cache\n", __func__);
-		return -EINVAL;
-	}
+	int ret = 0, i = 0;
+	struct scatterlist *sg = map_obj->dma_info.sg;
 
 	while (len) {
 		page = get_mapping_page(map_obj, pg_i);
 		if (!page) {
 			pr_err("%s: no page for %08lx\n", __func__, start);
-			return -EINVAL;
+			ret = -EINVAL;
+			goto out;
 		} else if (IS_ERR(page)) {
 			pr_err("%s: err page for %08lx(%lu)\n", __func__, start,
 			       PTR_ERR(page));
-			return PTR_ERR(page);
+			ret = PTR_ERR(page);
+			goto out;
 		}
 
 		offset = start & ~PAGE_MASK;
-		kaddr = kmap(page) + offset;
 		rest = min_t(ssize_t, PAGE_SIZE - offset, len);
-		mem_flush_cache(kaddr, rest, ftype);
 
-		kunmap(page);
+		sg_set_page(&sg[i], page, rest, offset);
+
 		len -= rest;
 		start += rest;
-		pg_i++;
+		pg_i++, i++;
 	}
 
+	if (i != map_obj->dma_info.num_pages) {
+		pr_err("%s: bad number of sg iterations\n", __func__);
+		ret = -EFAULT;
+		goto out;
+	}
+
+out:
+	return ret;
+}
+
+/* Cache operation against kernel address instead of users */
+static int memory_give_ownership(struct dmm_map_object *map_obj,
+		unsigned long start, ssize_t len, enum dma_data_direction dir)
+{
+	int pg_i, ret, sg_num;
+	struct scatterlist *sg;
+	unsigned long first_data_page = start >> PAGE_SHIFT;
+	unsigned long last_data_page = ((u32)(start + len - 1) >> PAGE_SHIFT);
+	/* calculating the number of pages this area spans */
+	unsigned long num_pages = last_data_page - first_data_page + 1;
+
+	pg_i = find_first_page_in_cache(map_obj, start);
+	if (pg_i < 0) {
+		pr_err("%s: failed to find first page in cache\n", __func__);
+		ret = -EINVAL;
+		goto out;
+	}
+
+	sg = kcalloc(num_pages, sizeof(*sg), GFP_KERNEL);
+	if (!sg) {
+		pr_err("%s: kcalloc failed\n", __func__);
+		ret = -ENOMEM;
+		goto out;
+	}
+
+	sg_init_table(sg, num_pages);
+
+	/* cleanup a previous sg allocation */
+	/* this may happen if application doesn't signal for e/o DMA */
+	kfree(map_obj->dma_info.sg);
+
+	map_obj->dma_info.sg = sg;
+	map_obj->dma_info.dir = dir;
+	map_obj->dma_info.num_pages = num_pages;
+
+	ret = build_dma_sg(map_obj, start, len, pg_i);
+	if (ret)
+		goto kfree_sg;
+
+	sg_num = dma_map_sg(bridge, sg, num_pages, dir);
+	if (sg_num < 1) {
+		pr_err("%s: dma_map_sg failed: %d\n", __func__, sg_num);
+		ret = -EFAULT;
+		goto kfree_sg;
+	}
+
+	pr_debug("%s: dma_map_sg mapped %d elements\n", __func__, sg_num);
+	map_obj->dma_info.sg_num = sg_num;
+
 	return 0;
+
+kfree_sg:
+	kfree(sg);
+	map_obj->dma_info.sg = NULL;
+out:
+	return ret;
 }
 
-static dsp_status proc_memory_sync(void *hprocessor, void *pmpu_addr,
-				   u32 ul_size, u32 ul_flags,
-				   enum dsp_flushtype FlushMemType)
+static int proc_begin_dma(void *hprocessor, void *pmpu_addr, u32 ul_size,
+				enum dma_data_direction dir)
 {
 	/* Keep STATUS here for future additions to this function */
 	dsp_status status = DSP_SOK;
@@ -664,7 +728,7 @@ static dsp_status proc_memory_sync(void *hprocessor, void *pmpu_addr,
 
 	pr_debug("%s: addr 0x%x, size 0x%x, type %d\n", __func__,
 							(u32)pmpu_addr,
-							ul_size, ul_flags);
+							ul_size, dir);
 
 	/* find requested memory are in cached mapping information */
 	map_obj = find_containing_mapping(pr_ctxt, (u32) pmpu_addr, ul_size);
@@ -673,7 +737,8 @@ static dsp_status proc_memory_sync(void *hprocessor, void *pmpu_addr,
 		status = -EFAULT;
 		goto err_out;
 	}
-	if (memory_sync_page(map_obj, (u32) pmpu_addr, ul_size, ul_flags)) {
+
+	if (memory_give_ownership(map_obj, (u32) pmpu_addr, ul_size, dir)) {
 		pr_err("%s: InValid address parameters %p %x\n",
 			       __func__, pmpu_addr, ul_size);
 		status = -EFAULT;
@@ -692,10 +757,9 @@ err_out:
 dsp_status proc_flush_memory(void *hprocessor, void *pmpu_addr,
 			     u32 ul_size, u32 ul_flags)
 {
-	enum dsp_flushtype mtype = PROC_WRITEBACK_INVALIDATE_MEM;
+	enum dma_data_direction dir = DMA_BIDIRECTIONAL;
 
-	return proc_memory_sync(hprocessor, pmpu_addr, ul_size, ul_flags,
-				mtype);
+	return proc_begin_dma(hprocessor, pmpu_addr, ul_size, dir);
 }
 
 /*
@@ -703,12 +767,11 @@ dsp_status proc_flush_memory(void *hprocessor, void *pmpu_addr,
  *  Purpose:
  *     Invalidates the memory specified
  */
-dsp_status proc_invalidate_memory(void *hprocessor, void *pmpu_addr,
-				  u32 ul_size)
+dsp_status proc_invalidate_memory(void *hprocessor, void *pmpu_addr, u32 size)
 {
-	enum dsp_flushtype mtype = PROC_INVALIDATE_MEM;
+	enum dma_data_direction dir = DMA_FROM_DEVICE;
 
-	return proc_memory_sync(hprocessor, pmpu_addr, ul_size, 0, mtype);
+	return proc_begin_dma(hprocessor, pmpu_addr, size, dir);
 }
 
 /*
-- 
1.7.0.4

[PATCH v3 5/7] DSPBRIDGE: remove mem_flush_cache

[Ohad Ben-Cohen]

Now that we use standard DMA API instead of low level cache
manipulation API, mem_flush_cache can be removed.

Signed-off-by: Ohad Ben-Cohen <ohad@wizery.com>
---
If you want, you can also reach me at <  ohadb at ti dot com  >.

 arch/arm/plat-omap/include/dspbridge/drv.h |   15 ------------
 drivers/dsp/bridge/rmgr/drv.c              |   33 ----------------------------
 2 files changed, 0 insertions(+), 48 deletions(-)

diff --git a/arch/arm/plat-omap/include/dspbridge/drv.h b/arch/arm/plat-omap/include/dspbridge/drv.h
index 3186935..2eeb47f 100644
--- a/arch/arm/plat-omap/include/dspbridge/drv.h
+++ b/arch/arm/plat-omap/include/dspbridge/drv.h
@@ -466,21 +466,6 @@ extern void *mem_alloc_phys_mem(IN u32 byte_size,
 				IN u32 ulAlign, OUT u32 *pPhysicalAddress);
 
 /*
- *  ======== mem_flush_cache ========
- *  Purpose:
- *      Performs system cache sync with discard
- *  Parameters:
- *      pMemBuf:    Pointer to memory region to be flushed.
- *      pMemBuf:    Size of the memory region to be flushed.
- *  Returns:
- *  Requires:
- *      MEM is initialized.
- *  Ensures:
- *      Cache is synchronized
- */
-extern void mem_flush_cache(void *pMemBuf, u32 byte_size, s32 FlushType);
-
-/*
  *  ======== mem_free_phys_mem ========
  *  Purpose:
  *      Free the given block of physically contiguous memory.
diff --git a/drivers/dsp/bridge/rmgr/drv.c b/drivers/dsp/bridge/rmgr/drv.c
index a8e711a..6ffae0b 100644
--- a/drivers/dsp/bridge/rmgr/drv.c
+++ b/drivers/dsp/bridge/rmgr/drv.c
@@ -1039,39 +1039,6 @@ void *mem_alloc_phys_mem(u32 byte_size, u32 ulAlign, OUT u32 * pPhysicalAddress)
 }
 
 /*
- *  ======== mem_flush_cache ========
- *  Purpose:
- *      Flush cache
- */
-void mem_flush_cache(void *pMemBuf, u32 byte_size, s32 FlushType)
-{
-	if (!pMemBuf)
-		return;
-
-	switch (FlushType) {
-		/* invalidate only */
-	case PROC_INVALIDATE_MEM:
-		dmac_inv_range(pMemBuf, pMemBuf + byte_size);
-		outer_inv_range(__pa((u32) pMemBuf), __pa((u32) pMemBuf +
-							  byte_size));
-		break;
-		/* writeback only */
-	case PROC_WRITEBACK_MEM:
-		dmac_clean_range(pMemBuf, pMemBuf + byte_size);
-		outer_clean_range(__pa((u32) pMemBuf), __pa((u32) pMemBuf +
-							    byte_size));
-		break;
-		/* writeback and invalidate */
-	case PROC_WRITEBACK_INVALIDATE_MEM:
-		dmac_flush_range(pMemBuf, pMemBuf + byte_size);
-		outer_flush_range(__pa((u32) pMemBuf), __pa((u32) pMemBuf +
-							    byte_size));
-		break;
-	}
-
-}
-
-/*
  *  ======== mem_free_phys_mem ========
  *  Purpose:
  *      Free the given block of physically contiguous memory.
-- 
1.7.0.4

[PATCH v3 6/7] DSPBRIDGE: add dspbridge API to mark end of DMA

[Ohad Ben-Cohen]

Add new dspbridge API that ends DMA transfers.

Signed-off-by: Ohad Ben-Cohen <ohad@wizery.com>
---
If you want, you can also reach me at <  ohadb at ti dot com  >.

 drivers/dsp/bridge/rmgr/proc.c |   68 ++++++++++++++++++++++++++++++++++++++++
 1 files changed, 68 insertions(+), 0 deletions(-)

diff --git a/drivers/dsp/bridge/rmgr/proc.c b/drivers/dsp/bridge/rmgr/proc.c
index 2710a11..e952d2e 100644
--- a/drivers/dsp/bridge/rmgr/proc.c
+++ b/drivers/dsp/bridge/rmgr/proc.c
@@ -653,6 +653,36 @@ out:
 	return ret;
 }
 
+static int memory_regain_ownership(struct dmm_map_object *map_obj,
+		unsigned long start, ssize_t len, enum dma_data_direction dir)
+{
+	int ret = 0;
+	unsigned long first_data_page = start >> PAGE_SHIFT;
+	unsigned long last_data_page = ((u32)(start + len - 1) >> PAGE_SHIFT);
+	/* calculating the number of pages this area spans */
+	unsigned long num_pages = last_data_page - first_data_page + 1;
+	struct bridge_dma_map_info *dma_info = &map_obj->dma_info;
+
+	if (!dma_info->sg)
+		goto out;
+
+	if (dma_info->dir != dir || dma_info->num_pages != num_pages) {
+		pr_err("%s: dma info doesn't match given params\n", __func__);
+		return -EINVAL;
+	}
+
+	dma_unmap_sg(bridge, dma_info->sg, num_pages, dma_info->dir);
+
+	pr_debug("%s: dma_map_sg unmapped\n", __func__);
+
+	kfree(dma_info->sg);
+
+	map_obj->dma_info.sg = NULL;
+
+out:
+	return ret;
+}
+
 /* Cache operation against kernel address instead of users */
 static int memory_give_ownership(struct dmm_map_object *map_obj,
 		unsigned long start, ssize_t len, enum dma_data_direction dir)
@@ -749,6 +779,44 @@ err_out:
 	return status;
 }
 
+static int proc_end_dma(void *hprocessor, void *pmpu_addr, u32 ul_size,
+			enum dma_data_direction dir)
+{
+	/* Keep STATUS here for future additions to this function */
+	int status = 0;
+	struct process_context *pr_ctxt = (struct process_context *) hprocessor;
+	struct dmm_map_object *map_obj;
+
+	DBC_REQUIRE(refs > 0);
+
+	if (!pr_ctxt) {
+		status = -EFAULT;
+		goto err_out;
+	}
+
+	pr_debug("%s: addr 0x%x, size 0x%x, type %d\n", __func__,
+							(u32)pmpu_addr,
+							ul_size, dir);
+
+	/* find requested memory are in cached mapping information */
+	map_obj = find_containing_mapping(pr_ctxt, (u32) pmpu_addr, ul_size);
+	if (!map_obj) {
+		pr_err("%s: find_containing_mapping failed\n", __func__);
+		status = -EFAULT;
+		goto err_out;
+	}
+
+	if (memory_regain_ownership(map_obj, (u32) pmpu_addr, ul_size, dir)) {
+		pr_err("%s: InValid address parameters %p %x\n",
+		       __func__, pmpu_addr, ul_size);
+		status = -EFAULT;
+		goto err_out;
+	}
+
+err_out:
+	return status;
+}
+
 /*
  *  ======== proc_flush_memory ========
  *  Purpose:
-- 
1.7.0.4

[PATCH v3 7/7] DSPBRIDGE: add new PROC_BEGINDMA and PROC_ENDDMA ioctls

[Ohad Ben-Cohen]

Add two new dspbridge ioctls that mark the
beginnind and end of a DMA transfer.
The direction of the DMA transfer is given as a parameter,
thus all three directions (DMA_BIDIRECTIONAL, DMA_TO_DEVICE
and DMA_FROM_DEVICE) are supported.

Signed-off-by: Ohad Ben-Cohen <ohad@wizery.com>
---
If you want, you can also reach me at <  ohadb at ti dot com  >.

 .../arm/plat-omap/include/dspbridge/dspapi-ioctl.h |    9 ++++++
 arch/arm/plat-omap/include/dspbridge/dspapi.h      |    2 +
 arch/arm/plat-omap/include/dspbridge/proc.h        |   29 +++++++++++++++++++
 drivers/dsp/bridge/pmgr/dspapi.c                   |   30 ++++++++++++++++++++
 drivers/dsp/bridge/rmgr/proc.c                     |    4 +-
 5 files changed, 72 insertions(+), 2 deletions(-)

diff --git a/arch/arm/plat-omap/include/dspbridge/dspapi-ioctl.h b/arch/arm/plat-omap/include/dspbridge/dspapi-ioctl.h
index 1780855..6ed2dcc 100644
--- a/arch/arm/plat-omap/include/dspbridge/dspapi-ioctl.h
+++ b/arch/arm/plat-omap/include/dspbridge/dspapi-ioctl.h
@@ -153,6 +153,13 @@ union Trapped_Args {
 		void *hprocessor;
 		void *pmpu_addr;
 		u32 ul_size;
+		u32 dir;
+	} args_proc_dma;
+
+	struct {
+		void *hprocessor;
+		void *pmpu_addr;
+		u32 ul_size;
 		u32 ul_flags;
 	} args_proc_flushmemory;
 
@@ -426,6 +433,8 @@ union Trapped_Args {
 #define PROC_FLUSHMEMORY	_IOW(DB, DB_IOC(DB_PROC, 14), unsigned long)
 #define PROC_STOP		_IOWR(DB, DB_IOC(DB_PROC, 15), unsigned long)
 #define PROC_INVALIDATEMEMORY	_IOW(DB, DB_IOC(DB_PROC, 16), unsigned long)
+#define PROC_BEGINDMA		_IOW(DB, DB_IOC(DB_PROC, 17), unsigned long)
+#define PROC_ENDDMA		_IOW(DB, DB_IOC(DB_PROC, 18), unsigned long)
 
 /* NODE Module */
 #define NODE_ALLOCATE		_IOWR(DB, DB_IOC(DB_NODE, 0), unsigned long)
diff --git a/arch/arm/plat-omap/include/dspbridge/dspapi.h b/arch/arm/plat-omap/include/dspbridge/dspapi.h
index 565c800..e821c7b 100644
--- a/arch/arm/plat-omap/include/dspbridge/dspapi.h
+++ b/arch/arm/plat-omap/include/dspbridge/dspapi.h
@@ -126,6 +126,8 @@ extern u32 procwrap_un_map(union Trapped_Args *args, void *pr_ctxt);
 extern u32 procwrap_flush_memory(union Trapped_Args *args, void *pr_ctxt);
 extern u32 procwrap_stop(union Trapped_Args *args, void *pr_ctxt);
 extern u32 procwrap_invalidate_memory(union Trapped_Args *args, void *pr_ctxt);
+extern u32 procwrap_begin_dma(union Trapped_Args *args, void *pr_ctxt);
+extern u32 procwrap_end_dma(union Trapped_Args *args, void *pr_ctxt);
 
 /* NODE wrapper functions */
 extern u32 nodewrap_allocate(union Trapped_Args *args, void *pr_ctxt);
diff --git a/arch/arm/plat-omap/include/dspbridge/proc.h b/arch/arm/plat-omap/include/dspbridge/proc.h
index 18b51c5..58fcdea 100644
--- a/arch/arm/plat-omap/include/dspbridge/proc.h
+++ b/arch/arm/plat-omap/include/dspbridge/proc.h
@@ -456,6 +456,35 @@ extern dsp_status proc_start(void *hprocessor);
 extern dsp_status proc_stop(void *hprocessor);
 
 /*
+ *  ======== proc_end_dma ========
+ *  Purpose:
+ *      Begin a DMA transfer
+ *  Parameters:
+ *      hprocessor      :   The processor handle.
+ *      pmpu_addr	:   Buffer start address
+ *      ul_size		:   Buffer size
+ *      dir		:   The direction of the transfer
+ *  Requires:
+ *      Memory was previously mapped.
+ */
+extern int proc_end_dma(void *hprocessor, void *pmpu_addr, u32 ul_size,
+						enum dma_data_direction dir);
+/*
+ *  ======== proc_begin_dma ========
+ *  Purpose:
+ *      Begin a DMA transfer
+ *  Parameters:
+ *      hprocessor      :   The processor handle.
+ *      pmpu_addr	:   Buffer start address
+ *      ul_size		:   Buffer size
+ *      dir		:   The direction of the transfer
+ *  Requires:
+ *      Memory was previously mapped.
+ */
+extern int proc_begin_dma(void *hprocessor, void *pmpu_addr, u32 ul_size,
+						enum dma_data_direction dir);
+
+/*
  *  ======== proc_flush_memory ========
  *  Purpose:
  *      Flushes a buffer from the MPU data cache.
diff --git a/drivers/dsp/bridge/pmgr/dspapi.c b/drivers/dsp/bridge/pmgr/dspapi.c
index cc64a99..2da2021 100644
--- a/drivers/dsp/bridge/pmgr/dspapi.c
+++ b/drivers/dsp/bridge/pmgr/dspapi.c
@@ -113,6 +113,8 @@ static struct api_cmd proc_cmd[] = {
 	{procwrap_flush_memory},	/* PROC_FLUSHMEMORY */
 	{procwrap_stop},	/* PROC_STOP */
 	{procwrap_invalidate_memory},	/* PROC_INVALIDATEMEMORY */
+	{procwrap_begin_dma},	/* PROC_BEGINDMA */
+	{procwrap_end_dma},	/* PROC_ENDDMA */
 };
 
 /* NODE wrapper functions */
@@ -677,6 +679,34 @@ u32 procwrap_enum_node_info(union Trapped_Args *args, void *pr_ctxt)
 	return status;
 }
 
+u32 procwrap_end_dma(union Trapped_Args *args, void *pr_ctxt)
+{
+	dsp_status status;
+
+	if (args->args_proc_dma.dir >= DMA_NONE)
+		return -EINVAL;
+
+	status = proc_end_dma(pr_ctxt,
+				   args->args_proc_dma.pmpu_addr,
+				   args->args_proc_dma.ul_size,
+				   args->args_proc_dma.dir);
+	return status;
+}
+
+u32 procwrap_begin_dma(union Trapped_Args *args, void *pr_ctxt)
+{
+	dsp_status status;
+
+	if (args->args_proc_dma.dir >= DMA_NONE)
+		return -EINVAL;
+
+	status = proc_begin_dma(pr_ctxt,
+				   args->args_proc_dma.pmpu_addr,
+				   args->args_proc_dma.ul_size,
+				   args->args_proc_dma.dir);
+	return status;
+}
+
 /*
  * ======== procwrap_flush_memory ========
  */
diff --git a/drivers/dsp/bridge/rmgr/proc.c b/drivers/dsp/bridge/rmgr/proc.c
index e952d2e..93db51b 100644
--- a/drivers/dsp/bridge/rmgr/proc.c
+++ b/drivers/dsp/bridge/rmgr/proc.c
@@ -741,7 +741,7 @@ out:
 	return ret;
 }
 
-static int proc_begin_dma(void *hprocessor, void *pmpu_addr, u32 ul_size,
+int proc_begin_dma(void *hprocessor, void *pmpu_addr, u32 ul_size,
 				enum dma_data_direction dir)
 {
 	/* Keep STATUS here for future additions to this function */
@@ -779,7 +779,7 @@ err_out:
 	return status;
 }
 
-static int proc_end_dma(void *hprocessor, void *pmpu_addr, u32 ul_size,
+int proc_end_dma(void *hprocessor, void *pmpu_addr, u32 ul_size,
 			enum dma_data_direction dir)
 {
 	/* Keep STATUS here for future additions to this function */
-- 
1.7.0.4

RE: [PATCH v3 0/7] DSPBRIDGE: fix mem+cache API issues

[Ramirez Luna, Omar]

>From: Ohad Ben-Cohen [mailto:ohad@wizery.com]
>
>This patchset introduces an approach to eliminate the direct calls
>to follow_page and to the low level cache APIs.
>
>The patchset works by caching the page information while memory
>is mapped, and then using that information later when needed
>instead of calling follow_page. The low level cache API is then replaced
>by standard DMA API.
>
>Changes from v2:
>
>* Fix rebase error that plagued the v2 series
>* Added some debug logs to help analyzing missing mapping issues
>
>Notes:
>1. The global bridge device struct is used by adding an 'extern'
>   to proc. This issue should be handled in a different patch series
>   (the struct should not be global. instead, it should be accessible
>   to the dspbridge code via one of the context objects. This way we
>   will also be able to transform pr_* prints to dev_* prints).
>2. The patchset was tested with testsuite, DMM sample app and varios
>    MM and recovery scenarios. Many thanks to Ivan Gomez Castellanos
>    for the help here. Also Many thanks to Felipe Contreras for the
>    thorough review, comments and testing.
>
>Thanks,
>Ohad.
>
>---
>If you want, you can also reach me at <  ohadb at ti dot com  >.
>
>Ohad Ben-Cohen (7):
>  DSPBRIDGE: enhance dmm_map_object
>  DSPBRIDGE: maintain mapping and page info
>  DSPBRIDGE: do not call follow_page
>  DSPBRIDGE: do not use low level cache manipulation API
>  DSPBRIDGE: remove mem_flush_cache
>  DSPBRIDGE: add dspbridge API to mark end of DMA
>  DSPBRIDGE: add new PROC_BEGINDMA and PROC_ENDDMA ioctls
>
> arch/arm/plat-omap/include/dspbridge/drv.h         |   32 +-
> .../arm/plat-omap/include/dspbridge/dspapi-ioctl.h |    9 +
> arch/arm/plat-omap/include/dspbridge/dspapi.h      |    2 +
> arch/arm/plat-omap/include/dspbridge/dspdefs.h     |    3 +-
> arch/arm/plat-omap/include/dspbridge/proc.h        |   29 ++
> drivers/dsp/bridge/core/io_sm.c                    |   11 +-
> drivers/dsp/bridge/core/tiomap3430.c               |    9 +-
> drivers/dsp/bridge/pmgr/dspapi.c                   |   34 ++-
> drivers/dsp/bridge/rmgr/drv.c                      |   33 --
> drivers/dsp/bridge/rmgr/proc.c                     |  410 ++++++++++++++++----
> 10 files changed, 433 insertions(+), 139 deletions(-)

Pushed to dspbridge.

- omar