erofs: tidy up z_erofs_lz4_handle_overlap()

 - Add some useful comments to explain inplace I/Os and decompression;

 - Rearrange the code to get rid of one unnecessary goto.

Reviewed-by: Chao Yu <chao@kernel.org>
Signed-off-by: Gao Xiang <hsiangkao@linux.alibaba.com>

Author: hsiangkao

fs/erofs/decompressor.c

@@ -105,65 +105,72 @@ static int z_erofs_lz4_prepare_dstpages(struct z_erofs_decompress_req *rq,
 	return kaddr ? 1 : 0;
 }
 
-static void *z_erofs_lz4_handle_overlap(struct z_erofs_decompress_req *rq,
+static void *z_erofs_lz4_handle_overlap(const struct z_erofs_decompress_req *rq,
 			void *inpage, void *out, unsigned int *inputmargin,
 			int *maptype, bool may_inplace)
 {
-	unsigned int oend, omargin, total, i;
+	unsigned int oend, omargin, cnt, i;
 	struct page **in;
-	void *src, *tmp;
-
-	if (rq->inplace_io) {
-		oend = rq->pageofs_out + rq->outputsize;
-		omargin = PAGE_ALIGN(oend) - oend;
-		if (rq->partial_decoding || !may_inplace ||
-		    omargin < LZ4_DECOMPRESS_INPLACE_MARGIN(rq->inputsize))
-			goto docopy;
+	void *src;
 
+	/*
+	 * If in-place I/O isn't used, for example, the bounce compressed cache
+	 * can hold data for incomplete read requests. Just map the compressed
+	 * buffer as well and decompress directly.
+	 */
+	if (!rq->inplace_io) {
+		if (rq->inpages <= 1) {
+			*maptype = 0;
+			return inpage;
+		}
+		kunmap_local(inpage);
+		src = erofs_vm_map_ram(rq->in, rq->inpages);
+		if (!src)
+			return ERR_PTR(-ENOMEM);
+		*maptype = 1;
+		return src;
+	}
+	/*
+	 * Then, deal with in-place I/Os. The reasons why in-place I/O is useful
+	 * are: (1) It minimizes memory footprint during the I/O submission,
+	 * which is useful for slow storage (including network devices and
+	 * low-end HDDs/eMMCs) but with a lot inflight I/Os; (2) If in-place
+	 * decompression can also be applied, it will reuse the unique buffer so
+	 * that no extra CPU D-cache is polluted with temporary compressed data
+	 * for extreme performance.
+	 */
+	oend = rq->pageofs_out + rq->outputsize;
+	omargin = PAGE_ALIGN(oend) - oend;
+	if (!rq->partial_decoding && may_inplace &&
+	    omargin >= LZ4_DECOMPRESS_INPLACE_MARGIN(rq->inputsize)) {
 		for (i = 0; i < rq->inpages; ++i)
 			if (rq->out[rq->outpages - rq->inpages + i] !=
 			    rq->in[i])
-				goto docopy;
-		kunmap_local(inpage);
-		*maptype = 3;
-		return out + ((rq->outpages - rq->inpages) << PAGE_SHIFT);
-	}
-
-	if (rq->inpages <= 1) {
-		*maptype = 0;
-		return inpage;
+				break;
+		if (i >= rq->inpages) {
+			kunmap_local(inpage);
+			*maptype = 3;
+			return out + ((rq->outpages - rq->inpages) << PAGE_SHIFT);
+		}
 	}
-	kunmap_local(inpage);
-	src = erofs_vm_map_ram(rq->in, rq->inpages);
-	if (!src)
-		return ERR_PTR(-ENOMEM);
-	*maptype = 1;
-	return src;
-
-docopy:
-	/* Or copy compressed data which can be overlapped to per-CPU buffer */
-	in = rq->in;
+	/*
+	 * If in-place decompression can't be applied, copy compressed data that
+	 * may potentially overlap during decompression to a per-CPU buffer.
+	 */
 	src = z_erofs_get_gbuf(rq->inpages);
 	if (!src) {
 		DBG_BUGON(1);
 		kunmap_local(inpage);
 		return ERR_PTR(-EFAULT);
 	}
 
-	tmp = src;
-	total = rq->inputsize;
-	while (total) {
-		unsigned int page_copycnt =
-			min_t(unsigned int, total, PAGE_SIZE - *inputmargin);
-
+	for (i = 0, in = rq->in; i < rq->inputsize; i += cnt, ++in) {
+		cnt = min_t(u32, rq->inputsize - i, PAGE_SIZE - *inputmargin);
 		if (!inpage)
 			inpage = kmap_local_page(*in);
-		memcpy(tmp, inpage + *inputmargin, page_copycnt);
+		memcpy(src + i, inpage + *inputmargin, cnt);
 		kunmap_local(inpage);
 		inpage = NULL;
-		tmp += page_copycnt;
-		total -= page_copycnt;
-		++in;
 		*inputmargin = 0;
 	}
 	*maptype = 2;