* [PATCH qemu-trad] HVM: atomically access pointers in bufioreq handling
@ 2015-06-22 11:50 Jan Beulich
2015-07-07 16:03 ` Ping: " Jan Beulich
0 siblings, 1 reply; 2+ messages in thread
From: Jan Beulich @ 2015-06-22 11:50 UTC (permalink / raw)
To: xen-devel; +Cc: Ian Jackson
[-- Attachment #1: Type: text/plain, Size: 2415 bytes --]
The number of slots per page being 511 (i.e. not a power of two) means
that the (32-bit) read and write indexes going beyond 2^32 will likely
disturb operation. The hypervisor side gets I/O req server creation
extended so we can indicate that we're using suitable atomic accesses
where needed (not all accesses to the two pointers really need to be
atomic), allowing it to atomically canonicalize both pointers when both
have gone through at least one cycle.
The Xen side counterpart (which is not a functional prereq to this
change, albeit the intention is for Xen to assume default servers
always use suitable atomic accesses) can be found at e.g.
http://lists.xenproject.org/archives/html/xen-devel/2015-06/msg02996.html
Signed-off-by: Jan Beulich <jbeulich@suse.com>
--- a/i386-dm/helper2.c
+++ b/i386-dm/helper2.c
@@ -493,10 +493,19 @@ static int __handle_buffered_iopage(CPUS
memset(&req, 0x00, sizeof(req));
- while (buffered_io_page->read_pointer !=
- buffered_io_page->write_pointer) {
- buf_req = &buffered_io_page->buf_ioreq[
- buffered_io_page->read_pointer % IOREQ_BUFFER_SLOT_NUM];
+ for (;;) {
+ uint32_t rdptr = buffered_io_page->read_pointer, wrptr;
+
+ xen_rmb();
+ wrptr = buffered_io_page->write_pointer;
+ xen_rmb();
+ if (rdptr != buffered_io_page->read_pointer) {
+ continue;
+ }
+ if (rdptr == wrptr) {
+ break;
+ }
+ buf_req = &buffered_io_page->buf_ioreq[rdptr % IOREQ_BUFFER_SLOT_NUM];
req.size = 1UL << buf_req->size;
req.count = 1;
req.addr = buf_req->addr;
@@ -508,15 +517,14 @@ static int __handle_buffered_iopage(CPUS
req.data_is_ptr = 0;
qw = (req.size == 8);
if (qw) {
- buf_req = &buffered_io_page->buf_ioreq[
- (buffered_io_page->read_pointer+1) % IOREQ_BUFFER_SLOT_NUM];
+ buf_req = &buffered_io_page->buf_ioreq[(rdptr + 1) %
+ IOREQ_BUFFER_SLOT_NUM];
req.data |= ((uint64_t)buf_req->data) << 32;
}
__handle_ioreq(env, &req);
- xen_mb();
- buffered_io_page->read_pointer += qw ? 2 : 1;
+ __sync_fetch_and_add(&buffered_io_page->read_pointer, qw + 1);
}
return req.count;
[-- Attachment #2: qemut-bufioreq-atomic-add.patch --]
[-- Type: text/plain, Size: 2464 bytes --]
HVM: atomically access pointers in bufioreq handling
The number of slots per page being 511 (i.e. not a power of two) means
that the (32-bit) read and write indexes going beyond 2^32 will likely
disturb operation. The hypervisor side gets I/O req server creation
extended so we can indicate that we're using suitable atomic accesses
where needed (not all accesses to the two pointers really need to be
atomic), allowing it to atomically canonicalize both pointers when both
have gone through at least one cycle.
The Xen side counterpart (which is not a functional prereq to this
change, albeit the intention is for Xen to assume default servers
always use suitable atomic accesses) can be found at e.g.
http://lists.xenproject.org/archives/html/xen-devel/2015-06/msg02996.html
Signed-off-by: Jan Beulich <jbeulich@suse.com>
--- a/i386-dm/helper2.c
+++ b/i386-dm/helper2.c
@@ -493,10 +493,19 @@ static int __handle_buffered_iopage(CPUS
memset(&req, 0x00, sizeof(req));
- while (buffered_io_page->read_pointer !=
- buffered_io_page->write_pointer) {
- buf_req = &buffered_io_page->buf_ioreq[
- buffered_io_page->read_pointer % IOREQ_BUFFER_SLOT_NUM];
+ for (;;) {
+ uint32_t rdptr = buffered_io_page->read_pointer, wrptr;
+
+ xen_rmb();
+ wrptr = buffered_io_page->write_pointer;
+ xen_rmb();
+ if (rdptr != buffered_io_page->read_pointer) {
+ continue;
+ }
+ if (rdptr == wrptr) {
+ break;
+ }
+ buf_req = &buffered_io_page->buf_ioreq[rdptr % IOREQ_BUFFER_SLOT_NUM];
req.size = 1UL << buf_req->size;
req.count = 1;
req.addr = buf_req->addr;
@@ -508,15 +517,14 @@ static int __handle_buffered_iopage(CPUS
req.data_is_ptr = 0;
qw = (req.size == 8);
if (qw) {
- buf_req = &buffered_io_page->buf_ioreq[
- (buffered_io_page->read_pointer+1) % IOREQ_BUFFER_SLOT_NUM];
+ buf_req = &buffered_io_page->buf_ioreq[(rdptr + 1) %
+ IOREQ_BUFFER_SLOT_NUM];
req.data |= ((uint64_t)buf_req->data) << 32;
}
__handle_ioreq(env, &req);
- xen_mb();
- buffered_io_page->read_pointer += qw ? 2 : 1;
+ __sync_fetch_and_add(&buffered_io_page->read_pointer, qw + 1);
}
return req.count;
[-- Attachment #3: Type: text/plain, Size: 126 bytes --]
_______________________________________________
Xen-devel mailing list
Xen-devel@lists.xen.org
http://lists.xen.org/xen-devel
^ permalink raw reply [flat|nested] 2+ messages in thread
* Ping: [PATCH qemu-trad] HVM: atomically access pointers in bufioreq handling
2015-06-22 11:50 [PATCH qemu-trad] HVM: atomically access pointers in bufioreq handling Jan Beulich
@ 2015-07-07 16:03 ` Jan Beulich
0 siblings, 0 replies; 2+ messages in thread
From: Jan Beulich @ 2015-07-07 16:03 UTC (permalink / raw)
To: Ian Jackson; +Cc: xen-devel
>>> On 22.06.15 at 13:50, <JBeulich@suse.com> wrote:
> The number of slots per page being 511 (i.e. not a power of two) means
> that the (32-bit) read and write indexes going beyond 2^32 will likely
> disturb operation. The hypervisor side gets I/O req server creation
> extended so we can indicate that we're using suitable atomic accesses
> where needed (not all accesses to the two pointers really need to be
> atomic), allowing it to atomically canonicalize both pointers when both
> have gone through at least one cycle.
>
> The Xen side counterpart (which is not a functional prereq to this
> change, albeit the intention is for Xen to assume default servers
> always use suitable atomic accesses) can be found at e.g.
> http://lists.xenproject.org/archives/html/xen-devel/2015-06/msg02996.html
>
> Signed-off-by: Jan Beulich <jbeulich@suse.com>
>
> --- a/i386-dm/helper2.c
> +++ b/i386-dm/helper2.c
> @@ -493,10 +493,19 @@ static int __handle_buffered_iopage(CPUS
>
> memset(&req, 0x00, sizeof(req));
>
> - while (buffered_io_page->read_pointer !=
> - buffered_io_page->write_pointer) {
> - buf_req = &buffered_io_page->buf_ioreq[
> - buffered_io_page->read_pointer % IOREQ_BUFFER_SLOT_NUM];
> + for (;;) {
> + uint32_t rdptr = buffered_io_page->read_pointer, wrptr;
> +
> + xen_rmb();
> + wrptr = buffered_io_page->write_pointer;
> + xen_rmb();
> + if (rdptr != buffered_io_page->read_pointer) {
> + continue;
> + }
> + if (rdptr == wrptr) {
> + break;
> + }
> + buf_req = &buffered_io_page->buf_ioreq[rdptr % IOREQ_BUFFER_SLOT_NUM];
> req.size = 1UL << buf_req->size;
> req.count = 1;
> req.addr = buf_req->addr;
> @@ -508,15 +517,14 @@ static int __handle_buffered_iopage(CPUS
> req.data_is_ptr = 0;
> qw = (req.size == 8);
> if (qw) {
> - buf_req = &buffered_io_page->buf_ioreq[
> - (buffered_io_page->read_pointer+1) % IOREQ_BUFFER_SLOT_NUM];
> + buf_req = &buffered_io_page->buf_ioreq[(rdptr + 1) %
> + IOREQ_BUFFER_SLOT_NUM];
> req.data |= ((uint64_t)buf_req->data) << 32;
> }
>
> __handle_ioreq(env, &req);
>
> - xen_mb();
> - buffered_io_page->read_pointer += qw ? 2 : 1;
> + __sync_fetch_and_add(&buffered_io_page->read_pointer, qw + 1);
> }
>
> return req.count;
^ permalink raw reply [flat|nested] 2+ messages in thread
end of thread, other threads:[~2015-07-07 16:03 UTC | newest]
Thread overview: 2+ messages (download: mbox.gz / follow: Atom feed)
-- links below jump to the message on this page --
2015-06-22 11:50 [PATCH qemu-trad] HVM: atomically access pointers in bufioreq handling Jan Beulich
2015-07-07 16:03 ` Ping: " Jan Beulich
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox;
as well as URLs for NNTP newsgroup(s).