This change enables the usage of persistent header buffers which
are allocated during process initialization and is mapped to CDSP.
This mechanism helps in reducing the RPC overhead for remote calls.
For any user process remote RPC call to CDSP, it is tried to use
persistent header buffers. In case all persistent buffers are
already in use, there is a fallback to cached buffer usage and
allocation using DMA API.
Co-developed-by: Anandu Krishnan E <[email protected]>
Signed-off-by: Anandu Krishnan E <[email protected]>
Signed-off-by: Ekansh Gupta <[email protected]>
---
drivers/misc/fastrpc.c | 215 +++++++++++++++++++++++++++++++++++++++----------
1 file changed, 173 insertions(+), 42 deletions(-)
diff --git a/drivers/misc/fastrpc.c b/drivers/misc/fastrpc.c
index a961a66..0132b63 100644
--- a/drivers/misc/fastrpc.c
+++ b/drivers/misc/fastrpc.c
@@ -44,6 +44,8 @@
/* Maximum buffers cached in cached buffer list */
#define FASTRPC_MAX_CACHED_BUFS (32)
#define FASTRPC_MAX_CACHE_BUF_SIZE (8*1024*1024)
+/* Max no. of persistent headers pre-allocated per user process */
+#define FASTRPC_MAX_PERSISTENT_HEADERS (25)
/* Add memory to static PD pool, protection thru XPU */
#define ADSP_MMAP_HEAP_ADDR 4
@@ -54,7 +56,9 @@
/* Add memory to userPD pool, for user heap */
#define ADSP_MMAP_ADD_PAGES 0x1000
/* Add memory to userPD pool, for LLC heap */
-#define ADSP_MMAP_ADD_PAGES_LLC 0x3000,
+#define ADSP_MMAP_ADD_PAGES_LLC 0x3000
+/* Map persistent header buffer on DSP */
+#define ADSP_MMAP_PERSIST_HDR 0x4000
#define DSP_UNSUPPORTED_API (0x80000414)
/* MAX NUMBER of DSP ATTRIBUTES SUPPORTED */
@@ -208,6 +212,7 @@ struct fastrpc_buf {
/* mmap support */
struct list_head node; /* list of user requested mmaps */
uintptr_t raddr;
+ bool in_use;
};
struct fastrpc_dma_buf_attachment {
@@ -303,11 +308,17 @@ struct fastrpc_user {
struct fastrpc_channel_ctx *cctx;
struct fastrpc_session_ctx *sctx;
struct fastrpc_buf *init_mem;
+ /* Pre-allocated header buffer */
+ struct fastrpc_buf *pers_hdr_buf;
+ /* Pre-allocated buffer divided into N chunks */
+ struct fastrpc_buf *hdr_bufs;
int tgid;
int pd;
/* total cached buffers */
u32 num_cached_buf;
+ /* total persistent headers */
+ u32 num_pers_hdrs;
bool is_secure_dev;
/* Lock for lists */
spinlock_t lock;
@@ -399,6 +410,37 @@ static int fastrpc_map_lookup(struct fastrpc_user *fl, int fd,
return ret;
}
+static bool fastrpc_get_persistent_buf(struct fastrpc_user *fl,
+ size_t size, int buf_type, struct fastrpc_buf **obuf)
+{
+ u32 i = 0;
+ bool found = false;
+ struct fastrpc_buf *buf = NULL;
+
+ spin_lock(&fl->lock);
+ /*
+ * Persistent header buffer can be used only if
+ * metadata length is less than 1 page size.
+ */
+ if (!fl->num_pers_hdrs || buf_type != METADATA_BUF || size > PAGE_SIZE) {
+ spin_unlock(&fl->lock);
+ return found;
+ }
+
+ for (i = 0; i < fl->num_pers_hdrs; i++) {
+ buf = &fl->hdr_bufs[i];
+ /* If buffer not in use, then assign it for requested alloc */
+ if (!buf->in_use) {
+ buf->in_use = true;
+ *obuf = buf;
+ found = true;
+ break;
+ }
+ }
+ spin_unlock(&fl->lock);
+ return found;
+}
+
static void __fastrpc_buf_free(struct fastrpc_buf *buf)
{
dma_free_coherent(buf->dev, buf->size, buf->virt,
@@ -430,6 +472,15 @@ static void fastrpc_cached_buf_list_add(struct fastrpc_buf *buf)
static void fastrpc_buf_free(struct fastrpc_buf *buf, bool cache)
{
+ struct fastrpc_user *fl = buf->fl;
+
+ if (buf->in_use) {
+ /* Don't free persistent header buf. Just mark as available */
+ spin_lock(&fl->lock);
+ buf->in_use = false;
+ spin_unlock(&fl->lock);
+ return;
+ }
if (cache)
fastrpc_cached_buf_list_add(buf);
else
@@ -523,6 +574,8 @@ static int fastrpc_buf_alloc(struct fastrpc_user *fl, struct device *dev,
int ret;
struct fastrpc_buf *buf;
+ if (fastrpc_get_persistent_buf(fl, size, buf_type, obuf))
+ return 0;
if (fastrpc_get_cached_buf(fl, size, buf_type, obuf))
return 0;
ret = __fastrpc_buf_alloc(fl, dev, size, obuf);
@@ -1305,6 +1358,107 @@ static int fastrpc_internal_invoke(struct fastrpc_user *fl, u32 kernel,
return err;
}
+static int fastrpc_mem_map_to_dsp(struct fastrpc_user *fl, int fd, int offset,
+ u32 flags, u32 va, u64 phys,
+ size_t size, uintptr_t *raddr)
+{
+ struct fastrpc_invoke_args args[4] = { [0 ... 3] = { 0 } };
+ struct fastrpc_mem_map_req_msg req_msg = { 0 };
+ struct fastrpc_mmap_rsp_msg rsp_msg = { 0 };
+ struct fastrpc_phy_page pages = { 0 };
+ struct device *dev = fl->sctx->dev;
+ int err = 0;
+ u32 sc;
+
+ req_msg.pgid = fl->tgid;
+ req_msg.fd = fd;
+ req_msg.offset = offset;
+ req_msg.vaddrin = va;
+ req_msg.flags = flags;
+ req_msg.num = sizeof(pages);
+ req_msg.data_len = 0;
+
+ args[0].ptr = (u64) (uintptr_t) &req_msg;
+ args[0].length = sizeof(req_msg);
+
+ pages.addr = phys;
+ pages.size = size;
+
+ args[1].ptr = (u64) (uintptr_t) &pages;
+ args[1].length = sizeof(pages);
+
+ args[2].ptr = (u64) (uintptr_t) &pages;
+ args[2].length = 0;
+
+ args[3].ptr = (u64) (uintptr_t) &rsp_msg;
+ args[3].length = sizeof(rsp_msg);
+
+ sc = FASTRPC_SCALARS(FASTRPC_RMID_INIT_MEM_MAP, 3, 1);
+ err = fastrpc_internal_invoke(fl, true, FASTRPC_INIT_HANDLE, sc, &args[0]);
+ if (err) {
+ dev_err(dev, "mem mmap error, fd %d, vaddr %llx, size %lld\n",
+ fd, va, size);
+ return err;
+ }
+ *raddr = rsp_msg.vaddr;
+
+ return 0;
+}
+
+static int fastrpc_create_persistent_headers(struct fastrpc_user *fl)
+{
+ int err = 0;
+ int i = 0;
+ u64 virtb = 0;
+ struct device *dev = fl->sctx->dev;
+ struct fastrpc_buf *hdr_bufs, *buf, *pers_hdr_buf = NULL;
+ u32 num_pers_hdrs = 0;
+ size_t hdr_buf_alloc_len = 0;
+
+ /*
+ * Pre-allocate memory for persistent header buffers based
+ * on concurrency info passed by user. Upper limit enforced.
+ */
+ num_pers_hdrs = FASTRPC_MAX_PERSISTENT_HEADERS;
+ hdr_buf_alloc_len = num_pers_hdrs * PAGE_SIZE;
+ err = fastrpc_buf_alloc(fl, dev, hdr_buf_alloc_len,
+ METADATA_BUF, &pers_hdr_buf);
+ if (err)
+ return err;
+
+ virtb = (u64) (uintptr_t)(pers_hdr_buf->virt);
+ err = fastrpc_mem_map_to_dsp(fl, -1, 0,
+ ADSP_MMAP_PERSIST_HDR, 0, (u64) (uintptr_t)(pers_hdr_buf->phys),
+ pers_hdr_buf->size, &pers_hdr_buf->raddr);
+ if (err)
+ goto err_dsp_map;
+
+ hdr_bufs = kcalloc(num_pers_hdrs, sizeof(struct fastrpc_buf),
+ GFP_KERNEL);
+ if (!hdr_bufs)
+ return -ENOMEM;
+
+ spin_lock(&fl->lock);
+ fl->pers_hdr_buf = pers_hdr_buf;
+ fl->num_pers_hdrs = num_pers_hdrs;
+ fl->hdr_bufs = hdr_bufs;
+ for (i = 0; i < num_pers_hdrs; i++) {
+ buf = &fl->hdr_bufs[i];
+ buf->fl = fl;
+ buf->virt = (void *)(virtb + (i * PAGE_SIZE));
+ buf->phys = pers_hdr_buf->phys + (i * PAGE_SIZE);
+ buf->size = PAGE_SIZE;
+ buf->type = pers_hdr_buf->type;
+ buf->in_use = false;
+ }
+ spin_unlock(&fl->lock);
+
+ return 0;
+err_dsp_map:
+ fastrpc_buf_free(pers_hdr_buf, 0);
+ return err;
+}
+
static bool is_session_rejected(struct fastrpc_user *fl, bool unsigned_pd_request)
{
/* Check if the device node is non-secure and channel is secure*/
@@ -1537,6 +1691,12 @@ static int fastrpc_init_create_process(struct fastrpc_user *fl,
if (err)
goto err_invoke;
+ if (fl->cctx->domain_id == CDSP_DOMAIN_ID) {
+ err = fastrpc_create_persistent_headers(fl);
+ if (err)
+ goto err_invoke;
+ }
+
kfree(args);
return 0;
@@ -1629,6 +1789,10 @@ static int fastrpc_device_release(struct inode *inode, struct file *file)
fastrpc_buf_free(buf, false);
}
+ if (fl->pers_hdr_buf)
+ fastrpc_buf_free(fl->pers_hdr_buf, false);
+ kfree(fl->hdr_bufs);
+
fastrpc_cached_buf_list_free(fl);
fastrpc_session_free(cctx, fl->sctx);
fastrpc_channel_ctx_put(cctx);
@@ -2089,16 +2253,11 @@ static int fastrpc_req_mem_unmap(struct fastrpc_user *fl, char __user *argp)
static int fastrpc_req_mem_map(struct fastrpc_user *fl, char __user *argp)
{
- struct fastrpc_invoke_args args[4] = { [0 ... 3] = { 0 } };
- struct fastrpc_mem_map_req_msg req_msg = { 0 };
- struct fastrpc_mmap_rsp_msg rsp_msg = { 0 };
struct fastrpc_mem_unmap req_unmap = { 0 };
- struct fastrpc_phy_page pages = { 0 };
- struct fastrpc_mem_map req;
+ struct fastrpc_mem_map req = {0};
struct device *dev = fl->sctx->dev;
struct fastrpc_map *map = NULL;
int err;
- u32 sc;
if (copy_from_user(&req, argp, sizeof(req)))
return -EFAULT;
@@ -2110,57 +2269,29 @@ static int fastrpc_req_mem_map(struct fastrpc_user *fl, char __user *argp)
return err;
}
- req_msg.pgid = fl->tgid;
- req_msg.fd = req.fd;
- req_msg.offset = req.offset;
- req_msg.vaddrin = req.vaddrin;
map->va = (void *) (uintptr_t) req.vaddrin;
- req_msg.flags = req.flags;
- req_msg.num = sizeof(pages);
- req_msg.data_len = 0;
-
- args[0].ptr = (u64) (uintptr_t) &req_msg;
- args[0].length = sizeof(req_msg);
-
- pages.addr = map->phys;
- pages.size = map->size;
-
- args[1].ptr = (u64) (uintptr_t) &pages;
- args[1].length = sizeof(pages);
-
- args[2].ptr = (u64) (uintptr_t) &pages;
- args[2].length = 0;
-
- args[3].ptr = (u64) (uintptr_t) &rsp_msg;
- args[3].length = sizeof(rsp_msg);
-
- sc = FASTRPC_SCALARS(FASTRPC_RMID_INIT_MEM_MAP, 3, 1);
- err = fastrpc_internal_invoke(fl, true, FASTRPC_INIT_HANDLE, sc, &args[0]);
+ /* map to dsp, get virtual adrress for the user*/
+ err = fastrpc_mem_map_to_dsp(fl, map->fd, req.offset,
+ req.flags, req.vaddrin, map->phys,
+ map->size, (uintptr_t *)&req.vaddrout);
if (err) {
- dev_err(dev, "mem mmap error, fd %d, vaddr %llx, size %lld\n",
- req.fd, req.vaddrin, map->size);
+ dev_err(dev, "failed to map buffer on dsp, fd = %d\n", map->fd);
goto err_invoke;
}
/* update the buffer to be able to deallocate the memory on the DSP */
- map->raddr = rsp_msg.vaddr;
-
- /* let the client know the address to use */
- req.vaddrout = rsp_msg.vaddr;
+ map->raddr = req.vaddrout;
if (copy_to_user((void __user *)argp, &req, sizeof(req))) {
/* unmap the memory and release the buffer */
- req_unmap.vaddr = (uintptr_t) rsp_msg.vaddr;
+ req_unmap.vaddr = (uintptr_t)req.vaddrout;
req_unmap.length = map->size;
fastrpc_req_mem_unmap_impl(fl, &req_unmap);
return -EFAULT;
}
-
return 0;
-
err_invoke:
fastrpc_map_put(map);
-
return err;
}
--
2.7.4
Hi Ekansh,
kernel test robot noticed the following build warnings:
[auto build test WARNING on char-misc/char-misc-testing]
[also build test WARNING on char-misc/char-misc-next char-misc/char-misc-linus linus/master v6.4-rc5 next-20230609]
[If your patch is applied to the wrong git tree, kindly drop us a note.
And when submitting patch, we suggest to use '--base' as documented in
https://git-scm.com/docs/git-format-patch#_base_tree_information]
url: https://github.com/intel-lab-lkp/linux/commits/Ekansh-Gupta/misc-fastrpc-Add-cached-buffer-support/20230609-184517
base: char-misc/char-misc-testing
patch link: https://lore.kernel.org/r/1686307187-15199-3-git-send-email-quic_ekangupt%40quicinc.com
patch subject: [PATCH v1 2/2] misc: fastrpc: Add persistent header buffer support
config: arm64-randconfig-r024-20230608 (https://download.01.org/0day-ci/archive/20230610/[email protected]/config)
compiler: clang version 14.0.6 (https://github.com/llvm/llvm-project.git f28c006a5895fc0e329fe15fead81e37457cb1d1)
reproduce (this is a W=1 build):
mkdir -p ~/bin
wget https://raw.githubusercontent.com/intel/lkp-tests/master/sbin/make.cross -O ~/bin/make.cross
chmod +x ~/bin/make.cross
# install arm64 cross compiling tool for clang build
# apt-get install binutils-aarch64-linux-gnu
git remote add char-misc https://git.kernel.org/pub/scm/linux/kernel/git/gregkh/char-misc.git
git fetch char-misc char-misc-testing
git checkout char-misc/char-misc-testing
b4 shazam https://lore.kernel.org/r/[email protected]
# save the config file
mkdir build_dir && cp config build_dir/.config
COMPILER_INSTALL_PATH=$HOME/0day COMPILER=clang ~/bin/make.cross W=1 O=build_dir ARCH=arm64 olddefconfig
COMPILER_INSTALL_PATH=$HOME/0day COMPILER=clang ~/bin/make.cross W=1 O=build_dir ARCH=arm64 SHELL=/bin/bash drivers/misc/
If you fix the issue in a separate patch/commit (i.e. not just a new version of
the same patch/commit), kindly add following tags
| Reported-by: kernel test robot <[email protected]>
| Closes: https://lore.kernel.org/oe-kbuild-all/[email protected]/
All warnings (new ones prefixed by >>):
>> drivers/misc/fastrpc.c:1400:8: warning: format specifies type 'unsigned long long' but the argument has type 'u32' (aka 'unsigned int') [-Wformat]
fd, va, size);
^~
include/linux/dev_printk.h:144:65: note: expanded from macro 'dev_err'
dev_printk_index_wrap(_dev_err, KERN_ERR, dev, dev_fmt(fmt), ##__VA_ARGS__)
~~~ ^~~~~~~~~~~
include/linux/dev_printk.h:110:23: note: expanded from macro 'dev_printk_index_wrap'
_p_func(dev, fmt, ##__VA_ARGS__); \
~~~ ^~~~~~~~~~~
>> drivers/misc/fastrpc.c:1400:12: warning: format specifies type 'long long' but the argument has type 'size_t' (aka 'unsigned long') [-Wformat]
fd, va, size);
^~~~
include/linux/dev_printk.h:144:65: note: expanded from macro 'dev_err'
dev_printk_index_wrap(_dev_err, KERN_ERR, dev, dev_fmt(fmt), ##__VA_ARGS__)
~~~ ^~~~~~~~~~~
include/linux/dev_printk.h:110:23: note: expanded from macro 'dev_printk_index_wrap'
_p_func(dev, fmt, ##__VA_ARGS__); \
~~~ ^~~~~~~~~~~
2 warnings generated.
vim +1400 drivers/misc/fastrpc.c
1360
1361 static int fastrpc_mem_map_to_dsp(struct fastrpc_user *fl, int fd, int offset,
1362 u32 flags, u32 va, u64 phys,
1363 size_t size, uintptr_t *raddr)
1364 {
1365 struct fastrpc_invoke_args args[4] = { [0 ... 3] = { 0 } };
1366 struct fastrpc_mem_map_req_msg req_msg = { 0 };
1367 struct fastrpc_mmap_rsp_msg rsp_msg = { 0 };
1368 struct fastrpc_phy_page pages = { 0 };
1369 struct device *dev = fl->sctx->dev;
1370 int err = 0;
1371 u32 sc;
1372
1373 req_msg.pgid = fl->tgid;
1374 req_msg.fd = fd;
1375 req_msg.offset = offset;
1376 req_msg.vaddrin = va;
1377 req_msg.flags = flags;
1378 req_msg.num = sizeof(pages);
1379 req_msg.data_len = 0;
1380
1381 args[0].ptr = (u64) (uintptr_t) &req_msg;
1382 args[0].length = sizeof(req_msg);
1383
1384 pages.addr = phys;
1385 pages.size = size;
1386
1387 args[1].ptr = (u64) (uintptr_t) &pages;
1388 args[1].length = sizeof(pages);
1389
1390 args[2].ptr = (u64) (uintptr_t) &pages;
1391 args[2].length = 0;
1392
1393 args[3].ptr = (u64) (uintptr_t) &rsp_msg;
1394 args[3].length = sizeof(rsp_msg);
1395
1396 sc = FASTRPC_SCALARS(FASTRPC_RMID_INIT_MEM_MAP, 3, 1);
1397 err = fastrpc_internal_invoke(fl, true, FASTRPC_INIT_HANDLE, sc, &args[0]);
1398 if (err) {
1399 dev_err(dev, "mem mmap error, fd %d, vaddr %llx, size %lld\n",
> 1400 fd, va, size);
1401 return err;
1402 }
1403 *raddr = rsp_msg.vaddr;
1404
1405 return 0;
1406 }
1407
--
0-DAY CI Kernel Test Service
https://github.com/intel/lkp-tests/wiki