[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]
Re: [Qemu-devel] [Qemu-block] [RFC PATCH] block: local qiov helper
From: |
Vladimir Sementsov-Ogievskiy |
Subject: |
Re: [Qemu-devel] [Qemu-block] [RFC PATCH] block: local qiov helper |
Date: |
Tue, 29 Jan 2019 10:24:08 +0000 |
29.01.2019 6:31, Stefan Hajnoczi wrote:
> On Fri, Jan 25, 2019 at 07:46:01PM +0300, Vladimir Sementsov-Ogievskiy wrote:
>> Hi all.
>>
>> What about such a simple helper for a very often patter around
>> qemu_iovec_init_external ?
>
> Sounds good, qemu_iovec_init() has 55 references vs
> qemu_iovec_init_external() with 51. It's worth making
> qemu_iovec_init_external() nicer to use.
>
>> If we like it, I'll update other callers of qemu_iovec_init_external.
>>
>> Possible interface change would be
>> LOCAL_QIOV(lc, buf, len);
>> instead of
>> LocalQiov lc = LOCAL_QIOV(lc, buf, len);
>>
>> or, may be, someone has a better idea?
>
> Bike-shedding territory, but I prefer LocalQiov lc = LOCAL_QIOV(lc, buf,
> len) because it reveals the type. This makes the code easier to read
> than just LOCAL_QIOV(lc, buf, len) by itself - the reader is forced to
> look up the macro definition to figure out what magic happens.
>
>> diff --git a/block/io.c b/block/io.c
>> index bd9d688f8b..c7d7b199c1 100644
>> --- a/block/io.c
>> +++ b/block/io.c
>> @@ -949,18 +949,13 @@ int bdrv_preadv(BdrvChild *child, int64_t offset,
>> QEMUIOVector *qiov)
>>
>> int bdrv_pread(BdrvChild *child, int64_t offset, void *buf, int bytes)
>> {
>> - QEMUIOVector qiov;
>> - struct iovec iov = {
>> - .iov_base = (void *)buf,
>> - .iov_len = bytes,
>> - };
>> + LocalQiov lq = LOCAL_QIOV(lq, buf, bytes);
>>
>> if (bytes < 0) {
>> return -EINVAL;
>> }
>>
>> - qemu_iovec_init_external(&qiov, &iov, 1);
>> - return bdrv_preadv(child, offset, &qiov);
>> + return bdrv_preadv(child, offset, &lq.qiov);
>
> I think it's unfortunate that LocalQiov is necessary since the caller
> only needs the qiov. Can we afford to embed the struct iovec into
> QEMUIOVector?
>
> That way callers don't need a separate LocalQiov type:
>
> QEMUIOVector qiov = QEMU_IOVEC_INIT_BUF(qiov, buf, bytes);
> ...
> return bdrv_preadv(child, offset, &qiov);
>
Hmm. In this case we definitely will have tiny extra memory usage, but we gain
beautiful
readability.
So, like this:
diff --git a/include/qemu/iov.h b/include/qemu/iov.h
index 5f433c7768..53de1b38bb 100644
--- a/include/qemu/iov.h
+++ b/include/qemu/iov.h
@@ -134,9 +134,31 @@ typedef struct QEMUIOVector {
struct iovec *iov;
int niov;
int nalloc;
- size_t size;
+ union {
+ struct {
+ void *__unused_iov_base;
+ size_t size;
+ };
+ struct iovec local_iov;
+ };
} QEMUIOVector;
+G_STATIC_ASSERT(offsetof(QEMUIOVector, size) ==
+ offsetof(QEMUIOVector, local_iov.iov_len));
+G_STATIC_ASSERT(sizeof(((QEMUIOVector *)NULL)->size) ==
+ sizeof(((QEMUIOVector *)NULL)->local_iov.iov_len));
+
+#define QEMU_IOVEC_INIT_BUF(self, buf, len) \
+{ \
+ .iov = &self.local_iov, \
+ .niov = 1, \
+ .nalloc = -1, \
+ .local_iov = { \
+ .iov_base = (void *)(buf), \
+ .iov_len = len \
+ } \
+}
+
void qemu_iovec_init(QEMUIOVector *qiov, int alloc_hint);
void qemu_iovec_init_external(QEMUIOVector *qiov, struct iovec *iov, int
niov);
void qemu_iovec_add(QEMUIOVector *qiov, void *base, size_t len);
diff --git a/block/io.c b/block/io.c
index bd9d688f8b..39a1a848af 100644
--- a/block/io.c
+++ b/block/io.c
@@ -949,17 +949,12 @@ int bdrv_preadv(BdrvChild *child, int64_t offset,
QEMUIOVector *qiov)
int bdrv_pread(BdrvChild *child, int64_t offset, void *buf, int bytes)
{
- QEMUIOVector qiov;
- struct iovec iov = {
- .iov_base = (void *)buf,
- .iov_len = bytes,
- };
+ QEMUIOVector qiov = QEMU_IOVEC_INIT_BUF(qiov, buf, bytes);
if (bytes < 0) {
return -EINVAL;
}
- qemu_iovec_init_external(&qiov, &iov, 1);
return bdrv_preadv(child, offset, &qiov);
}
Ok?
It's also possible to unite nalloc and local_iov.iov_base, to save 4 bytes, and
we'll have to add
void qemu_iovec_add(QEMUIOVector *qiov, void *base, size_t len)
{
- assert(qiov->nalloc != -1);
+ assert(qiov->iov != &qiov->local_iov && qiov->nalloc != -1);
But I think it's not worth it.
--
Best regards,
Vladimir