Hello Maxime Ripard,
This is a semi-automatic email about new static checker warnings.
The patch 67a6eedc4d2c: "dmaengine: xdmac: Add scatter gathered
memset support" from Jul 6, 2015, leads to the following Smatch
complaint:
drivers/dma/at_xdmac.c:1330 at_xdmac_prep_dma_memset_sg()
error: we previously assumed 'psg' could be null (see line 1242)
drivers/dma/at_xdmac.c
1241 pstride = stride;
1242 if (psg)
^^^
psg is non-NULL if we have gone through the loop one iteration.
1243 stride = sg_dma_address(sg) -
1244 (sg_dma_address(psg) + sg_dma_len(psg));
1245
1246 /*
1247 * The scatterlist API gives us only the address and
1248 * length of each elements.
1249 *
1250 * Unfortunately, we don't have the stride, which we
1251 * will need to compute.
1252 *
1253 * That make us end up in a situation like this one:
1254 * len stride len stride len
1255 * +-------+ +-------+ +-------+
1256 * | N-2 | | N-1 | | N |
1257 * +-------+ +-------+ +-------+
1258 *
1259 * We need all these three elements (N-2, N-1 and N)
1260 * to actually take the decision on whether we need to
1261 * queue N-1 or reuse N-2.
1262 *
1263 * We will only consider N if it is the last element.
1264 */
1265 if (ppdesc && pdesc) {
1266 if ((stride == pstride) &&
1267 (sg_dma_len(ppsg) == sg_dma_len(psg))) {
1268 dev_dbg(chan2dev(chan),
1269 "%s: desc 0x%p can be merged with desc 0x%p\n",
1270 __func__, pdesc, ppdesc);
1271
1272 /*
1273 * Increment the block count of the
1274 * N-2 descriptor
1275 */
1276 at_xdmac_increment_block_count(chan, ppdesc);
1277 ppdesc->lld.mbr_dus = stride;
1278
1279 /*
1280 * Put back the N-1 descriptor in the
1281 * free descriptor list
1282 */
1283 list_add_tail(&pdesc->desc_node,
1284 &atchan->free_descs_list);
1285
1286 /*
1287 * Make our N-1 descriptor pointer
1288 * point to the N-2 since they were
1289 * actually merged.
1290 */
1291 pdesc = ppdesc;
1292
1293 /*
1294 * Rule out the case where we don't have
1295 * pstride computed yet (our second sg
1296 * element)
1297 *
1298 * We also want to catch the case where there
1299 * would be a negative stride,
1300 */
1301 } else if (pstride ||
1302 sg_dma_address(sg) < sg_dma_address(psg)) {
1303 /*
1304 * Queue the N-1 descriptor after the
1305 * N-2
1306 */
1307 at_xdmac_queue_desc(chan, ppdesc, pdesc);
1308
1309 /*
1310 * Add the N-1 descriptor to the list
1311 * of the descriptors used for this
1312 * transfer
1313 */
1314 list_add_tail(&desc->desc_node,
1315 &first->descs_list);
1316 dev_dbg(chan2dev(chan),
1317 "%s: add desc 0x%p to descs_list 0x%p\n",
1318 __func__, desc, first);
1319 }
1320 }
1321
1322 /*
1323 * If we are the last element, just see if we have the
1324 * same size than the previous element.
1325 *
1326 * If so, we can merge it with the previous descriptor
1327 * since we don't care about the stride anymore.
1328 */
1329 if ((i == (sg_len - 1)) &&
^^^^^^^^^^^^^^^^^^
1330 sg_dma_len(ppsg) == sg_dma_len(psg)) {
^^^^^^^^^^^^^^^
I guess the concern is if there is only one item on the list. Not sure
if that can happen or not...
1331 dev_dbg(chan2dev(chan),
1332 "%s: desc 0x%p can be merged with desc 0x%p\n",
regards,
dan carpenter
--
To unsubscribe from this list: send the line "unsubscribe dmaengine" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at http://vger.kernel.org/majordomo-info.html
