Re: [RFC PATCH 1/12] x86/Hyper-V: Add visibility parameter for vmbus_establish_gpadl()

From: Tianyu Lan
Date: Fri Mar 05 2021 - 01:06:38 EST


Hi Vitaly:
Thanks for your review.

On 3/4/2021 12:27 AM, Vitaly Kuznetsov wrote:
Tianyu Lan <ltykernel@xxxxxxxxx> writes:

From: Tianyu Lan <Tianyu.Lan@xxxxxxxxxxxxx>

Add visibility parameter for vmbus_establish_gpadl() and prepare
to change host visibility when create gpadl for buffer.


"No functional change" as you don't actually use the parameter.

Yes, will add it into commit log.


Signed-off-by: Sunil Muthuswamy <sunilmut@xxxxxxxxxxxxx>
Co-Developed-by: Sunil Muthuswamy <sunilmut@xxxxxxxxxxxxx>
Signed-off-by: Tianyu Lan <Tianyu.Lan@xxxxxxxxxxxxx>

Nit: Sunil's SoB looks misleading because the patch is from you,
Co-Developed-by should be sufficient.


Will update.

---
arch/x86/include/asm/hyperv-tlfs.h | 9 +++++++++
drivers/hv/channel.c | 20 +++++++++++---------
drivers/net/hyperv/netvsc.c | 8 ++++++--
drivers/uio/uio_hv_generic.c | 7 +++++--
include/linux/hyperv.h | 3 ++-
5 files changed, 33 insertions(+), 14 deletions(-)

diff --git a/arch/x86/include/asm/hyperv-tlfs.h b/arch/x86/include/asm/hyperv-tlfs.h
index e6cd3fee562b..fb1893a4c32b 100644
--- a/arch/x86/include/asm/hyperv-tlfs.h
+++ b/arch/x86/include/asm/hyperv-tlfs.h
@@ -236,6 +236,15 @@ enum hv_isolation_type {
/* TSC invariant control */
#define HV_X64_MSR_TSC_INVARIANT_CONTROL 0x40000118
+/* Hyper-V GPA map flags */
+#define HV_MAP_GPA_PERMISSIONS_NONE 0x0
+#define HV_MAP_GPA_READABLE 0x1
+#define HV_MAP_GPA_WRITABLE 0x2
+
+#define VMBUS_PAGE_VISIBLE_READ_ONLY HV_MAP_GPA_READABLE
+#define VMBUS_PAGE_VISIBLE_READ_WRITE (HV_MAP_GPA_READABLE|HV_MAP_GPA_WRITABLE)
+#define VMBUS_PAGE_NOT_VISIBLE HV_MAP_GPA_PERMISSIONS_NONE
+

Are these x86-only? If not, then we should probably move these defines
to include/asm-generic/hyperv-tlfs.h. In case they are, we should do
something as we're using them from arch neutral places.

Also, could you please add a comment stating that these flags define
host's visibility of a page and not guest's (this seems to be not
obvious at least to me).





/*
* Declare the MSR used to setup pages used to communicate with the hypervisor.
*/
diff --git a/drivers/hv/channel.c b/drivers/hv/channel.c
index 0bd202de7960..daa21cc72beb 100644
--- a/drivers/hv/channel.c
+++ b/drivers/hv/channel.c
@@ -242,7 +242,7 @@ EXPORT_SYMBOL_GPL(vmbus_send_modifychannel);
*/
static int create_gpadl_header(enum hv_gpadl_type type, void *kbuffer,
u32 size, u32 send_offset,
- struct vmbus_channel_msginfo **msginfo)
+ struct vmbus_channel_msginfo **msginfo, u32 visibility)
{
int i;
int pagecount;
@@ -391,7 +391,7 @@ static int create_gpadl_header(enum hv_gpadl_type type, void *kbuffer,
static int __vmbus_establish_gpadl(struct vmbus_channel *channel,
enum hv_gpadl_type type, void *kbuffer,
u32 size, u32 send_offset,
- u32 *gpadl_handle)
+ u32 *gpadl_handle, u32 visibility)
{
struct vmbus_channel_gpadl_header *gpadlmsg;
struct vmbus_channel_gpadl_body *gpadl_body;
@@ -405,7 +405,8 @@ static int __vmbus_establish_gpadl(struct vmbus_channel *channel,
next_gpadl_handle =
(atomic_inc_return(&vmbus_connection.next_gpadl_handle) - 1);
- ret = create_gpadl_header(type, kbuffer, size, send_offset, &msginfo);
+ ret = create_gpadl_header(type, kbuffer, size, send_offset,
+ &msginfo, visibility);
if (ret)
return ret;
@@ -496,10 +497,10 @@ static int __vmbus_establish_gpadl(struct vmbus_channel *channel,
* @gpadl_handle: some funky thing
*/
int vmbus_establish_gpadl(struct vmbus_channel *channel, void *kbuffer,
- u32 size, u32 *gpadl_handle)
+ u32 size, u32 *gpadl_handle, u32 visibility)
{
return __vmbus_establish_gpadl(channel, HV_GPADL_BUFFER, kbuffer, size,
- 0U, gpadl_handle);
+ 0U, gpadl_handle, visibility);
}
EXPORT_SYMBOL_GPL(vmbus_establish_gpadl);
@@ -610,10 +611,11 @@ static int __vmbus_open(struct vmbus_channel *newchannel,
newchannel->ringbuffer_gpadlhandle = 0;
err = __vmbus_establish_gpadl(newchannel, HV_GPADL_RING,
- page_address(newchannel->ringbuffer_page),
- (send_pages + recv_pages) << PAGE_SHIFT,
- newchannel->ringbuffer_send_offset << PAGE_SHIFT,
- &newchannel->ringbuffer_gpadlhandle);
+ page_address(newchannel->ringbuffer_page),
+ (send_pages + recv_pages) << PAGE_SHIFT,
+ newchannel->ringbuffer_send_offset << PAGE_SHIFT,
+ &newchannel->ringbuffer_gpadlhandle,
+ VMBUS_PAGE_VISIBLE_READ_WRITE);

Nit: I liked the original alignment more and we can avoid the unneeded
code churn.

if (err)
goto error_clean_ring;
diff --git a/drivers/net/hyperv/netvsc.c b/drivers/net/hyperv/netvsc.c
index 2353623259f3..bb72c7578330 100644
--- a/drivers/net/hyperv/netvsc.c
+++ b/drivers/net/hyperv/netvsc.c
@@ -333,7 +333,8 @@ static int netvsc_init_buf(struct hv_device *device,
*/
ret = vmbus_establish_gpadl(device->channel, net_device->recv_buf,
buf_size,
- &net_device->recv_buf_gpadl_handle);
+ &net_device->recv_buf_gpadl_handle,
+ VMBUS_PAGE_VISIBLE_READ_WRITE);
if (ret != 0) {
netdev_err(ndev,
"unable to establish receive buffer's gpadl\n");
@@ -422,10 +423,13 @@ static int netvsc_init_buf(struct hv_device *device,
/* Establish the gpadl handle for this buffer on this
* channel. Note: This call uses the vmbus connection rather
* than the channel to establish the gpadl handle.
+ * Send buffer should theoretically be only marked as "read-only", but
+ * the netvsp for some reason needs write capabilities on it.
*/
ret = vmbus_establish_gpadl(device->channel, net_device->send_buf,
buf_size,
- &net_device->send_buf_gpadl_handle);
+ &net_device->send_buf_gpadl_handle,
+ VMBUS_PAGE_VISIBLE_READ_WRITE);
if (ret != 0) {
netdev_err(ndev,
"unable to establish send buffer's gpadl\n");
diff --git a/drivers/uio/uio_hv_generic.c b/drivers/uio/uio_hv_generic.c
index 0330ba99730e..813a7bee5139 100644
--- a/drivers/uio/uio_hv_generic.c
+++ b/drivers/uio/uio_hv_generic.c
@@ -29,6 +29,7 @@
#include <linux/hyperv.h>
#include <linux/vmalloc.h>
#include <linux/slab.h>
+#include <asm/mshyperv.h>
#include "../hv/hyperv_vmbus.h"
@@ -295,7 +296,8 @@ hv_uio_probe(struct hv_device *dev,
}
ret = vmbus_establish_gpadl(channel, pdata->recv_buf,
- RECV_BUFFER_SIZE, &pdata->recv_gpadl);
+ RECV_BUFFER_SIZE, &pdata->recv_gpadl,
+ VMBUS_PAGE_VISIBLE_READ_WRITE);
if (ret)
goto fail_close;
@@ -315,7 +317,8 @@ hv_uio_probe(struct hv_device *dev,
}
ret = vmbus_establish_gpadl(channel, pdata->send_buf,
- SEND_BUFFER_SIZE, &pdata->send_gpadl);
+ SEND_BUFFER_SIZE, &pdata->send_gpadl,
+ VMBUS_PAGE_VISIBLE_READ_ONLY);

Actually, this is the only place where you use 'READ_ONLY' mapping --
which makes me wonder if it's actually worth it or we can hard-code
VMBUS_PAGE_VISIBLE_READ_WRITE for now and avoid this additional
parameter.


Another option is to set host visibility out of vmbus_establish_gpadl(). There are three places calling vmbus_establish_gpadl(). Vmbus, netvsc and uio drivers.

if (ret)
goto fail_close;
diff --git a/include/linux/hyperv.h b/include/linux/hyperv.h
index f1d74dcf0353..016fdca20d6e 100644
--- a/include/linux/hyperv.h
+++ b/include/linux/hyperv.h
@@ -1179,7 +1179,8 @@ extern int vmbus_sendpacket_mpb_desc(struct vmbus_channel *channel,
extern int vmbus_establish_gpadl(struct vmbus_channel *channel,
void *kbuffer,
u32 size,
- u32 *gpadl_handle);
+ u32 *gpadl_handle,
+ u32 visibility);
extern int vmbus_teardown_gpadl(struct vmbus_channel *channel,
u32 gpadl_handle);