media: venus: don't abuse dma_alloc for non-DMA allocations
authorStanimir Varbanov <stanimir.varbanov@linaro.org>
Wed, 19 Jul 2017 11:51:37 +0000 (07:51 -0400)
committerMauro Carvalho Chehab <mchehab@s-opensource.com>
Wed, 26 Jul 2017 10:14:32 +0000 (06:14 -0400)
In venus_boot(), we pass a pointer to a phys_addr_t
into dmam_alloc_coherent, which the compiler warns about:

platform/qcom/venus/firmware.c: In function 'venus_boot':
platform/qcom/venus/firmware.c:63:49: error: passing argument 3 of 'dmam_alloc_coherent' from incompatible pointer type [-Werror=incompatible-pointer-types]

To avoid the error refactor venus_boot function by discard
dma_alloc_coherent invocation because we don't want to map the
memory for the device.  Something more, the usage of
DMA mapping API is actually wrong and the current
implementation relies on several bugs in DMA mapping code.
When these bugs are fixed that will break firmware loading,
so fix this now to avoid future troubles.

The meaning of venus_boot is to copy the content of the
firmware buffer into reserved (and memblock removed)
block of memory and pass that physical address to the
trusted zone for authentication and mapping through iommu
form the secure world. After iommu mapping is done the iova
is passed as ane entry point to the remote processor.

After this change memory-region property is parsed manually
and the physical address is memremap to CPU, call mdt_load to
load firmware segments into proper places and unmap
reserved memory.

Fixes: af2c3834c8ca ("[media] media: venus: adding core part and helper functions")

Signed-off-by: Stanimir Varbanov <stanimir.varbanov@linaro.org>
Reviewed-by: Arnd Bergmann <arnd@arndb.de>
Signed-off-by: Hans Verkuil <hans.verkuil@cisco.com>
Signed-off-by: Mauro Carvalho Chehab <mchehab@s-opensource.com>
drivers/media/platform/qcom/venus/core.c
drivers/media/platform/qcom/venus/core.h
drivers/media/platform/qcom/venus/firmware.c
drivers/media/platform/qcom/venus/firmware.h

index 47f79637938c31b2632e671394f2051a61f18944..41eef376eb2da69dfa710a1d5113b123998e1b4f 100644 (file)
@@ -76,7 +76,7 @@ static void venus_sys_error_handler(struct work_struct *work)
        hfi_core_deinit(core, true);
        hfi_destroy(core);
        mutex_lock(&core->lock);
-       venus_shutdown(&core->dev_fw);
+       venus_shutdown(core->dev);
 
        pm_runtime_put_sync(core->dev);
 
@@ -84,7 +84,7 @@ static void venus_sys_error_handler(struct work_struct *work)
 
        pm_runtime_get_sync(core->dev);
 
-       ret |= venus_boot(core->dev, &core->dev_fw, core->res->fwname);
+       ret |= venus_boot(core->dev, core->res->fwname);
 
        ret |= hfi_core_resume(core, true);
 
@@ -207,7 +207,7 @@ static int venus_probe(struct platform_device *pdev)
        if (ret < 0)
                goto err_runtime_disable;
 
-       ret = venus_boot(dev, &core->dev_fw, core->res->fwname);
+       ret = venus_boot(dev, core->res->fwname);
        if (ret)
                goto err_runtime_disable;
 
@@ -238,7 +238,7 @@ err_dev_unregister:
 err_core_deinit:
        hfi_core_deinit(core, false);
 err_venus_shutdown:
-       venus_shutdown(&core->dev_fw);
+       venus_shutdown(dev);
 err_runtime_disable:
        pm_runtime_set_suspended(dev);
        pm_runtime_disable(dev);
@@ -259,7 +259,7 @@ static int venus_remove(struct platform_device *pdev)
        WARN_ON(ret);
 
        hfi_destroy(core);
-       venus_shutdown(&core->dev_fw);
+       venus_shutdown(dev);
        of_platform_depopulate(dev);
 
        pm_runtime_put_sync(dev);
index e542700eee32147ee7e58cceb7cd8e17c4d3d995..cba092bcb76ddaf5e43f6e499e8365824fbe8e07 100644 (file)
@@ -101,7 +101,6 @@ struct venus_core {
        struct device *dev;
        struct device *dev_dec;
        struct device *dev_enc;
-       struct device dev_fw;
        struct mutex lock;
        struct list_head instances;
        atomic_t insts_count;
index 1ede92300dd6e9951cb610c1be24d8f03f04720d..521d4b36c0904d1e6f25e8db688216108ad61c40 100644 (file)
  *
  */
 
-#include <linux/dma-mapping.h>
+#include <linux/device.h>
 #include <linux/firmware.h>
 #include <linux/kernel.h>
+#include <linux/io.h>
 #include <linux/of.h>
-#include <linux/of_reserved_mem.h>
-#include <linux/slab.h>
+#include <linux/of_address.h>
 #include <linux/qcom_scm.h>
+#include <linux/sizes.h>
 #include <linux/soc/qcom/mdt_loader.h>
 
 #include "firmware.h"
 
 #define VENUS_PAS_ID                   9
-#define VENUS_FW_MEM_SIZE              SZ_8M
+#define VENUS_FW_MEM_SIZE              (6 * SZ_1M)
 
-static void device_release_dummy(struct device *dev)
-{
-       of_reserved_mem_device_release(dev);
-}
-
-int venus_boot(struct device *parent, struct device *fw_dev, const char *fwname)
+int venus_boot(struct device *dev, const char *fwname)
 {
        const struct firmware *mdt;
+       struct device_node *node;
        phys_addr_t mem_phys;
+       struct resource r;
        ssize_t fw_size;
        size_t mem_size;
        void *mem_va;
@@ -43,66 +41,58 @@ int venus_boot(struct device *parent, struct device *fw_dev, const char *fwname)
        if (!IS_ENABLED(CONFIG_QCOM_MDT_LOADER) || !qcom_scm_is_available())
                return -EPROBE_DEFER;
 
-       fw_dev->parent = parent;
-       fw_dev->release = device_release_dummy;
+       node = of_parse_phandle(dev->of_node, "memory-region", 0);
+       if (!node) {
+               dev_err(dev, "no memory-region specified\n");
+               return -EINVAL;
+       }
 
-       ret = dev_set_name(fw_dev, "%s:%s", dev_name(parent), "firmware");
+       ret = of_address_to_resource(node, 0, &r);
        if (ret)
                return ret;
 
-       ret = device_register(fw_dev);
-       if (ret < 0)
-               return ret;
+       mem_phys = r.start;
+       mem_size = resource_size(&r);
 
-       ret = of_reserved_mem_device_init_by_idx(fw_dev, parent->of_node, 0);
-       if (ret)
-               goto err_unreg_device;
+       if (mem_size < VENUS_FW_MEM_SIZE)
+               return -EINVAL;
 
-       mem_size = VENUS_FW_MEM_SIZE;
-
-       mem_va = dmam_alloc_coherent(fw_dev, mem_size, &mem_phys, GFP_KERNEL);
+       mem_va = memremap(r.start, mem_size, MEMREMAP_WC);
        if (!mem_va) {
-               ret = -ENOMEM;
-               goto err_unreg_device;
+               dev_err(dev, "unable to map memory region: %pa+%zx\n",
+                       &r.start, mem_size);
+               return -ENOMEM;
        }
 
-       ret = request_firmware(&mdt, fwname, fw_dev);
+       ret = request_firmware(&mdt, fwname, dev);
        if (ret < 0)
-               goto err_unreg_device;
+               goto err_unmap;
 
        fw_size = qcom_mdt_get_size(mdt);
        if (fw_size < 0) {
                ret = fw_size;
                release_firmware(mdt);
-               goto err_unreg_device;
+               goto err_unmap;
        }
 
-       ret = qcom_mdt_load(fw_dev, mdt, fwname, VENUS_PAS_ID, mem_va, mem_phys,
+       ret = qcom_mdt_load(dev, mdt, fwname, VENUS_PAS_ID, mem_va, mem_phys,
                            mem_size);
 
        release_firmware(mdt);
 
        if (ret)
-               goto err_unreg_device;
+               goto err_unmap;
 
        ret = qcom_scm_pas_auth_and_reset(VENUS_PAS_ID);
        if (ret)
-               goto err_unreg_device;
-
-       return 0;
+               goto err_unmap;
 
-err_unreg_device:
-       device_unregister(fw_dev);
+err_unmap:
+       memunmap(mem_va);
        return ret;
 }
 
-int venus_shutdown(struct device *fw_dev)
+int venus_shutdown(struct device *dev)
 {
-       int ret;
-
-       ret = qcom_scm_pas_shutdown(VENUS_PAS_ID);
-       device_unregister(fw_dev);
-       memset(fw_dev, 0, sizeof(*fw_dev));
-
-       return ret;
+       return qcom_scm_pas_shutdown(VENUS_PAS_ID);
 }
index f81a98979798c9b3136f4db5f7ed11c14eb3a04f..428efb56d3391d14f66aa8938e691cc226e42868 100644 (file)
@@ -16,8 +16,7 @@
 
 struct device;
 
-int venus_boot(struct device *parent, struct device *fw_dev,
-              const char *fwname);
-int venus_shutdown(struct device *fw_dev);
+int venus_boot(struct device *dev, const char *fwname);
+int venus_shutdown(struct device *dev);
 
 #endif