brintos

brintos / linux-shallow public Read only

0
0
Text · 21.6 KiB · e8f488a Raw
434 lines · plain
1================================2Coherent Accelerator (CXL) Flash3================================4 5Introduction6============7 8    The IBM Power architecture provides support for CAPI (Coherent9    Accelerator Power Interface), which is available to certain PCIe slots10    on Power 8 systems. CAPI can be thought of as a special tunneling11    protocol through PCIe that allow PCIe adapters to look like special12    purpose co-processors which can read or write an application's13    memory and generate page faults. As a result, the host interface to14    an adapter running in CAPI mode does not require the data buffers to15    be mapped to the device's memory (IOMMU bypass) nor does it require16    memory to be pinned.17 18    On Linux, Coherent Accelerator (CXL) kernel services present CAPI19    devices as a PCI device by implementing a virtual PCI host bridge.20    This abstraction simplifies the infrastructure and programming21    model, allowing for drivers to look similar to other native PCI22    device drivers.23 24    CXL provides a mechanism by which user space applications can25    directly talk to a device (network or storage) bypassing the typical26    kernel/device driver stack. The CXL Flash Adapter Driver enables a27    user space application direct access to Flash storage.28 29    The CXL Flash Adapter Driver is a kernel module that sits in the30    SCSI stack as a low level device driver (below the SCSI disk and31    protocol drivers) for the IBM CXL Flash Adapter. This driver is32    responsible for the initialization of the adapter, setting up the33    special path for user space access, and performing error recovery. It34    communicates directly the Flash Accelerator Functional Unit (AFU)35    as described in Documentation/arch/powerpc/cxl.rst.36 37    The cxlflash driver supports two, mutually exclusive, modes of38    operation at the device (LUN) level:39 40        - Any flash device (LUN) can be configured to be accessed as a41          regular disk device (i.e.: /dev/sdc). This is the default mode.42 43        - Any flash device (LUN) can be configured to be accessed from44          user space with a special block library. This mode further45          specifies the means of accessing the device and provides for46          either raw access to the entire LUN (referred to as direct47          or physical LUN access) or access to a kernel/AFU-mediated48          partition of the LUN (referred to as virtual LUN access). The49          segmentation of a disk device into virtual LUNs is assisted50          by special translation services provided by the Flash AFU.51 52Overview53========54 55    The Coherent Accelerator Interface Architecture (CAIA) introduces a56    concept of a master context. A master typically has special privileges57    granted to it by the kernel or hypervisor allowing it to perform AFU58    wide management and control. The master may or may not be involved59    directly in each user I/O, but at the minimum is involved in the60    initial setup before the user application is allowed to send requests61    directly to the AFU.62 63    The CXL Flash Adapter Driver establishes a master context with the64    AFU. It uses memory mapped I/O (MMIO) for this control and setup. The65    Adapter Problem Space Memory Map looks like this::66 67                     +-------------------------------+68                     |    512 * 64 KB User MMIO      |69                     |        (per context)          |70                     |       User Accessible         |71                     +-------------------------------+72                     |    512 * 128 B per context    |73                     |    Provisioning and Control   |74                     |   Trusted Process accessible  |75                     +-------------------------------+76                     |         64 KB Global          |77                     |   Trusted Process accessible  |78                     +-------------------------------+79 80    This driver configures itself into the SCSI software stack as an81    adapter driver. The driver is the only entity that is considered a82    Trusted Process to program the Provisioning and Control and Global83    areas in the MMIO Space shown above.  The master context driver84    discovers all LUNs attached to the CXL Flash adapter and instantiates85    scsi block devices (/dev/sdb, /dev/sdc etc.) for each unique LUN86    seen from each path.87 88    Once these scsi block devices are instantiated, an application89    written to a specification provided by the block library may get90    access to the Flash from user space (without requiring a system call).91 92    This master context driver also provides a series of ioctls for this93    block library to enable this user space access.  The driver supports94    two modes for accessing the block device.95 96    The first mode is called a virtual mode. In this mode a single scsi97    block device (/dev/sdb) may be carved up into any number of distinct98    virtual LUNs. The virtual LUNs may be resized as long as the sum of99    the sizes of all the virtual LUNs, along with the meta-data associated100    with it does not exceed the physical capacity.101 102    The second mode is called the physical mode. In this mode a single103    block device (/dev/sdb) may be opened directly by the block library104    and the entire space for the LUN is available to the application.105 106    Only the physical mode provides persistence of the data.  i.e. The107    data written to the block device will survive application exit and108    restart and also reboot. The virtual LUNs do not persist (i.e. do109    not survive after the application terminates or the system reboots).110 111 112Block library API113=================114 115    Applications intending to get access to the CXL Flash from user116    space should use the block library, as it abstracts the details of117    interfacing directly with the cxlflash driver that are necessary for118    performing administrative actions (i.e.: setup, tear down, resize).119    The block library can be thought of as a 'user' of services,120    implemented as IOCTLs, that are provided by the cxlflash driver121    specifically for devices (LUNs) operating in user space access122    mode. While it is not a requirement that applications understand123    the interface between the block library and the cxlflash driver,124    a high-level overview of each supported service (IOCTL) is provided125    below.126 127    The block library can be found on GitHub:128    http://github.com/open-power/capiflash129 130 131CXL Flash Driver LUN IOCTLs132===========================133 134    Users, such as the block library, that wish to interface with a flash135    device (LUN) via user space access need to use the services provided136    by the cxlflash driver. As these services are implemented as ioctls,137    a file descriptor handle must first be obtained in order to establish138    the communication channel between a user and the kernel.  This file139    descriptor is obtained by opening the device special file associated140    with the scsi disk device (/dev/sdb) that was created during LUN141    discovery. As per the location of the cxlflash driver within the142    SCSI protocol stack, this open is actually not seen by the cxlflash143    driver. Upon successful open, the user receives a file descriptor144    (herein referred to as fd1) that should be used for issuing the145    subsequent ioctls listed below.146 147    The structure definitions for these IOCTLs are available in:148    uapi/scsi/cxlflash_ioctl.h149 150DK_CXLFLASH_ATTACH151------------------152 153    This ioctl obtains, initializes, and starts a context using the CXL154    kernel services. These services specify a context id (u16) by which155    to uniquely identify the context and its allocated resources. The156    services additionally provide a second file descriptor (herein157    referred to as fd2) that is used by the block library to initiate158    memory mapped I/O (via mmap()) to the CXL flash device and poll for159    completion events. This file descriptor is intentionally installed by160    this driver and not the CXL kernel services to allow for intermediary161    notification and access in the event of a non-user-initiated close(),162    such as a killed process. This design point is described in further163    detail in the description for the DK_CXLFLASH_DETACH ioctl.164 165    There are a few important aspects regarding the "tokens" (context id166    and fd2) that are provided back to the user:167 168        - These tokens are only valid for the process under which they169          were created. The child of a forked process cannot continue170          to use the context id or file descriptor created by its parent171          (see DK_CXLFLASH_VLUN_CLONE for further details).172 173        - These tokens are only valid for the lifetime of the context and174          the process under which they were created. Once either is175          destroyed, the tokens are to be considered stale and subsequent176          usage will result in errors.177 178	- A valid adapter file descriptor (fd2 >= 0) is only returned on179	  the initial attach for a context. Subsequent attaches to an180	  existing context (DK_CXLFLASH_ATTACH_REUSE_CONTEXT flag present)181	  do not provide the adapter file descriptor as it was previously182	  made known to the application.183 184        - When a context is no longer needed, the user shall detach from185          the context via the DK_CXLFLASH_DETACH ioctl. When this ioctl186	  returns with a valid adapter file descriptor and the return flag187	  DK_CXLFLASH_APP_CLOSE_ADAP_FD is present, the application _must_188	  close the adapter file descriptor following a successful detach.189 190	- When this ioctl returns with a valid fd2 and the return flag191	  DK_CXLFLASH_APP_CLOSE_ADAP_FD is present, the application _must_192	  close fd2 in the following circumstances:193 194		+ Following a successful detach of the last user of the context195		+ Following a successful recovery on the context's original fd2196		+ In the child process of a fork(), following a clone ioctl,197		  on the fd2 associated with the source context198 199        - At any time, a close on fd2 will invalidate the tokens. Applications200	  should exercise caution to only close fd2 when appropriate (outlined201	  in the previous bullet) to avoid premature loss of I/O.202 203DK_CXLFLASH_USER_DIRECT204-----------------------205    This ioctl is responsible for transitioning the LUN to direct206    (physical) mode access and configuring the AFU for direct access from207    user space on a per-context basis. Additionally, the block size and208    last logical block address (LBA) are returned to the user.209 210    As mentioned previously, when operating in user space access mode,211    LUNs may be accessed in whole or in part. Only one mode is allowed212    at a time and if one mode is active (outstanding references exist),213    requests to use the LUN in a different mode are denied.214 215    The AFU is configured for direct access from user space by adding an216    entry to the AFU's resource handle table. The index of the entry is217    treated as a resource handle that is returned to the user. The user218    is then able to use the handle to reference the LUN during I/O.219 220DK_CXLFLASH_USER_VIRTUAL221------------------------222    This ioctl is responsible for transitioning the LUN to virtual mode223    of access and configuring the AFU for virtual access from user space224    on a per-context basis. Additionally, the block size and last logical225    block address (LBA) are returned to the user.226 227    As mentioned previously, when operating in user space access mode,228    LUNs may be accessed in whole or in part. Only one mode is allowed229    at a time and if one mode is active (outstanding references exist),230    requests to use the LUN in a different mode are denied.231 232    The AFU is configured for virtual access from user space by adding233    an entry to the AFU's resource handle table. The index of the entry234    is treated as a resource handle that is returned to the user. The235    user is then able to use the handle to reference the LUN during I/O.236 237    By default, the virtual LUN is created with a size of 0. The user238    would need to use the DK_CXLFLASH_VLUN_RESIZE ioctl to adjust the grow239    the virtual LUN to a desired size. To avoid having to perform this240    resize for the initial creation of the virtual LUN, the user has the241    option of specifying a size as part of the DK_CXLFLASH_USER_VIRTUAL242    ioctl, such that when success is returned to the user, the243    resource handle that is provided is already referencing provisioned244    storage. This is reflected by the last LBA being a non-zero value.245 246    When a LUN is accessible from more than one port, this ioctl will247    return with the DK_CXLFLASH_ALL_PORTS_ACTIVE return flag set. This248    provides the user with a hint that I/O can be retried in the event249    of an I/O error as the LUN can be reached over multiple paths.250 251DK_CXLFLASH_VLUN_RESIZE252-----------------------253    This ioctl is responsible for resizing a previously created virtual254    LUN and will fail if invoked upon a LUN that is not in virtual255    mode. Upon success, an updated last LBA is returned to the user256    indicating the new size of the virtual LUN associated with the257    resource handle.258 259    The partitioning of virtual LUNs is jointly mediated by the cxlflash260    driver and the AFU. An allocation table is kept for each LUN that is261    operating in the virtual mode and used to program a LUN translation262    table that the AFU references when provided with a resource handle.263 264    This ioctl can return -EAGAIN if an AFU sync operation takes too long.265    In addition to returning a failure to user, cxlflash will also schedule266    an asynchronous AFU reset. Should the user choose to retry the operation,267    it is expected to succeed. If this ioctl fails with -EAGAIN, the user268    can either retry the operation or treat it as a failure.269 270DK_CXLFLASH_RELEASE271-------------------272    This ioctl is responsible for releasing a previously obtained273    reference to either a physical or virtual LUN. This can be274    thought of as the inverse of the DK_CXLFLASH_USER_DIRECT or275    DK_CXLFLASH_USER_VIRTUAL ioctls. Upon success, the resource handle276    is no longer valid and the entry in the resource handle table is277    made available to be used again.278 279    As part of the release process for virtual LUNs, the virtual LUN280    is first resized to 0 to clear out and free the translation tables281    associated with the virtual LUN reference.282 283DK_CXLFLASH_DETACH284------------------285    This ioctl is responsible for unregistering a context with the286    cxlflash driver and release outstanding resources that were287    not explicitly released via the DK_CXLFLASH_RELEASE ioctl. Upon288    success, all "tokens" which had been provided to the user from the289    DK_CXLFLASH_ATTACH onward are no longer valid.290 291    When the DK_CXLFLASH_APP_CLOSE_ADAP_FD flag was returned on a successful292    attach, the application _must_ close the fd2 associated with the context293    following the detach of the final user of the context.294 295DK_CXLFLASH_VLUN_CLONE296----------------------297    This ioctl is responsible for cloning a previously created298    context to a more recently created context. It exists solely to299    support maintaining user space access to storage after a process300    forks. Upon success, the child process (which invoked the ioctl)301    will have access to the same LUNs via the same resource handle(s)302    as the parent, but under a different context.303 304    Context sharing across processes is not supported with CXL and305    therefore each fork must be met with establishing a new context306    for the child process. This ioctl simplifies the state management307    and playback required by a user in such a scenario. When a process308    forks, child process can clone the parents context by first creating309    a context (via DK_CXLFLASH_ATTACH) and then using this ioctl to310    perform the clone from the parent to the child.311 312    The clone itself is fairly simple. The resource handle and lun313    translation tables are copied from the parent context to the child's314    and then synced with the AFU.315 316    When the DK_CXLFLASH_APP_CLOSE_ADAP_FD flag was returned on a successful317    attach, the application _must_ close the fd2 associated with the source318    context (still resident/accessible in the parent process) following the319    clone. This is to avoid a stale entry in the file descriptor table of the320    child process.321 322    This ioctl can return -EAGAIN if an AFU sync operation takes too long.323    In addition to returning a failure to user, cxlflash will also schedule324    an asynchronous AFU reset. Should the user choose to retry the operation,325    it is expected to succeed. If this ioctl fails with -EAGAIN, the user326    can either retry the operation or treat it as a failure.327 328DK_CXLFLASH_VERIFY329------------------330    This ioctl is used to detect various changes such as the capacity of331    the disk changing, the number of LUNs visible changing, etc. In cases332    where the changes affect the application (such as a LUN resize), the333    cxlflash driver will report the changed state to the application.334 335    The user calls in when they want to validate that a LUN hasn't been336    changed in response to a check condition. As the user is operating out337    of band from the kernel, they will see these types of events without338    the kernel's knowledge. When encountered, the user's architected339    behavior is to call in to this ioctl, indicating what they want to340    verify and passing along any appropriate information. For now, only341    verifying a LUN change (ie: size different) with sense data is342    supported.343 344DK_CXLFLASH_RECOVER_AFU345-----------------------346    This ioctl is used to drive recovery (if such an action is warranted)347    of a specified user context. Any state associated with the user context348    is re-established upon successful recovery.349 350    User contexts are put into an error condition when the device needs to351    be reset or is terminating. Users are notified of this error condition352    by seeing all 0xF's on an MMIO read. Upon encountering this, the353    architected behavior for a user is to call into this ioctl to recover354    their context. A user may also call into this ioctl at any time to355    check if the device is operating normally. If a failure is returned356    from this ioctl, the user is expected to gracefully clean up their357    context via release/detach ioctls. Until they do, the context they358    hold is not relinquished. The user may also optionally exit the process359    at which time the context/resources they held will be freed as part of360    the release fop.361 362    When the DK_CXLFLASH_APP_CLOSE_ADAP_FD flag was returned on a successful363    attach, the application _must_ unmap and close the fd2 associated with the364    original context following this ioctl returning success and indicating that365    the context was recovered (DK_CXLFLASH_RECOVER_AFU_CONTEXT_RESET).366 367DK_CXLFLASH_MANAGE_LUN368----------------------369    This ioctl is used to switch a LUN from a mode where it is available370    for file-system access (legacy), to a mode where it is set aside for371    exclusive user space access (superpipe). In case a LUN is visible372    across multiple ports and adapters, this ioctl is used to uniquely373    identify each LUN by its World Wide Node Name (WWNN).374 375 376CXL Flash Driver Host IOCTLs377============================378 379    Each host adapter instance that is supported by the cxlflash driver380    has a special character device associated with it to enable a set of381    host management function. These character devices are hosted in a382    class dedicated for cxlflash and can be accessed via `/dev/cxlflash/*`.383 384    Applications can be written to perform various functions using the385    host ioctl APIs below.386 387    The structure definitions for these IOCTLs are available in:388    uapi/scsi/cxlflash_ioctl.h389 390HT_CXLFLASH_LUN_PROVISION391-------------------------392    This ioctl is used to create and delete persistent LUNs on cxlflash393    devices that lack an external LUN management interface. It is only394    valid when used with AFUs that support the LUN provision capability.395 396    When sufficient space is available, LUNs can be created by specifying397    the target port to host the LUN and a desired size in 4K blocks. Upon398    success, the LUN ID and WWID of the created LUN will be returned and399    the SCSI bus can be scanned to detect the change in LUN topology. Note400    that partial allocations are not supported. Should a creation fail due401    to a space issue, the target port can be queried for its current LUN402    geometry.403 404    To remove a LUN, the device must first be disassociated from the Linux405    SCSI subsystem. The LUN deletion can then be initiated by specifying a406    target port and LUN ID. Upon success, the LUN geometry associated with407    the port will be updated to reflect new number of provisioned LUNs and408    available capacity.409 410    To query the LUN geometry of a port, the target port is specified and411    upon success, the following information is presented:412 413        - Maximum number of provisioned LUNs allowed for the port414        - Current number of provisioned LUNs for the port415        - Maximum total capacity of provisioned LUNs for the port (4K blocks)416        - Current total capacity of provisioned LUNs for the port (4K blocks)417 418    With this information, the number of available LUNs and capacity can be419    can be calculated.420 421HT_CXLFLASH_AFU_DEBUG422---------------------423    This ioctl is used to debug AFUs by supporting a command pass-through424    interface. It is only valid when used with AFUs that support the AFU425    debug capability.426 427    With exception of buffer management, AFU debug commands are opaque to428    cxlflash and treated as pass-through. For debug commands that do require429    data transfer, the user supplies an adequately sized data buffer and must430    specify the data transfer direction with respect to the host. There is a431    maximum transfer size of 256K imposed. Note that partial read completions432    are not supported - when errors are experienced with a host read data433    transfer, the data buffer is not copied back to the user.434