[RFC v2 00/11] Hyper-V: Support PAGE_SIZE larger than 4K

[RFC v2 00/11] Hyper-V: Support PAGE_SIZE larger than 4K

[Boqun Feng]

This patchset add the necessary changes to support guests whose page
size is larger than 4K.

Previous version:
v1: https://lore.kernel.org/lkml/20200721014135.84140-1-boqun.feng@gmail.com/

Changes since v1:

*	Introduce a hv_ring_gpadl_send_offset() to improve the
	readability as per suggestion from Michael.

*	Use max(..., 2 * PAGE_SIZE) instead of hard-coding size for
	inputvsc ringbuffer to align with other ringbuffer settinngs

*	Calculate the exact size of storvsc payload (other than a
	maximum size) to save memory in storvsc_queuecommand() as per
	suggestion from Michael.

*	Use "unsigned int" for loop index inside a page, so that we can
	have the compiler's help for optimization in PAGE_SIZE ==
	HV_HYP_PAGE_SIZE case as per suggestion from Michael.

*	Rebase on to v5.9-rc2 with Michael's latest core support
	patchset[1]


Hyper-V always uses 4K as the page size and expects the same page size
when communicating with guests. That is, all the "pfn"s in the
hypervisor-guest communication protocol are the page numbers in the unit
of HV_HYP_PAGE_SIZE rather than PAGE_SIZE. To support guests with larger
page size, we need to convert between these two page sizes correctly in
the hypervisor-guest communication, which is basically what this
patchset does.

In this conversion, one challenge is how to handle the ringbuffer. A
ringbuffer has two parts: a header and a data part, both of which want
to be PAGE_SIZE aligned in the guest, because we use the "double
mapping" trick to map the data part twice in the guest virtual address
space for faster wrap-around and ease to process data in place. However,
the Hyper-V hypervisor always treats the ringbuffer headers as 4k pages.
To overcome this gap, we enlarge the hv_ring_buffer structure to be
always PAGE_SIZE aligned, and introduce the gpadl type concept to allow
vmbus_establish_gpadl() to handle ringbuffer cases specially. Note that
gpadl type is only meaningful to the guest, there is no such concept in
Hyper-V hypervisor.

This patchset consists of 11 patches:

Patch 1~4: Introduce the types of gpadl, so that we can handle
	   ringbuffer when PAGE_SIZE != HV_HYP_PAGE_SIZE, and also fix
	   a few places where we should use HV_HYP_PAGE_SIZE other than
	   PAGE_SIZE.

Patch 5~6: Add a few helper functions to help calculate the hvpfn (page
	   number in the unit of HV_HYP_PAGE_SIZE) and other related
	   data. So that we can use them in the code of drivers.

Patch 7~11: Use the helpers and change the driver code accordingly to
	    make net/input/util/storage driver work with PAGE_SIZE !=
	    HV_HYP_PAGE_SIZE

I've done some tests with PAGE_SIZE=64k and PAGE_SIZE=16k configurations
on ARM64 guests (with Michael's patchset[1] for ARM64 Hyper-V guest
support), nothing major breaks yet ;-) (I could observe an error caused
by unaligned firmware data, but it's better to have it fixed in the
Hyper-V). I also have done a build and boot test on x86, everything
worked well.

Looking forwards to comments and suggestions!

Regards,
Boqun

[1]: https://lore.kernel.org/lkml/1598287583-71762-1-git-send-email-mikelley@microsoft.com/


Boqun Feng (11):
  Drivers: hv: vmbus: Always use HV_HYP_PAGE_SIZE for gpadl
  Drivers: hv: vmbus: Move __vmbus_open()
  Drivers: hv: vmbus: Introduce types of GPADL
  Drivers: hv: Use HV_HYP_PAGE in hv_synic_enable_regs()
  Drivers: hv: vmbus: Move virt_to_hvpfn() to hyperv header
  hv: hyperv.h: Introduce some hvpfn helper functions
  hv_netvsc: Use HV_HYP_PAGE_SIZE for Hyper-V communication
  Input: hyperv-keyboard: Make ringbuffer at least take two pages
  HID: hyperv: Make ringbuffer at least take two pages
  Driver: hv: util: Make ringbuffer at least take two pages
  scsi: storvsc: Support PAGE_SIZE larger than 4K

 drivers/hid/hid-hyperv.c              |   4 +-
 drivers/hv/channel.c                  | 462 ++++++++++++++++----------
 drivers/hv/hv.c                       |   4 +-
 drivers/hv/hv_util.c                  |  16 +-
 drivers/input/serio/hyperv-keyboard.c |   4 +-
 drivers/net/hyperv/netvsc.c           |   2 +-
 drivers/net/hyperv/netvsc_drv.c       |  46 +--
 drivers/net/hyperv/rndis_filter.c     |  12 +-
 drivers/scsi/storvsc_drv.c            |  60 +++-
 include/linux/hyperv.h                |  63 +++-
 10 files changed, 447 insertions(+), 226 deletions(-)

-- 
2.28.0

[RFC v2 01/11] Drivers: hv: vmbus: Always use HV_HYP_PAGE_SIZE for gpadl

[Boqun Feng]

Since the hypervisor always uses 4K as its page size, the size of PFNs
used for gpadl should be HV_HYP_PAGE_SIZE rather than PAGE_SIZE, so
adjust this accordingly as the preparation for supporting 16K/64K page
size guests. No functional changes on x86, since PAGE_SIZE is always 4k
(equals to HV_HYP_PAGE_SIZE).

Signed-off-by: Boqun Feng <boqun.feng@gmail.com>
---
 drivers/hv/channel.c | 13 +++++--------
 1 file changed, 5 insertions(+), 8 deletions(-)

diff --git a/drivers/hv/channel.c b/drivers/hv/channel.c
index 3ebda7707e46..4d0f8e5a88d6 100644
--- a/drivers/hv/channel.c
+++ b/drivers/hv/channel.c
@@ -22,9 +22,6 @@
 
 #include "hyperv_vmbus.h"
 
-#define NUM_PAGES_SPANNED(addr, len) \
-((PAGE_ALIGN(addr + len) >> PAGE_SHIFT) - (addr >> PAGE_SHIFT))
-
 static unsigned long virt_to_hvpfn(void *addr)
 {
 	phys_addr_t paddr;
@@ -35,7 +32,7 @@ static unsigned long virt_to_hvpfn(void *addr)
 	else
 		paddr = __pa(addr);
 
-	return  paddr >> PAGE_SHIFT;
+	return  paddr >> HV_HYP_PAGE_SHIFT;
 }
 
 /*
@@ -330,7 +327,7 @@ static int create_gpadl_header(void *kbuffer, u32 size,
 
 	int pfnsum, pfncount, pfnleft, pfncurr, pfnsize;
 
-	pagecount = size >> PAGE_SHIFT;
+	pagecount = size >> HV_HYP_PAGE_SHIFT;
 
 	/* do we need a gpadl body msg */
 	pfnsize = MAX_SIZE_CHANNEL_MESSAGE -
@@ -360,7 +357,7 @@ static int create_gpadl_header(void *kbuffer, u32 size,
 		gpadl_header->range[0].byte_count = size;
 		for (i = 0; i < pfncount; i++)
 			gpadl_header->range[0].pfn_array[i] = virt_to_hvpfn(
-				kbuffer + PAGE_SIZE * i);
+				kbuffer + HV_HYP_PAGE_SIZE * i);
 		*msginfo = msgheader;
 
 		pfnsum = pfncount;
@@ -412,7 +409,7 @@ static int create_gpadl_header(void *kbuffer, u32 size,
 			 */
 			for (i = 0; i < pfncurr; i++)
 				gpadl_body->pfn[i] = virt_to_hvpfn(
-					kbuffer + PAGE_SIZE * (pfnsum + i));
+					kbuffer + HV_HYP_PAGE_SIZE * (pfnsum + i));
 
 			/* add to msg header */
 			list_add_tail(&msgbody->msglistentry,
@@ -441,7 +438,7 @@ static int create_gpadl_header(void *kbuffer, u32 size,
 		gpadl_header->range[0].byte_count = size;
 		for (i = 0; i < pagecount; i++)
 			gpadl_header->range[0].pfn_array[i] = virt_to_hvpfn(
-				kbuffer + PAGE_SIZE * i);
+				kbuffer + HV_HYP_PAGE_SIZE * i);
 
 		*msginfo = msgheader;
 	}
-- 
2.28.0

[RFC v2 02/11] Drivers: hv: vmbus: Move __vmbus_open()

[Boqun Feng]

Pure function movement, no functional changes. The move is made, because
in a later change, __vmbus_open() will rely on some static functions
afterwards, so we sperate the move and the modification of
__vmbus_open() in two patches to make it easy to review.

Signed-off-by: Boqun Feng <boqun.feng@gmail.com>
Reviewed-by: Wei Liu <wei.liu@kernel.org>
---
 drivers/hv/channel.c | 309 ++++++++++++++++++++++---------------------
 1 file changed, 155 insertions(+), 154 deletions(-)

diff --git a/drivers/hv/channel.c b/drivers/hv/channel.c
index 4d0f8e5a88d6..1cbe8fc931fc 100644
--- a/drivers/hv/channel.c
+++ b/drivers/hv/channel.c
@@ -109,160 +109,6 @@ int vmbus_alloc_ring(struct vmbus_channel *newchannel,
 }
 EXPORT_SYMBOL_GPL(vmbus_alloc_ring);
 
-static int __vmbus_open(struct vmbus_channel *newchannel,
-		       void *userdata, u32 userdatalen,
-		       void (*onchannelcallback)(void *context), void *context)
-{
-	struct vmbus_channel_open_channel *open_msg;
-	struct vmbus_channel_msginfo *open_info = NULL;
-	struct page *page = newchannel->ringbuffer_page;
-	u32 send_pages, recv_pages;
-	unsigned long flags;
-	int err;
-
-	if (userdatalen > MAX_USER_DEFINED_BYTES)
-		return -EINVAL;
-
-	send_pages = newchannel->ringbuffer_send_offset;
-	recv_pages = newchannel->ringbuffer_pagecount - send_pages;
-
-	if (newchannel->state != CHANNEL_OPEN_STATE)
-		return -EINVAL;
-
-	newchannel->state = CHANNEL_OPENING_STATE;
-	newchannel->onchannel_callback = onchannelcallback;
-	newchannel->channel_callback_context = context;
-
-	err = hv_ringbuffer_init(&newchannel->outbound, page, send_pages);
-	if (err)
-		goto error_clean_ring;
-
-	err = hv_ringbuffer_init(&newchannel->inbound,
-				 &page[send_pages], recv_pages);
-	if (err)
-		goto error_clean_ring;
-
-	/* Establish the gpadl for the ring buffer */
-	newchannel->ringbuffer_gpadlhandle = 0;
-
-	err = vmbus_establish_gpadl(newchannel,
-				    page_address(newchannel->ringbuffer_page),
-				    (send_pages + recv_pages) << PAGE_SHIFT,
-				    &newchannel->ringbuffer_gpadlhandle);
-	if (err)
-		goto error_clean_ring;
-
-	/* Create and init the channel open message */
-	open_info = kmalloc(sizeof(*open_info) +
-			   sizeof(struct vmbus_channel_open_channel),
-			   GFP_KERNEL);
-	if (!open_info) {
-		err = -ENOMEM;
-		goto error_free_gpadl;
-	}
-
-	init_completion(&open_info->waitevent);
-	open_info->waiting_channel = newchannel;
-
-	open_msg = (struct vmbus_channel_open_channel *)open_info->msg;
-	open_msg->header.msgtype = CHANNELMSG_OPENCHANNEL;
-	open_msg->openid = newchannel->offermsg.child_relid;
-	open_msg->child_relid = newchannel->offermsg.child_relid;
-	open_msg->ringbuffer_gpadlhandle = newchannel->ringbuffer_gpadlhandle;
-	open_msg->downstream_ringbuffer_pageoffset = newchannel->ringbuffer_send_offset;
-	open_msg->target_vp = hv_cpu_number_to_vp_number(newchannel->target_cpu);
-
-	if (userdatalen)
-		memcpy(open_msg->userdata, userdata, userdatalen);
-
-	spin_lock_irqsave(&vmbus_connection.channelmsg_lock, flags);
-	list_add_tail(&open_info->msglistentry,
-		      &vmbus_connection.chn_msg_list);
-	spin_unlock_irqrestore(&vmbus_connection.channelmsg_lock, flags);
-
-	if (newchannel->rescind) {
-		err = -ENODEV;
-		goto error_free_info;
-	}
-
-	err = vmbus_post_msg(open_msg,
-			     sizeof(struct vmbus_channel_open_channel), true);
-
-	trace_vmbus_open(open_msg, err);
-
-	if (err != 0)
-		goto error_clean_msglist;
-
-	wait_for_completion(&open_info->waitevent);
-
-	spin_lock_irqsave(&vmbus_connection.channelmsg_lock, flags);
-	list_del(&open_info->msglistentry);
-	spin_unlock_irqrestore(&vmbus_connection.channelmsg_lock, flags);
-
-	if (newchannel->rescind) {
-		err = -ENODEV;
-		goto error_free_info;
-	}
-
-	if (open_info->response.open_result.status) {
-		err = -EAGAIN;
-		goto error_free_info;
-	}
-
-	newchannel->state = CHANNEL_OPENED_STATE;
-	kfree(open_info);
-	return 0;
-
-error_clean_msglist:
-	spin_lock_irqsave(&vmbus_connection.channelmsg_lock, flags);
-	list_del(&open_info->msglistentry);
-	spin_unlock_irqrestore(&vmbus_connection.channelmsg_lock, flags);
-error_free_info:
-	kfree(open_info);
-error_free_gpadl:
-	vmbus_teardown_gpadl(newchannel, newchannel->ringbuffer_gpadlhandle);
-	newchannel->ringbuffer_gpadlhandle = 0;
-error_clean_ring:
-	hv_ringbuffer_cleanup(&newchannel->outbound);
-	hv_ringbuffer_cleanup(&newchannel->inbound);
-	newchannel->state = CHANNEL_OPEN_STATE;
-	return err;
-}
-
-/*
- * vmbus_connect_ring - Open the channel but reuse ring buffer
- */
-int vmbus_connect_ring(struct vmbus_channel *newchannel,
-		       void (*onchannelcallback)(void *context), void *context)
-{
-	return  __vmbus_open(newchannel, NULL, 0, onchannelcallback, context);
-}
-EXPORT_SYMBOL_GPL(vmbus_connect_ring);
-
-/*
- * vmbus_open - Open the specified channel.
- */
-int vmbus_open(struct vmbus_channel *newchannel,
-	       u32 send_ringbuffer_size, u32 recv_ringbuffer_size,
-	       void *userdata, u32 userdatalen,
-	       void (*onchannelcallback)(void *context), void *context)
-{
-	int err;
-
-	err = vmbus_alloc_ring(newchannel, send_ringbuffer_size,
-			       recv_ringbuffer_size);
-	if (err)
-		return err;
-
-	err = __vmbus_open(newchannel, userdata, userdatalen,
-			   onchannelcallback, context);
-	if (err)
-		vmbus_free_ring(newchannel);
-
-	return err;
-}
-EXPORT_SYMBOL_GPL(vmbus_open);
-
 /* Used for Hyper-V Socket: a guest client's connect() to the host */
 int vmbus_send_tl_connect_request(const guid_t *shv_guest_servie_id,
 				  const guid_t *shv_host_servie_id)
@@ -556,6 +402,161 @@ int vmbus_establish_gpadl(struct vmbus_channel *channel, void *kbuffer,
 }
 EXPORT_SYMBOL_GPL(vmbus_establish_gpadl);
 
+static int __vmbus_open(struct vmbus_channel *newchannel,
+		       void *userdata, u32 userdatalen,
+		       void (*onchannelcallback)(void *context), void *context)
+{
+	struct vmbus_channel_open_channel *open_msg;
+	struct vmbus_channel_msginfo *open_info = NULL;
+	struct page *page = newchannel->ringbuffer_page;
+	u32 send_pages, recv_pages;
+	unsigned long flags;
+	int err;
+
+	if (userdatalen > MAX_USER_DEFINED_BYTES)
+		return -EINVAL;
+
+	send_pages = newchannel->ringbuffer_send_offset;
+	recv_pages = newchannel->ringbuffer_pagecount - send_pages;
+
+	if (newchannel->state != CHANNEL_OPEN_STATE)
+		return -EINVAL;
+
+	newchannel->state = CHANNEL_OPENING_STATE;
+	newchannel->onchannel_callback = onchannelcallback;
+	newchannel->channel_callback_context = context;
+
+	err = hv_ringbuffer_init(&newchannel->outbound, page, send_pages);
+	if (err)
+		goto error_clean_ring;
+
+	err = hv_ringbuffer_init(&newchannel->inbound,
+				 &page[send_pages], recv_pages);
+	if (err)
+		goto error_clean_ring;
+
+	/* Establish the gpadl for the ring buffer */
+	newchannel->ringbuffer_gpadlhandle = 0;
+
+	err = vmbus_establish_gpadl(newchannel,
+				    page_address(newchannel->ringbuffer_page),
+				    (send_pages + recv_pages) << PAGE_SHIFT,
+				    &newchannel->ringbuffer_gpadlhandle);
+	if (err)
+		goto error_clean_ring;
+
+	/* Create and init the channel open message */
+	open_info = kmalloc(sizeof(*open_info) +
+			   sizeof(struct vmbus_channel_open_channel),
+			   GFP_KERNEL);
+	if (!open_info) {
+		err = -ENOMEM;
+		goto error_free_gpadl;
+	}
+
+	init_completion(&open_info->waitevent);
+	open_info->waiting_channel = newchannel;
+
+	open_msg = (struct vmbus_channel_open_channel *)open_info->msg;
+	open_msg->header.msgtype = CHANNELMSG_OPENCHANNEL;
+	open_msg->openid = newchannel->offermsg.child_relid;
+	open_msg->child_relid = newchannel->offermsg.child_relid;
+	open_msg->ringbuffer_gpadlhandle = newchannel->ringbuffer_gpadlhandle;
+	open_msg->downstream_ringbuffer_pageoffset = newchannel->ringbuffer_send_offset;
+	open_msg->target_vp = hv_cpu_number_to_vp_number(newchannel->target_cpu);
+
+	if (userdatalen)
+		memcpy(open_msg->userdata, userdata, userdatalen);
+
+	spin_lock_irqsave(&vmbus_connection.channelmsg_lock, flags);
+	list_add_tail(&open_info->msglistentry,
+		      &vmbus_connection.chn_msg_list);
+	spin_unlock_irqrestore(&vmbus_connection.channelmsg_lock, flags);
+
+	if (newchannel->rescind) {
+		err = -ENODEV;
+		goto error_free_info;
+	}
+
+	err = vmbus_post_msg(open_msg,
+			     sizeof(struct vmbus_channel_open_channel), true);
+
+	trace_vmbus_open(open_msg, err);
+
+	if (err != 0)
+		goto error_clean_msglist;
+
+	wait_for_completion(&open_info->waitevent);
+
+	spin_lock_irqsave(&vmbus_connection.channelmsg_lock, flags);
+	list_del(&open_info->msglistentry);
+	spin_unlock_irqrestore(&vmbus_connection.channelmsg_lock, flags);
+
+	if (newchannel->rescind) {
+		err = -ENODEV;
+		goto error_free_info;
+	}
+
+	if (open_info->response.open_result.status) {
+		err = -EAGAIN;
+		goto error_free_info;
+	}
+
+	newchannel->state = CHANNEL_OPENED_STATE;
+	kfree(open_info);
+	return 0;
+
+error_clean_msglist:
+	spin_lock_irqsave(&vmbus_connection.channelmsg_lock, flags);
+	list_del(&open_info->msglistentry);
+	spin_unlock_irqrestore(&vmbus_connection.channelmsg_lock, flags);
+error_free_info:
+	kfree(open_info);
+error_free_gpadl:
+	vmbus_teardown_gpadl(newchannel, newchannel->ringbuffer_gpadlhandle);
+	newchannel->ringbuffer_gpadlhandle = 0;
+error_clean_ring:
+	hv_ringbuffer_cleanup(&newchannel->outbound);
+	hv_ringbuffer_cleanup(&newchannel->inbound);
+	newchannel->state = CHANNEL_OPEN_STATE;
+	return err;
+}
+
+/*
+ * vmbus_connect_ring - Open the channel but reuse ring buffer
+ */
+int vmbus_connect_ring(struct vmbus_channel *newchannel,
+		       void (*onchannelcallback)(void *context), void *context)
+{
+	return  __vmbus_open(newchannel, NULL, 0, onchannelcallback, context);
+}
+EXPORT_SYMBOL_GPL(vmbus_connect_ring);
+
+/*
+ * vmbus_open - Open the specified channel.
+ */
+int vmbus_open(struct vmbus_channel *newchannel,
+	       u32 send_ringbuffer_size, u32 recv_ringbuffer_size,
+	       void *userdata, u32 userdatalen,
+	       void (*onchannelcallback)(void *context), void *context)
+{
+	int err;
+
+	err = vmbus_alloc_ring(newchannel, send_ringbuffer_size,
+			       recv_ringbuffer_size);
+	if (err)
+		return err;
+
+	err = __vmbus_open(newchannel, userdata, userdatalen,
+			   onchannelcallback, context);
+	if (err)
+		vmbus_free_ring(newchannel);
+
+	return err;
+}
+EXPORT_SYMBOL_GPL(vmbus_open);
+
+
 /*
  * vmbus_teardown_gpadl -Teardown the specified GPADL handle
  */
-- 
2.28.0

[RFC v2 03/11] Drivers: hv: vmbus: Introduce types of GPADL

[Boqun Feng]

This patch introduces two types of GPADL: HV_GPADL_{BUFFER, RING}. The
types of GPADL are purely the concept in the guest, IOW the hypervisor
treat them as the same.

The reason of introducing the types of GPADL is to support guests whose
page size is not 4k (the page size of Hyper-V hypervisor). In these
guests, both the headers and the data parts of the ringbuffers need to
be aligned to the PAGE_SIZE, because 1) some of the ringbuffers will be
mapped into userspace and 2) we use "double mapping" mechanism to
support fast wrap-around, and "double mapping" relies on ringbuffers
being page-aligned. However, the Hyper-V hypervisor only uses 4k
(HV_HYP_PAGE_SIZE) headers. Our solution to this is that we always make
the headers of ringbuffers take one guest page and when GPADL is
established between the guest and hypervisor, the only first 4k of
header is used. To handle this special case, we need the types of GPADL
to differ different guest memory usage for GPADL.

Type enum is introduced along with several general interfaces to
describe the differences between normal buffer GPADL and ringbuffer
GPADL.

Signed-off-by: Boqun Feng <boqun.feng@gmail.com>
---
 drivers/hv/channel.c   | 159 +++++++++++++++++++++++++++++++++++------
 include/linux/hyperv.h |  44 +++++++++++-
 2 files changed, 182 insertions(+), 21 deletions(-)

diff --git a/drivers/hv/channel.c b/drivers/hv/channel.c
index 1cbe8fc931fc..7c443fd567e4 100644
--- a/drivers/hv/channel.c
+++ b/drivers/hv/channel.c
@@ -35,6 +35,98 @@ static unsigned long virt_to_hvpfn(void *addr)
 	return  paddr >> HV_HYP_PAGE_SHIFT;
 }
 
+/*
+ * hv_gpadl_size - Return the real size of a gpadl, the size that Hyper-V uses
+ *
+ * For BUFFER gpadl, Hyper-V uses the exact same size as the guest does.
+ *
+ * For RING gpadl, in each ring, the guest uses one PAGE_SIZE as the header
+ * (because of the alignment requirement), however, the hypervisor only
+ * uses the first HV_HYP_PAGE_SIZE as the header, therefore leaving a
+ * (PAGE_SIZE - HV_HYP_PAGE_SIZE) gap. And since there are two rings in a
+ * ringbuffer, So the total size for a RING gpadl that Hyper-V uses is the
+ * total size that the guest uses minus twice of the gap size.
+ */
+static inline u32 hv_gpadl_size(enum hv_gpadl_type type, u32 size)
+{
+	switch (type) {
+	case HV_GPADL_BUFFER:
+		return size;
+	case HV_GPADL_RING:
+		/* The size of a ringbuffer must be page-aligned */
+		BUG_ON(size % PAGE_SIZE);
+		/*
+		 * Two things to notice here:
+		 * 1) We're processing two ring buffers as a unit
+		 * 2) We're skipping any space larger than HV_HYP_PAGE_SIZE in
+		 * the first guest-size page of each of the two ring buffers.
+		 * So we effectively subtract out two guest-size pages, and add
+		 * back two Hyper-V size pages.
+		 */
+		return size - 2 * (PAGE_SIZE - HV_HYP_PAGE_SIZE);
+	}
+	BUG();
+	return 0;
+}
+
+/*
+ * hv_ring_gpadl_send_offset - Calculate the send offset in a ring gpadl based
+ *                             on the offset in the guest
+ *
+ * @send_offset: the offset (in bytes) where the send ringbuffer starts in the
+ *               virtual address space of the guest
+ */
+static inline u32 hv_ring_gpadl_send_offset(u32 send_offset)
+{ 
+
+	/*
+	 * For RING gpadl, in each ring, the guest uses one PAGE_SIZE as the
+	 * header (because of the alignment requirement), however, the
+	 * hypervisor only uses the first HV_HYP_PAGE_SIZE as the header,
+	 * therefore leaving a (PAGE_SIZE - HV_HYP_PAGE_SIZE) gap.
+	 *
+	 * And to calculate the effective send offset in gpadl, we need to
+	 * substract this gap.
+	 */
+	return send_offset - (PAGE_SIZE - HV_HYP_PAGE_SIZE);
+}
+
+/*
+ * hv_gpadl_hvpfn - Return the Hyper-V page PFN of the @i th Hyper-V page in
+ *                  the gpadl
+ *
+ * @type: the type of the gpadl
+ * @kbuffer: the pointer to the gpadl in the guest
+ * @size: the total size (in bytes) of the gpadl
+ * @send_offset: the offset (in bytes) where the send ringbuffer starts in the
+ *               virtual address space of the guest
+ * @i: the index
+ */
+static inline u64 hv_gpadl_hvpfn(enum hv_gpadl_type type, void *kbuffer,
+				 u32 size, u32 send_offset, int i)
+{
+	int send_idx = hv_ring_gpadl_send_offset(send_offset) >> HV_HYP_PAGE_SHIFT;
+	unsigned long delta = 0UL;
+
+	switch (type) {
+	case HV_GPADL_BUFFER:
+		break;
+	case HV_GPADL_RING:
+		if (i == 0)
+			delta = 0;
+		else if (i <= send_idx)
+			delta = PAGE_SIZE - HV_HYP_PAGE_SIZE;
+		else
+			delta = 2 * (PAGE_SIZE - HV_HYP_PAGE_SIZE);
+		break;
+	default:
+		BUG();
+		break;
+	}
+
+	return virt_to_hvpfn(kbuffer + delta + (HV_HYP_PAGE_SIZE * i));
+}
+
 /*
  * vmbus_setevent- Trigger an event notification on the specified
  * channel.
@@ -160,7 +252,8 @@ EXPORT_SYMBOL_GPL(vmbus_send_modifychannel);
 /*
  * create_gpadl_header - Creates a gpadl for the specified buffer
  */
-static int create_gpadl_header(void *kbuffer, u32 size,
+static int create_gpadl_header(enum hv_gpadl_type type, void *kbuffer,
+			       u32 size, u32 send_offset,
 			       struct vmbus_channel_msginfo **msginfo)
 {
 	int i;
@@ -173,7 +266,7 @@ static int create_gpadl_header(void *kbuffer, u32 size,
 
 	int pfnsum, pfncount, pfnleft, pfncurr, pfnsize;
 
-	pagecount = size >> HV_HYP_PAGE_SHIFT;
+	pagecount = hv_gpadl_size(type, size) >> HV_HYP_PAGE_SHIFT;
 
 	/* do we need a gpadl body msg */
 	pfnsize = MAX_SIZE_CHANNEL_MESSAGE -
@@ -200,10 +293,10 @@ static int create_gpadl_header(void *kbuffer, u32 size,
 		gpadl_header->range_buflen = sizeof(struct gpa_range) +
 					 pagecount * sizeof(u64);
 		gpadl_header->range[0].byte_offset = 0;
-		gpadl_header->range[0].byte_count = size;
+		gpadl_header->range[0].byte_count = hv_gpadl_size(type, size);
 		for (i = 0; i < pfncount; i++)
-			gpadl_header->range[0].pfn_array[i] = virt_to_hvpfn(
-				kbuffer + HV_HYP_PAGE_SIZE * i);
+			gpadl_header->range[0].pfn_array[i] = hv_gpadl_hvpfn(
+				type, kbuffer, size, send_offset, i);
 		*msginfo = msgheader;
 
 		pfnsum = pfncount;
@@ -254,8 +347,8 @@ static int create_gpadl_header(void *kbuffer, u32 size,
 			 * so the hypervisor guarantees that this is ok.
 			 */
 			for (i = 0; i < pfncurr; i++)
-				gpadl_body->pfn[i] = virt_to_hvpfn(
-					kbuffer + HV_HYP_PAGE_SIZE * (pfnsum + i));
+				gpadl_body->pfn[i] = hv_gpadl_hvpfn(type,
+					kbuffer, size, send_offset, pfnsum + i);
 
 			/* add to msg header */
 			list_add_tail(&msgbody->msglistentry,
@@ -281,10 +374,10 @@ static int create_gpadl_header(void *kbuffer, u32 size,
 		gpadl_header->range_buflen = sizeof(struct gpa_range) +
 					 pagecount * sizeof(u64);
 		gpadl_header->range[0].byte_offset = 0;
-		gpadl_header->range[0].byte_count = size;
+		gpadl_header->range[0].byte_count = hv_gpadl_size(type, size);
 		for (i = 0; i < pagecount; i++)
-			gpadl_header->range[0].pfn_array[i] = virt_to_hvpfn(
-				kbuffer + HV_HYP_PAGE_SIZE * i);
+			gpadl_header->range[0].pfn_array[i] = hv_gpadl_hvpfn(
+				type, kbuffer, size, send_offset, i);
 
 		*msginfo = msgheader;
 	}
@@ -297,15 +390,20 @@ static int create_gpadl_header(void *kbuffer, u32 size,
 }
 
 /*
- * vmbus_establish_gpadl - Establish a GPADL for the specified buffer
+ * __vmbus_establish_gpadl - Establish a GPADL for a buffer or ringbuffer
  *
  * @channel: a channel
+ * @type: the type of the corresponding GPADL, only meaningful for the guest.
  * @kbuffer: from kmalloc or vmalloc
  * @size: page-size multiple
+ * @send_offset: the offset (in bytes) where the send ring buffer starts,
+ * 		 should be 0 for BUFFER type gpadl
  * @gpadl_handle: some funky thing
  */
-int vmbus_establish_gpadl(struct vmbus_channel *channel, void *kbuffer,
-			       u32 size, u32 *gpadl_handle)
+static int __vmbus_establish_gpadl(struct vmbus_channel *channel,
+				   enum hv_gpadl_type type, void *kbuffer,
+				   u32 size, u32 send_offset,
+				   u32 *gpadl_handle)
 {
 	struct vmbus_channel_gpadl_header *gpadlmsg;
 	struct vmbus_channel_gpadl_body *gpadl_body;
@@ -319,7 +417,7 @@ int vmbus_establish_gpadl(struct vmbus_channel *channel, void *kbuffer,
 	next_gpadl_handle =
 		(atomic_inc_return(&vmbus_connection.next_gpadl_handle) - 1);
 
-	ret = create_gpadl_header(kbuffer, size, &msginfo);
+	ret = create_gpadl_header(type, kbuffer, size, send_offset, &msginfo);
 	if (ret)
 		return ret;
 
@@ -400,6 +498,21 @@ int vmbus_establish_gpadl(struct vmbus_channel *channel, void *kbuffer,
 	kfree(msginfo);
 	return ret;
 }
+
+/*
+ * vmbus_establish_gpadl - Establish a GPADL for the specified buffer
+ *
+ * @channel: a channel
+ * @kbuffer: from kmalloc or vmalloc
+ * @size: page-size multiple
+ * @gpadl_handle: some funky thing
+ */
+int vmbus_establish_gpadl(struct vmbus_channel *channel, void *kbuffer,
+			  u32 size, u32 *gpadl_handle)
+{
+	return __vmbus_establish_gpadl(channel, HV_GPADL_BUFFER, kbuffer, size,
+				       0U, gpadl_handle);
+}
 EXPORT_SYMBOL_GPL(vmbus_establish_gpadl);
 
 static int __vmbus_open(struct vmbus_channel *newchannel,
@@ -438,10 +551,11 @@ static int __vmbus_open(struct vmbus_channel *newchannel,
 	/* Establish the gpadl for the ring buffer */
 	newchannel->ringbuffer_gpadlhandle = 0;
 
-	err = vmbus_establish_gpadl(newchannel,
-				    page_address(newchannel->ringbuffer_page),
-				    (send_pages + recv_pages) << PAGE_SHIFT,
-				    &newchannel->ringbuffer_gpadlhandle);
+	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);
 	if (err)
 		goto error_clean_ring;
 
@@ -462,7 +576,13 @@ static int __vmbus_open(struct vmbus_channel *newchannel,
 	open_msg->openid = newchannel->offermsg.child_relid;
 	open_msg->child_relid = newchannel->offermsg.child_relid;
 	open_msg->ringbuffer_gpadlhandle = newchannel->ringbuffer_gpadlhandle;
-	open_msg->downstream_ringbuffer_pageoffset = newchannel->ringbuffer_send_offset;
+	/*
+	 * The unit of ->downstream_ringbuffer_pageoffset is HV_HYP_PAGE and
+	 * the unit of ->ringbuffer_send_offset is PAGE, so here we first
+	 * calculate it into bytes and then convert into HV_HYP_PAGE.
+	 */
+	open_msg->downstream_ringbuffer_pageoffset =
+		hv_ring_gpadl_send_offset(newchannel->ringbuffer_send_offset << PAGE_SHIFT) >> HV_HYP_PAGE_SHIFT;
 	open_msg->target_vp = hv_cpu_number_to_vp_number(newchannel->target_cpu);
 
 	if (userdatalen)
@@ -556,7 +676,6 @@ int vmbus_open(struct vmbus_channel *newchannel,
 }
 EXPORT_SYMBOL_GPL(vmbus_open);
 
-
 /*
  * vmbus_teardown_gpadl -Teardown the specified GPADL handle
  */
diff --git a/include/linux/hyperv.h b/include/linux/hyperv.h
index 38100e80360a..7d16dd28aa48 100644
--- a/include/linux/hyperv.h
+++ b/include/linux/hyperv.h
@@ -29,6 +29,48 @@
 
 #pragma pack(push, 1)
 
+/*
+ * Types for GPADL, decides is how GPADL header is created.
+ *
+ * It doesn't make much difference between BUFFER and RING if PAGE_SIZE is the
+ * same as HV_HYP_PAGE_SIZE.
+ *
+ * If PAGE_SIZE is bigger than HV_HYP_PAGE_SIZE, the headers of ring buffers
+ * will be of PAGE_SIZE, however, only the first HV_HYP_PAGE will be put
+ * into gpadl, therefore the number for HV_HYP_PAGE and the indexes of each
+ * HV_HYP_PAGE will be different between different types of GPADL, for example
+ * if PAGE_SIZE is 64K:
+ *
+ * BUFFER:
+ *
+ * gva:    |--       64k      --|--       64k      --| ... |
+ * gpa:    | 4k | 4k | ... | 4k | 4k | 4k | ... | 4k |
+ * index:  0    1    2     15   16   17   18 .. 31   32 ...
+ *         |    |    ...   |    |    |   ...    |   ...
+ *         v    V          V    V    V          V
+ * gpadl:  | 4k | 4k | ... | 4k | 4k | 4k | ... | 4k | ... |
+ * index:  0    1    2 ... 15   16   17   18 .. 31   32 ...
+ *
+ * RING:
+ *
+ *         | header  |           data           | header  |     data      |
+ * gva:    |-- 64k --|--       64k      --| ... |-- 64k --|-- 64k --| ... |
+ * gpa:    | 4k | .. | 4k | 4k | ... | 4k | ... | 4k | .. | 4k | .. | ... |
+ * index:  0    1    16   17   18    31   ...   n   n+1  n+16 ...         2n
+ *         |         /    /          /          |         /               /
+ *         |        /    /          /           |        /               /
+ *         |       /    /   ...    /    ...     |       /      ...      /
+ *         |      /    /          /             |      /               /
+ *         |     /    /          /              |     /               /
+ *         V    V    V          V               V    V               v
+ * gpadl:  | 4k | 4k |   ...    |    ...        | 4k | 4k |  ...     |
+ * index:  0    1    2   ...    16   ...       n-15 n-14 n-13  ...  2n-30
+ */
+enum hv_gpadl_type {
+	HV_GPADL_BUFFER,
+	HV_GPADL_RING
+};
+
 /* Single-page buffer */
 struct hv_page_buffer {
 	u32 len;
@@ -111,7 +153,7 @@ struct hv_ring_buffer {
 	} feature_bits;
 
 	/* Pad it to PAGE_SIZE so that data starts on page boundary */
-	u8	reserved2[4028];
+	u8	reserved2[PAGE_SIZE - 68];
 
 	/*
 	 * Ring data starts here + RingDataStartOffset
-- 
2.28.0

[RFC v2 04/11] Drivers: hv: Use HV_HYP_PAGE in hv_synic_enable_regs()

[Boqun Feng]

Both the base_*_gpa should use the guest page number in Hyper-V page, so
use HV_HYP_PAGE instead of PAGE.

Signed-off-by: Boqun Feng <boqun.feng@gmail.com>
---
 drivers/hv/hv.c | 4 ++--
 1 file changed, 2 insertions(+), 2 deletions(-)

diff --git a/drivers/hv/hv.c b/drivers/hv/hv.c
index 7499079f4077..8ac8bbf5b5aa 100644
--- a/drivers/hv/hv.c
+++ b/drivers/hv/hv.c
@@ -165,7 +165,7 @@ void hv_synic_enable_regs(unsigned int cpu)
 	hv_get_simp(simp.as_uint64);
 	simp.simp_enabled = 1;
 	simp.base_simp_gpa = virt_to_phys(hv_cpu->synic_message_page)
-		>> PAGE_SHIFT;
+		>> HV_HYP_PAGE_SHIFT;
 
 	hv_set_simp(simp.as_uint64);
 
@@ -173,7 +173,7 @@ void hv_synic_enable_regs(unsigned int cpu)
 	hv_get_siefp(siefp.as_uint64);
 	siefp.siefp_enabled = 1;
 	siefp.base_siefp_gpa = virt_to_phys(hv_cpu->synic_event_page)
-		>> PAGE_SHIFT;
+		>> HV_HYP_PAGE_SHIFT;
 
 	hv_set_siefp(siefp.as_uint64);
 
-- 
2.28.0

[RFC v2 05/11] Drivers: hv: vmbus: Move virt_to_hvpfn() to hyperv header

[Boqun Feng]

There will be more places other than vmbus where we need to calculate
the Hyper-V page PFN from a virtual address, so move virt_to_hvpfn() to
hyperv generic header.

Signed-off-by: Boqun Feng <boqun.feng@gmail.com>
---
 drivers/hv/channel.c   | 13 -------------
 include/linux/hyperv.h | 15 +++++++++++++++
 2 files changed, 15 insertions(+), 13 deletions(-)

diff --git a/drivers/hv/channel.c b/drivers/hv/channel.c
index 7c443fd567e4..74a8f49ab76a 100644
--- a/drivers/hv/channel.c
+++ b/drivers/hv/channel.c
@@ -22,19 +22,6 @@
 
 #include "hyperv_vmbus.h"
 
-static unsigned long virt_to_hvpfn(void *addr)
-{
-	phys_addr_t paddr;
-
-	if (is_vmalloc_addr(addr))
-		paddr = page_to_phys(vmalloc_to_page(addr)) +
-					 offset_in_page(addr);
-	else
-		paddr = __pa(addr);
-
-	return  paddr >> HV_HYP_PAGE_SHIFT;
-}
-
 /*
  * hv_gpadl_size - Return the real size of a gpadl, the size that Hyper-V uses
  *
diff --git a/include/linux/hyperv.h b/include/linux/hyperv.h
index 7d16dd28aa48..6f4831212979 100644
--- a/include/linux/hyperv.h
+++ b/include/linux/hyperv.h
@@ -14,6 +14,7 @@
 
 #include <uapi/linux/hyperv.h>
 
+#include <linux/mm.h>
 #include <linux/types.h>
 #include <linux/scatterlist.h>
 #include <linux/list.h>
@@ -23,6 +24,7 @@
 #include <linux/mod_devicetable.h>
 #include <linux/interrupt.h>
 #include <linux/reciprocal_div.h>
+#include <asm/hyperv-tlfs.h>
 
 #define MAX_PAGE_BUFFER_COUNT				32
 #define MAX_MULTIPAGE_BUFFER_COUNT			32 /* 128K */
@@ -1672,4 +1674,17 @@ struct hyperv_pci_block_ops {
 
 extern struct hyperv_pci_block_ops hvpci_block_ops;
 
+static inline unsigned long virt_to_hvpfn(void *addr)
+{
+	phys_addr_t paddr;
+
+	if (is_vmalloc_addr(addr))
+		paddr = page_to_phys(vmalloc_to_page(addr)) +
+				     offset_in_page(addr);
+	else
+		paddr = __pa(addr);
+
+	return  paddr >> HV_HYP_PAGE_SHIFT;
+}
+
 #endif /* _HYPERV_H */
-- 
2.28.0

[RFC v2 06/11] hv: hyperv.h: Introduce some hvpfn helper functions

[Boqun Feng]

When a guest communicate with the hypervisor, it must use HV_HYP_PAGE to
calculate PFN, so introduce a few hvpfn helper functions as the
counterpart of the page helper functions. This is the preparation for
supporting guest whose PAGE_SIZE is not 4k.

Signed-off-by: Boqun Feng <boqun.feng@gmail.com>
---
 include/linux/hyperv.h | 4 ++++
 1 file changed, 4 insertions(+)

diff --git a/include/linux/hyperv.h b/include/linux/hyperv.h
index 6f4831212979..54e84ba6b554 100644
--- a/include/linux/hyperv.h
+++ b/include/linux/hyperv.h
@@ -1687,4 +1687,8 @@ static inline unsigned long virt_to_hvpfn(void *addr)
 	return  paddr >> HV_HYP_PAGE_SHIFT;
 }
 
+#define offset_in_hvpage(ptr)	((unsigned long)(ptr) & ~HV_HYP_PAGE_MASK)
+#define HVPFN_UP(x)	(((x) + HV_HYP_PAGE_SIZE-1) >> HV_HYP_PAGE_SHIFT)
+#define page_to_hvpfn(page)	((page_to_pfn(page) << PAGE_SHIFT) >> HV_HYP_PAGE_SHIFT)
+
 #endif /* _HYPERV_H */
-- 
2.28.0

[RFC v2 07/11] hv_netvsc: Use HV_HYP_PAGE_SIZE for Hyper-V communication

[Boqun Feng]

When communicating with Hyper-V, HV_HYP_PAGE_SIZE should be used since
that's the page size used by Hyper-V and Hyper-V expects all
page-related data using the unit of HY_HYP_PAGE_SIZE, for example, the
"pfn" in hv_page_buffer is actually the HV_HYP_PAGE (i.e. the Hyper-V
page) number.

In order to support guest whose page size is not 4k, we need to make
hv_netvsc always use HV_HYP_PAGE_SIZE for Hyper-V communication.

Signed-off-by: Boqun Feng <boqun.feng@gmail.com>
---
 drivers/net/hyperv/netvsc.c       |  2 +-
 drivers/net/hyperv/netvsc_drv.c   | 46 +++++++++++++++----------------
 drivers/net/hyperv/rndis_filter.c | 12 ++++----
 3 files changed, 30 insertions(+), 30 deletions(-)

diff --git a/drivers/net/hyperv/netvsc.c b/drivers/net/hyperv/netvsc.c
index 41f5cf0bb997..1d6f2256da6b 100644
--- a/drivers/net/hyperv/netvsc.c
+++ b/drivers/net/hyperv/netvsc.c
@@ -794,7 +794,7 @@ static void netvsc_copy_to_send_buf(struct netvsc_device *net_device,
 	}
 
 	for (i = 0; i < page_count; i++) {
-		char *src = phys_to_virt(pb[i].pfn << PAGE_SHIFT);
+		char *src = phys_to_virt(pb[i].pfn << HV_HYP_PAGE_SHIFT);
 		u32 offset = pb[i].offset;
 		u32 len = pb[i].len;
 
diff --git a/drivers/net/hyperv/netvsc_drv.c b/drivers/net/hyperv/netvsc_drv.c
index 64b0a74c1523..61ea568e1ddf 100644
--- a/drivers/net/hyperv/netvsc_drv.c
+++ b/drivers/net/hyperv/netvsc_drv.c
@@ -373,32 +373,29 @@ static u16 netvsc_select_queue(struct net_device *ndev, struct sk_buff *skb,
 	return txq;
 }
 
-static u32 fill_pg_buf(struct page *page, u32 offset, u32 len,
+static u32 fill_pg_buf(unsigned long hvpfn, u32 offset, u32 len,
 		       struct hv_page_buffer *pb)
 {
 	int j = 0;
 
-	/* Deal with compound pages by ignoring unused part
-	 * of the page.
-	 */
-	page += (offset >> PAGE_SHIFT);
-	offset &= ~PAGE_MASK;
+	hvpfn += offset >> HV_HYP_PAGE_SHIFT;
+	offset = offset & ~HV_HYP_PAGE_MASK;
 
 	while (len > 0) {
 		unsigned long bytes;
 
-		bytes = PAGE_SIZE - offset;
+		bytes = HV_HYP_PAGE_SIZE - offset;
 		if (bytes > len)
 			bytes = len;
-		pb[j].pfn = page_to_pfn(page);
+		pb[j].pfn = hvpfn;
 		pb[j].offset = offset;
 		pb[j].len = bytes;
 
 		offset += bytes;
 		len -= bytes;
 
-		if (offset == PAGE_SIZE && len) {
-			page++;
+		if (offset == HV_HYP_PAGE_SIZE && len) {
+			hvpfn++;
 			offset = 0;
 			j++;
 		}
@@ -421,23 +418,26 @@ static u32 init_page_array(void *hdr, u32 len, struct sk_buff *skb,
 	 * 2. skb linear data
 	 * 3. skb fragment data
 	 */
-	slots_used += fill_pg_buf(virt_to_page(hdr),
-				  offset_in_page(hdr),
-				  len, &pb[slots_used]);
+	slots_used += fill_pg_buf(virt_to_hvpfn(hdr),
+				  offset_in_hvpage(hdr),
+				  len,
+				  &pb[slots_used]);
 
 	packet->rmsg_size = len;
 	packet->rmsg_pgcnt = slots_used;
 
-	slots_used += fill_pg_buf(virt_to_page(data),
-				offset_in_page(data),
-				skb_headlen(skb), &pb[slots_used]);
+	slots_used += fill_pg_buf(virt_to_hvpfn(data),
+				  offset_in_hvpage(data),
+				  skb_headlen(skb),
+				  &pb[slots_used]);
 
 	for (i = 0; i < frags; i++) {
 		skb_frag_t *frag = skb_shinfo(skb)->frags + i;
 
-		slots_used += fill_pg_buf(skb_frag_page(frag),
-					skb_frag_off(frag),
-					skb_frag_size(frag), &pb[slots_used]);
+		slots_used += fill_pg_buf(page_to_hvpfn(skb_frag_page(frag)),
+					  skb_frag_off(frag),
+					  skb_frag_size(frag),
+					  &pb[slots_used]);
 	}
 	return slots_used;
 }
@@ -453,8 +453,8 @@ static int count_skb_frag_slots(struct sk_buff *skb)
 		unsigned long offset = skb_frag_off(frag);
 
 		/* Skip unused frames from start of page */
-		offset &= ~PAGE_MASK;
-		pages += PFN_UP(offset + size);
+		offset &= ~HV_HYP_PAGE_MASK;
+		pages += HVPFN_UP(offset + size);
 	}
 	return pages;
 }
@@ -462,12 +462,12 @@ static int count_skb_frag_slots(struct sk_buff *skb)
 static int netvsc_get_slots(struct sk_buff *skb)
 {
 	char *data = skb->data;
-	unsigned int offset = offset_in_page(data);
+	unsigned int offset = offset_in_hvpage(data);
 	unsigned int len = skb_headlen(skb);
 	int slots;
 	int frag_slots;
 
-	slots = DIV_ROUND_UP(offset + len, PAGE_SIZE);
+	slots = DIV_ROUND_UP(offset + len, HV_HYP_PAGE_SIZE);
 	frag_slots = count_skb_frag_slots(skb);
 	return slots + frag_slots;
 }
diff --git a/drivers/net/hyperv/rndis_filter.c b/drivers/net/hyperv/rndis_filter.c
index b81ceba38218..acc8d957bbfc 100644
--- a/drivers/net/hyperv/rndis_filter.c
+++ b/drivers/net/hyperv/rndis_filter.c
@@ -25,7 +25,7 @@
 
 static void rndis_set_multicast(struct work_struct *w);
 
-#define RNDIS_EXT_LEN PAGE_SIZE
+#define RNDIS_EXT_LEN HV_HYP_PAGE_SIZE
 struct rndis_request {
 	struct list_head list_ent;
 	struct completion  wait_event;
@@ -215,18 +215,18 @@ static int rndis_filter_send_request(struct rndis_device *dev,
 	packet->page_buf_cnt = 1;
 
 	pb[0].pfn = virt_to_phys(&req->request_msg) >>
-					PAGE_SHIFT;
+					HV_HYP_PAGE_SHIFT;
 	pb[0].len = req->request_msg.msg_len;
 	pb[0].offset =
-		(unsigned long)&req->request_msg & (PAGE_SIZE - 1);
+		(unsigned long)&req->request_msg & (HV_HYP_PAGE_SIZE - 1);
 
 	/* Add one page_buf when request_msg crossing page boundary */
-	if (pb[0].offset + pb[0].len > PAGE_SIZE) {
+	if (pb[0].offset + pb[0].len > HV_HYP_PAGE_SIZE) {
 		packet->page_buf_cnt++;
-		pb[0].len = PAGE_SIZE -
+		pb[0].len = HV_HYP_PAGE_SIZE -
 			pb[0].offset;
 		pb[1].pfn = virt_to_phys((void *)&req->request_msg
-			+ pb[0].len) >> PAGE_SHIFT;
+			+ pb[0].len) >> HV_HYP_PAGE_SHIFT;
 		pb[1].offset = 0;
 		pb[1].len = req->request_msg.msg_len -
 			pb[0].len;
-- 
2.28.0

[RFC v2 08/11] Input: hyperv-keyboard: Make ringbuffer at least take two pages

[Boqun Feng]

When PAGE_SIZE > HV_HYP_PAGE_SIZE, we need the ringbuffer size to be at
least 2 * PAGE_SIZE: one page for the header and at least one page of
the data part (because of the alignment requirement for double mapping).

So make sure the ringbuffer sizes to be at least 2 * PAGE_SIZE when
using vmbus_open() to establish the vmbus connection.

Signed-off-by: Boqun Feng <boqun.feng@gmail.com>
---
 drivers/input/serio/hyperv-keyboard.c | 4 ++--
 1 file changed, 2 insertions(+), 2 deletions(-)

diff --git a/drivers/input/serio/hyperv-keyboard.c b/drivers/input/serio/hyperv-keyboard.c
index df4e9f6f4529..77ba57ba2691 100644
--- a/drivers/input/serio/hyperv-keyboard.c
+++ b/drivers/input/serio/hyperv-keyboard.c
@@ -75,8 +75,8 @@ struct synth_kbd_keystroke {
 
 #define HK_MAXIMUM_MESSAGE_SIZE 256
 
-#define KBD_VSC_SEND_RING_BUFFER_SIZE		(40 * 1024)
-#define KBD_VSC_RECV_RING_BUFFER_SIZE		(40 * 1024)
+#define KBD_VSC_SEND_RING_BUFFER_SIZE		max(40 * 1024, 2 * PAGE_SIZE)
+#define KBD_VSC_RECV_RING_BUFFER_SIZE		max(40 * 1024, 2 * PAGE_SIZE)
 
 #define XTKBD_EMUL0     0xe0
 #define XTKBD_EMUL1     0xe1
-- 
2.28.0

[RFC v2 09/11] HID: hyperv: Make ringbuffer at least take two pages

[Boqun Feng]

When PAGE_SIZE > HV_HYP_PAGE_SIZE, we need the ringbuffer size to be at
least 2 * PAGE_SIZE: one page for the header and at least one page of
the data part (because of the alignment requirement for double mapping).

So make sure the ringbuffer sizes to be at least 2 * PAGE_SIZE when
using vmbus_open() to establish the vmbus connection.

Signed-off-by: Boqun Feng <boqun.feng@gmail.com>
---
 drivers/hid/hid-hyperv.c | 4 ++--
 1 file changed, 2 insertions(+), 2 deletions(-)

diff --git a/drivers/hid/hid-hyperv.c b/drivers/hid/hid-hyperv.c
index 0b6ee1dee625..dff032f17ad0 100644
--- a/drivers/hid/hid-hyperv.c
+++ b/drivers/hid/hid-hyperv.c
@@ -104,8 +104,8 @@ struct synthhid_input_report {
 
 #pragma pack(pop)
 
-#define INPUTVSC_SEND_RING_BUFFER_SIZE		(40 * 1024)
-#define INPUTVSC_RECV_RING_BUFFER_SIZE		(40 * 1024)
+#define INPUTVSC_SEND_RING_BUFFER_SIZE		max(40 * 1024, 2 * PAGE_SIZE)
+#define INPUTVSC_RECV_RING_BUFFER_SIZE		max(40 * 1024, 2 * PAGE_SIZE)
 
 
 enum pipe_prot_msg_type {
-- 
2.28.0

[RFC v2 10/11] Driver: hv: util: Make ringbuffer at least take two pages

[Boqun Feng]

When PAGE_SIZE > HV_HYP_PAGE_SIZE, we need the ringbuffer size to be at
least 2 * PAGE_SIZE: one page for the header and at least one page of
the data part (because of the alignment requirement for double mapping).

So make sure the ringbuffer sizes to be at least 2 * PAGE_SIZE when
using vmbus_open() to establish the vmbus connection.

Signed-off-by: Boqun Feng <boqun.feng@gmail.com>
---
 drivers/hv/hv_util.c | 16 ++++++++++++----
 1 file changed, 12 insertions(+), 4 deletions(-)

diff --git a/drivers/hv/hv_util.c b/drivers/hv/hv_util.c
index 92ee0fe4c919..73a77bead2be 100644
--- a/drivers/hv/hv_util.c
+++ b/drivers/hv/hv_util.c
@@ -461,6 +461,14 @@ static void heartbeat_onchannelcallback(void *context)
 	}
 }
 
+/*
+ * The size of each ring should be at least 2 * PAGE_SIZE, because we need one
+ * page for the header and at least another page (because of the alignment
+ * requirement for double mapping) for data part.
+ */
+#define HV_UTIL_RING_SEND_SIZE max(4 * HV_HYP_PAGE_SIZE, 2 * PAGE_SIZE)
+#define HV_UTIL_RING_RECV_SIZE max(4 * HV_HYP_PAGE_SIZE, 2 * PAGE_SIZE)
+
 static int util_probe(struct hv_device *dev,
 			const struct hv_vmbus_device_id *dev_id)
 {
@@ -491,8 +499,8 @@ static int util_probe(struct hv_device *dev,
 
 	hv_set_drvdata(dev, srv);
 
-	ret = vmbus_open(dev->channel, 4 * HV_HYP_PAGE_SIZE,
-			 4 * HV_HYP_PAGE_SIZE, NULL, 0, srv->util_cb,
+	ret = vmbus_open(dev->channel, HV_UTIL_RING_SEND_SIZE,
+			 HV_UTIL_RING_RECV_SIZE, NULL, 0, srv->util_cb,
 			 dev->channel);
 	if (ret)
 		goto error;
@@ -551,8 +559,8 @@ static int util_resume(struct hv_device *dev)
 			return ret;
 	}
 
-	ret = vmbus_open(dev->channel, 4 * HV_HYP_PAGE_SIZE,
-			 4 * HV_HYP_PAGE_SIZE, NULL, 0, srv->util_cb,
+	ret = vmbus_open(dev->channel, HV_UTIL_RING_SEND_SIZE,
+			 HV_UTIL_RING_RECV_SIZE, NULL, 0, srv->util_cb,
 			 dev->channel);
 	return ret;
 }
-- 
2.28.0

[RFC v2 11/11] scsi: storvsc: Support PAGE_SIZE larger than 4K

[Boqun Feng]

Hyper-V always use 4k page size (HV_HYP_PAGE_SIZE), so when
communicating with Hyper-V, a guest should always use HV_HYP_PAGE_SIZE
as the unit for page related data. For storvsc, the data is
vmbus_packet_mpb_array. And since in scsi_cmnd, sglist of pages (in unit
of PAGE_SIZE) is used, we need convert pages in the sglist of scsi_cmnd
into Hyper-V pages in vmbus_packet_mpb_array.

This patch does the conversion by dividing pages in sglist into Hyper-V
pages, offset and indexes in vmbus_packet_mpb_array are recalculated
accordingly.

Signed-off-by: Boqun Feng <boqun.feng@gmail.com>
---
 drivers/scsi/storvsc_drv.c | 60 ++++++++++++++++++++++++++++++++++----
 1 file changed, 54 insertions(+), 6 deletions(-)

diff --git a/drivers/scsi/storvsc_drv.c b/drivers/scsi/storvsc_drv.c
index 8f5f5dc863a4..3f6610717d4e 100644
--- a/drivers/scsi/storvsc_drv.c
+++ b/drivers/scsi/storvsc_drv.c
@@ -1739,23 +1739,71 @@ static int storvsc_queuecommand(struct Scsi_Host *host, struct scsi_cmnd *scmnd)
 	payload_sz = sizeof(cmd_request->mpb);
 
 	if (sg_count) {
-		if (sg_count > MAX_PAGE_BUFFER_COUNT) {
+		unsigned int hvpg_idx = 0;
+		unsigned int j = 0;
+		unsigned long hvpg_offset = sgl->offset & ~HV_HYP_PAGE_MASK;
+		unsigned int hvpg_count = HVPFN_UP(hvpg_offset + length);
 
-			payload_sz = (sg_count * sizeof(u64) +
+		if (hvpg_count > MAX_PAGE_BUFFER_COUNT) {
+
+			payload_sz = (hvpg_count * sizeof(u64) +
 				      sizeof(struct vmbus_packet_mpb_array));
 			payload = kzalloc(payload_sz, GFP_ATOMIC);
 			if (!payload)
 				return SCSI_MLQUEUE_DEVICE_BUSY;
 		}
 
+		/*
+		 * sgl is a list of PAGEs, and payload->range.pfn_array
+		 * expects the page number in the unit of HV_HYP_PAGE_SIZE (the
+		 * page size that Hyper-V uses, so here we need to divide PAGEs
+		 * into HV_HYP_PAGE in case that PAGE_SIZE > HV_HYP_PAGE_SIZE.
+		 */
 		payload->range.len = length;
-		payload->range.offset = sgl[0].offset;
+		payload->range.offset = sgl[0].offset & ~HV_HYP_PAGE_MASK;
+		hvpg_idx = sgl[0].offset >> HV_HYP_PAGE_SHIFT;
 
 		cur_sgl = sgl;
-		for (i = 0; i < sg_count; i++) {
-			payload->range.pfn_array[i] =
-				page_to_pfn(sg_page((cur_sgl)));
+		for (i = 0, j = 0; i < sg_count; i++) {
+			/*
+			 * "PAGE_SIZE / HV_HYP_PAGE_SIZE - hvpg_idx" is the #
+			 * of HV_HYP_PAGEs in the current PAGE.
+			 *
+			 * "hvpg_count - j" is the # of unhandled HV_HYP_PAGEs.
+			 *
+			 * As shown in the following, the minimal of both is
+			 * the # of HV_HYP_PAGEs, we need to handle in this
+			 * PAGE.
+			 *
+			 * |------------------ PAGE ----------------------|
+			 * |   PAGE_SIZE / HV_HYP_PAGE_SIZE in total      |
+			 * |hvpg|hvpg| ...                 |hvpg|... |hvpg|
+			 *           ^                     ^
+			 *         hvpg_idx                |
+			 *           ^                     |
+			 *           +---(hvpg_count - j)--+
+			 *
+			 * or
+			 *
+			 * |------------------ PAGE ----------------------|
+			 * |   PAGE_SIZE / HV_HYP_PAGE_SIZE in total      |
+			 * |hvpg|hvpg| ...                 |hvpg|... |hvpg|
+			 *           ^                                           ^
+			 *         hvpg_idx                                      |
+			 *           ^                                           |
+			 *           +---(hvpg_count - j)------------------------+
+			 */
+			unsigned int nr_hvpg = min((unsigned int)(PAGE_SIZE / HV_HYP_PAGE_SIZE) - hvpg_idx,
+						   hvpg_count - j);
+			unsigned int k;
+
+			for (k = 0; k < nr_hvpg; k++) {
+				payload->range.pfn_array[j] =
+					page_to_hvpfn(sg_page((cur_sgl))) + hvpg_idx + k;
+				j++;
+			}
 			cur_sgl = sg_next(cur_sgl);
+			hvpg_idx = 0;
 		}
 	}
 
-- 
2.28.0

Re: [RFC v2 09/11] HID: hyperv: Make ringbuffer at least take two pages

[Jiri Kosina]

On Wed, 2 Sep 2020, Boqun Feng wrote:

> When PAGE_SIZE > HV_HYP_PAGE_SIZE, we need the ringbuffer size to be at
> least 2 * PAGE_SIZE: one page for the header and at least one page of
> the data part (because of the alignment requirement for double mapping).
> 
> So make sure the ringbuffer sizes to be at least 2 * PAGE_SIZE when
> using vmbus_open() to establish the vmbus connection.
> 
> Signed-off-by: Boqun Feng <boqun.feng@gmail.com>

Acked-by: Jiri Kosina <jkosina@suse.cz>

-- 
Jiri Kosina
SUSE Labs

RE: [RFC v2 02/11] Drivers: hv: vmbus: Move __vmbus_open()

[Michael Kelley]

From: Boqun Feng <boqun.feng@gmail.com> Sent: Tuesday, September 1, 2020 8:01 PM
> 
> Pure function movement, no functional changes. The move is made, because
> in a later change, __vmbus_open() will rely on some static functions
> afterwards, so we sperate the move and the modification of

s/sperate/separate/

> __vmbus_open() in two patches to make it easy to review.
> 
> Signed-off-by: Boqun Feng <boqun.feng@gmail.com>
> Reviewed-by: Wei Liu <wei.liu@kernel.org>
> ---
>  drivers/hv/channel.c | 309 ++++++++++++++++++++++---------------------
>  1 file changed, 155 insertions(+), 154 deletions(-)
> 

RE: [RFC v2 03/11] Drivers: hv: vmbus: Introduce types of GPADL

[Michael Kelley]

From: Boqun Feng <boqun.feng@gmail.com> Sent: Tuesday, September 1, 2020 8:01 PM
> 
> This patch introduces two types of GPADL: HV_GPADL_{BUFFER, RING}. The
> types of GPADL are purely the concept in the guest, IOW the hypervisor
> treat them as the same.
> 
> The reason of introducing the types of GPADL is to support guests whose

s/of/for/

> page size is not 4k (the page size of Hyper-V hypervisor). In these
> guests, both the headers and the data parts of the ringbuffers need to
> be aligned to the PAGE_SIZE, because 1) some of the ringbuffers will be
> mapped into userspace and 2) we use "double mapping" mechanism to
> support fast wrap-around, and "double mapping" relies on ringbuffers
> being page-aligned. However, the Hyper-V hypervisor only uses 4k
> (HV_HYP_PAGE_SIZE) headers. Our solution to this is that we always make
> the headers of ringbuffers take one guest page and when GPADL is
> established between the guest and hypervisor, the only first 4k of
> header is used. To handle this special case, we need the types of GPADL
> to differ different guest memory usage for GPADL.
> 
> Type enum is introduced along with several general interfaces to
> describe the differences between normal buffer GPADL and ringbuffer
> GPADL.
> 
> Signed-off-by: Boqun Feng <boqun.feng@gmail.com>
> ---
>  drivers/hv/channel.c   | 159 +++++++++++++++++++++++++++++++++++------
>  include/linux/hyperv.h |  44 +++++++++++-
>  2 files changed, 182 insertions(+), 21 deletions(-)
> 
> diff --git a/drivers/hv/channel.c b/drivers/hv/channel.c
> index 1cbe8fc931fc..7c443fd567e4 100644
> --- a/drivers/hv/channel.c
> +++ b/drivers/hv/channel.c
> @@ -35,6 +35,98 @@ static unsigned long virt_to_hvpfn(void *addr)
>  	return  paddr >> HV_HYP_PAGE_SHIFT;
>  }
> 
> +/*
> + * hv_gpadl_size - Return the real size of a gpadl, the size that Hyper-V uses
> + *
> + * For BUFFER gpadl, Hyper-V uses the exact same size as the guest does.
> + *
> + * For RING gpadl, in each ring, the guest uses one PAGE_SIZE as the header
> + * (because of the alignment requirement), however, the hypervisor only
> + * uses the first HV_HYP_PAGE_SIZE as the header, therefore leaving a
> + * (PAGE_SIZE - HV_HYP_PAGE_SIZE) gap. And since there are two rings in a
> + * ringbuffer, So the total size for a RING gpadl that Hyper-V uses is the

Unneeded word "So"

> + * total size that the guest uses minus twice of the gap size.
> + */
> +static inline u32 hv_gpadl_size(enum hv_gpadl_type type, u32 size)
> +{
> +	switch (type) {
> +	case HV_GPADL_BUFFER:
> +		return size;
> +	case HV_GPADL_RING:
> +		/* The size of a ringbuffer must be page-aligned */
> +		BUG_ON(size % PAGE_SIZE);
> +		/*
> +		 * Two things to notice here:
> +		 * 1) We're processing two ring buffers as a unit
> +		 * 2) We're skipping any space larger than HV_HYP_PAGE_SIZE in
> +		 * the first guest-size page of each of the two ring buffers.
> +		 * So we effectively subtract out two guest-size pages, and add
> +		 * back two Hyper-V size pages.
> +		 */
> +		return size - 2 * (PAGE_SIZE - HV_HYP_PAGE_SIZE);
> +	}
> +	BUG();
> +	return 0;
> +}
> +
> +/*
> + * hv_ring_gpadl_send_offset - Calculate the send offset in a ring gpadl based
> + *                             on the offset in the guest
> + *
> + * @send_offset: the offset (in bytes) where the send ringbuffer starts in the
> + *               virtual address space of the guest
> + */
> +static inline u32 hv_ring_gpadl_send_offset(u32 send_offset)
> +{
> +
> +	/*
> +	 * For RING gpadl, in each ring, the guest uses one PAGE_SIZE as the
> +	 * header (because of the alignment requirement), however, the
> +	 * hypervisor only uses the first HV_HYP_PAGE_SIZE as the header,
> +	 * therefore leaving a (PAGE_SIZE - HV_HYP_PAGE_SIZE) gap.
> +	 *
> +	 * And to calculate the effective send offset in gpadl, we need to
> +	 * substract this gap.
> +	 */
> +	return send_offset - (PAGE_SIZE - HV_HYP_PAGE_SIZE);
> +}
> +
> +/*
> + * hv_gpadl_hvpfn - Return the Hyper-V page PFN of the @i th Hyper-V page in
> + *                  the gpadl
> + *
> + * @type: the type of the gpadl
> + * @kbuffer: the pointer to the gpadl in the guest
> + * @size: the total size (in bytes) of the gpadl
> + * @send_offset: the offset (in bytes) where the send ringbuffer starts in the
> + *               virtual address space of the guest
> + * @i: the index
> + */
> +static inline u64 hv_gpadl_hvpfn(enum hv_gpadl_type type, void *kbuffer,
> +				 u32 size, u32 send_offset, int i)
> +{
> +	int send_idx = hv_ring_gpadl_send_offset(send_offset) >> HV_HYP_PAGE_SHIFT;
> +	unsigned long delta = 0UL;
> +
> +	switch (type) {
> +	case HV_GPADL_BUFFER:
> +		break;
> +	case HV_GPADL_RING:
> +		if (i == 0)
> +			delta = 0;
> +		else if (i <= send_idx)
> +			delta = PAGE_SIZE - HV_HYP_PAGE_SIZE;
> +		else
> +			delta = 2 * (PAGE_SIZE - HV_HYP_PAGE_SIZE);
> +		break;
> +	default:
> +		BUG();
> +		break;
> +	}
> +
> +	return virt_to_hvpfn(kbuffer + delta + (HV_HYP_PAGE_SIZE * i));
> +}
> +
>  /*
>   * vmbus_setevent- Trigger an event notification on the specified
>   * channel.
> @@ -160,7 +252,8 @@ EXPORT_SYMBOL_GPL(vmbus_send_modifychannel);
>  /*
>   * create_gpadl_header - Creates a gpadl for the specified buffer
>   */
> -static int create_gpadl_header(void *kbuffer, u32 size,
> +static int create_gpadl_header(enum hv_gpadl_type type, void *kbuffer,
> +			       u32 size, u32 send_offset,
>  			       struct vmbus_channel_msginfo **msginfo)
>  {
>  	int i;
> @@ -173,7 +266,7 @@ static int create_gpadl_header(void *kbuffer, u32 size,
> 
>  	int pfnsum, pfncount, pfnleft, pfncurr, pfnsize;
> 
> -	pagecount = size >> HV_HYP_PAGE_SHIFT;
> +	pagecount = hv_gpadl_size(type, size) >> HV_HYP_PAGE_SHIFT;
> 
>  	/* do we need a gpadl body msg */
>  	pfnsize = MAX_SIZE_CHANNEL_MESSAGE -
> @@ -200,10 +293,10 @@ static int create_gpadl_header(void *kbuffer, u32 size,
>  		gpadl_header->range_buflen = sizeof(struct gpa_range) +
>  					 pagecount * sizeof(u64);
>  		gpadl_header->range[0].byte_offset = 0;
> -		gpadl_header->range[0].byte_count = size;
> +		gpadl_header->range[0].byte_count = hv_gpadl_size(type, size);
>  		for (i = 0; i < pfncount; i++)
> -			gpadl_header->range[0].pfn_array[i] = virt_to_hvpfn(
> -				kbuffer + HV_HYP_PAGE_SIZE * i);
> +			gpadl_header->range[0].pfn_array[i] = hv_gpadl_hvpfn(
> +				type, kbuffer, size, send_offset, i);
>  		*msginfo = msgheader;
> 
>  		pfnsum = pfncount;
> @@ -254,8 +347,8 @@ static int create_gpadl_header(void *kbuffer, u32 size,
>  			 * so the hypervisor guarantees that this is ok.
>  			 */
>  			for (i = 0; i < pfncurr; i++)
> -				gpadl_body->pfn[i] = virt_to_hvpfn(
> -					kbuffer + HV_HYP_PAGE_SIZE * (pfnsum + i));
> +				gpadl_body->pfn[i] = hv_gpadl_hvpfn(type,
> +					kbuffer, size, send_offset, pfnsum + i);
> 
>  			/* add to msg header */
>  			list_add_tail(&msgbody->msglistentry,
> @@ -281,10 +374,10 @@ static int create_gpadl_header(void *kbuffer, u32 size,
>  		gpadl_header->range_buflen = sizeof(struct gpa_range) +
>  					 pagecount * sizeof(u64);
>  		gpadl_header->range[0].byte_offset = 0;
> -		gpadl_header->range[0].byte_count = size;
> +		gpadl_header->range[0].byte_count = hv_gpadl_size(type, size);
>  		for (i = 0; i < pagecount; i++)
> -			gpadl_header->range[0].pfn_array[i] = virt_to_hvpfn(
> -				kbuffer + HV_HYP_PAGE_SIZE * i);
> +			gpadl_header->range[0].pfn_array[i] = hv_gpadl_hvpfn(
> +				type, kbuffer, size, send_offset, i);
> 
>  		*msginfo = msgheader;
>  	}
> @@ -297,15 +390,20 @@ static int create_gpadl_header(void *kbuffer, u32 size,
>  }
> 
>  /*
> - * vmbus_establish_gpadl - Establish a GPADL for the specified buffer
> + * __vmbus_establish_gpadl - Establish a GPADL for a buffer or ringbuffer
>   *
>   * @channel: a channel
> + * @type: the type of the corresponding GPADL, only meaningful for the guest.
>   * @kbuffer: from kmalloc or vmalloc
>   * @size: page-size multiple
> + * @send_offset: the offset (in bytes) where the send ring buffer starts,
> + * 		 should be 0 for BUFFER type gpadl
>   * @gpadl_handle: some funky thing
>   */
> -int vmbus_establish_gpadl(struct vmbus_channel *channel, void *kbuffer,
> -			       u32 size, u32 *gpadl_handle)
> +static int __vmbus_establish_gpadl(struct vmbus_channel *channel,
> +				   enum hv_gpadl_type type, void *kbuffer,
> +				   u32 size, u32 send_offset,
> +				   u32 *gpadl_handle)
>  {
>  	struct vmbus_channel_gpadl_header *gpadlmsg;
>  	struct vmbus_channel_gpadl_body *gpadl_body;
> @@ -319,7 +417,7 @@ int vmbus_establish_gpadl(struct vmbus_channel *channel, void
> *kbuffer,
>  	next_gpadl_handle =
>  		(atomic_inc_return(&vmbus_connection.next_gpadl_handle) - 1);
> 
> -	ret = create_gpadl_header(kbuffer, size, &msginfo);
> +	ret = create_gpadl_header(type, kbuffer, size, send_offset, &msginfo);
>  	if (ret)
>  		return ret;
> 
> @@ -400,6 +498,21 @@ int vmbus_establish_gpadl(struct vmbus_channel *channel, void
> *kbuffer,
>  	kfree(msginfo);
>  	return ret;
>  }
> +
> +/*
> + * vmbus_establish_gpadl - Establish a GPADL for the specified buffer
> + *
> + * @channel: a channel
> + * @kbuffer: from kmalloc or vmalloc
> + * @size: page-size multiple
> + * @gpadl_handle: some funky thing
> + */
> +int vmbus_establish_gpadl(struct vmbus_channel *channel, void *kbuffer,
> +			  u32 size, u32 *gpadl_handle)
> +{
> +	return __vmbus_establish_gpadl(channel, HV_GPADL_BUFFER, kbuffer, size,
> +				       0U, gpadl_handle);
> +}
>  EXPORT_SYMBOL_GPL(vmbus_establish_gpadl);
> 
>  static int __vmbus_open(struct vmbus_channel *newchannel,
> @@ -438,10 +551,11 @@ static int __vmbus_open(struct vmbus_channel *newchannel,
>  	/* Establish the gpadl for the ring buffer */
>  	newchannel->ringbuffer_gpadlhandle = 0;
> 
> -	err = vmbus_establish_gpadl(newchannel,
> -				    page_address(newchannel->ringbuffer_page),
> -				    (send_pages + recv_pages) << PAGE_SHIFT,
> -				    &newchannel->ringbuffer_gpadlhandle);
> +	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);
>  	if (err)
>  		goto error_clean_ring;
> 
> @@ -462,7 +576,13 @@ static int __vmbus_open(struct vmbus_channel *newchannel,
>  	open_msg->openid = newchannel->offermsg.child_relid;
>  	open_msg->child_relid = newchannel->offermsg.child_relid;
>  	open_msg->ringbuffer_gpadlhandle = newchannel->ringbuffer_gpadlhandle;
> -	open_msg->downstream_ringbuffer_pageoffset = newchannel-
> >ringbuffer_send_offset;
> +	/*
> +	 * The unit of ->downstream_ringbuffer_pageoffset is HV_HYP_PAGE and
> +	 * the unit of ->ringbuffer_send_offset is PAGE, so here we first
> +	 * calculate it into bytes and then convert into HV_HYP_PAGE.
> +	 */
> +	open_msg->downstream_ringbuffer_pageoffset =
> +		hv_ring_gpadl_send_offset(newchannel->ringbuffer_send_offset << PAGE_SHIFT) >> HV_HYP_PAGE_SHIFT;

Line length?

>  	open_msg->target_vp = hv_cpu_number_to_vp_number(newchannel->target_cpu);
> 
>  	if (userdatalen)
> @@ -556,7 +676,6 @@ int vmbus_open(struct vmbus_channel *newchannel,
>  }
>  EXPORT_SYMBOL_GPL(vmbus_open);
> 
> -
>  /*
>   * vmbus_teardown_gpadl -Teardown the specified GPADL handle
>   */
> diff --git a/include/linux/hyperv.h b/include/linux/hyperv.h
> index 38100e80360a..7d16dd28aa48 100644
> --- a/include/linux/hyperv.h
> +++ b/include/linux/hyperv.h
> @@ -29,6 +29,48 @@
> 
>  #pragma pack(push, 1)
> 
> +/*
> + * Types for GPADL, decides is how GPADL header is created.
> + *
> + * It doesn't make much difference between BUFFER and RING if PAGE_SIZE is the
> + * same as HV_HYP_PAGE_SIZE.
> + *
> + * If PAGE_SIZE is bigger than HV_HYP_PAGE_SIZE, the headers of ring buffers
> + * will be of PAGE_SIZE, however, only the first HV_HYP_PAGE will be put
> + * into gpadl, therefore the number for HV_HYP_PAGE and the indexes of each
> + * HV_HYP_PAGE will be different between different types of GPADL, for example
> + * if PAGE_SIZE is 64K:
> + *
> + * BUFFER:
> + *
> + * gva:    |--       64k      --|--       64k      --| ... |
> + * gpa:    | 4k | 4k | ... | 4k | 4k | 4k | ... | 4k |
> + * index:  0    1    2     15   16   17   18 .. 31   32 ...
> + *         |    |    ...   |    |    |   ...    |   ...
> + *         v    V          V    V    V          V
> + * gpadl:  | 4k | 4k | ... | 4k | 4k | 4k | ... | 4k | ... |
> + * index:  0    1    2 ... 15   16   17   18 .. 31   32 ...
> + *
> + * RING:
> + *
> + *         | header  |           data           | header  |     data      |
> + * gva:    |-- 64k --|--       64k      --| ... |-- 64k --|-- 64k --| ... |
> + * gpa:    | 4k | .. | 4k | 4k | ... | 4k | ... | 4k | .. | 4k | .. | ... |
> + * index:  0    1    16   17   18    31   ...   n   n+1  n+16 ...         2n
> + *         |         /    /          /          |         /               /
> + *         |        /    /          /           |        /               /
> + *         |       /    /   ...    /    ...     |       /      ...      /
> + *         |      /    /          /             |      /               /
> + *         |     /    /          /              |     /               /
> + *         V    V    V          V               V    V               v
> + * gpadl:  | 4k | 4k |   ...    |    ...        | 4k | 4k |  ...     |
> + * index:  0    1    2   ...    16   ...       n-15 n-14 n-13  ...  2n-30
> + */
> +enum hv_gpadl_type {
> +	HV_GPADL_BUFFER,
> +	HV_GPADL_RING
> +};
> +
>  /* Single-page buffer */
>  struct hv_page_buffer {
>  	u32 len;
> @@ -111,7 +153,7 @@ struct hv_ring_buffer {
>  	} feature_bits;
> 
>  	/* Pad it to PAGE_SIZE so that data starts on page boundary */
> -	u8	reserved2[4028];
> +	u8	reserved2[PAGE_SIZE - 68];
> 
>  	/*
>  	 * Ring data starts here + RingDataStartOffset
> --
> 2.28.0

RE: [RFC v2 07/11] hv_netvsc: Use HV_HYP_PAGE_SIZE for Hyper-V communication

[Michael Kelley]

From: Boqun Feng <boqun.feng@gmail.com> Sent: Tuesday, September 1, 2020 8:01 PM
> 
> When communicating with Hyper-V, HV_HYP_PAGE_SIZE should be used since
> that's the page size used by Hyper-V and Hyper-V expects all
> page-related data using the unit of HY_HYP_PAGE_SIZE, for example, the
> "pfn" in hv_page_buffer is actually the HV_HYP_PAGE (i.e. the Hyper-V
> page) number.
> 
> In order to support guest whose page size is not 4k, we need to make
> hv_netvsc always use HV_HYP_PAGE_SIZE for Hyper-V communication.
> 
> Signed-off-by: Boqun Feng <boqun.feng@gmail.com>
> ---
>  drivers/net/hyperv/netvsc.c       |  2 +-
>  drivers/net/hyperv/netvsc_drv.c   | 46 +++++++++++++++----------------
>  drivers/net/hyperv/rndis_filter.c | 12 ++++----
>  3 files changed, 30 insertions(+), 30 deletions(-)
> 
> diff --git a/drivers/net/hyperv/netvsc.c b/drivers/net/hyperv/netvsc.c
> index 41f5cf0bb997..1d6f2256da6b 100644
> --- a/drivers/net/hyperv/netvsc.c
> +++ b/drivers/net/hyperv/netvsc.c
> @@ -794,7 +794,7 @@ static void netvsc_copy_to_send_buf(struct netvsc_device
> *net_device,
>  	}
> 
>  	for (i = 0; i < page_count; i++) {
> -		char *src = phys_to_virt(pb[i].pfn << PAGE_SHIFT);
> +		char *src = phys_to_virt(pb[i].pfn << HV_HYP_PAGE_SHIFT);
>  		u32 offset = pb[i].offset;
>  		u32 len = pb[i].len;
> 
> diff --git a/drivers/net/hyperv/netvsc_drv.c b/drivers/net/hyperv/netvsc_drv.c
> index 64b0a74c1523..61ea568e1ddf 100644
> --- a/drivers/net/hyperv/netvsc_drv.c
> +++ b/drivers/net/hyperv/netvsc_drv.c
> @@ -373,32 +373,29 @@ static u16 netvsc_select_queue(struct net_device *ndev, struct
> sk_buff *skb,
>  	return txq;
>  }
> 
> -static u32 fill_pg_buf(struct page *page, u32 offset, u32 len,
> +static u32 fill_pg_buf(unsigned long hvpfn, u32 offset, u32 len,
>  		       struct hv_page_buffer *pb)
>  {
>  	int j = 0;
> 
> -	/* Deal with compound pages by ignoring unused part
> -	 * of the page.
> -	 */
> -	page += (offset >> PAGE_SHIFT);
> -	offset &= ~PAGE_MASK;
> +	hvpfn += offset >> HV_HYP_PAGE_SHIFT;
> +	offset = offset & ~HV_HYP_PAGE_MASK;
> 
>  	while (len > 0) {
>  		unsigned long bytes;
> 
> -		bytes = PAGE_SIZE - offset;
> +		bytes = HV_HYP_PAGE_SIZE - offset;
>  		if (bytes > len)
>  			bytes = len;
> -		pb[j].pfn = page_to_pfn(page);
> +		pb[j].pfn = hvpfn;
>  		pb[j].offset = offset;
>  		pb[j].len = bytes;
> 
>  		offset += bytes;
>  		len -= bytes;
> 
> -		if (offset == PAGE_SIZE && len) {
> -			page++;
> +		if (offset == HV_HYP_PAGE_SIZE && len) {
> +			hvpfn++;
>  			offset = 0;
>  			j++;
>  		}
> @@ -421,23 +418,26 @@ static u32 init_page_array(void *hdr, u32 len, struct sk_buff *skb,
>  	 * 2. skb linear data
>  	 * 3. skb fragment data
>  	 */
> -	slots_used += fill_pg_buf(virt_to_page(hdr),
> -				  offset_in_page(hdr),
> -				  len, &pb[slots_used]);
> +	slots_used += fill_pg_buf(virt_to_hvpfn(hdr),
> +				  offset_in_hvpage(hdr),
> +				  len,
> +				  &pb[slots_used]);
> 
>  	packet->rmsg_size = len;
>  	packet->rmsg_pgcnt = slots_used;
> 
> -	slots_used += fill_pg_buf(virt_to_page(data),
> -				offset_in_page(data),
> -				skb_headlen(skb), &pb[slots_used]);
> +	slots_used += fill_pg_buf(virt_to_hvpfn(data),
> +				  offset_in_hvpage(data),
> +				  skb_headlen(skb),
> +				  &pb[slots_used]);
> 
>  	for (i = 0; i < frags; i++) {
>  		skb_frag_t *frag = skb_shinfo(skb)->frags + i;
> 
> -		slots_used += fill_pg_buf(skb_frag_page(frag),
> -					skb_frag_off(frag),
> -					skb_frag_size(frag), &pb[slots_used]);
> +		slots_used += fill_pg_buf(page_to_hvpfn(skb_frag_page(frag)),
> +					  skb_frag_off(frag),
> +					  skb_frag_size(frag),
> +					  &pb[slots_used]);
>  	}
>  	return slots_used;
>  }
> @@ -453,8 +453,8 @@ static int count_skb_frag_slots(struct sk_buff *skb)
>  		unsigned long offset = skb_frag_off(frag);
> 
>  		/* Skip unused frames from start of page */
> -		offset &= ~PAGE_MASK;
> -		pages += PFN_UP(offset + size);
> +		offset &= ~HV_HYP_PAGE_MASK;
> +		pages += HVPFN_UP(offset + size);
>  	}
>  	return pages;
>  }
> @@ -462,12 +462,12 @@ static int count_skb_frag_slots(struct sk_buff *skb)
>  static int netvsc_get_slots(struct sk_buff *skb)
>  {
>  	char *data = skb->data;
> -	unsigned int offset = offset_in_page(data);
> +	unsigned int offset = offset_in_hvpage(data);
>  	unsigned int len = skb_headlen(skb);
>  	int slots;
>  	int frag_slots;
> 
> -	slots = DIV_ROUND_UP(offset + len, PAGE_SIZE);
> +	slots = DIV_ROUND_UP(offset + len, HV_HYP_PAGE_SIZE);
>  	frag_slots = count_skb_frag_slots(skb);
>  	return slots + frag_slots;
>  }
> diff --git a/drivers/net/hyperv/rndis_filter.c b/drivers/net/hyperv/rndis_filter.c
> index b81ceba38218..acc8d957bbfc 100644
> --- a/drivers/net/hyperv/rndis_filter.c
> +++ b/drivers/net/hyperv/rndis_filter.c
> @@ -25,7 +25,7 @@
> 
>  static void rndis_set_multicast(struct work_struct *w);
> 
> -#define RNDIS_EXT_LEN PAGE_SIZE
> +#define RNDIS_EXT_LEN HV_HYP_PAGE_SIZE
>  struct rndis_request {
>  	struct list_head list_ent;
>  	struct completion  wait_event;
> @@ -215,18 +215,18 @@ static int rndis_filter_send_request(struct rndis_device *dev,
>  	packet->page_buf_cnt = 1;
> 
>  	pb[0].pfn = virt_to_phys(&req->request_msg) >>
> -					PAGE_SHIFT;
> +					HV_HYP_PAGE_SHIFT;
>  	pb[0].len = req->request_msg.msg_len;
>  	pb[0].offset =
> -		(unsigned long)&req->request_msg & (PAGE_SIZE - 1);
> +		(unsigned long)&req->request_msg & (HV_HYP_PAGE_SIZE - 1);

Use offset_in_hvpage() as defined in patch 6 of the series?

> 
>  	/* Add one page_buf when request_msg crossing page boundary */
> -	if (pb[0].offset + pb[0].len > PAGE_SIZE) {
> +	if (pb[0].offset + pb[0].len > HV_HYP_PAGE_SIZE) {
>  		packet->page_buf_cnt++;
> -		pb[0].len = PAGE_SIZE -
> +		pb[0].len = HV_HYP_PAGE_SIZE -
>  			pb[0].offset;
>  		pb[1].pfn = virt_to_phys((void *)&req->request_msg
> -			+ pb[0].len) >> PAGE_SHIFT;
> +			+ pb[0].len) >> HV_HYP_PAGE_SHIFT;
>  		pb[1].offset = 0;
>  		pb[1].len = req->request_msg.msg_len -
>  			pb[0].len;
> --
> 2.28.0

RE: [RFC v2 11/11] scsi: storvsc: Support PAGE_SIZE larger than 4K

[Michael Kelley]

From: Boqun Feng <boqun.feng@gmail.com> Sent: Tuesday, September 1, 2020 8:01 PM
> 
> Hyper-V always use 4k page size (HV_HYP_PAGE_SIZE), so when
> communicating with Hyper-V, a guest should always use HV_HYP_PAGE_SIZE
> as the unit for page related data. For storvsc, the data is
> vmbus_packet_mpb_array. And since in scsi_cmnd, sglist of pages (in unit
> of PAGE_SIZE) is used, we need convert pages in the sglist of scsi_cmnd
> into Hyper-V pages in vmbus_packet_mpb_array.
> 
> This patch does the conversion by dividing pages in sglist into Hyper-V
> pages, offset and indexes in vmbus_packet_mpb_array are recalculated
> accordingly.
> 
> Signed-off-by: Boqun Feng <boqun.feng@gmail.com>
> ---
>  drivers/scsi/storvsc_drv.c | 60 ++++++++++++++++++++++++++++++++++----
>  1 file changed, 54 insertions(+), 6 deletions(-)
> 
> diff --git a/drivers/scsi/storvsc_drv.c b/drivers/scsi/storvsc_drv.c
> index 8f5f5dc863a4..3f6610717d4e 100644
> --- a/drivers/scsi/storvsc_drv.c
> +++ b/drivers/scsi/storvsc_drv.c
> @@ -1739,23 +1739,71 @@ static int storvsc_queuecommand(struct Scsi_Host *host, struct
> scsi_cmnd *scmnd)
>  	payload_sz = sizeof(cmd_request->mpb);
> 
>  	if (sg_count) {
> -		if (sg_count > MAX_PAGE_BUFFER_COUNT) {
> +		unsigned int hvpg_idx = 0;
> +		unsigned int j = 0;
> +		unsigned long hvpg_offset = sgl->offset & ~HV_HYP_PAGE_MASK;
> +		unsigned int hvpg_count = HVPFN_UP(hvpg_offset + length);
> 
> -			payload_sz = (sg_count * sizeof(u64) +
> +		if (hvpg_count > MAX_PAGE_BUFFER_COUNT) {
> +
> +			payload_sz = (hvpg_count * sizeof(u64) +
>  				      sizeof(struct vmbus_packet_mpb_array));
>  			payload = kzalloc(payload_sz, GFP_ATOMIC);
>  			if (!payload)
>  				return SCSI_MLQUEUE_DEVICE_BUSY;
>  		}
> 
> +		/*
> +		 * sgl is a list of PAGEs, and payload->range.pfn_array
> +		 * expects the page number in the unit of HV_HYP_PAGE_SIZE (the
> +		 * page size that Hyper-V uses, so here we need to divide PAGEs
> +		 * into HV_HYP_PAGE in case that PAGE_SIZE > HV_HYP_PAGE_SIZE.
> +		 */
>  		payload->range.len = length;
> -		payload->range.offset = sgl[0].offset;
> +		payload->range.offset = sgl[0].offset & ~HV_HYP_PAGE_MASK;
> +		hvpg_idx = sgl[0].offset >> HV_HYP_PAGE_SHIFT;
> 
>  		cur_sgl = sgl;
> -		for (i = 0; i < sg_count; i++) {
> -			payload->range.pfn_array[i] =
> -				page_to_pfn(sg_page((cur_sgl)));
> +		for (i = 0, j = 0; i < sg_count; i++) {
> +			/*
> +			 * "PAGE_SIZE / HV_HYP_PAGE_SIZE - hvpg_idx" is the #
> +			 * of HV_HYP_PAGEs in the current PAGE.
> +			 *
> +			 * "hvpg_count - j" is the # of unhandled HV_HYP_PAGEs.
> +			 *
> +			 * As shown in the following, the minimal of both is
> +			 * the # of HV_HYP_PAGEs, we need to handle in this
> +			 * PAGE.
> +			 *
> +			 * |------------------ PAGE ----------------------|
> +			 * |   PAGE_SIZE / HV_HYP_PAGE_SIZE in total      |
> +			 * |hvpg|hvpg| ...                 |hvpg|... |hvpg|
> +			 *           ^                     ^
> +			 *         hvpg_idx                |
> +			 *           ^                     |
> +			 *           +---(hvpg_count - j)--+
> +			 *
> +			 * or
> +			 *
> +			 * |------------------ PAGE ----------------------|
> +			 * |   PAGE_SIZE / HV_HYP_PAGE_SIZE in total      |
> +			 * |hvpg|hvpg| ...                 |hvpg|... |hvpg|
> +			 *           ^                                           ^
> +			 *         hvpg_idx                                      |
> +			 *           ^                                           |
> +			 *           +---(hvpg_count - j)------------------------+
> +			 */
> +			unsigned int nr_hvpg = min((unsigned int)(PAGE_SIZE / HV_HYP_PAGE_SIZE) - hvpg_idx,
> +						   hvpg_count - j);
> +			unsigned int k;
> +
> +			for (k = 0; k < nr_hvpg; k++) {
> +				payload->range.pfn_array[j] =
> +					page_to_hvpfn(sg_page((cur_sgl))) + hvpg_idx + k;
> +				j++;
> +			}
>  			cur_sgl = sg_next(cur_sgl);
> +			hvpg_idx = 0;
>  		}

This code works; I don't see any errors.  But I think it can be made simpler based
on doing two things:
1)  Rather than iterating over the sg_count, and having to calculate nr_hvpg on
each iteration, base the exit decision on having filled up the pfn_array[].  You've
already calculated the exact size of the array that is needed given the data
length, so it's easy to exit when the array is full.
2) In the inner loop, iterate from hvpg_idx to PAGE_SIZE/HV_HYP_PAGE_SIZE
rather than from 0 to a calculated value.

Also, as an optimization, pull page_to_hvpfn(sg_page((cur_sgl)) out of the
inner loop.

I think this code does it (though I haven't tested it):

                for (j = 0; ; sgl = sg_next(sgl)) {
                        unsigned int k;
                        unsigned long pfn;

                        pfn = page_to_hvpfn(sg_page(sgl));
                        for (k = hvpg_idx; k < (unsigned int)(PAGE_SIZE /HV_HYP_PAGE_SIZE); k++) {
                                payload->range.pfn_array[j] = pfn + k;
                                if (++j == hvpg_count)
                                        goto done;
                        }
                        hvpg_idx = 0;
                }
done:

This approach also makes the limit of the inner loop a constant, and that
constant will be 1 when page size is 4K.  So the compiler should be able to
optimize away the loop in that case.

Michael






>  	}
> 
> --
> 2.28.0

Re: [RFC v2 11/11] scsi: storvsc: Support PAGE_SIZE larger than 4K

[Boqun Feng]

On Sat, Sep 05, 2020 at 02:55:48AM +0000, Michael Kelley wrote:
> From: Boqun Feng <boqun.feng@gmail.com> Sent: Tuesday, September 1, 2020 8:01 PM
> > 
> > Hyper-V always use 4k page size (HV_HYP_PAGE_SIZE), so when
> > communicating with Hyper-V, a guest should always use HV_HYP_PAGE_SIZE
> > as the unit for page related data. For storvsc, the data is
> > vmbus_packet_mpb_array. And since in scsi_cmnd, sglist of pages (in unit
> > of PAGE_SIZE) is used, we need convert pages in the sglist of scsi_cmnd
> > into Hyper-V pages in vmbus_packet_mpb_array.
> > 
> > This patch does the conversion by dividing pages in sglist into Hyper-V
> > pages, offset and indexes in vmbus_packet_mpb_array are recalculated
> > accordingly.
> > 
> > Signed-off-by: Boqun Feng <boqun.feng@gmail.com>
> > ---
> >  drivers/scsi/storvsc_drv.c | 60 ++++++++++++++++++++++++++++++++++----
> >  1 file changed, 54 insertions(+), 6 deletions(-)
> > 
> > diff --git a/drivers/scsi/storvsc_drv.c b/drivers/scsi/storvsc_drv.c
> > index 8f5f5dc863a4..3f6610717d4e 100644
> > --- a/drivers/scsi/storvsc_drv.c
> > +++ b/drivers/scsi/storvsc_drv.c
> > @@ -1739,23 +1739,71 @@ static int storvsc_queuecommand(struct Scsi_Host *host, struct
> > scsi_cmnd *scmnd)
> >  	payload_sz = sizeof(cmd_request->mpb);
> > 
> >  	if (sg_count) {
> > -		if (sg_count > MAX_PAGE_BUFFER_COUNT) {
> > +		unsigned int hvpg_idx = 0;
> > +		unsigned int j = 0;
> > +		unsigned long hvpg_offset = sgl->offset & ~HV_HYP_PAGE_MASK;
> > +		unsigned int hvpg_count = HVPFN_UP(hvpg_offset + length);
> > 
> > -			payload_sz = (sg_count * sizeof(u64) +
> > +		if (hvpg_count > MAX_PAGE_BUFFER_COUNT) {
> > +
> > +			payload_sz = (hvpg_count * sizeof(u64) +
> >  				      sizeof(struct vmbus_packet_mpb_array));
> >  			payload = kzalloc(payload_sz, GFP_ATOMIC);
> >  			if (!payload)
> >  				return SCSI_MLQUEUE_DEVICE_BUSY;
> >  		}
> > 
> > +		/*
> > +		 * sgl is a list of PAGEs, and payload->range.pfn_array
> > +		 * expects the page number in the unit of HV_HYP_PAGE_SIZE (the
> > +		 * page size that Hyper-V uses, so here we need to divide PAGEs
> > +		 * into HV_HYP_PAGE in case that PAGE_SIZE > HV_HYP_PAGE_SIZE.
> > +		 */
> >  		payload->range.len = length;
> > -		payload->range.offset = sgl[0].offset;
> > +		payload->range.offset = sgl[0].offset & ~HV_HYP_PAGE_MASK;
> > +		hvpg_idx = sgl[0].offset >> HV_HYP_PAGE_SHIFT;
> > 
> >  		cur_sgl = sgl;
> > -		for (i = 0; i < sg_count; i++) {
> > -			payload->range.pfn_array[i] =
> > -				page_to_pfn(sg_page((cur_sgl)));
> > +		for (i = 0, j = 0; i < sg_count; i++) {
> > +			/*
> > +			 * "PAGE_SIZE / HV_HYP_PAGE_SIZE - hvpg_idx" is the #
> > +			 * of HV_HYP_PAGEs in the current PAGE.
> > +			 *
> > +			 * "hvpg_count - j" is the # of unhandled HV_HYP_PAGEs.
> > +			 *
> > +			 * As shown in the following, the minimal of both is
> > +			 * the # of HV_HYP_PAGEs, we need to handle in this
> > +			 * PAGE.
> > +			 *
> > +			 * |------------------ PAGE ----------------------|
> > +			 * |   PAGE_SIZE / HV_HYP_PAGE_SIZE in total      |
> > +			 * |hvpg|hvpg| ...                 |hvpg|... |hvpg|
> > +			 *           ^                     ^
> > +			 *         hvpg_idx                |
> > +			 *           ^                     |
> > +			 *           +---(hvpg_count - j)--+
> > +			 *
> > +			 * or
> > +			 *
> > +			 * |------------------ PAGE ----------------------|
> > +			 * |   PAGE_SIZE / HV_HYP_PAGE_SIZE in total      |
> > +			 * |hvpg|hvpg| ...                 |hvpg|... |hvpg|
> > +			 *           ^                                           ^
> > +			 *         hvpg_idx                                      |
> > +			 *           ^                                           |
> > +			 *           +---(hvpg_count - j)------------------------+
> > +			 */
> > +			unsigned int nr_hvpg = min((unsigned int)(PAGE_SIZE / HV_HYP_PAGE_SIZE) - hvpg_idx,
> > +						   hvpg_count - j);
> > +			unsigned int k;
> > +
> > +			for (k = 0; k < nr_hvpg; k++) {
> > +				payload->range.pfn_array[j] =
> > +					page_to_hvpfn(sg_page((cur_sgl))) + hvpg_idx + k;
> > +				j++;
> > +			}
> >  			cur_sgl = sg_next(cur_sgl);
> > +			hvpg_idx = 0;
> >  		}
> 
> This code works; I don't see any errors.  But I think it can be made simpler based
> on doing two things:
> 1)  Rather than iterating over the sg_count, and having to calculate nr_hvpg on
> each iteration, base the exit decision on having filled up the pfn_array[].  You've
> already calculated the exact size of the array that is needed given the data
> length, so it's easy to exit when the array is full.
> 2) In the inner loop, iterate from hvpg_idx to PAGE_SIZE/HV_HYP_PAGE_SIZE
> rather than from 0 to a calculated value.
> 
> Also, as an optimization, pull page_to_hvpfn(sg_page((cur_sgl)) out of the
> inner loop.
> 
> I think this code does it (though I haven't tested it):
> 
>                 for (j = 0; ; sgl = sg_next(sgl)) {
>                         unsigned int k;
>                         unsigned long pfn;
> 
>                         pfn = page_to_hvpfn(sg_page(sgl));
>                         for (k = hvpg_idx; k < (unsigned int)(PAGE_SIZE /HV_HYP_PAGE_SIZE); k++) {
>                                 payload->range.pfn_array[j] = pfn + k;
>                                 if (++j == hvpg_count)
>                                         goto done;
>                         }
>                         hvpg_idx = 0;
>                 }
> done:
> 
> This approach also makes the limit of the inner loop a constant, and that
> constant will be 1 when page size is 4K.  So the compiler should be able to
> optimize away the loop in that case.
> 

Good point! I like your suggestion, and after thinking a bit harder
based on your approach, I come up with the following:

#define HV_HYP_PAGES_IN_PAGE ((unsigned int)(PAGE_SIZE / HV_HYP_PAGE_SIZE))

		for (j = 0; j < hvpg_count; j++) {
			unsigned int k = (j + hvpg_idx) % HV_HYP_PAGES_IN_PAGE;

			/*
			 * Two cases that we need to fetch a page:
			 * a) j == 0: the first step or
			 * b) k == 0: when we reach the boundary of a
			 * page.
			 * 
			if (k == 0 || j == 0) {
				pfn = page_to_hvpfn(sg_page(cur_sgl));
				cur_sgl = sg_next(cur_sgl);
			}

			payload->range.pfn_arrary[j] = pfn + k;
		}

, given the HV_HYP_PAGES_IN_PAGE is always a power of 2, so I think
compilers could easily optimize the "%" into bit masking operation. And
when HV_HYP_PAGES_IN_PAGE is 1, I think compilers can easily figure out
k is always zero, then the if-statement can be optimized as always
taken. And that gives us the same code as before ;-)

Thoughts? I will try with a test to see if I'm missing something subtle.

Thanks for looking into this!

Regards,
Boqun


> Michael
> 
> 
> 
> 
> 
> 
> >  	}
> > 
> > --
> > 2.28.0
> 

RE: [RFC v2 11/11] scsi: storvsc: Support PAGE_SIZE larger than 4K

[Michael Kelley]

From: Boqun Feng <boqun.feng@gmail.com> Sent: Saturday, September 5, 2020 7:15 AM
> 
> On Sat, Sep 05, 2020 at 02:55:48AM +0000, Michael Kelley wrote:
> > From: Boqun Feng <boqun.feng@gmail.com> Sent: Tuesday, September 1, 2020 8:01 PM
> > >
> > > Hyper-V always use 4k page size (HV_HYP_PAGE_SIZE), so when
> > > communicating with Hyper-V, a guest should always use HV_HYP_PAGE_SIZE
> > > as the unit for page related data. For storvsc, the data is
> > > vmbus_packet_mpb_array. And since in scsi_cmnd, sglist of pages (in unit
> > > of PAGE_SIZE) is used, we need convert pages in the sglist of scsi_cmnd
> > > into Hyper-V pages in vmbus_packet_mpb_array.
> > >
> > > This patch does the conversion by dividing pages in sglist into Hyper-V
> > > pages, offset and indexes in vmbus_packet_mpb_array are recalculated
> > > accordingly.
> > >
> > > Signed-off-by: Boqun Feng <boqun.feng@gmail.com>
> > > ---
> > >  drivers/scsi/storvsc_drv.c | 60 ++++++++++++++++++++++++++++++++++----
> > >  1 file changed, 54 insertions(+), 6 deletions(-)
> > >
> > > diff --git a/drivers/scsi/storvsc_drv.c b/drivers/scsi/storvsc_drv.c
> > > index 8f5f5dc863a4..3f6610717d4e 100644
> > > --- a/drivers/scsi/storvsc_drv.c
> > > +++ b/drivers/scsi/storvsc_drv.c
> > > @@ -1739,23 +1739,71 @@ static int storvsc_queuecommand(struct Scsi_Host *host,
> struct
> > > scsi_cmnd *scmnd)
> > >  	payload_sz = sizeof(cmd_request->mpb);
> > >
> > >  	if (sg_count) {
> > > -		if (sg_count > MAX_PAGE_BUFFER_COUNT) {
> > > +		unsigned int hvpg_idx = 0;
> > > +		unsigned int j = 0;
> > > +		unsigned long hvpg_offset = sgl->offset & ~HV_HYP_PAGE_MASK;
> > > +		unsigned int hvpg_count = HVPFN_UP(hvpg_offset + length);
> > >
> > > -			payload_sz = (sg_count * sizeof(u64) +
> > > +		if (hvpg_count > MAX_PAGE_BUFFER_COUNT) {
> > > +
> > > +			payload_sz = (hvpg_count * sizeof(u64) +
> > >  				      sizeof(struct vmbus_packet_mpb_array));
> > >  			payload = kzalloc(payload_sz, GFP_ATOMIC);
> > >  			if (!payload)
> > >  				return SCSI_MLQUEUE_DEVICE_BUSY;
> > >  		}
> > >
> > > +		/*
> > > +		 * sgl is a list of PAGEs, and payload->range.pfn_array
> > > +		 * expects the page number in the unit of HV_HYP_PAGE_SIZE (the
> > > +		 * page size that Hyper-V uses, so here we need to divide PAGEs
> > > +		 * into HV_HYP_PAGE in case that PAGE_SIZE > HV_HYP_PAGE_SIZE.
> > > +		 */
> > >  		payload->range.len = length;
> > > -		payload->range.offset = sgl[0].offset;
> > > +		payload->range.offset = sgl[0].offset & ~HV_HYP_PAGE_MASK;
> > > +		hvpg_idx = sgl[0].offset >> HV_HYP_PAGE_SHIFT;
> > >
> > >  		cur_sgl = sgl;
> > > -		for (i = 0; i < sg_count; i++) {
> > > -			payload->range.pfn_array[i] =
> > > -				page_to_pfn(sg_page((cur_sgl)));
> > > +		for (i = 0, j = 0; i < sg_count; i++) {
> > > +			/*
> > > +			 * "PAGE_SIZE / HV_HYP_PAGE_SIZE - hvpg_idx" is the #
> > > +			 * of HV_HYP_PAGEs in the current PAGE.
> > > +			 *
> > > +			 * "hvpg_count - j" is the # of unhandled HV_HYP_PAGEs.
> > > +			 *
> > > +			 * As shown in the following, the minimal of both is
> > > +			 * the # of HV_HYP_PAGEs, we need to handle in this
> > > +			 * PAGE.
> > > +			 *
> > > +			 * |------------------ PAGE ----------------------|
> > > +			 * |   PAGE_SIZE / HV_HYP_PAGE_SIZE in total      |
> > > +			 * |hvpg|hvpg| ...                 |hvpg|... |hvpg|
> > > +			 *           ^                     ^
> > > +			 *         hvpg_idx                |
> > > +			 *           ^                     |
> > > +			 *           +---(hvpg_count - j)--+
> > > +			 *
> > > +			 * or
> > > +			 *
> > > +			 * |------------------ PAGE ----------------------|
> > > +			 * |   PAGE_SIZE / HV_HYP_PAGE_SIZE in total      |
> > > +			 * |hvpg|hvpg| ...                 |hvpg|... |hvpg|
> > > +			 *           ^                                           ^
> > > +			 *         hvpg_idx                                      |
> > > +			 *           ^                                           |
> > > +			 *           +---(hvpg_count - j)------------------------+
> > > +			 */
> > > +			unsigned int nr_hvpg = min((unsigned int)(PAGE_SIZE /
> HV_HYP_PAGE_SIZE) - hvpg_idx,
> > > +						   hvpg_count - j);
> > > +			unsigned int k;
> > > +
> > > +			for (k = 0; k < nr_hvpg; k++) {
> > > +				payload->range.pfn_array[j] =
> > > +					page_to_hvpfn(sg_page((cur_sgl))) + hvpg_idx + k;
> > > +				j++;
> > > +			}
> > >  			cur_sgl = sg_next(cur_sgl);
> > > +			hvpg_idx = 0;
> > >  		}
> >
> > This code works; I don't see any errors.  But I think it can be made simpler based
> > on doing two things:
> > 1)  Rather than iterating over the sg_count, and having to calculate nr_hvpg on
> > each iteration, base the exit decision on having filled up the pfn_array[].  You've
> > already calculated the exact size of the array that is needed given the data
> > length, so it's easy to exit when the array is full.
> > 2) In the inner loop, iterate from hvpg_idx to PAGE_SIZE/HV_HYP_PAGE_SIZE
> > rather than from 0 to a calculated value.
> >
> > Also, as an optimization, pull page_to_hvpfn(sg_page((cur_sgl)) out of the
> > inner loop.
> >
> > I think this code does it (though I haven't tested it):
> >
> >                 for (j = 0; ; sgl = sg_next(sgl)) {
> >                         unsigned int k;
> >                         unsigned long pfn;
> >
> >                         pfn = page_to_hvpfn(sg_page(sgl));
> >                         for (k = hvpg_idx; k < (unsigned int)(PAGE_SIZE /HV_HYP_PAGE_SIZE); k++) {
> >                                 payload->range.pfn_array[j] = pfn + k;
> >                                 if (++j == hvpg_count)
> >                                         goto done;
> >                         }
> >                         hvpg_idx = 0;
> >                 }
> > done:
> >
> > This approach also makes the limit of the inner loop a constant, and that
> > constant will be 1 when page size is 4K.  So the compiler should be able to
> > optimize away the loop in that case.
> >
> 
> Good point! I like your suggestion, and after thinking a bit harder
> based on your approach, I come up with the following:
> 
> #define HV_HYP_PAGES_IN_PAGE ((unsigned int)(PAGE_SIZE / HV_HYP_PAGE_SIZE))
> 
> 		for (j = 0; j < hvpg_count; j++) {
> 			unsigned int k = (j + hvpg_idx) % HV_HYP_PAGES_IN_PAGE;
> 
> 			/*
> 			 * Two cases that we need to fetch a page:
> 			 * a) j == 0: the first step or
> 			 * b) k == 0: when we reach the boundary of a
> 			 * page.
> 			 *
> 			if (k == 0 || j == 0) {
> 				pfn = page_to_hvpfn(sg_page(cur_sgl));
> 				cur_sgl = sg_next(cur_sgl);
> 			}
> 
> 			payload->range.pfn_arrary[j] = pfn + k;
> 		}
> 
> , given the HV_HYP_PAGES_IN_PAGE is always a power of 2, so I think
> compilers could easily optimize the "%" into bit masking operation. And
> when HV_HYP_PAGES_IN_PAGE is 1, I think compilers can easily figure out
> k is always zero, then the if-statement can be optimized as always
> taken. And that gives us the same code as before ;-)
> 
> Thoughts? I will try with a test to see if I'm missing something subtle.
> 
> Thanks for looking into this!
> 

Your newest version looks right to me -- very clever!  I like it even better 
than my version.

Michael

Re: [RFC v2 03/11] Drivers: hv: vmbus: Introduce types of GPADL

[Boqun Feng]

On Sat, Sep 05, 2020 at 12:19:08AM +0000, Michael Kelley wrote:
[...]
> > 
> > @@ -462,7 +576,13 @@ static int __vmbus_open(struct vmbus_channel *newchannel,
> >  	open_msg->openid = newchannel->offermsg.child_relid;
> >  	open_msg->child_relid = newchannel->offermsg.child_relid;
> >  	open_msg->ringbuffer_gpadlhandle = newchannel->ringbuffer_gpadlhandle;
> > -	open_msg->downstream_ringbuffer_pageoffset = newchannel-
> > >ringbuffer_send_offset;
> > +	/*
> > +	 * The unit of ->downstream_ringbuffer_pageoffset is HV_HYP_PAGE and
> > +	 * the unit of ->ringbuffer_send_offset is PAGE, so here we first
> > +	 * calculate it into bytes and then convert into HV_HYP_PAGE.
> > +	 */
> > +	open_msg->downstream_ringbuffer_pageoffset =
> > +		hv_ring_gpadl_send_offset(newchannel->ringbuffer_send_offset << PAGE_SHIFT) >> HV_HYP_PAGE_SHIFT;
> 
> Line length?
> 

Thanks for the review! I've resolved all your comments on wording for
patch #2 and #4 in my local branch. For this line length issue, I fix it
with two changes:

1)	both the callsite of hv_ring_gpadl_send_offset() use ">> ..."
	to calculate the index in HV_HYP_PAGE, so I change the function
	to return offset in unit of HV_HYP_PAGE instead of bytes, and
	that can save us the ">> ..." here.

2)	newchannel->ringbuffer_send_offset is read in the previous code
	of the function into local variable "send_pages", so I use it
	to replace the "newchannel->ringbuffer_send_offset" here.

now the code is:

	open_msg->downstream_ringbuffer_pageoffset =
		hv_ring_gpadl_send_hvpgoffset(send_pages << PAGE_SHIFT);

Regards,
Boqun


> >  	open_msg->target_vp = hv_cpu_number_to_vp_number(newchannel->target_cpu);
> > 
> >  	if (userdatalen)
> > @@ -556,7 +676,6 @@ int vmbus_open(struct vmbus_channel *newchannel,
> >  }
> >  EXPORT_SYMBOL_GPL(vmbus_open);
> > 

Re: [RFC v2 07/11] hv_netvsc: Use HV_HYP_PAGE_SIZE for Hyper-V communication

[Boqun Feng]

On Sat, Sep 05, 2020 at 12:30:48AM +0000, Michael Kelley wrote:
> From: Boqun Feng <boqun.feng@gmail.com> Sent: Tuesday, September 1, 2020 8:01 PM
[...]
> >  struct rndis_request {
> >  	struct list_head list_ent;
> >  	struct completion  wait_event;
> > @@ -215,18 +215,18 @@ static int rndis_filter_send_request(struct rndis_device *dev,
> >  	packet->page_buf_cnt = 1;
> > 
> >  	pb[0].pfn = virt_to_phys(&req->request_msg) >>
> > -					PAGE_SHIFT;
> > +					HV_HYP_PAGE_SHIFT;
> >  	pb[0].len = req->request_msg.msg_len;
> >  	pb[0].offset =
> > -		(unsigned long)&req->request_msg & (PAGE_SIZE - 1);
> > +		(unsigned long)&req->request_msg & (HV_HYP_PAGE_SIZE - 1);
> 
> Use offset_in_hvpage() as defined in patch 6 of the series?
> 

Fair enough, I will use offset_in_hvpage() in the next version

Regards,
Boqun

> > 
> >  	/* Add one page_buf when request_msg crossing page boundary */
> > -	if (pb[0].offset + pb[0].len > PAGE_SIZE) {
> > +	if (pb[0].offset + pb[0].len > HV_HYP_PAGE_SIZE) {
> >  		packet->page_buf_cnt++;
> > -		pb[0].len = PAGE_SIZE -
> > +		pb[0].len = HV_HYP_PAGE_SIZE -
> >  			pb[0].offset;
> >  		pb[1].pfn = virt_to_phys((void *)&req->request_msg
> > -			+ pb[0].len) >> PAGE_SHIFT;
> > +			+ pb[0].len) >> HV_HYP_PAGE_SHIFT;
> >  		pb[1].offset = 0;
> >  		pb[1].len = req->request_msg.msg_len -
> >  			pb[0].len;
> > --
> > 2.28.0
>