displif: Protocol version 2
PATCH 1/2 xen/displif: Protocol version 2
In reply to Oleksandr AndrushchenkoFrom: Oleksandr Andrushchenko <[email protected]>
1. Change protocol version from string to integer
Version string, which is in fact an integer, is hard to handle in the code that supports different protocol versions. To simplify that make the version an integer.
2. Pass buffer offset with XENDISPL_OP_DBUF_CREATE
There are cases when display data buffer is created with non-zero offset to the data start. Handle such cases and provide that offset while creating a display buffer.
3. Add XENDISPL_OP_GET_EDID command
Add an optional request for reading Extended Display Identification Data (EDID) structure which allows better configuration of the display connectors over the configuration set in XenStore. With this change connectors may have multiple resolutions defined with respect to detailed timing definitions and additional properties normally provided by displays.
If this request is not supported by the backend then visible area is defined by the relevant XenStore's "resolution" property.
If backend provides extended display identification data (EDID) with XENDISPL_OP_GET_EDID request then EDID values must take precedence over the resolutions defined in XenStore.
4. Bump protocol version to 2.
--- xen/include/public/io/displif.h | 83 +++++++++++++++++++++++++++++++-- 1 file changed, 80 insertions(+), 3 deletions(-)
xen/include/public/io/displif.h +80 -3
@@ -38,7 +38,7 @@* Protocol version ****************************************************************************** */ -#define XENDISPL_PROTOCOL_VERSION "1" +#define XENDISPL_PROTOCOL_VERSION 2/* ******************************************************************************
@@ -202,6 +202,9 @@* Width and height of the connector in pixels separated by * XENDISPL_RESOLUTION_SEPARATOR. This defines visible area of the * display. +* If backend provides extended display identification data (EDID) with +* XENDISPL_OP_GET_EDID request then EDID values must take precedence +* over the resolutions defined here. * *------------------ Connector Request Transport Parameters ------------------- *
@@ -349,6 +352,7 @@#define XENDISPL_OP_FB_DETACH 0x13#define XENDISPL_OP_SET_CONFIG 0x14#define XENDISPL_OP_PG_FLIP 0x15 +#define XENDISPL_OP_GET_EDID 0x16/* ******************************************************************************
@@ -377,6 +381,10 @@#define XENDISPL_FIELD_BE_ALLOC "be-alloc"#define XENDISPL_FIELD_UNIQUE_ID "unique-id" +#define XENDISPL_EDID_BLOCK_SIZE 128 +#define XENDISPL_EDID_BLOCK_COUNT 256 +#define XENDISPL_EDID_MAX_SIZE (XENDISPL_EDID_BLOCK_SIZE * XENDISPL_EDID_BLOCK_COUNT) +/* ****************************************************************************** * STATUS RETURN CODES
@@ -451,7 +459,9 @@* +----------------+----------------+----------------+----------------+ * | gref_directory | 40 * +----------------+----------------+----------------+----------------+ -* | reserved | 44 +* | data_ofs | 44 +* +----------------+----------------+----------------+----------------+ +* | reserved | 48 * +----------------+----------------+----------------+----------------+ * |/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/| * +----------------+----------------+----------------+----------------+
@@ -494,6 +504,7 @@* buffer size (buffer_sz) exceeds what can be addressed by this single page, * then reference to the next page must be supplied (see gref_dir_next_page * below) +* data_ofs - uint32_t, offset of the data in the buffer, octets */ #define XENDISPL_DBUF_FLG_REQ_ALLOC (1 << 0)
@@ -506,6 +517,7 @@ struct xendispl_dbuf_create_req {uint32_t buffer_sz;uint32_t flags;grant_ref_t gref_directory; +uint32_t data_ofs; };/*
@@ -731,6 +743,42 @@ struct xendispl_page_flip_req {uint64_t fb_cookie; }; +/* +* Request EDID - request EDID describing current connector: +* 0 1 2 3 octet +* +----------------+----------------+----------------+----------------+ +* | id | _OP_GET_EDID | reserved | 4 +* +----------------+----------------+----------------+----------------+ +* | buffer_sz | 8 +* +----------------+----------------+----------------+----------------+ +* | gref_directory | 12 +* +----------------+----------------+----------------+----------------+ +* | reserved | 16 +* +----------------+----------------+----------------+----------------+ +* |/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/| +* +----------------+----------------+----------------+----------------+ +* | reserved | 64 +* +----------------+----------------+----------------+----------------+ +* +* Notes: +* - This request is optional and if not supported then visible area +* is defined by the relevant XenStore's "resolution" property. +* - Shared buffer, allocated for EDID storage, must not be less then +* XENDISPL_EDID_MAX_SIZE octets. +* +* buffer_sz - uint32_t, buffer size to be allocated, octets +* gref_directory - grant_ref_t, a reference to the first shared page +* describing EDID buffer references. See XENDISPL_OP_DBUF_CREATE for +* grant page directory structure (struct xendispl_page_directory). +* +* See response format for this request. +*/ + +struct xendispl_get_edid_req { +uint32_t buffer_sz; +grant_ref_t gref_directory; +}; +/* *---------------------------------- Responses -------------------------------- *
@@ -753,6 +801,31 @@ struct xendispl_page_flip_req {* id - uint16_t, private guest value, echoed from request * status - int32_t, response status, zero on success and -XEN_EXX on failure * +* +* Get EDID response - response for XENDISPL_OP_GET_EDID: +* 0 1 2 3 octet +* +----------------+----------------+----------------+----------------+ +* | id | operation | reserved | 4 +* +----------------+----------------+----------------+----------------+ +* | status | 8 +* +----------------+----------------+----------------+----------------+ +* | edid_sz | 12 +* +----------------+----------------+----------------+----------------+ +* | reserved | 16 +* +----------------+----------------+----------------+----------------+ +* |/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/| +* +----------------+----------------+----------------+----------------+ +* | reserved | 64 +* +----------------+----------------+----------------+----------------+ +* +* edid_sz - uint32_t, size of the EDID, octets +*/ + +struct xendispl_get_edid_resp { +uint32_t edid_sz; +}; + +/* *----------------------------------- Events ---------------------------------- * * Events are sent via a shared page allocated by the front and propagated by
@@ -804,6 +877,7 @@ struct xendispl_req {struct xendispl_fb_detach_req fb_detach;struct xendispl_set_config_req set_config;struct xendispl_page_flip_req pg_flip; +struct xendispl_get_edid_req get_edid;uint8_t reserved[56 ]; } op; };
@@ -813,7 +887,10 @@ struct xendispl_resp {uint8_t operation;uint8_t reserved;int32_t status; -uint8_t reserved1[56 ]; +union { +struct xendispl_get_edid_resp get_edid; +uint8_t reserved1[56 ]; + } op; };struct xendispl_evt {
Signature · 1 line
2.17.1
cover displif: Protocol version 2
From: Oleksandr Andrushchenko <[email protected]>
Hello all,
this series extends the existing displif protocol with new request and adds additional parameter to the exiting one. It also provides support for the new parameter in libgnttab via ioctl. The relevant changes in the backend can be found at [1] and the frontend at [2].
List of changes:
1. Change protocol version from string to integer
Version string, which is in fact an integer, is hard to handle in the code that supports different protocol versions. To simplify that make the version an integer.
2. Pass buffer offset with XENDISPL_OP_DBUF_CREATE
There are cases when display data buffer is created with non-zero offset to the data start. Handle such cases and provide that offset while creating a display buffer. Add the corresponding filed to the protocol and handle it in libgnttab.
This change is required for bringing virtual display on iMX8 platform which uses offset of 64 bytes for the buffers allocated on GPU side and then imported into PV DRM frontend. Otherwise the final picture looks shifted.
3. Add XENDISPL_OP_GET_EDID command
Add an optional request for reading Extended Display Identification Data (EDID) structure which allows better configuration of the display connectors over the configuration set in XenStore. With this change connectors may have multiple resolutions defined with respect to detailed timing definitions and additional properties normally provided by displays.
If this request is not supported by the backend then visible area is defined by the relevant XenStore's "resolution" property.
If backend provides extended display identification data (EDID) with XENDISPL_OP_GET_EDID request then EDID values must take precedence over the resolutions defined in XenStore.
4. Bump protocol version to 2.
Open questions and notes on the changes:
1. gnttab minor version change from 2 to 3
As per my understanding it is required to bump the version when I add the new functionality, but I would like to hear from the maintainers on that.
2. gnttab and version 2 of the ioctls
Because we add an additional parameter (data offset) and the structures used to pass ioctl arguments cannot be extended (there are no enough reserved fields), so there are to ways to solve that:
- break the existing API and add data_ofs field into the existing structures
- create a copy of the ioctl (v2) with the additional parameter.
It seems to be easier to go the first way, but this breaks things, so I decided to introduce v2 of the same ioctl which behaves gracefully with respect to the existing users, but adds some amount of copy-paste code (I was trying to minimize copy-paste so it is easier to maintain, but the result looked ugly to me because of lots of "if (protocol v1)" constructs.
Please note that struct ioctl_gntdev_dmabuf_imp_to_refs, e.g. version 1 of the ioctl, has uint32_t reserved field which can be used for the data offset, but its counterpart (ioctl_gntdev_dmabuf_exp_from_refs) doesn't have any, so it seems not to be aligned to only introduce version 2 of the ioctl for the later.
The other question is if to keep that reserved field in ioctl_gntdev_dmabuf_imp_to_refs_v2 structure or drop it.
3. displif protocol version string to int conversion
The existing protocol defines its version as a string "1" which is ok, but makes it not so easy to be directly used by C/C++ preprocessor which would like to see an integer for constructs like "#if XENDISPL_PROTOCOL_VERSION > 2"
At the same time this change may break the existing users of the protocol which still expect version as a string. I tried something like
#define STR_HELPER(x) #x
#define STR(x) STR_HELPER(x)
#define XENDISPL_PROTOCOL_VERSION_INT 1
#define XENDISPL_PROTOCOL_VERSION STR(XENDISPL_PROTOCOL_VERSION_INT)
but not sure if this is the right and good way to solve the issue, so in this series I have deliberately changed the protocol version to int.
Other possible way could be that every user of the header has its local copy (this is what we now use in the display backend). This way the local copy can be changed in a way suitable for the concrete user. This cannot be done (?) for the Linux Kernel though.
Thank you,
Oleksandr
[1] https://github.com/xen-troops/displ_be
[2] https://github.com/xen-troops/linux/pull/87
Oleksandr Andrushchenko (2):
xen/displif: Protocol version 2
libgnttab: Add support for Linux dma-buf offset
tools/include/xen-sys/Linux/gntdev.h | 53 +++++++++++++++++ tools/libs/gnttab/Makefile | 2 +- tools/libs/gnttab/freebsd.c | 15 +++++ tools/libs/gnttab/gnttab_core.c | 17 ++++++ tools/libs/gnttab/include/xengnttab.h | 13 ++++ tools/libs/gnttab/libxengnttab.map | 6 ++ tools/libs/gnttab/linux.c | 86 +++++++++++++++++++++++++++ tools/libs/gnttab/minios.c | 15 +++++ tools/libs/gnttab/private.h | 9 +++ xen/include/public/io/displif.h | 83 +++++++++++++++++++++++++- 10 files changed, 295 insertions(+), 4 deletions(-)
Signature · 1 line
2.17.1
PATCH 2/2 libgnttab: Add support for Linux dma-buf offset
From: Oleksandr Andrushchenko <[email protected]>
Add version 2 of the dma-buf ioctls which adds data_ofs parameter.
dma-buf is backed by a scatter-gather table and has offset parameter which tells where the actual data starts. Relevant ioctls are extended to support that offset:
- when dma-buf is created (exported) from grant references then data_ofs is used to set the offset field in the scatter list of the new dma-buf
- when dma-buf is imported and grant references provided then data_ofs is used to report that offset to user-space
--- tools/include/xen-sys/Linux/gntdev.h | 53 +++++++++++++++++ tools/libs/gnttab/Makefile | 2 +- tools/libs/gnttab/freebsd.c | 15 +++++ tools/libs/gnttab/gnttab_core.c | 17 ++++++ tools/libs/gnttab/include/xengnttab.h | 13 ++++ tools/libs/gnttab/libxengnttab.map | 6 ++ tools/libs/gnttab/linux.c | 86 +++++++++++++++++++++++++++ tools/libs/gnttab/minios.c | 15 +++++ tools/libs/gnttab/private.h | 9 +++ 9 files changed, 215 insertions(+), 1 deletion(-)
tools/include/xen-sys/Linux/gntdev.h +53 -0
@@ -274,4 +274,57 @@ struct ioctl_gntdev_dmabuf_imp_release { uint32_t reserved; }; +/* +* Version 2 of the ioctls adds @data_ofs parameter. +* +* dma-buf is backed by a scatter-gather table and has offset +* parameter which tells where the actual data starts. +* Relevant ioctls are extended to support that offset: +* - when dma-buf is created (exported) from grant references then +* @data_ofs is used to set the offset field in the scatter list +* of the new dma-buf +* - when dma-buf is imported and grant references are provided then +* @data_ofs is used to report that offset to user-space +*/ +#define IOCTL_GNTDEV_DMABUF_EXP_FROM_REFS_V2 \ + _IOC(_IOC_NONE, 'G', 13, \ + sizeof(struct ioctl_gntdev_dmabuf_exp_from_refs_v2)) +struct ioctl_gntdev_dmabuf_exp_from_refs_v2 { +/* IN parameters. */ +/* Specific options for this dma-buf: see GNTDEV_DMA_FLAG_XXX. */ +uint32_t flags; +/* Number of grant references in @refs array. */ +uint32_t count; +/* Offset of the data in the dma-buf. */ +uint32_t data_ofs; +/* OUT parameters. */ +/* File descriptor of the dma-buf. */ +uint32_t fd; +/* The domain ID of the grant references to be mapped. */ +uint32_t domid; +/* Variable IN parameter. */ +/* Array of grant references of size @count. */ +uint32_t refs[1 ]; +}; + +#define IOCTL_GNTDEV_DMABUF_IMP_TO_REFS_V2 \ + _IOC(_IOC_NONE, 'G', 14, \ + sizeof(struct ioctl_gntdev_dmabuf_imp_to_refs_v2)) +struct ioctl_gntdev_dmabuf_imp_to_refs_v2 { +/* IN parameters. */ +/* File descriptor of the dma-buf. */ +uint32_t fd; +/* Number of grant references in @refs array. */ +uint32_t count; +/* The domain ID for which references to be granted. */ +uint32_t domid; +/* Reserved - must be zero. */ +uint32_t reserved; +/* OUT parameters. */ +/* Offset of the data in the dma-buf. */ +uint32_t data_ofs; +/* Array of grant references of size @count. */ +uint32_t refs[1 ]; +}; +#endif /* __LINUX_PUBLIC_GNTDEV_H__ */
tools/libs/gnttab/Makefile +1 -1
@@ -2,7 +2,7 @@ XEN_ROOT = $(CURDIR)/../../.. include $(XEN_ROOT)/tools/Rules.mk MAJOR = 1 -MINOR = 2 +MINOR = 3 LIBNAME := gnttab USELIBS := toollog toolcore
tools/libs/gnttab/freebsd.c +15 -0
@@ -319,6 +319,14 @@ int osdep_gnttab_dmabuf_exp_from_refs(xengnttab_handle *xgt, uint32_t domid,abort (); } +int osdep_gnttab_dmabuf_exp_from_refs_v2 (xengnttab_handle *xgt,uint32_t domid, +uint32_t flags,uint32_t count, +const uint32_t *refs, +uint32_t *dmabuf_fd,uint32_t data_ofs) +{ +abort (); +} +int osdep_gnttab_dmabuf_exp_wait_released (xengnttab_handle *xgt,uint32_t fd,uint32_t wait_to_ms) {
@@ -331,6 +339,13 @@ int osdep_gnttab_dmabuf_imp_to_refs(xengnttab_handle *xgt, uint32_t domid,abort (); } +int osdep_gnttab_dmabuf_imp_to_refs_v2 (xengnttab_handle *xgt,uint32_t domid, +uint32_t fd,uint32_t count, +uint32_t *refs,uint32_t *data_ofs) +{ +abort (); +} +int osdep_gnttab_dmabuf_imp_release (xengnttab_handle *xgt,uint32_t fd) {abort ();
tools/libs/gnttab/gnttab_core.c +17 -0
@@ -144,6 +144,15 @@ int xengnttab_dmabuf_exp_from_refs(xengnttab_handle *xgt, uint32_t domid, refs, fd); } +int xengnttab_dmabuf_exp_from_refs_v2 (xengnttab_handle *xgt,uint32_t domid, +uint32_t flags,uint32_t count, +const uint32_t *refs,uint32_t *fd, +uint32_t data_ofs) +{ +return osdep_gnttab_dmabuf_exp_from_refs_v2 (xgt, domid, flags, count, + refs, fd, data_ofs); +} +int xengnttab_dmabuf_exp_wait_released (xengnttab_handle *xgt,uint32_t fd,uint32_t wait_to_ms) {
@@ -156,6 +165,14 @@ int xengnttab_dmabuf_imp_to_refs(xengnttab_handle *xgt, uint32_t domid,return osdep_gnttab_dmabuf_imp_to_refs (xgt, domid, fd, count, refs); } +int xengnttab_dmabuf_imp_to_refs_v2 (xengnttab_handle *xgt,uint32_t domid, +uint32_t fd,uint32_t count,uint32_t *refs, +uint32_t *data_ofs) +{ +return osdep_gnttab_dmabuf_imp_to_refs_v2 (xgt, domid, fd, count, refs, + data_ofs); +} +int xengnttab_dmabuf_imp_release (xengnttab_handle *xgt,uint32_t fd) {return osdep_gnttab_dmabuf_imp_release (xgt, fd);
tools/libs/gnttab/include/xengnttab.h +13 -0
@@ -322,12 +322,19 @@ int xengnttab_grant_copy(xengnttab_handle *xgt,* Returns 0 if dma-buf was successfully created and the corresponding * dma-buf's file descriptor is returned in @fd. * +* Version 2 also accepts @data_ofs offset of the data in the buffer. +* * [1] https://elixir.bootlin.com/linux/latest/source/Documentation/driver-api/dma-buf.rst */ int xengnttab_dmabuf_exp_from_refs (xengnttab_handle *xgt,uint32_t domid,uint32_t flags,uint32_t count,const uint32_t *refs,uint32_t *fd); +int xengnttab_dmabuf_exp_from_refs_v2 (xengnttab_handle *xgt,uint32_t domid, +uint32_t flags,uint32_t count, +const uint32_t *refs,uint32_t *fd, +uint32_t data_ofs); +/* * This will block until the dma-buf with the file descriptor @fd is * released. This is only valid for buffers created with
@@ -345,10 +352,16 @@ int xengnttab_dmabuf_exp_wait_released(xengnttab_handle *xgt, uint32_t fd,/* * Import a dma-buf with file descriptor @fd and export granted references * to the pages of that dma-buf into array @refs of size @count. +* +* Version 2 also provides @data_ofs offset of the data in the buffer. */ int xengnttab_dmabuf_imp_to_refs (xengnttab_handle *xgt,uint32_t domid,uint32_t fd,uint32_t count,uint32_t *refs); +int xengnttab_dmabuf_imp_to_refs_v2 (xengnttab_handle *xgt,uint32_t domid, +uint32_t fd,uint32_t count,uint32_t *refs, +uint32_t *data_ofs); +/* * This will close all references to an imported buffer, so it can be * released by the owner. This is only valid for buffers created with
tools/libs/gnttab/libxengnttab.map +6 -0
@@ -36,3 +36,9 @@ VERS_1.2 { xengnttab_dmabuf_imp_to_refs; xengnttab_dmabuf_imp_release; } VERS_1.1; + +VERS_1.3 { + global: + xengnttab_dmabuf_exp_from_refs_v2; + xengnttab_dmabuf_imp_to_refs_v2; +} VERS_1.2;
tools/libs/gnttab/linux.c +86 -0
98 lines not on this page · raw message
tools/libs/gnttab/minios.c +15 -0
@@ -120,6 +120,14 @@ int osdep_gnttab_dmabuf_exp_from_refs(xengnttab_handle *xgt, uint32_t domid,return -1 ; } +int osdep_gnttab_dmabuf_exp_from_refs_v2 (xengnttab_handle *xgt,uint32_t domid, +uint32_t flags,uint32_t count, +const uint32_t *refs,uint32_t *fd, +uint32_t data_ofs) +{ +return -1 ; +} +int osdep_gnttab_dmabuf_exp_wait_released (xengnttab_handle *xgt,uint32_t fd,uint32_t wait_to_ms) {
@@ -133,6 +141,13 @@ int osdep_gnttab_dmabuf_imp_to_refs(xengnttab_handle *xgt, uint32_t domid,return -1 ; } +int osdep_gnttab_dmabuf_imp_to_refs_v2 (xengnttab_handle *xgt,uint32_t domid, +uint32_t fd,uint32_t count, +uint32_t *refs,uint32_t *data_ofs) +{ +return -1 ; +} +int osdep_gnttab_dmabuf_imp_release (xengnttab_handle *xgt,uint32_t fd) {return -1 ;
tools/libs/gnttab/private.h +9 -0
@@ -39,6 +39,11 @@ int osdep_gnttab_dmabuf_exp_from_refs(xengnttab_handle *xgt, uint32_t domid, uint32_t flags, uint32_t count, const uint32_t *refs, uint32_t *fd); +int osdep_gnttab_dmabuf_exp_from_refs_v2 (xengnttab_handle *xgt,uint32_t domid, +uint32_t flags,uint32_t count, +const uint32_t *refs,uint32_t *fd, +uint32_t data_ofs); +int osdep_gnttab_dmabuf_exp_wait_released (xengnttab_handle *xgt,uint32_t fd,uint32_t wait_to_ms);
@@ -46,6 +51,10 @@ int osdep_gnttab_dmabuf_imp_to_refs(xengnttab_handle *xgt, uint32_t domid, uint32_t fd, uint32_t count, uint32_t *refs); +int osdep_gnttab_dmabuf_imp_to_refs_v2 (xengnttab_handle *xgt,uint32_t domid, +uint32_t fd,uint32_t count, +uint32_t *refs,uint32_t *data_ofs); +int osdep_gnttab_dmabuf_imp_release (xengnttab_handle *xgt,uint32_t fd);int osdep_gntshr_open (xengntshr_handle *xgs);
Signature · 1 line
2.17.1
ping
Quoted 123 lines · Oleksandr Andrushchenko
From: Oleksandr Andrushchenko <[email protected]>
Hello all,
this series extends the existing displif protocol with new request and adds additional parameter to the exiting one. It also provides support for the new parameter in libgnttab via ioctl. The relevant changes in the backend can be found at [1] and the frontend at [2].
List of changes:
1. Change protocol version from string to integer
Version string, which is in fact an integer, is hard to handle in the code that supports different protocol versions. To simplify that make the version an integer.
2. Pass buffer offset with
XENDISPL_OP_DBUF_CREATEThere are cases when display data buffer is created with non-zero offset to the data start. Handle such cases and provide that offset while creating a display buffer. Add the corresponding filed to the protocol and handle it in libgnttab.
This change is required for bringing virtual display on iMX8 platform which uses offset of 64 bytes for the buffers allocated on GPU side and then imported into PV DRM frontend. Otherwise the final picture looks shifted.3. Add
XENDISPL_OP_GET_EDIDcommandAdd an optional request for reading Extended Display Identification Data (EDID) structure which allows better configuration of the display connectors over the configuration set in XenStore. With this change connectors may have multiple resolutions defined with respect to detailed timing definitions and additional properties normally provided by displays.
If this request is not supported by the backend then visible area is defined by the relevant XenStore's "resolution" property.
If backend provides extended display identification data (EDID) with
XENDISPL_OP_GET_EDIDrequest then EDID values must take precedence over the resolutions defined in XenStore.4. Bump protocol version to 2.
Open questions and notes on the changes:
1. gnttab minor version change from 2 to 3
As per my understanding it is required to bump the version when I add the new functionality, but I would like to hear from the maintainers on that.2. gnttab and version 2 of the ioctls
Because we add an additional parameter (data offset) and the structures used to pass ioctl arguments cannot be extended (there are no enough reserved fields), so there are to ways to solve that:
- break the existing API and adddata_ofsfield into the existing structures
- create a copy of the ioctl (v2) with the additional parameter.It seems to be easier to go the first way, but this breaks things, so I decided to introduce v2 of the same ioctl which behaves gracefully with respect to the existing users, but adds some amount of copy-paste code (I was trying to minimize copy-paste so it is easier to maintain, but the result looked ugly to me because of lots of "if (protocol v1)" constructs.
Please note that struct
ioctl_gntdev_dmabuf_imp_to_refs, e.g. version 1 of the ioctl, hasuint32_treserved field which can be used for the data offset, but its counterpart (ioctl_gntdev_dmabuf_exp_from_refs) doesn't have any, so it seems not to be aligned to only introduce version 2 of the ioctl for the later.The other question is if to keep that reserved field in
ioctl_gntdev_dmabuf_imp_to_refs_v2structure or drop it.3. displif protocol version string to int conversion
The existing protocol defines its version as a string "1" which is ok, but makes it not so easy to be directly used by C/C++ preprocessor which would like to see an integer for constructs like "#ifXENDISPL_PROTOCOL_VERSION> 2"At the same time this change may break the existing users of the protocol which still expect version as a string. I tried something like
#define
STR_HELPER(x)#x
#define STR(x)STR_HELPER(x)#define
XENDISPL_PROTOCOL_VERSION_INT1
#defineXENDISPL_PROTOCOL_VERSIONSTR(XENDISPL_PROTOCOL_VERSION_INT)but not sure if this is the right and good way to solve the issue, so in this series I have deliberately changed the protocol version to int.
Other possible way could be that every user of the header has its local copy (this is what we now use in the display backend). This way the local copy can be changed in a way suitable for the concrete user. This cannot be done (?) for the Linux Kernel though.Thank you,
Oleksandr[1] https://github.com/xen-troops/displ_be
[2] https://github.com/xen-troops/linux/pull/87Oleksandr Andrushchenko (2):
xen/displif: Protocol version 2
libgnttab: Add support for Linux dma-buf offsettools/include/xen-sys/Linux/gntdev.h | 53 +++++++++++++++++ tools/libs/gnttab/Makefile | 2 +- tools/libs/gnttab/freebsd.c | 15 +++++ tools/libs/gnttab/gnttab_core.c | 17 ++++++ tools/libs/gnttab/include/xengnttab.h | 13 ++++ tools/libs/gnttab/libxengnttab.map | 6 ++ tools/libs/gnttab/linux.c | 86 +++++++++++++++++++++++++++ tools/libs/gnttab/minios.c | 15 +++++ tools/libs/gnttab/private.h | 9 +++ xen/include/public/io/displif.h | 83 +++++++++++++++++++++++++- 10 files changed, 295 insertions(+), 4 deletions(-)
Re: [PATCH 1/2] xen/displif: Protocol version 2
In reply to Oleksandr AndrushchenkoQuoted 42 lines · Oleksandr Andrushchenko
From: Oleksandr Andrushchenko <[email protected]>
1. Change protocol version from string to integer
Version string, which is in fact an integer, is hard to handle in the code that supports different protocol versions. To simplify that make the version an integer.
2. Pass buffer offset with
XENDISPL_OP_DBUF_CREATEThere are cases when display data buffer is created with non-zero offset to the data start. Handle such cases and provide that offset while creating a display buffer.
3. Add
XENDISPL_OP_GET_EDIDcommandAdd an optional request for reading Extended Display Identification Data (EDID) structure which allows better configuration of the display connectors over the configuration set in XenStore. With this change connectors may have multiple resolutions defined with respect to detailed timing definitions and additional properties normally provided by displays.
If this request is not supported by the backend then visible area is defined by the relevant XenStore's "resolution" property.
If backend provides extended display identification data (EDID) with
XENDISPL_OP_GET_EDIDrequest then EDID values must take precedence over the resolutions defined in XenStore.4. Bump protocol version to 2.
Signed-off-by Oleksandr Andrushchenko <[email protected]>--- xen/include/public/io/displif.h | 83 +++++++++++++++++++++++++++++++-- 1 file changed, 80 insertions(+), 3 deletions(-)xen/include/public/io/displif.h +0
-0
xen/include/public/io/displif.h line 41
Quoted 5 lines
xen/include/public/io/displif.h +1
-1@@ -38,7 +38,7 @@* Protocol version ****************************************************************************** */ -#define XENDISPL_PROTOCOL_VERSION "1" +#define XENDISPL_PROTOCOL_VERSION 2
This is not very nice regarding compatibility.
Can't you add a new macro for the integer value?
And please add comments further down which additions are related to the new version.
Juergen
Re: [PATCH 1/2] xen/displif: Protocol version 2
Quoted 51 lines · Jürgen Groß
Quoted 42 lines · Oleksandr Andrushchenko
From: Oleksandr Andrushchenko <[email protected]>
1. Change protocol version from string to integer
Version string, which is in fact an integer, is hard to handle in the code that supports different protocol versions. To simplify that make the version an integer.
2. Pass buffer offset with
XENDISPL_OP_DBUF_CREATEThere are cases when display data buffer is created with non-zero offset to the data start. Handle such cases and provide that offset while creating a display buffer.
3. Add
XENDISPL_OP_GET_EDIDcommandAdd an optional request for reading Extended Display Identification Data (EDID) structure which allows better configuration of the display connectors over the configuration set in XenStore. With this change connectors may have multiple resolutions defined with respect to detailed timing definitions and additional properties normally provided by displays.
If this request is not supported by the backend then visible area is defined by the relevant XenStore's "resolution" property.
If backend provides extended display identification data (EDID) with
XENDISPL_OP_GET_EDIDrequest then EDID values must take precedence over the resolutions defined in XenStore.4. Bump protocol version to 2.
Signed-off-by Oleksandr Andrushchenko <[email protected]>---
xen/include/public/io/displif.h| 83 +++++++++++++++++++++++++++++++--1 file changed, 80 insertions(+), 3 deletions(-)xen/include/public/io/displif.h +0
-0xen/include/public/io/displif.h line 41
Quoted 5 lines
xen/include/public/io/displif.h +1
-1@@ -38,7 +38,7 @@* Protocol version ****************************************************************************** */ -#define XENDISPL_PROTOCOL_VERSION "1" +#define XENDISPL_PROTOCOL_VERSION 2This is not very nice regarding compatibility.
Quoted 1 line
Can't you add a new macro for the integer value?
We can leave it as is, e.g. define XENDISPL_PROTOCOL_VERSION as "2",
so we do not break the existing users. Then if every user of the header has
its local copy (this is what we now use in the display backend), then that local copy can be changed in a way suitable for the concrete user, e.g. "2"
redefined to 2. This cannot be done (?) for the Linux Kernel though.
Or we can have
#define XENDISPL_PROTOCOL_VERSION "2"
#define XENDISPL_PROTOCOL_VERSION_INT 2
Jurgen, what's your preference here?
Quoted 3 lines
And please add comments further down which additions are related to the new version.
I will after the review is done and other comments are fixed
Quoted 3 lines
Juergen
Thank you,
Oleksandr
Re: [PATCH 1/2] xen/displif: Protocol version 2
Quoted 63 lines · Oleksandr Andrushchenko
Quoted 51 lines · Jürgen Groß
Quoted 42 lines · Oleksandr Andrushchenko
From: Oleksandr Andrushchenko <[email protected]>
1. Change protocol version from string to integer
Version string, which is in fact an integer, is hard to handle in the code that supports different protocol versions. To simplify that make the version an integer.
2. Pass buffer offset with
XENDISPL_OP_DBUF_CREATEThere are cases when display data buffer is created with non-zero offset to the data start. Handle such cases and provide that offset while creating a display buffer.
3. Add
XENDISPL_OP_GET_EDIDcommandAdd an optional request for reading Extended Display Identification Data (EDID) structure which allows better configuration of the display connectors over the configuration set in XenStore. With this change connectors may have multiple resolutions defined with respect to detailed timing definitions and additional properties normally provided by displays.
If this request is not supported by the backend then visible area is defined by the relevant XenStore's "resolution" property.
If backend provides extended display identification data (EDID) with
XENDISPL_OP_GET_EDIDrequest then EDID values must take precedence over the resolutions defined in XenStore.4. Bump protocol version to 2.
Signed-off-by Oleksandr Andrushchenko <[email protected]>--- xen/include/public/io/displif.h | 83 +++++++++++++++++++++++++++++++-- 1 file changed, 80 insertions(+), 3 deletions(-)xen/include/public/io/displif.h +0
-0xen/include/public/io/displif.h line 41
Quoted 5 lines
xen/include/public/io/displif.h +1
-1@@ -38,7 +38,7 @@* Protocol version ****************************************************************************** */ -#define XENDISPL_PROTOCOL_VERSION "1" +#define XENDISPL_PROTOCOL_VERSION 2This is not very nice regarding compatibility.
Quoted 1 line
Can't you add a new macro for the integer value?
We can leave it as is, e.g. define
XENDISPL_PROTOCOL_VERSIONas "2",so we do not break the existing users. Then if every user of the header has
its local copy (this is what we now use in the display backend), then that local copy can be changed in a way suitable for the concrete user, e.g. "2"
redefined to 2. This cannot be done (?) for the Linux Kernel though.
Quoted 7 lines
Or we can have
#define XENDISPL_PROTOCOL_VERSION "2"
#define XENDISPL_PROTOCOL_VERSION_INT 2
Jurgen, what's your preference here?
I would prefer the latter, as it avoids the need to modify the header when copying it to a local project.
Juergen
Re: [PATCH 1/2] xen/displif: Protocol version 2
Quoted 66 lines · Jürgen Groß
Quoted 65 lines · Oleksandr Andrushchenko
Quoted 51 lines · Jürgen Groß
Quoted 42 lines · Oleksandr Andrushchenko
From: Oleksandr Andrushchenko <[email protected]>
1. Change protocol version from string to integer
Version string, which is in fact an integer, is hard to handle in the code that supports different protocol versions. To simplify that make the version an integer.
2. Pass buffer offset with
XENDISPL_OP_DBUF_CREATEThere are cases when display data buffer is created with non-zero offset to the data start. Handle such cases and provide that offset while creating a display buffer.
3. Add
XENDISPL_OP_GET_EDIDcommandAdd an optional request for reading Extended Display Identification Data (EDID) structure which allows better configuration of the display connectors over the configuration set in XenStore. With this change connectors may have multiple resolutions defined with respect to detailed timing definitions and additional properties normally provided by displays.
If this request is not supported by the backend then visible area is defined by the relevant XenStore's "resolution" property.
If backend provides extended display identification data (EDID) with
XENDISPL_OP_GET_EDIDrequest then EDID values must take precedence over the resolutions defined in XenStore.4. Bump protocol version to 2.
Signed-off-by Oleksandr Andrushchenko <[email protected]>---
xen/include/public/io/displif.h| 83 +++++++++++++++++++++++++++++++--1 file changed, 80 insertions(+), 3 deletions(-)xen/include/public/io/displif.h +0
-0xen/include/public/io/displif.h line 41
Quoted 5 lines
xen/include/public/io/displif.h +1
-1@@ -38,7 +38,7 @@* Protocol version ****************************************************************************** */ -#define XENDISPL_PROTOCOL_VERSION "1" +#define XENDISPL_PROTOCOL_VERSION 2This is not very nice regarding compatibility.
Quoted 1 line
Can't you add a new macro for the integer value?
We can leave it as is, e.g. define
XENDISPL_PROTOCOL_VERSIONas "2",so we do not break the existing users. Then if every user of the header has
its local copy (this is what we now use in the display backend), then that local copy can be changed in a way suitable for the concrete user, e.g. "2"
redefined to 2. This cannot be done (?) for the Linux Kernel though.
Or we can have
Quoted 9 lines
Quoted 5 lines
#define XENDISPL_PROTOCOL_VERSION "2"
#define XENDISPL_PROTOCOL_VERSION_INT 2
Jurgen, what's your preference here?
I would prefer the latter, as it avoids the need to modify the header when copying it to a local project.
Ok, I'll have 2 definitions then
Anything else (beside the comments on new functionality) I can add/change
before sending v2 of the patch?
Quoted 2 lines
Juergen
Re: [PATCH 1/2] xen/displif: Protocol version 2
Quoted 75 lines · Oleksandr Andrushchenko
Quoted 74 lines · Jürgen Groß
Quoted 72 lines · Oleksandr Andrushchenko
Quoted 47 lines · Jürgen Groß
Quoted 46 lines · Oleksandr Andrushchenko
From: Oleksandr Andrushchenko <[email protected]>
1. Change protocol version from string to integer
Version string, which is in fact an integer, is hard to handle in the code that supports different protocol versions. To simplify that make the version an integer.
2. Pass buffer offset with
XENDISPL_OP_DBUF_CREATEThere are cases when display data buffer is created with non-zero offset to the data start. Handle such cases and provide that offset while creating a display buffer.
3. Add
XENDISPL_OP_GET_EDIDcommandAdd an optional request for reading Extended Display Identification Data (EDID) structure which allows better configuration of the display connectors over the configuration set in XenStore. With this change connectors may have multiple resolutions defined with respect to detailed timing definitions and additional properties normally provided by displays.
If this request is not supported by the backend then visible area is defined by the relevant XenStore's "resolution" property.
If backend provides extended display identification data (EDID) with
XENDISPL_OP_GET_EDIDrequest then EDID values must take precedence over the resolutions defined in XenStore.4. Bump protocol version to 2.
Signed-off-by Oleksandr Andrushchenko <[email protected]>---
xen/include/public/io/displif.h| 83
+++++++++++++++++++++++++++++++--1 file changed, 80 insertions(+), 3 deletions(-)xen/include/public/io/displif.h +0
-0@@ -38,7 +38,7 @@ * Protocol version ******************************************************************************Quoted 7 lines
xen/include/public/io/displif.h line 41
Quoted 3 lines
xen/include/public/io/displif.h +1
-1*/ -#define XENDISPL_PROTOCOL_VERSION "1" +#define XENDISPL_PROTOCOL_VERSION 2This is not very nice regarding compatibility.
Can't you add a new macro for the integer value?
We can leave it as is, e.g. define
XENDISPL_PROTOCOL_VERSIONas "2",so we do not break the existing users. Then if every user of the header has
its local copy (this is what we now use in the display backend), then that local copy can be changed in a way suitable for the concrete user, e.g. "2"
redefined to 2. This cannot be done (?) for the Linux Kernel though.
Or we can have
#define XENDISPL_PROTOCOL_VERSION "2"
#define XENDISPL_PROTOCOL_VERSION_INT 2
Jurgen, what's your preference here?
Quoted 8 lines
Quoted 3 lines
I would prefer the latter, as it avoids the need to modify the header when copying it to a local project.
Ok, I'll have 2 definitions then
Anything else (beside the comments on new functionality) I can add/change
before sending v2 of the patch?
I would have said so. :-)
Juergen
Re: [PATCH 1/2] xen/displif: Protocol version 2
Quoted 77 lines · Jürgen Groß
Quoted 76 lines · Oleksandr Andrushchenko
Quoted 75 lines · Jürgen Groß
Quoted 63 lines · Oleksandr Andrushchenko
Quoted 51 lines · Jürgen Groß
Quoted 42 lines · Oleksandr Andrushchenko
From: Oleksandr Andrushchenko <[email protected]>
1. Change protocol version from string to integer
Version string, which is in fact an integer, is hard to handle in the code that supports different protocol versions. To simplify that make the version an integer.
2. Pass buffer offset with
XENDISPL_OP_DBUF_CREATEThere are cases when display data buffer is created with non-zero offset to the data start. Handle such cases and provide that offset while creating a display buffer.
3. Add
XENDISPL_OP_GET_EDIDcommandAdd an optional request for reading Extended Display Identification Data (EDID) structure which allows better configuration of the display connectors over the configuration set in XenStore. With this change connectors may have multiple resolutions defined with respect to detailed timing definitions and additional properties normally provided by displays.
If this request is not supported by the backend then visible area is defined by the relevant XenStore's "resolution" property.
If backend provides extended display identification data (EDID) with
XENDISPL_OP_GET_EDIDrequest then EDID values must take precedence over the resolutions defined in XenStore.4. Bump protocol version to 2.
Signed-off-by Oleksandr Andrushchenko <[email protected]>---
xen/include/public/io/displif.h| 83 +++++++++++++++++++++++++++++++--1 file changed, 80 insertions(+), 3 deletions(-)xen/include/public/io/displif.h +0
-0xen/include/public/io/displif.h line 41
Quoted 5 lines
xen/include/public/io/displif.h +1
-1@@ -38,7 +38,7 @@* Protocol version ****************************************************************************** */ -#define XENDISPL_PROTOCOL_VERSION "1" +#define XENDISPL_PROTOCOL_VERSION 2This is not very nice regarding compatibility.
Quoted 1 line
Can't you add a new macro for the integer value?
We can leave it as is, e.g. define
XENDISPL_PROTOCOL_VERSIONas "2",so we do not break the existing users. Then if every user of the header has
its local copy (this is what we now use in the display backend), then that local copy can be changed in a way suitable for the concrete user, e.g. "2"
redefined to 2. This cannot be done (?) for the Linux Kernel though.
Quoted 7 lines
Or we can have
#define XENDISPL_PROTOCOL_VERSION "2"
#define XENDISPL_PROTOCOL_VERSION_INT 2
Jurgen, what's your preference here?
I would prefer the latter, as it avoids the need to modify the header when copying it to a local project.
Quoted 7 lines
Quoted 5 lines
Ok, I'll have 2 definitions then
Anything else (beside the comments on new functionality) I can add/change
before sending v2 of the patch?
I would have said so. :-)
Thank you,
the series also has a patch for libgnttab. Do you want me to send the protocol patch
separately or should we wait for Ian and Wei to review? These changes are related,
thus I sent then togheter
Quoted 3 lines
Juergen
Thank you,
Oleksandr
Re: [PATCH 1/2] xen/displif: Protocol version 2
Quoted 86 lines · Oleksandr Andrushchenko
Quoted 77 lines · Jürgen Groß
Quoted 76 lines · Oleksandr Andrushchenko
Quoted 75 lines · Jürgen Groß
Quoted 63 lines · Oleksandr Andrushchenko
Quoted 51 lines · Jürgen Groß
Quoted 42 lines · Oleksandr Andrushchenko
From: Oleksandr Andrushchenko <[email protected]>
1. Change protocol version from string to integer
Version string, which is in fact an integer, is hard to handle in the code that supports different protocol versions. To simplify that make the version an integer.
2. Pass buffer offset with
XENDISPL_OP_DBUF_CREATEThere are cases when display data buffer is created with non-zero offset to the data start. Handle such cases and provide that offset while creating a display buffer.
3. Add
XENDISPL_OP_GET_EDIDcommandAdd an optional request for reading Extended Display Identification Data (EDID) structure which allows better configuration of the display connectors over the configuration set in XenStore. With this change connectors may have multiple resolutions defined with respect to detailed timing definitions and additional properties normally provided by displays.
If this request is not supported by the backend then visible area is defined by the relevant XenStore's "resolution" property.
If backend provides extended display identification data (EDID) with
XENDISPL_OP_GET_EDIDrequest then EDID values must take precedence over the resolutions defined in XenStore.4. Bump protocol version to 2.
Signed-off-by Oleksandr Andrushchenko <[email protected]>--- xen/include/public/io/displif.h | 83 +++++++++++++++++++++++++++++++-- 1 file changed, 80 insertions(+), 3 deletions(-)xen/include/public/io/displif.h +0
-0xen/include/public/io/displif.h line 41
Quoted 5 lines
xen/include/public/io/displif.h +1
-1@@ -38,7 +38,7 @@* Protocol version ****************************************************************************** */ -#define XENDISPL_PROTOCOL_VERSION "1" +#define XENDISPL_PROTOCOL_VERSION 2This is not very nice regarding compatibility.
Quoted 1 line
Can't you add a new macro for the integer value?
We can leave it as is, e.g. define
XENDISPL_PROTOCOL_VERSIONas "2",so we do not break the existing users. Then if every user of the header has
its local copy (this is what we now use in the display backend), then that local copy can be changed in a way suitable for the concrete user, e.g. "2"
redefined to 2. This cannot be done (?) for the Linux Kernel though.
Quoted 7 lines
Or we can have
#define XENDISPL_PROTOCOL_VERSION "2"
#define XENDISPL_PROTOCOL_VERSION_INT 2
Jurgen, what's your preference here?
I would prefer the latter, as it avoids the need to modify the header when copying it to a local project.
Quoted 7 lines
Quoted 5 lines
Ok, I'll have 2 definitions then
Anything else (beside the comments on new functionality) I can add/change
before sending v2 of the patch?
I would have said so. :-)
Quoted 7 lines
Thank you,
the series also has a patch for libgnttab. Do you want me to send the protocol patch
separately or should we wait for Ian and Wei to review? These changes are related,
thus I sent then togheter
This is something to ask the tools maintainers IMO.
Juergen
Re: [PATCH 2/2] libgnttab: Add support for Linux dma-buf offset
In reply to Oleksandr AndrushchenkoIan, Wei, would you mind looking at the below please?
Thank you in advance,
Oleksandr
Quoted 388 lines · Oleksandr Andrushchenko
From: Oleksandr Andrushchenko <[email protected]>
Add version 2 of the dma-buf ioctls which adds
data_ofsparameter.dma-buf is backed by a scatter-gather table and has offset parameter which tells where the actual data starts. Relevant ioctls are extended to support that offset:
- when dma-buf is created (exported) from grant references thendata_ofsis used to set the offset field in the scatter list of the new dma-buf
- when dma-buf is imported and grant references provided thendata_ofsis used to report that offset to user-spaceSigned-off-by Oleksandr Andrushchenko <[email protected]>--- tools/include/xen-sys/Linux/gntdev.h | 53 +++++++++++++++++ tools/libs/gnttab/Makefile | 2 +- tools/libs/gnttab/freebsd.c | 15 +++++ tools/libs/gnttab/gnttab_core.c | 17 ++++++ tools/libs/gnttab/include/xengnttab.h | 13 ++++ tools/libs/gnttab/libxengnttab.map | 6 ++ tools/libs/gnttab/linux.c | 86 +++++++++++++++++++++++++++ tools/libs/gnttab/minios.c | 15 +++++ tools/libs/gnttab/private.h | 9 +++ 9 files changed, 215 insertions(+), 1 deletion(-)tools/include/xen-sys/Linux/gntdev.h +53
-0@@ -274,4 +274,57 @@ struct ioctl_gntdev_dmabuf_imp_release { uint32_t reserved; }; +/* +* Version 2 of the ioctls adds @data_ofs parameter. +* +* dma-buf is backed by a scatter-gather table and has offset +* parameter which tells where the actual data starts. +* Relevant ioctls are extended to support that offset: +* - when dma-buf is created (exported) from grant references then +* @data_ofs is used to set the offset field in the scatter list +* of the new dma-buf +* - when dma-buf is imported and grant references are provided then +* @data_ofs is used to report that offset to user-space +*/ +#define IOCTL_GNTDEV_DMABUF_EXP_FROM_REFS_V2 \ + _IOC(_IOC_NONE, 'G', 13, \ + sizeof(struct ioctl_gntdev_dmabuf_exp_from_refs_v2)) +struct ioctl_gntdev_dmabuf_exp_from_refs_v2 { +/* IN parameters. */ +/* Specific options for this dma-buf: see GNTDEV_DMA_FLAG_XXX. */ +uint32_t flags; +/* Number of grant references in @refs array. */ +uint32_t count; +/* Offset of the data in the dma-buf. */ +uint32_t data_ofs; +/* OUT parameters. */ +/* File descriptor of the dma-buf. */ +uint32_t fd; +/* The domain ID of the grant references to be mapped. */ +uint32_t domid; +/* Variable IN parameter. */ +/* Array of grant references of size @count. */ +uint32_t refs[1 ]; +}; + +#define IOCTL_GNTDEV_DMABUF_IMP_TO_REFS_V2 \ + _IOC(_IOC_NONE, 'G', 14, \ + sizeof(struct ioctl_gntdev_dmabuf_imp_to_refs_v2)) +struct ioctl_gntdev_dmabuf_imp_to_refs_v2 { +/* IN parameters. */ +/* File descriptor of the dma-buf. */ +uint32_t fd; +/* Number of grant references in @refs array. */ +uint32_t count; +/* The domain ID for which references to be granted. */ +uint32_t domid; +/* Reserved - must be zero. */ +uint32_t reserved; +/* OUT parameters. */ +/* Offset of the data in the dma-buf. */ +uint32_t data_ofs; +/* Array of grant references of size @count. */ +uint32_t refs[1 ]; +}; +#endif /* __LINUX_PUBLIC_GNTDEV_H__ */ tools/libs/gnttab/Makefile +1
-1@@ -2,7 +2,7 @@ XEN_ROOT = $(CURDIR)/../../.. include $(XEN_ROOT)/tools/Rules.mk MAJOR = 1 -MINOR = 2 +MINOR = 3 LIBNAME := gnttab USELIBS := toollog toolcoretools/libs/gnttab/freebsd.c +15
-0@@ -319,6 +319,14 @@ int osdep_gnttab_dmabuf_exp_from_refs(xengnttab_handle *xgt, uint32_t domid,abort (); } +int osdep_gnttab_dmabuf_exp_from_refs_v2 (xengnttab_handle *xgt,uint32_t domid, +uint32_t flags,uint32_t count, +const uint32_t *refs, +uint32_t *dmabuf_fd,uint32_t data_ofs) +{ +abort (); +} +int osdep_gnttab_dmabuf_exp_wait_released (xengnttab_handle *xgt,uint32_t fd,uint32_t wait_to_ms) {@@ -331,6 +339,13 @@ int osdep_gnttab_dmabuf_imp_to_refs(xengnttab_handle *xgt, uint32_t domid,abort (); } +int osdep_gnttab_dmabuf_imp_to_refs_v2 (xengnttab_handle *xgt,uint32_t domid, +uint32_t fd,uint32_t count, +uint32_t *refs,uint32_t *data_ofs) +{ +abort (); +} +int osdep_gnttab_dmabuf_imp_release (xengnttab_handle *xgt,uint32_t fd) {abort ();tools/libs/gnttab/gnttab_core.c +17
-0@@ -144,6 +144,15 @@ int xengnttab_dmabuf_exp_from_refs(xengnttab_handle *xgt, uint32_t domid, refs, fd); } +int xengnttab_dmabuf_exp_from_refs_v2 (xengnttab_handle *xgt,uint32_t domid, +uint32_t flags,uint32_t count, +const uint32_t *refs,uint32_t *fd, +uint32_t data_ofs) +{ +return osdep_gnttab_dmabuf_exp_from_refs_v2 (xgt, domid, flags, count, + refs, fd, data_ofs); +} +int xengnttab_dmabuf_exp_wait_released (xengnttab_handle *xgt,uint32_t fd,uint32_t wait_to_ms) {@@ -156,6 +165,14 @@ int xengnttab_dmabuf_imp_to_refs(xengnttab_handle *xgt, uint32_t domid,return osdep_gnttab_dmabuf_imp_to_refs (xgt, domid, fd, count, refs); } +int xengnttab_dmabuf_imp_to_refs_v2 (xengnttab_handle *xgt,uint32_t domid, +uint32_t fd,uint32_t count,uint32_t *refs, +uint32_t *data_ofs) +{ +return osdep_gnttab_dmabuf_imp_to_refs_v2 (xgt, domid, fd, count, refs, + data_ofs); +} +int xengnttab_dmabuf_imp_release (xengnttab_handle *xgt,uint32_t fd) {return osdep_gnttab_dmabuf_imp_release (xgt, fd);tools/libs/gnttab/include/xengnttab.h +13
-0@@ -322,12 +322,19 @@ int xengnttab_grant_copy(xengnttab_handle *xgt,* Returns 0 if dma-buf was successfully created and the corresponding * dma-buf's file descriptor is returned in @fd. * +* Version 2 also accepts @data_ofs offset of the data in the buffer. +* * [1] https://urldefense.com/v3/__https://elixir.bootlin.com/linux/latest/source/Documentation/driver-api/dma-buf.rst__;!!GF_29dbcQIUBPA!nBABkPpEyQW1_5nPE9nbyCbEaCvRjXQxOBKRpRSIGUgAdqcc0VCm4jL-9cCabcWNDS4bc_DR6Q$ */ int xengnttab_dmabuf_exp_from_refs (xengnttab_handle *xgt,uint32_t domid,uint32_t flags,uint32_t count,const uint32_t *refs,uint32_t *fd); +int xengnttab_dmabuf_exp_from_refs_v2 (xengnttab_handle *xgt,uint32_t domid, +uint32_t flags,uint32_t count, +const uint32_t *refs,uint32_t *fd, +uint32_t data_ofs); +/* * This will block until the dma-buf with the file descriptor @fd is * released. This is only valid for buffers created with @@ -345,10 +352,16 @@ int xengnttab_dmabuf_exp_wait_released(xengnttab_handle *xgt, uint32_t fd,/* * Import a dma-buf with file descriptor @fd and export granted references * to the pages of that dma-buf into array @refs of size @count. +* +* Version 2 also provides @data_ofs offset of the data in the buffer. */ int xengnttab_dmabuf_imp_to_refs (xengnttab_handle *xgt,uint32_t domid,uint32_t fd,uint32_t count,uint32_t *refs); +int xengnttab_dmabuf_imp_to_refs_v2 (xengnttab_handle *xgt,uint32_t domid, +uint32_t fd,uint32_t count,uint32_t *refs, +uint32_t *data_ofs); +/* * This will close all references to an imported buffer, so it can be * released by the owner. This is only valid for buffers created with tools/libs/gnttab/libxengnttab.map +6
-0@@ -36,3 +36,9 @@ VERS_1.2 { xengnttab_dmabuf_imp_to_refs; xengnttab_dmabuf_imp_release; } VERS_1.1; + +VERS_1.3 { + global: + xengnttab_dmabuf_exp_from_refs_v2; + xengnttab_dmabuf_imp_to_refs_v2; +} VERS_1.2;tools/libs/gnttab/linux.c +86
-0@@ -352,6 +352,51 @@ out:return rc; } +int osdep_gnttab_dmabuf_exp_from_refs_v2 (xengnttab_handle *xgt,uint32_t domid, +uint32_t flags,uint32_t count, +const uint32_t *refs, +uint32_t *dmabuf_fd, +uint32_t data_ofs) +{ +struct ioctl_gntdev_dmabuf_exp_from_refs_v2 *from_refs_v2 =NULL ; +int rc =-1 ; + +if ( !count ) + { + errno =EINVAL ; + goto out; + } + + from_refs_v2 =malloc (sizeof (*from_refs_v2) + + (count -1 ) *sizeof (from_refs_v2->refs[0 ])); +if ( !from_refs_v2 ) + { + errno =ENOMEM ; + goto out; + } + + from_refs_v2->flags = flags; + from_refs_v2->count = count; + from_refs_v2->domid = domid; + from_refs_v2->data_ofs = data_ofs; + +memcpy (from_refs_v2->refs, refs, count *sizeof (from_refs_v2->refs[0 ])); + +if ( (rc =ioctl (xgt->fd,IOCTL_GNTDEV_DMABUF_EXP_FROM_REFS_V2 , + from_refs_v2)) ) + { +GTERROR (xgt->logger,"ioctl DMABUF_EXP_FROM_REFS_V2 failed" ); + goto out; + } + + *dmabuf_fd = from_refs_v2->fd; + rc =0 ; + +out: +free (from_refs_v2); +return rc; +} +int osdep_gnttab_dmabuf_exp_wait_released (xengnttab_handle *xgt,uint32_t fd,uint32_t wait_to_ms) {@@ -413,6 +458,47 @@ out:return rc; } +int osdep_gnttab_dmabuf_imp_to_refs_v2 (xengnttab_handle *xgt,uint32_t domid, +uint32_t fd,uint32_t count, +uint32_t *refs, +uint32_t *data_ofs) +{ +struct ioctl_gntdev_dmabuf_imp_to_refs_v2 *to_refs_v2 =NULL ; +int rc =-1 ; + +if ( !count ) + { + errno =EINVAL ; + goto out; + } + + to_refs_v2 =malloc (sizeof (*to_refs_v2) + + (count -1 ) *sizeof (to_refs_v2->refs[0 ])); +if ( !to_refs_v2 ) + { + errno =ENOMEM ; + goto out; + } + + to_refs_v2->fd = fd; + to_refs_v2->count = count; + to_refs_v2->domid = domid; + +if ( (rc =ioctl (xgt->fd,IOCTL_GNTDEV_DMABUF_IMP_TO_REFS_V2 , to_refs_v2)) ) + { +GTERROR (xgt->logger,"ioctl DMABUF_IMP_TO_REFS_V2 failed" ); + goto out; + } + +memcpy (refs, to_refs_v2->refs, count *sizeof (*refs)); + *data_ofs = to_refs_v2->data_ofs; + rc =0 ; + +out: +free (to_refs_v2); +return rc; +} +int osdep_gnttab_dmabuf_imp_release (xengnttab_handle *xgt,uint32_t fd) {struct ioctl_gntdev_dmabuf_imp_release release;tools/libs/gnttab/minios.c +15
-0@@ -120,6 +120,14 @@ int osdep_gnttab_dmabuf_exp_from_refs(xengnttab_handle *xgt, uint32_t domid,return -1 ; } +int osdep_gnttab_dmabuf_exp_from_refs_v2 (xengnttab_handle *xgt,uint32_t domid, +uint32_t flags,uint32_t count, +const uint32_t *refs,uint32_t *fd, +uint32_t data_ofs) +{ +return -1 ; +} +int osdep_gnttab_dmabuf_exp_wait_released (xengnttab_handle *xgt,uint32_t fd,uint32_t wait_to_ms) {@@ -133,6 +141,13 @@ int osdep_gnttab_dmabuf_imp_to_refs(xengnttab_handle *xgt, uint32_t domid,return -1 ; } +int osdep_gnttab_dmabuf_imp_to_refs_v2 (xengnttab_handle *xgt,uint32_t domid, +uint32_t fd,uint32_t count, +uint32_t *refs,uint32_t *data_ofs) +{ +return -1 ; +} +int osdep_gnttab_dmabuf_imp_release (xengnttab_handle *xgt,uint32_t fd) {return -1 ;tools/libs/gnttab/private.h +9
-0@@ -39,6 +39,11 @@ int osdep_gnttab_dmabuf_exp_from_refs(xengnttab_handle *xgt, uint32_t domid, uint32_t flags, uint32_t count, const uint32_t *refs, uint32_t *fd); +int osdep_gnttab_dmabuf_exp_from_refs_v2 (xengnttab_handle *xgt,uint32_t domid, +uint32_t flags,uint32_t count, +const uint32_t *refs,uint32_t *fd, +uint32_t data_ofs); +int osdep_gnttab_dmabuf_exp_wait_released (xengnttab_handle *xgt,uint32_t fd,uint32_t wait_to_ms);@@ -46,6 +51,10 @@ int osdep_gnttab_dmabuf_imp_to_refs(xengnttab_handle *xgt, uint32_t domid, uint32_t fd, uint32_t count, uint32_t *refs); +int osdep_gnttab_dmabuf_imp_to_refs_v2 (xengnttab_handle *xgt,uint32_t domid, +uint32_t fd,uint32_t count, +uint32_t *refs,uint32_t *data_ofs); +int osdep_gnttab_dmabuf_imp_release (xengnttab_handle *xgt,uint32_t fd);int osdep_gntshr_open (xengntshr_handle *xgs);
Re: [PATCH 2/2] libgnttab: Add support for Linux dma-buf offset
In reply to Oleksandr AndrushchenkoHello, Ian, Wei!
Initially I have sent this patch as a part of the series for the display protocol changes,
but later decided to split. At the moment the protocol changes are already
accepted, but this part is still missing feedback and is still wanted.
I would really appreciate if you could have a look at the change below.
Thank you in advance,
Oleksandr Andrushchenko
Quoted 388 lines · Oleksandr Andrushchenko
From: Oleksandr Andrushchenko <[email protected]>
Add version 2 of the dma-buf ioctls which adds
data_ofsparameter.dma-buf is backed by a scatter-gather table and has offset parameter which tells where the actual data starts. Relevant ioctls are extended to support that offset:
- when dma-buf is created (exported) from grant references thendata_ofsis used to set the offset field in the scatter list of the new dma-buf
- when dma-buf is imported and grant references provided thendata_ofsis used to report that offset to user-spaceSigned-off-by Oleksandr Andrushchenko <[email protected]>--- tools/include/xen-sys/Linux/gntdev.h | 53 +++++++++++++++++ tools/libs/gnttab/Makefile | 2 +- tools/libs/gnttab/freebsd.c | 15 +++++ tools/libs/gnttab/gnttab_core.c | 17 ++++++ tools/libs/gnttab/include/xengnttab.h | 13 ++++ tools/libs/gnttab/libxengnttab.map | 6 ++ tools/libs/gnttab/linux.c | 86 +++++++++++++++++++++++++++ tools/libs/gnttab/minios.c | 15 +++++ tools/libs/gnttab/private.h | 9 +++ 9 files changed, 215 insertions(+), 1 deletion(-)tools/include/xen-sys/Linux/gntdev.h +53
-0@@ -274,4 +274,57 @@ struct ioctl_gntdev_dmabuf_imp_release { uint32_t reserved; }; +/* +* Version 2 of the ioctls adds @data_ofs parameter. +* +* dma-buf is backed by a scatter-gather table and has offset +* parameter which tells where the actual data starts. +* Relevant ioctls are extended to support that offset: +* - when dma-buf is created (exported) from grant references then +* @data_ofs is used to set the offset field in the scatter list +* of the new dma-buf +* - when dma-buf is imported and grant references are provided then +* @data_ofs is used to report that offset to user-space +*/ +#define IOCTL_GNTDEV_DMABUF_EXP_FROM_REFS_V2 \ + _IOC(_IOC_NONE, 'G', 13, \ + sizeof(struct ioctl_gntdev_dmabuf_exp_from_refs_v2)) +struct ioctl_gntdev_dmabuf_exp_from_refs_v2 { +/* IN parameters. */ +/* Specific options for this dma-buf: see GNTDEV_DMA_FLAG_XXX. */ +uint32_t flags; +/* Number of grant references in @refs array. */ +uint32_t count; +/* Offset of the data in the dma-buf. */ +uint32_t data_ofs; +/* OUT parameters. */ +/* File descriptor of the dma-buf. */ +uint32_t fd; +/* The domain ID of the grant references to be mapped. */ +uint32_t domid; +/* Variable IN parameter. */ +/* Array of grant references of size @count. */ +uint32_t refs[1 ]; +}; + +#define IOCTL_GNTDEV_DMABUF_IMP_TO_REFS_V2 \ + _IOC(_IOC_NONE, 'G', 14, \ + sizeof(struct ioctl_gntdev_dmabuf_imp_to_refs_v2)) +struct ioctl_gntdev_dmabuf_imp_to_refs_v2 { +/* IN parameters. */ +/* File descriptor of the dma-buf. */ +uint32_t fd; +/* Number of grant references in @refs array. */ +uint32_t count; +/* The domain ID for which references to be granted. */ +uint32_t domid; +/* Reserved - must be zero. */ +uint32_t reserved; +/* OUT parameters. */ +/* Offset of the data in the dma-buf. */ +uint32_t data_ofs; +/* Array of grant references of size @count. */ +uint32_t refs[1 ]; +}; +#endif /* __LINUX_PUBLIC_GNTDEV_H__ */ tools/libs/gnttab/Makefile +1
-1@@ -2,7 +2,7 @@ XEN_ROOT = $(CURDIR)/../../.. include $(XEN_ROOT)/tools/Rules.mk MAJOR = 1 -MINOR = 2 +MINOR = 3 LIBNAME := gnttab USELIBS := toollog toolcoretools/libs/gnttab/freebsd.c +15
-0@@ -319,6 +319,14 @@ int osdep_gnttab_dmabuf_exp_from_refs(xengnttab_handle *xgt, uint32_t domid,abort (); } +int osdep_gnttab_dmabuf_exp_from_refs_v2 (xengnttab_handle *xgt,uint32_t domid, +uint32_t flags,uint32_t count, +const uint32_t *refs, +uint32_t *dmabuf_fd,uint32_t data_ofs) +{ +abort (); +} +int osdep_gnttab_dmabuf_exp_wait_released (xengnttab_handle *xgt,uint32_t fd,uint32_t wait_to_ms) {@@ -331,6 +339,13 @@ int osdep_gnttab_dmabuf_imp_to_refs(xengnttab_handle *xgt, uint32_t domid,abort (); } +int osdep_gnttab_dmabuf_imp_to_refs_v2 (xengnttab_handle *xgt,uint32_t domid, +uint32_t fd,uint32_t count, +uint32_t *refs,uint32_t *data_ofs) +{ +abort (); +} +int osdep_gnttab_dmabuf_imp_release (xengnttab_handle *xgt,uint32_t fd) {abort ();tools/libs/gnttab/gnttab_core.c +17
-0@@ -144,6 +144,15 @@ int xengnttab_dmabuf_exp_from_refs(xengnttab_handle *xgt, uint32_t domid, refs, fd); } +int xengnttab_dmabuf_exp_from_refs_v2 (xengnttab_handle *xgt,uint32_t domid, +uint32_t flags,uint32_t count, +const uint32_t *refs,uint32_t *fd, +uint32_t data_ofs) +{ +return osdep_gnttab_dmabuf_exp_from_refs_v2 (xgt, domid, flags, count, + refs, fd, data_ofs); +} +int xengnttab_dmabuf_exp_wait_released (xengnttab_handle *xgt,uint32_t fd,uint32_t wait_to_ms) {@@ -156,6 +165,14 @@ int xengnttab_dmabuf_imp_to_refs(xengnttab_handle *xgt, uint32_t domid,return osdep_gnttab_dmabuf_imp_to_refs (xgt, domid, fd, count, refs); } +int xengnttab_dmabuf_imp_to_refs_v2 (xengnttab_handle *xgt,uint32_t domid, +uint32_t fd,uint32_t count,uint32_t *refs, +uint32_t *data_ofs) +{ +return osdep_gnttab_dmabuf_imp_to_refs_v2 (xgt, domid, fd, count, refs, + data_ofs); +} +int xengnttab_dmabuf_imp_release (xengnttab_handle *xgt,uint32_t fd) {return osdep_gnttab_dmabuf_imp_release (xgt, fd);tools/libs/gnttab/include/xengnttab.h +13
-0@@ -322,12 +322,19 @@ int xengnttab_grant_copy(xengnttab_handle *xgt,* Returns 0 if dma-buf was successfully created and the corresponding * dma-buf's file descriptor is returned in @fd. * +* Version 2 also accepts @data_ofs offset of the data in the buffer. +* * [1] https://elixir.bootlin.com/linux/latest/source/Documentation/driver-api/dma-buf.rst */ int xengnttab_dmabuf_exp_from_refs (xengnttab_handle *xgt,uint32_t domid,uint32_t flags,uint32_t count,const uint32_t *refs,uint32_t *fd); +int xengnttab_dmabuf_exp_from_refs_v2 (xengnttab_handle *xgt,uint32_t domid, +uint32_t flags,uint32_t count, +const uint32_t *refs,uint32_t *fd, +uint32_t data_ofs); +/* * This will block until the dma-buf with the file descriptor @fd is * released. This is only valid for buffers created with @@ -345,10 +352,16 @@ int xengnttab_dmabuf_exp_wait_released(xengnttab_handle *xgt, uint32_t fd,/* * Import a dma-buf with file descriptor @fd and export granted references * to the pages of that dma-buf into array @refs of size @count. +* +* Version 2 also provides @data_ofs offset of the data in the buffer. */ int xengnttab_dmabuf_imp_to_refs (xengnttab_handle *xgt,uint32_t domid,uint32_t fd,uint32_t count,uint32_t *refs); +int xengnttab_dmabuf_imp_to_refs_v2 (xengnttab_handle *xgt,uint32_t domid, +uint32_t fd,uint32_t count,uint32_t *refs, +uint32_t *data_ofs); +/* * This will close all references to an imported buffer, so it can be * released by the owner. This is only valid for buffers created with tools/libs/gnttab/libxengnttab.map +6
-0@@ -36,3 +36,9 @@ VERS_1.2 { xengnttab_dmabuf_imp_to_refs; xengnttab_dmabuf_imp_release; } VERS_1.1; + +VERS_1.3 { + global: + xengnttab_dmabuf_exp_from_refs_v2; + xengnttab_dmabuf_imp_to_refs_v2; +} VERS_1.2;tools/libs/gnttab/linux.c +86
-0@@ -352,6 +352,51 @@ out:return rc; } +int osdep_gnttab_dmabuf_exp_from_refs_v2 (xengnttab_handle *xgt,uint32_t domid, +uint32_t flags,uint32_t count, +const uint32_t *refs, +uint32_t *dmabuf_fd, +uint32_t data_ofs) +{ +struct ioctl_gntdev_dmabuf_exp_from_refs_v2 *from_refs_v2 =NULL ; +int rc =-1 ; + +if ( !count ) + { + errno =EINVAL ; + goto out; + } + + from_refs_v2 =malloc (sizeof (*from_refs_v2) + + (count -1 ) *sizeof (from_refs_v2->refs[0 ])); +if ( !from_refs_v2 ) + { + errno =ENOMEM ; + goto out; + } + + from_refs_v2->flags = flags; + from_refs_v2->count = count; + from_refs_v2->domid = domid; + from_refs_v2->data_ofs = data_ofs; + +memcpy (from_refs_v2->refs, refs, count *sizeof (from_refs_v2->refs[0 ])); + +if ( (rc =ioctl (xgt->fd,IOCTL_GNTDEV_DMABUF_EXP_FROM_REFS_V2 , + from_refs_v2)) ) + { +GTERROR (xgt->logger,"ioctl DMABUF_EXP_FROM_REFS_V2 failed" ); + goto out; + } + + *dmabuf_fd = from_refs_v2->fd; + rc =0 ; + +out: +free (from_refs_v2); +return rc; +} +int osdep_gnttab_dmabuf_exp_wait_released (xengnttab_handle *xgt,uint32_t fd,uint32_t wait_to_ms) {@@ -413,6 +458,47 @@ out:return rc; } +int osdep_gnttab_dmabuf_imp_to_refs_v2 (xengnttab_handle *xgt,uint32_t domid, +uint32_t fd,uint32_t count, +uint32_t *refs, +uint32_t *data_ofs) +{ +struct ioctl_gntdev_dmabuf_imp_to_refs_v2 *to_refs_v2 =NULL ; +int rc =-1 ; + +if ( !count ) + { + errno =EINVAL ; + goto out; + } + + to_refs_v2 =malloc (sizeof (*to_refs_v2) + + (count -1 ) *sizeof (to_refs_v2->refs[0 ])); +if ( !to_refs_v2 ) + { + errno =ENOMEM ; + goto out; + } + + to_refs_v2->fd = fd; + to_refs_v2->count = count; + to_refs_v2->domid = domid; + +if ( (rc =ioctl (xgt->fd,IOCTL_GNTDEV_DMABUF_IMP_TO_REFS_V2 , to_refs_v2)) ) + { +GTERROR (xgt->logger,"ioctl DMABUF_IMP_TO_REFS_V2 failed" ); + goto out; + } + +memcpy (refs, to_refs_v2->refs, count *sizeof (*refs)); + *data_ofs = to_refs_v2->data_ofs; + rc =0 ; + +out: +free (to_refs_v2); +return rc; +} +int osdep_gnttab_dmabuf_imp_release (xengnttab_handle *xgt,uint32_t fd) {struct ioctl_gntdev_dmabuf_imp_release release;tools/libs/gnttab/minios.c +15
-0@@ -120,6 +120,14 @@ int osdep_gnttab_dmabuf_exp_from_refs(xengnttab_handle *xgt, uint32_t domid,return -1 ; } +int osdep_gnttab_dmabuf_exp_from_refs_v2 (xengnttab_handle *xgt,uint32_t domid, +uint32_t flags,uint32_t count, +const uint32_t *refs,uint32_t *fd, +uint32_t data_ofs) +{ +return -1 ; +} +int osdep_gnttab_dmabuf_exp_wait_released (xengnttab_handle *xgt,uint32_t fd,uint32_t wait_to_ms) {@@ -133,6 +141,13 @@ int osdep_gnttab_dmabuf_imp_to_refs(xengnttab_handle *xgt, uint32_t domid,return -1 ; } +int osdep_gnttab_dmabuf_imp_to_refs_v2 (xengnttab_handle *xgt,uint32_t domid, +uint32_t fd,uint32_t count, +uint32_t *refs,uint32_t *data_ofs) +{ +return -1 ; +} +int osdep_gnttab_dmabuf_imp_release (xengnttab_handle *xgt,uint32_t fd) {return -1 ;tools/libs/gnttab/private.h +9
-0@@ -39,6 +39,11 @@ int osdep_gnttab_dmabuf_exp_from_refs(xengnttab_handle *xgt, uint32_t domid, uint32_t flags, uint32_t count, const uint32_t *refs, uint32_t *fd); +int osdep_gnttab_dmabuf_exp_from_refs_v2 (xengnttab_handle *xgt,uint32_t domid, +uint32_t flags,uint32_t count, +const uint32_t *refs,uint32_t *fd, +uint32_t data_ofs); +int osdep_gnttab_dmabuf_exp_wait_released (xengnttab_handle *xgt,uint32_t fd,uint32_t wait_to_ms);@@ -46,6 +51,10 @@ int osdep_gnttab_dmabuf_imp_to_refs(xengnttab_handle *xgt, uint32_t domid, uint32_t fd, uint32_t count, uint32_t *refs); +int osdep_gnttab_dmabuf_imp_to_refs_v2 (xengnttab_handle *xgt,uint32_t domid, +uint32_t fd,uint32_t count, +uint32_t *refs,uint32_t *data_ofs); +int osdep_gnttab_dmabuf_imp_release (xengnttab_handle *xgt,uint32_t fd);int osdep_gntshr_open (xengntshr_handle *xgs);
Re: [PATCH 2/2] libgnttab: Add support for Linux dma-buf offset
In reply to Oleksandr AndrushchenkoOleksandr Andrushchenko writes ("[PATCH 2/2] libgnttab: Add support for Linux dma-buf offset"):
Quoted 4 lines
From: Oleksandr Andrushchenko <[email protected]>
Add version 2 of the dma-buf ioctls which adds
data_ofsparameter.
Quoted 8 lines
dma-buf is backed by a scatter-gather table and has offset parameter which tells where the actual data starts. Relevant ioctls are extended to support that offset:
- when dma-buf is created (exported) from grant references thendata_ofsis used to set the offset field in the scatter list of the new dma-buf
- when dma-buf is imported and grant references provided thendata_ofsis used to report that offset to user-space
Thanks. I'm not a DMA expert, but I think this is probably going in roughly the right direction. I will probably want a review from a DMA expert too, but let me get on with my questions:
When you say "the protocol changes are already accepted" I think you mean the Linux ioctl changes ? If not, what *do* you mean ?
Quoted 4 lines
tools/include/xen-sys/Linux/gntdev.h +4
-0+/* +* Version 2 of the ioctls adds @data_ofs parameter. +* +* dma-buf is backed by a scatter-gather table and has offset
tools/include/xen-sys/Linux/gntdev.h struct ioctl_gntdev_dmabuf_imp_release { · line 290
Quoted 10 lines
tools/include/xen-sys/Linux/gntdev.h +10
-0+* parameter which tells where the actual data starts. +* Relevant ioctls are extended to support that offset: +* - when dma-buf is created (exported) from grant references then +* @data_ofs is used to set the offset field in the scatter list +* of the new dma-buf +* - when dma-buf is imported and grant references are provided then +* @data_ofs is used to report that offset to user-space +*/ +#define IOCTL_GNTDEV_DMABUF_EXP_FROM_REFS_V2 \ + _IOC(_IOC_NONE, 'G', 13, \
I think this was copied from a Linux header file ? If so please quote the precise file and revision in the commit message. And be sure to copy the copyright informtaion appropriately.
tools/libs/gnttab/freebsd.c int osdep_gnttab_dmabuf_exp_from_refs(xengnttab_handle *xgt, uint32_t domid, · line 327
Quoted 6 lines
tools/libs/gnttab/freebsd.c +6
-0+int osdep_gnttab_dmabuf_exp_from_refs_v2 (xengnttab_handle *xgt,uint32_t domid, +uint32_t flags,uint32_t count, +const uint32_t *refs, +uint32_t *dmabuf_fd,uint32_t data_ofs) +{ +abort ();
I'm pretty sure this is wrong.
This leads me to ask about compatibility, both across versions of the various components, and API compatibility across different platforms.
libxengnttab is supposed to have a stable API and ABI. This means that old programs should work with the new library - which I think you have achieved.
But I think it also means that it should work with new programs, and the new library, on old kernels. What is your compatibility story here ? What is the intended mode of use by an application ?
And the same application code should be useable, so far as possible, across different plaatforms that support Xen.
What fallback would be possible for application do if the v2 function is not available ? I think that fallback action needs to be selectable at runtime, to support new userspace on old kernels.
What architectures is the new Linux ioctl available on ?
Quoted 11 lines
tools/libs/gnttab/include/xengnttab.h +3
-0@@ -322,12 +322,19 @@ int xengnttab_grant_copy(xengnttab_handle *xgt,* Returns 0 if dma-buf was successfully created and the corresponding * dma-buf's file descriptor is returned in @fd. * + +* Version 2 also accepts @data_ofs offset of the data in the buffer. +*
tools/libs/gnttab/include/xengnttab.h int xengnttab_grant_copy(xengnttab_handle *xgt, · line 337
Quoted 10 lines
tools/libs/gnttab/include/xengnttab.h +4
-0* [1] https://elixir.bootlin.com/linux/latest/source/Documentation/driver-api/dma-buf.rst */ int xengnttab_dmabuf_exp_from_refs (xengnttab_handle *xgt,uint32_t domid,uint32_t flags,uint32_t count,const uint32_t *refs,uint32_t *fd); +int xengnttab_dmabuf_exp_from_refs_v2 (xengnttab_handle *xgt,uint32_t domid, +uint32_t flags,uint32_t count, +const uint32_t *refs,uint32_t *fd, +uint32_t data_ofs);
I think the information about the meaning of @data_ofs must be in the doc comment. Indeed, that should be the primary location.
Conversely there is no need to duplicate information between the patch contents, and the commit message.
Is _v2 really the best name for this ? Are we likely to want to extend this again in future ? Perhaps it should be called ..._offset or something ? Please think about this and tell me your opinion.
Quoted 26 lines
tools/libs/gnttab/include/xengnttab.h +26
-0+int osdep_gnttab_dmabuf_exp_from_refs_v2 (xengnttab_handle *xgt,uint32_t domid, +uint32_t flags,uint32_t count, +const uint32_t *refs, +uint32_t *dmabuf_fd, +uint32_t data_ofs) +{ +struct ioctl_gntdev_dmabuf_exp_from_refs_v2 *from_refs_v2 =NULL ; +int rc =-1 ; + +if ( !count ) + { + errno =EINVAL ; + goto out; + } + + from_refs_v2 =malloc (sizeof (*from_refs_v2) + + (count -1 ) *sizeof (from_refs_v2->refs[0 ])); +if ( !from_refs_v2 ) + { + errno =ENOMEM ; + goto out; + } + + from_refs_v2->flags = flags; + from_refs_v2->count = count; + from_refs_v2->domid = domid;
tools/libs/gnttab/include/xengnttab.h int xengnttab_grant_copy(xengnttab_handle *xgt,
Quoted 10 lines
tools/libs/gnttab/include/xengnttab.h +10
-0+ from_refs_v2->data_ofs = data_ofs; + +memcpy (from_refs_v2->refs, refs, count *sizeof (from_refs_v2->refs[0 ])); + +if ( (rc =ioctl (xgt->fd,IOCTL_GNTDEV_DMABUF_EXP_FROM_REFS_V2 , + from_refs_v2)) ) + { +GTERROR (xgt->logger,"ioctl DMABUF_EXP_FROM_REFS_V2 failed" ); + goto out; + }
This seems just a fairly obvious wrapper for this ioctl. I think it would be best for me to review this in detail with reference to the ioctl documentation (which you helpfully refer to - thank you!) after I see the answers to my other questions.
tools/libs/gnttab/include/xengnttab.h
Quoted 5 lines
tools/libs/gnttab/include/xengnttab.h +5
-0+int osdep_gnttab_dmabuf_imp_to_refs_v2 (xengnttab_handle *xgt,uint32_t domid, +uint32_t fd,uint32_t count, +uint32_t *refs, +uint32_t *data_ofs) +{
This function is very similar to the previous one. I'm uncomfortable with the duplication, but I see that
osdep_gnttab_dmabuf_{imp_to,exp_from}_refs
are very duplicative already, so I am also somewhat uncomfortable with asking you to clean this up with refactoring. But perhaps if you felt like thinking about combioning some of this, that might be nice.
What do my co-maintainers think ?
Regards,
Ian.
Re: [PATCH 2/2] libgnttab: Add support for Linux dma-buf offset
Hi,
Quoted 13 lines · Ian Jackson
Oleksandr Andrushchenko writes ("[PATCH 2/2] libgnttab: Add support for Linux dma-buf offset"):
Quoted 4 lines
From: Oleksandr Andrushchenko <[email protected]>
Add version 2 of the dma-buf ioctls which adds
data_ofsparameter.Quoted 8 lines
dma-buf is backed by a scatter-gather table and has offset parameter which tells where the actual data starts. Relevant ioctls are extended to support that offset:
- when dma-buf is created (exported) from grant references thendata_ofsis used to set the offset field in the scatter list of the new dma-buf
- when dma-buf is imported and grant references provided thendata_ofsis used to report that offset to user-space
Quoted 6 lines
Thanks. I'm not a DMA expert, but I think this is probably going in roughly the right direction. I will probably want a review from a DMA expert too, but let me get on with my questions:
When you say "the protocol changes are already accepted" I think you mean the Linux ioctl changes ? If not, what *do* you mean ?
I mean that the relevant protocol changes are already part of both Xen [1]
and Linux trees [2]. What is missing is ioctl implementation in the kernel and
its support in Xen' tools. This is why I have marked the patch as RFC in order
to get some view on the matter from Xen community. Once we agree on the
naming, structure etc. I'll send patches for both Xen and Linux
Quoted 17 lines
Quoted 5 lines
tools/include/xen-sys/Linux/gntdev.h +4
-0+/* +* Version 2 of the ioctls adds @data_ofs parameter. +* +* dma-buf is backed by a scatter-gather table and has offset tools/include/xen-sys/Linux/gntdev.h struct ioctl_gntdev_dmabuf_imp_release { · line 290
Quoted 10 lines
tools/include/xen-sys/Linux/gntdev.h +10
-0+* parameter which tells where the actual data starts. +* Relevant ioctls are extended to support that offset: +* - when dma-buf is created (exported) from grant references then +* @data_ofs is used to set the offset field in the scatter list +* of the new dma-buf +* - when dma-buf is imported and grant references are provided then +* @data_ofs is used to report that offset to user-space +*/ +#define IOCTL_GNTDEV_DMABUF_EXP_FROM_REFS_V2 \ + _IOC(_IOC_NONE, 'G', 13, \I think this was copied from a Linux header file ? If so please quote the precise file and revision in the commit message.
This is not upstream yet, please see explanation above
Quoted 10 lines
And be sure to
copy the copyright informtaion appropriately.tools/libs/gnttab/freebsd.c int osdep_gnttab_dmabuf_exp_from_refs(xengnttab_handle *xgt, uint32_t domid, · line 327
Quoted 6 lines
tools/libs/gnttab/freebsd.c +6
-0+int osdep_gnttab_dmabuf_exp_from_refs_v2 (xengnttab_handle *xgt,uint32_t domid, +uint32_t flags,uint32_t count, +const uint32_t *refs, +uint32_t *dmabuf_fd,uint32_t data_ofs) +{ +abort ();I'm pretty sure this is wrong.
First of all, Linux dma-bufs are only supported on Linux, so neither FreeBSD nor Mini-OS
will have that. If you are referring to "abort()" here, so I am just aligning to what previously
was there, e.g. all non-relevant dma-buf OS specifics were implemented like that.
Quoted 7 lines
This leads me to ask about compatibility, both across versions of the various components, and API compatibility across different platforms.
libxengnttab is supposed to have a stable API and ABI. This means that old programs should work with the new library - which I think you have achieved.
Yes
Quoted 4 lines
But I think it also means that it should work with new programs, and the new library, on old kernels. What is your compatibility story here ? What is the intended mode of use by an application ?
Well, this is a tough story. If we have new software and new library, but old
kernel it means that the offset we are trying to get with the new ioctl will be
unavailable to that new software. In most cases we can use offset of 0, but some
platforms (iMX8) use offset of 64. So, we can workaround that for most(?) platforms
by reporting offset 0, but some platforms will fail. I am not sure if this is good to state that
this combination of software (as described above) "will mostly work" or just let
the system fail at run-time, by letting Linux return ENOTSUPP for the new ioctl.
By fail I mean that the display backend may decide if to use the previous version of the ioctl
without the offset field.
Quoted 7 lines
And the same application code should be useable, so far as possible, across different plaatforms that support Xen.
What fallback would be possible for application do if the v2 function is not available ? I think that fallback action needs to be selectable at runtime, to support new userspace on old kernels.
Well, as I said before, for the platforms with offset 0 we are "fine" ignoring the offset and
using v1 of the ioctl without the offset field. For the platforms with non-zero offset it results
at least in slight screen distortion and they do need v2 of the ioctl
Quoted 2 lines
What architectures is the new Linux ioctl available on ?
x86/ARM
Quoted 18 lines
Quoted 18 lines
tools/libs/gnttab/include/xengnttab.h +3
-0@@ -322,12 +322,19 @@ int xengnttab_grant_copy(xengnttab_handle *xgt,* Returns 0 if dma-buf was successfully created and the corresponding * dma-buf's file descriptor is returned in @fd. * + +* Version 2 also accepts @data_ofs offset of the data in the buffer. +* * [1] https://urldefense.com/v3/__https://elixir.bootlin.com/linux/latest/source/Documentation/driver-api/dma-buf.rst__;!!GF_29dbcQIUBPA!ia7gsD5o9vPtgQzm3pUYmcPEaUmaODZRUkyniq74vNDZkbz9zGbqe-zCUesHHF3-ckRWLuIBKg$ [elixir[.]bootlin[.]com] */ int xengnttab_dmabuf_exp_from_refs (xengnttab_handle *xgt,uint32_t domid,uint32_t flags,uint32_t count,const uint32_t *refs,uint32_t *fd);
Quoted 6 lines
tools/libs/gnttab/include/xengnttab.h int xengnttab_grant_copy(xengnttab_handle *xgt, · line 336
Quoted 4 lines
tools/libs/gnttab/include/xengnttab.h +4
-0+int xengnttab_dmabuf_exp_from_refs_v2 (xengnttab_handle *xgt,uint32_t domid, +uint32_t flags,uint32_t count, +const uint32_t *refs,uint32_t *fd, +uint32_t data_ofs);I think the information about the meaning of @data_ofs must be in the doc comment. Indeed, that should be the primary location.
Sure
Quoted 3 lines
Conversely there is no need to duplicate information between the patch contents, and the commit message.
It's just a me that always wants the doc at handy location so I don't need to dig for
the commit messages? But at the same time the commit message should allow one
quickly understand what's in there. So, I would prefer to have more description in the
patch then
Quoted 4 lines
Is
_v2really the best name for this ? Are we likely to want to extend this again in future ? Perhaps it should be called ..._offset or something ? Please think about this and tell me your opinion.
I don't actually like v2. Neither I can produce anything more cute ;)
On the other hand it is easier to understand that v2 is actually extends/removes/changes
something that was here before. Say, if you have 2 ioctls yyy and ddd you need to compare
the two to understand what is more relevant at the moment. Having explicit version in the
name leaves no doubt about what is newer.
Quoted 33 lines
Quoted 33 lines
tools/libs/gnttab/include/xengnttab.h +32
-0+int osdep_gnttab_dmabuf_exp_from_refs_v2 (xengnttab_handle *xgt,uint32_t domid, +uint32_t flags,uint32_t count, +const uint32_t *refs, +uint32_t *dmabuf_fd, +uint32_t data_ofs) +{ +struct ioctl_gntdev_dmabuf_exp_from_refs_v2 *from_refs_v2 =NULL ; +int rc =-1 ; + +if ( !count ) + { + errno =EINVAL ; + goto out; + } + + from_refs_v2 =malloc (sizeof (*from_refs_v2) + + (count -1 ) *sizeof (from_refs_v2->refs[0 ])); +if ( !from_refs_v2 ) + { + errno =ENOMEM ; + goto out; + } + + from_refs_v2->flags = flags; + from_refs_v2->count = count; + from_refs_v2->domid = domid; + from_refs_v2->data_ofs = data_ofs; + +memcpy (from_refs_v2->refs, refs, count *sizeof (from_refs_v2->refs[0 ])); + +if ( (rc =ioctl (xgt->fd,IOCTL_GNTDEV_DMABUF_EXP_FROM_REFS_V2 , + from_refs_v2)) )
Quoted 8 lines
tools/libs/gnttab/include/xengnttab.h
Quoted 4 lines
tools/libs/gnttab/include/xengnttab.h +4
-0+ { +GTERROR (xgt->logger,"ioctl DMABUF_EXP_FROM_REFS_V2 failed" ); + goto out; + }This seems just a fairly obvious wrapper for this ioctl. I think it would be best for me to review this in detail with reference to the ioctl documentation (which you helpfully refer to - thank you!) after I see the answers to my other questions.
Well, I have little to add as the only change and the reason is that scatter-gather table's
offset must be honored which was not a problem until we faced iMX8 platform which has
that offset non-zero. Frankly, lots of software assumes it is zero...
Quoted 12 lines
tools/libs/gnttab/include/xengnttab.h
Quoted 6 lines
tools/libs/gnttab/include/xengnttab.h +5
-0+int osdep_gnttab_dmabuf_imp_to_refs_v2 (xengnttab_handle *xgt,uint32_t domid, +uint32_t fd,uint32_t count, +uint32_t *refs, +uint32_t *data_ofs) +{This function is very similar to the previous one. I'm uncomfortable with the duplication, but I see that
osdep_gnttab_dmabuf_{imp_to,exp_from}_refs
are very duplicative already, so I am also somewhat uncomfortable with asking you to clean this up with refactoring. But perhaps if you felt like thinking about combioning some of this, that might be nice.
I hate having code duplication as well: less code less maintenance. But in this case
the common code makes that function full of "if"s so finally I gave up and make a copy-paste.
No strong opinion here: if you think "if"s are still better I'll rework that
Quoted 6 lines
What do my co-maintainers think ?
Regards,
Ian.
Thank you for the review and your time,
Oleksandr
[1] https://xenbits.xen.org/gitweb/?p=xen.git;a=commit;h=c27a184225eab54d20435c8cab5ad0ef384dc2c0
Re: [PATCH 2/2] libgnttab: Add support for Linux dma-buf offset
Hello, all!
I would like to bring back this old thread as it seems it has stuck long time ago
without clear NAck or Ack. I didn't rebase the changes because the change itself
requires answers on the way we should go here: new ioctl (seems to be better) or
extension of the existing one (not so great)
Thank you in advance,
Oleksandr
Quoted 230 lines · Oleksandr Andrushchenko
Hi,
Quoted 13 lines · Ian Jackson
Oleksandr Andrushchenko writes ("[PATCH 2/2] libgnttab: Add support for Linux dma-buf offset"):
Quoted 4 lines
From: Oleksandr Andrushchenko <[email protected]>
Add version 2 of the dma-buf ioctls which adds
data_ofsparameter.Quoted 8 lines
dma-buf is backed by a scatter-gather table and has offset parameter which tells where the actual data starts. Relevant ioctls are extended to support that offset:
- when dma-buf is created (exported) from grant references thendata_ofsis used to set the offset field in the scatter list of the new dma-buf
- when dma-buf is imported and grant references provided thendata_ofsis used to report that offset to user-spaceQuoted 6 lines
Thanks. I'm not a DMA expert, but I think this is probably going in roughly the right direction. I will probably want a review from a DMA expert too, but let me get on with my questions:
When you say "the protocol changes are already accepted" I think you mean the Linux ioctl changes ? If not, what *do* you mean ?
I mean that the relevant protocol changes are already part of both Xen [1]
and Linux trees [2]. What is missing is ioctl implementation in the kernel and
its support in Xen' tools. This is why I have marked the patch as RFC in order
to get some view on the matter from Xen community. Once we agree on the
naming, structure etc. I'll send patches for both Xen and Linux
Quoted 8 lines
Quoted 8 lines
+/*
+ * Version 2 of the ioctls adds @data_ofs parameter.
+ *
+ * dma-buf is backed by a scatter-gather table and has offset
+ * parameter which tells where the actual data starts.
+ * Relevant ioctls are extended to support that offset:
+ * - when dma-buf is created (exported) from grant references then
+ * @data_ofs is used to set the offset field in the scatter listQuoted 8 lines
tools/include/xen-sys/Linux/gntdev.h struct ioctl_gntdev_dmabuf_imp_release { · line 290
Quoted 6 lines
+ * of the new dma-buf
+ * - when dma-buf is imported and grant references are provided then
+ * @data_ofs is used to report that offset to user-space
+ */tools/include/xen-sys/Linux/gntdev.h +2
-0+#define IOCTL_GNTDEV_DMABUF_EXP_FROM_REFS_V2 \ + _IOC(_IOC_NONE, 'G', 13, \I think this was copied from a Linux header file ? If so please quote the precise file and revision in the commit message.
This is not upstream yet, please see explanation above
Quoted 10 lines
And be sure to
copy the copyright informtaion appropriately.tools/libs/gnttab/freebsd.c int osdep_gnttab_dmabuf_exp_from_refs(xengnttab_handle *xgt, uint32_t domid, · line 327
Quoted 6 lines
tools/libs/gnttab/freebsd.c +6
-0+int osdep_gnttab_dmabuf_exp_from_refs_v2 (xengnttab_handle *xgt,uint32_t domid, +uint32_t flags,uint32_t count, +const uint32_t *refs, +uint32_t *dmabuf_fd,uint32_t data_ofs) +{ +abort ();I'm pretty sure this is wrong.
First of all, Linux dma-bufs are only supported on Linux, so neither FreeBSD nor Mini-OS
will have that. If you are referring to "
abort()" here, so I am just aligning to what previouslywas there, e.g. all non-relevant dma-buf OS specifics were implemented like that.
Quoted 6 lines
This leads me to ask about compatibility, both across versions of the various components, and API compatibility across different platforms.
libxengnttab is supposed to have a stable API and ABI. This means that old programs should work with the new library - which I think you have achieved.
Yes
Quoted 3 lines
But I think it also means that it should work with new programs, and the new library, on old kernels. What is your compatibility story here ? What is the intended mode of use by an application ?
Well, this is a tough story. If we have new software and new library, but old
kernel it means that the offset we are trying to get with the new ioctl will be
unavailable to that new software. In most cases we can use offset of 0, but some
platforms (iMX8) use offset of 64. So, we can workaround that for most(?) platforms
by reporting offset 0, but some platforms will fail. I am not sure if this is good to state that
this combination of software (as described above) "will mostly work" or just let
the system fail at run-time, by letting Linux return ENOTSUPP for the new ioctl.
By fail I mean that the display backend may decide if to use the previous version of the ioctl
without the offset field.
Quoted 6 lines
And the same application code should be useable, so far as possible, across different plaatforms that support Xen.
What fallback would be possible for application do if the v2 function is not available ? I think that fallback action needs to be selectable at runtime, to support new userspace on old kernels.
Well, as I said before, for the platforms with offset 0 we are "fine" ignoring the offset and
using v1 of the ioctl without the offset field. For the platforms with non-zero offset it results
at least in slight screen distortion and they do need v2 of the ioctl
Quoted 1 line
What architectures is the new Linux ioctl available on ?
x86/ARM
Quoted 17 lines
Quoted 17 lines
tools/libs/gnttab/include/xengnttab.h +3
-0@@ -322,12 +322,19 @@ int xengnttab_grant_copy(xengnttab_handle *xgt,* Returns 0 if dma-buf was successfully created and the corresponding * dma-buf's file descriptor is returned in @fd. * + +* Version 2 also accepts @data_ofs offset of the data in the buffer. +* * [1] https://urldefense.com/v3/__https://elixir.bootlin.com/linux/latest/source/Documentation/driver-api/dma-buf.rst__;!!GF_29dbcQIUBPA!ia7gsD5o9vPtgQzm3pUYmcPEaUmaODZRUkyniq74vNDZkbz9zGbqe-zCUesHHF3-ckRWLuIBKg$ [elixir[.]bootlin[.]com] */ int xengnttab_dmabuf_exp_from_refs (xengnttab_handle *xgt,uint32_t domid,uint32_t flags,uint32_t count,const uint32_t *refs,uint32_t *fd);Quoted 6 lines
tools/libs/gnttab/include/xengnttab.h int xengnttab_grant_copy(xengnttab_handle *xgt, · line 336
Quoted 4 lines
tools/libs/gnttab/include/xengnttab.h +4
-0+int xengnttab_dmabuf_exp_from_refs_v2 (xengnttab_handle *xgt,uint32_t domid, +uint32_t flags,uint32_t count, +const uint32_t *refs,uint32_t *fd, +uint32_t data_ofs);I think the information about the meaning of @data_ofs must be in the doc comment. Indeed, that should be the primary location.
Sure
Quoted 2 lines
Conversely there is no need to duplicate information between the patch contents, and the commit message.
It's just a me that always wants the doc at handy location so I don't need to dig for
the commit messages? But at the same time the commit message should allow one
quickly understand what's in there. So, I would prefer to have more description in the
patch then
Quoted 3 lines
Is
_v2really the best name for this ? Are we likely to want to extend this again in future ? Perhaps it should be called ..._offset or something ? Please think about this and tell me your opinion.I don't actually like v2. Neither I can produce anything more cute ;)
On the other hand it is easier to understand that v2 is actually extends/removes/changes
something that was here before. Say, if you have 2 ioctls yyy and ddd you need to compare
the two to understand what is more relevant at the moment. Having explicit version in the
name leaves no doubt about what is newer.
Quoted 32 lines
Quoted 32 lines
tools/libs/gnttab/include/xengnttab.h +32
-0+int osdep_gnttab_dmabuf_exp_from_refs_v2 (xengnttab_handle *xgt,uint32_t domid, +uint32_t flags,uint32_t count, +const uint32_t *refs, +uint32_t *dmabuf_fd, +uint32_t data_ofs) +{ +struct ioctl_gntdev_dmabuf_exp_from_refs_v2 *from_refs_v2 =NULL ; +int rc =-1 ; + +if ( !count ) + { + errno =EINVAL ; + goto out; + } + + from_refs_v2 =malloc (sizeof (*from_refs_v2) + + (count -1 ) *sizeof (from_refs_v2->refs[0 ])); +if ( !from_refs_v2 ) + { + errno =ENOMEM ; + goto out; + } + + from_refs_v2->flags = flags; + from_refs_v2->count = count; + from_refs_v2->domid = domid; + from_refs_v2->data_ofs = data_ofs; + +memcpy (from_refs_v2->refs, refs, count *sizeof (from_refs_v2->refs[0 ])); + +if ( (rc =ioctl (xgt->fd,IOCTL_GNTDEV_DMABUF_EXP_FROM_REFS_V2 , + from_refs_v2)) )Quoted 8 lines
tools/libs/gnttab/include/xengnttab.h
Quoted 4 lines
tools/libs/gnttab/include/xengnttab.h +4
-0+ { +GTERROR (xgt->logger,"ioctl DMABUF_EXP_FROM_REFS_V2 failed" ); + goto out; + }This seems just a fairly obvious wrapper for this ioctl. I think it would be best for me to review this in detail with reference to the ioctl documentation (which you helpfully refer to - thank you!) after I see the answers to my other questions.
Well, I have little to add as the only change and the reason is that scatter-gather table's
offset must be honored which was not a problem until we faced iMX8 platform which has
that offset non-zero. Frankly, lots of software assumes it is zero...
Quoted 11 lines
tools/libs/gnttab/include/xengnttab.h
Quoted 5 lines
tools/libs/gnttab/include/xengnttab.h +5
-0+int osdep_gnttab_dmabuf_imp_to_refs_v2 (xengnttab_handle *xgt,uint32_t domid, +uint32_t fd,uint32_t count, +uint32_t *refs, +uint32_t *data_ofs) +{This function is very similar to the previous one. I'm uncomfortable with the duplication, but I see that
osdep_gnttab_dmabuf_{imp_to,exp_from}_refs
are very duplicative already, so I am also somewhat uncomfortable with asking you to clean this up with refactoring. But perhaps if you felt like thinking about combioning some of this, that might be nice.I hate having code duplication as well: less code less maintenance. But in this case
the common code makes that function full of "if"s so finally I gave up and make a copy-paste.
No strong opinion here: if you think "if"s are still better I'll rework that
Quoted 5 lines
What do my co-maintainers think ?
Regards,
Ian.Thank you for the review and your time,
Oleksandr
[1] https://xenbits.xen.org/gitweb/?p=xen.git;a=commit;h=c27a184225eab54d20435c8cab5ad0ef384dc2c0