diff options
Diffstat (limited to 'Documentation/block')
| -rw-r--r-- | Documentation/block/00-INDEX | 34 | ||||
| -rw-r--r-- | Documentation/block/bfq-iosched.txt | 561 | ||||
| -rw-r--r-- | Documentation/block/biodoc.txt | 1165 | ||||
| -rw-r--r-- | Documentation/block/biovecs.txt | 119 | ||||
| -rw-r--r-- | Documentation/block/capability.txt | 15 | ||||
| -rw-r--r-- | Documentation/block/cfq-iosched.txt | 291 | ||||
| -rw-r--r-- | Documentation/block/cmdline-partition.txt | 46 | ||||
| -rw-r--r-- | Documentation/block/data-integrity.txt | 281 | ||||
| -rw-r--r-- | Documentation/block/deadline-iosched.txt | 75 | ||||
| -rw-r--r-- | Documentation/block/ioprio.txt | 183 | ||||
| -rw-r--r-- | Documentation/block/kyber-iosched.txt | 14 | ||||
| -rw-r--r-- | Documentation/block/null_blk.txt | 94 | ||||
| -rw-r--r-- | Documentation/block/pr.txt | 119 | ||||
| -rw-r--r-- | Documentation/block/queue-sysfs.txt | 197 | ||||
| -rw-r--r-- | Documentation/block/request.txt | 88 | ||||
| -rw-r--r-- | Documentation/block/stat.txt | 86 | ||||
| -rw-r--r-- | Documentation/block/switching-sched.txt | 37 | ||||
| -rw-r--r-- | Documentation/block/writeback_cache_control.txt | 86 |
18 files changed, 3491 insertions, 0 deletions
diff --git a/Documentation/block/00-INDEX b/Documentation/block/00-INDEX new file mode 100644 index 000000000..8d55b4bbb --- /dev/null +++ b/Documentation/block/00-INDEX @@ -0,0 +1,34 @@ +00-INDEX + - This file +bfq-iosched.txt + - BFQ IO scheduler and its tunables +biodoc.txt + - Notes on the Generic Block Layer Rewrite in Linux 2.5 +biovecs.txt + - Immutable biovecs and biovec iterators +capability.txt + - Generic Block Device Capability (/sys/block/<device>/capability) +cfq-iosched.txt + - CFQ IO scheduler tunables +cmdline-partition.txt + - how to specify block device partitions on kernel command line +data-integrity.txt + - Block data integrity +deadline-iosched.txt + - Deadline IO scheduler tunables +ioprio.txt + - Block io priorities (in CFQ scheduler) +pr.txt + - Block layer support for Persistent Reservations +null_blk.txt + - Null block for block-layer benchmarking. +queue-sysfs.txt + - Queue's sysfs entries +request.txt + - The members of struct request (in include/linux/blkdev.h) +stat.txt + - Block layer statistics in /sys/block/<device>/stat +switching-sched.txt + - Switching I/O schedulers at runtime +writeback_cache_control.txt + - Control of volatile write back caches diff --git a/Documentation/block/bfq-iosched.txt b/Documentation/block/bfq-iosched.txt new file mode 100644 index 000000000..8d8d8f06c --- /dev/null +++ b/Documentation/block/bfq-iosched.txt @@ -0,0 +1,561 @@ +BFQ (Budget Fair Queueing) +========================== + +BFQ is a proportional-share I/O scheduler, with some extra +low-latency capabilities. In addition to cgroups support (blkio or io +controllers), BFQ's main features are: +- BFQ guarantees a high system and application responsiveness, and a + low latency for time-sensitive applications, such as audio or video + players; +- BFQ distributes bandwidth, and not just time, among processes or + groups (switching back to time distribution when needed to keep + throughput high). + +In its default configuration, BFQ privileges latency over +throughput. So, when needed for achieving a lower latency, BFQ builds +schedules that may lead to a lower throughput. If your main or only +goal, for a given device, is to achieve the maximum-possible +throughput at all times, then do switch off all low-latency heuristics +for that device, by setting low_latency to 0. See Section 3 for +details on how to configure BFQ for the desired tradeoff between +latency and throughput, or on how to maximize throughput. + +BFQ has a non-null overhead, which limits the maximum IOPS that a CPU +can process for a device scheduled with BFQ. To give an idea of the +limits on slow or average CPUs, here are, first, the limits of BFQ for +three different CPUs, on, respectively, an average laptop, an old +desktop, and a cheap embedded system, in case full hierarchical +support is enabled (i.e., CONFIG_BFQ_GROUP_IOSCHED is set), but +CONFIG_DEBUG_BLK_CGROUP is not set (Section 4-2): +- Intel i7-4850HQ: 400 KIOPS +- AMD A8-3850: 250 KIOPS +- ARM CortexTM-A53 Octa-core: 80 KIOPS + +If CONFIG_DEBUG_BLK_CGROUP is set (and of course full hierarchical +support is enabled), then the sustainable throughput with BFQ +decreases, because all blkio.bfq* statistics are created and updated +(Section 4-2). For BFQ, this leads to the following maximum +sustainable throughputs, on the same systems as above: +- Intel i7-4850HQ: 310 KIOPS +- AMD A8-3850: 200 KIOPS +- ARM CortexTM-A53 Octa-core: 56 KIOPS + +BFQ works for multi-queue devices too. + +The table of contents follow. Impatients can just jump to Section 3. + +CONTENTS + +1. When may BFQ be useful? + 1-1 Personal systems + 1-2 Server systems +2. How does BFQ work? +3. What are BFQ's tunables and how to properly configure BFQ? +4. BFQ group scheduling + 4-1 Service guarantees provided + 4-2 Interface + +1. When may BFQ be useful? +========================== + +BFQ provides the following benefits on personal and server systems. + +1-1 Personal systems +-------------------- + +Low latency for interactive applications + +Regardless of the actual background workload, BFQ guarantees that, for +interactive tasks, the storage device is virtually as responsive as if +it was idle. For example, even if one or more of the following +background workloads are being executed: +- one or more large files are being read, written or copied, +- a tree of source files is being compiled, +- one or more virtual machines are performing I/O, +- a software update is in progress, +- indexing daemons are scanning filesystems and updating their + databases, +starting an application or loading a file from within an application +takes about the same time as if the storage device was idle. As a +comparison, with CFQ, NOOP or DEADLINE, and in the same conditions, +applications experience high latencies, or even become unresponsive +until the background workload terminates (also on SSDs). + +Low latency for soft real-time applications + +Also soft real-time applications, such as audio and video +players/streamers, enjoy a low latency and a low drop rate, regardless +of the background I/O workload. As a consequence, these applications +do not suffer from almost any glitch due to the background workload. + +Higher speed for code-development tasks + +If some additional workload happens to be executed in parallel, then +BFQ executes the I/O-related components of typical code-development +tasks (compilation, checkout, merge, ...) much more quickly than CFQ, +NOOP or DEADLINE. + +High throughput + +On hard disks, BFQ achieves up to 30% higher throughput than CFQ, and +up to 150% higher throughput than DEADLINE and NOOP, with all the +sequential workloads considered in our tests. With random workloads, +and with all the workloads on flash-based devices, BFQ achieves, +instead, about the same throughput as the other schedulers. + +Strong fairness, bandwidth and delay guarantees + +BFQ distributes the device throughput, and not just the device time, +among I/O-bound applications in proportion their weights, with any +workload and regardless of the device parameters. From these bandwidth +guarantees, it is possible to compute tight per-I/O-request delay +guarantees by a simple formula. If not configured for strict service +guarantees, BFQ switches to time-based resource sharing (only) for +applications that would otherwise cause a throughput loss. + +1-2 Server systems +------------------ + +Most benefits for server systems follow from the same service +properties as above. In particular, regardless of whether additional, +possibly heavy workloads are being served, BFQ guarantees: + +. audio and video-streaming with zero or very low jitter and drop + rate; + +. fast retrieval of WEB pages and embedded objects; + +. real-time recording of data in live-dumping applications (e.g., + packet logging); + +. responsiveness in local and remote access to a server. + + +2. How does BFQ work? +===================== + +BFQ is a proportional-share I/O scheduler, whose general structure, +plus a lot of code, are borrowed from CFQ. + +- Each process doing I/O on a device is associated with a weight and a + (bfq_)queue. + +- BFQ grants exclusive access to the device, for a while, to one queue + (process) at a time, and implements this service model by + associating every queue with a budget, measured in number of + sectors. + + - After a queue is granted access to the device, the budget of the + queue is decremented, on each request dispatch, by the size of the + request. + + - The in-service queue is expired, i.e., its service is suspended, + only if one of the following events occurs: 1) the queue finishes + its budget, 2) the queue empties, 3) a "budget timeout" fires. + + - The budget timeout prevents processes doing random I/O from + holding the device for too long and dramatically reducing + throughput. + + - Actually, as in CFQ, a queue associated with a process issuing + sync requests may not be expired immediately when it empties. In + contrast, BFQ may idle the device for a short time interval, + giving the process the chance to go on being served if it issues + a new request in time. Device idling typically boosts the + throughput on rotational devices and on non-queueing flash-based + devices, if processes do synchronous and sequential I/O. In + addition, under BFQ, device idling is also instrumental in + guaranteeing the desired throughput fraction to processes + issuing sync requests (see the description of the slice_idle + tunable in this document, or [1, 2], for more details). + + - With respect to idling for service guarantees, if several + processes are competing for the device at the same time, but + all processes and groups have the same weight, then BFQ + guarantees the expected throughput distribution without ever + idling the device. Throughput is thus as high as possible in + this common scenario. + + - On flash-based storage with internal queueing of commands + (typically NCQ), device idling happens to be always detrimental + for throughput. So, with these devices, BFQ performs idling + only when strictly needed for service guarantees, i.e., for + guaranteeing low latency or fairness. In these cases, overall + throughput may be sub-optimal. No solution currently exists to + provide both strong service guarantees and optimal throughput + on devices with internal queueing. + + - If low-latency mode is enabled (default configuration), BFQ + executes some special heuristics to detect interactive and soft + real-time applications (e.g., video or audio players/streamers), + and to reduce their latency. The most important action taken to + achieve this goal is to give to the queues associated with these + applications more than their fair share of the device + throughput. For brevity, we call just "weight-raising" the whole + sets of actions taken by BFQ to privilege these queues. In + particular, BFQ provides a milder form of weight-raising for + interactive applications, and a stronger form for soft real-time + applications. + + - BFQ automatically deactivates idling for queues born in a burst of + queue creations. In fact, these queues are usually associated with + the processes of applications and services that benefit mostly + from a high throughput. Examples are systemd during boot, or git + grep. + + - As CFQ, BFQ merges queues performing interleaved I/O, i.e., + performing random I/O that becomes mostly sequential if + merged. Differently from CFQ, BFQ achieves this goal with a more + reactive mechanism, called Early Queue Merge (EQM). EQM is so + responsive in detecting interleaved I/O (cooperating processes), + that it enables BFQ to achieve a high throughput, by queue + merging, even for queues for which CFQ needs a different + mechanism, preemption, to get a high throughput. As such EQM is a + unified mechanism to achieve a high throughput with interleaved + I/O. + + - Queues are scheduled according to a variant of WF2Q+, named + B-WF2Q+, and implemented using an augmented rb-tree to preserve an + O(log N) overall complexity. See [2] for more details. B-WF2Q+ is + also ready for hierarchical scheduling, details in Section 4. + + - B-WF2Q+ guarantees a tight deviation with respect to an ideal, + perfectly fair, and smooth service. In particular, B-WF2Q+ + guarantees that each queue receives a fraction of the device + throughput proportional to its weight, even if the throughput + fluctuates, and regardless of: the device parameters, the current + workload and the budgets assigned to the queue. + + - The last, budget-independence, property (although probably + counterintuitive in the first place) is definitely beneficial, for + the following reasons: + + - First, with any proportional-share scheduler, the maximum + deviation with respect to an ideal service is proportional to + the maximum budget (slice) assigned to queues. As a consequence, + BFQ can keep this deviation tight not only because of the + accurate service of B-WF2Q+, but also because BFQ *does not* + need to assign a larger budget to a queue to let the queue + receive a higher fraction of the device throughput. + + - Second, BFQ is free to choose, for every process (queue), the + budget that best fits the needs of the process, or best + leverages the I/O pattern of the process. In particular, BFQ + updates queue budgets with a simple feedback-loop algorithm that + allows a high throughput to be achieved, while still providing + tight latency guarantees to time-sensitive applications. When + the in-service queue expires, this algorithm computes the next + budget of the queue so as to: + + - Let large budgets be eventually assigned to the queues + associated with I/O-bound applications performing sequential + I/O: in fact, the longer these applications are served once + got access to the device, the higher the throughput is. + + - Let small budgets be eventually assigned to the queues + associated with time-sensitive applications (which typically + perform sporadic and short I/O), because, the smaller the + budget assigned to a queue waiting for service is, the sooner + B-WF2Q+ will serve that queue (Subsec 3.3 in [2]). + +- If several processes are competing for the device at the same time, + but all processes and groups have the same weight, then BFQ + guarantees the expected throughput distribution without ever idling + the device. It uses preemption instead. Throughput is then much + higher in this common scenario. + +- ioprio classes are served in strict priority order, i.e., + lower-priority queues are not served as long as there are + higher-priority queues. Among queues in the same class, the + bandwidth is distributed in proportion to the weight of each + queue. A very thin extra bandwidth is however guaranteed to + the Idle class, to prevent it from starving. + + +3. What are BFQ's tunables and how to properly configure BFQ? +============================================================= + +Most BFQ tunables affect service guarantees (basically latency and +fairness) and throughput. For full details on how to choose the +desired tradeoff between service guarantees and throughput, see the +parameters slice_idle, strict_guarantees and low_latency. For details +on how to maximise throughput, see slice_idle, timeout_sync and +max_budget. The other performance-related parameters have been +inherited from, and have been preserved mostly for compatibility with +CFQ. So far, no performance improvement has been reported after +changing the latter parameters in BFQ. + +In particular, the tunables back_seek-max, back_seek_penalty, +fifo_expire_async and fifo_expire_sync below are the same as in +CFQ. Their description is just copied from that for CFQ. Some +considerations in the description of slice_idle are copied from CFQ +too. + +per-process ioprio and weight +----------------------------- + +Unless the cgroups interface is used (see "4. BFQ group scheduling"), +weights can be assigned to processes only indirectly, through I/O +priorities, and according to the relation: +weight = (IOPRIO_BE_NR - ioprio) * 10. + +Beware that, if low-latency is set, then BFQ automatically raises the +weight of the queues associated with interactive and soft real-time +applications. Unset this tunable if you need/want to control weights. + +slice_idle +---------- + +This parameter specifies how long BFQ should idle for next I/O +request, when certain sync BFQ queues become empty. By default +slice_idle is a non-zero value. Idling has a double purpose: boosting +throughput and making sure that the desired throughput distribution is +respected (see the description of how BFQ works, and, if needed, the +papers referred there). + +As for throughput, idling can be very helpful on highly seeky media +like single spindle SATA/SAS disks where we can cut down on overall +number of seeks and see improved throughput. + +Setting slice_idle to 0 will remove all the idling on queues and one +should see an overall improved throughput on faster storage devices +like multiple SATA/SAS disks in hardware RAID configuration, as well +as flash-based storage with internal command queueing (and +parallelism). + +So depending on storage and workload, it might be useful to set +slice_idle=0. In general for SATA/SAS disks and software RAID of +SATA/SAS disks keeping slice_idle enabled should be useful. For any +configurations where there are multiple spindles behind single LUN +(Host based hardware RAID controller or for storage arrays), or with +flash-based fast storage, setting slice_idle=0 might end up in better +throughput and acceptable latencies. + +Idling is however necessary to have service guarantees enforced in +case of differentiated weights or differentiated I/O-request lengths. +To see why, suppose that a given BFQ queue A must get several I/O +requests served for each request served for another queue B. Idling +ensures that, if A makes a new I/O request slightly after becoming +empty, then no request of B is dispatched in the middle, and thus A +does not lose the possibility to get more than one request dispatched +before the next request of B is dispatched. Note that idling +guarantees the desired differentiated treatment of queues only in +terms of I/O-request dispatches. To guarantee that the actual service +order then corresponds to the dispatch order, the strict_guarantees +tunable must be set too. + +There is an important flipside for idling: apart from the above cases +where it is beneficial also for throughput, idling can severely impact +throughput. One important case is random workload. Because of this +issue, BFQ tends to avoid idling as much as possible, when it is not +beneficial also for throughput (as detailed in Section 2). As a +consequence of this behavior, and of further issues described for the +strict_guarantees tunable, short-term service guarantees may be +occasionally violated. And, in some cases, these guarantees may be +more important than guaranteeing maximum throughput. For example, in +video playing/streaming, a very low drop rate may be more important +than maximum throughput. In these cases, consider setting the +strict_guarantees parameter. + +strict_guarantees +----------------- + +If this parameter is set (default: unset), then BFQ + +- always performs idling when the in-service queue becomes empty; + +- forces the device to serve one I/O request at a time, by dispatching a + new request only if there is no outstanding request. + +In the presence of differentiated weights or I/O-request sizes, both +the above conditions are needed to guarantee that every BFQ queue +receives its allotted share of the bandwidth. The first condition is +needed for the reasons explained in the description of the slice_idle +tunable. The second condition is needed because all modern storage +devices reorder internally-queued requests, which may trivially break +the service guarantees enforced by the I/O scheduler. + +Setting strict_guarantees may evidently affect throughput. + +back_seek_max +------------- + +This specifies, given in Kbytes, the maximum "distance" for backward seeking. +The distance is the amount of space from the current head location to the +sectors that are backward in terms of distance. + +This parameter allows the scheduler to anticipate requests in the "backward" +direction and consider them as being the "next" if they are within this +distance from the current head location. + +back_seek_penalty +----------------- + +This parameter is used to compute the cost of backward seeking. If the +backward distance of request is just 1/back_seek_penalty from a "front" +request, then the seeking cost of two requests is considered equivalent. + +So scheduler will not bias toward one or the other request (otherwise scheduler +will bias toward front request). Default value of back_seek_penalty is 2. + +fifo_expire_async +----------------- + +This parameter is used to set the timeout of asynchronous requests. Default +value of this is 248ms. + +fifo_expire_sync +---------------- + +This parameter is used to set the timeout of synchronous requests. Default +value of this is 124ms. In case to favor synchronous requests over asynchronous +one, this value should be decreased relative to fifo_expire_async. + +low_latency +----------- + +This parameter is used to enable/disable BFQ's low latency mode. By +default, low latency mode is enabled. If enabled, interactive and soft +real-time applications are privileged and experience a lower latency, +as explained in more detail in the description of how BFQ works. + +DISABLE this mode if you need full control on bandwidth +distribution. In fact, if it is enabled, then BFQ automatically +increases the bandwidth share of privileged applications, as the main +means to guarantee a lower latency to them. + +In addition, as already highlighted at the beginning of this document, +DISABLE this mode if your only goal is to achieve a high throughput. +In fact, privileging the I/O of some application over the rest may +entail a lower throughput. To achieve the highest-possible throughput +on a non-rotational device, setting slice_idle to 0 may be needed too +(at the cost of giving up any strong guarantee on fairness and low +latency). + +timeout_sync +------------ + +Maximum amount of device time that can be given to a task (queue) once +it has been selected for service. On devices with costly seeks, +increasing this time usually increases maximum throughput. On the +opposite end, increasing this time coarsens the granularity of the +short-term bandwidth and latency guarantees, especially if the +following parameter is set to zero. + +max_budget +---------- + +Maximum amount of service, measured in sectors, that can be provided +to a BFQ queue once it is set in service (of course within the limits +of the above timeout). According to what said in the description of +the algorithm, larger values increase the throughput in proportion to +the percentage of sequential I/O requests issued. The price of larger +values is that they coarsen the granularity of short-term bandwidth +and latency guarantees. + +The default value is 0, which enables auto-tuning: BFQ sets max_budget +to the maximum number of sectors that can be served during +timeout_sync, according to the estimated peak rate. + +For specific devices, some users have occasionally reported to have +reached a higher throughput by setting max_budget explicitly, i.e., by +setting max_budget to a higher value than 0. In particular, they have +set max_budget to higher values than those to which BFQ would have set +it with auto-tuning. An alternative way to achieve this goal is to +just increase the value of timeout_sync, leaving max_budget equal to 0. + +weights +------- + +Read-only parameter, used to show the weights of the currently active +BFQ queues. + + +4. Group scheduling with BFQ +============================ + +BFQ supports both cgroups-v1 and cgroups-v2 io controllers, namely +blkio and io. In particular, BFQ supports weight-based proportional +share. To activate cgroups support, set BFQ_GROUP_IOSCHED. + +4-1 Service guarantees provided +------------------------------- + +With BFQ, proportional share means true proportional share of the +device bandwidth, according to group weights. For example, a group +with weight 200 gets twice the bandwidth, and not just twice the time, +of a group with weight 100. + +BFQ supports hierarchies (group trees) of any depth. Bandwidth is +distributed among groups and processes in the expected way: for each +group, the children of the group share the whole bandwidth of the +group in proportion to their weights. In particular, this implies +that, for each leaf group, every process of the group receives the +same share of the whole group bandwidth, unless the ioprio of the +process is modified. + +The resource-sharing guarantee for a group may partially or totally +switch from bandwidth to time, if providing bandwidth guarantees to +the group lowers the throughput too much. This switch occurs on a +per-process basis: if a process of a leaf group causes throughput loss +if served in such a way to receive its share of the bandwidth, then +BFQ switches back to just time-based proportional share for that +process. + +4-2 Interface +------------- + +To get proportional sharing of bandwidth with BFQ for a given device, +BFQ must of course be the active scheduler for that device. + +Within each group directory, the names of the files associated with +BFQ-specific cgroup parameters and stats begin with the "bfq." +prefix. So, with cgroups-v1 or cgroups-v2, the full prefix for +BFQ-specific files is "blkio.bfq." or "io.bfq." For example, the group +parameter to set the weight of a group with BFQ is blkio.bfq.weight +or io.bfq.weight. + +As for cgroups-v1 (blkio controller), the exact set of stat files +created, and kept up-to-date by bfq, depends on whether +CONFIG_DEBUG_BLK_CGROUP is set. If it is set, then bfq creates all +the stat files documented in +Documentation/cgroup-v1/blkio-controller.txt. If, instead, +CONFIG_DEBUG_BLK_CGROUP is not set, then bfq creates only the files +blkio.bfq.io_service_bytes +blkio.bfq.io_service_bytes_recursive +blkio.bfq.io_serviced +blkio.bfq.io_serviced_recursive + +The value of CONFIG_DEBUG_BLK_CGROUP greatly influences the maximum +throughput sustainable with bfq, because updating the blkio.bfq.* +stats is rather costly, especially for some of the stats enabled by +CONFIG_DEBUG_BLK_CGROUP. + +Parameters to set +----------------- + +For each group, there is only the following parameter to set. + +weight (namely blkio.bfq.weight or io.bfq-weight): the weight of the +group inside its parent. Available values: 1..10000 (default 100). The +linear mapping between ioprio and weights, described at the beginning +of the tunable section, is still valid, but all weights higher than +IOPRIO_BE_NR*10 are mapped to ioprio 0. + +Recall that, if low-latency is set, then BFQ automatically raises the +weight of the queues associated with interactive and soft real-time +applications. Unset this tunable if you need/want to control weights. + + +[1] P. Valente, A. Avanzini, "Evolution of the BFQ Storage I/O + Scheduler", Proceedings of the First Workshop on Mobile System + Technologies (MST-2015), May 2015. + http://algogroup.unimore.it/people/paolo/disk_sched/mst-2015.pdf + +[2] P. Valente and M. Andreolini, "Improving Application + Responsiveness with the BFQ Disk I/O Scheduler", Proceedings of + the 5th Annual International Systems and Storage Conference + (SYSTOR '12), June 2012. + Slightly extended version: + http://algogroup.unimore.it/people/paolo/disk_sched/bfq-v1-suite- + results.pdf diff --git a/Documentation/block/biodoc.txt b/Documentation/block/biodoc.txt new file mode 100644 index 000000000..207eca58e --- /dev/null +++ b/Documentation/block/biodoc.txt @@ -0,0 +1,1165 @@ + Notes on the Generic Block Layer Rewrite in Linux 2.5 + ===================================================== + +Notes Written on Jan 15, 2002: + Jens Axboe <jens.axboe@oracle.com> + Suparna Bhattacharya <suparna@in.ibm.com> + +Last Updated May 2, 2002 +September 2003: Updated I/O Scheduler portions + Nick Piggin <npiggin@kernel.dk> + +Introduction: + +These are some notes describing some aspects of the 2.5 block layer in the +context of the bio rewrite. The idea is to bring out some of the key +changes and a glimpse of the rationale behind those changes. + +Please mail corrections & suggestions to suparna@in.ibm.com. + +Credits: +--------- + +2.5 bio rewrite: + Jens Axboe <jens.axboe@oracle.com> + +Many aspects of the generic block layer redesign were driven by and evolved +over discussions, prior patches and the collective experience of several +people. See sections 8 and 9 for a list of some related references. + +The following people helped with review comments and inputs for this +document: + Christoph Hellwig <hch@infradead.org> + Arjan van de Ven <arjanv@redhat.com> + Randy Dunlap <rdunlap@xenotime.net> + Andre Hedrick <andre@linux-ide.org> + +The following people helped with fixes/contributions to the bio patches +while it was still work-in-progress: + David S. Miller <davem@redhat.com> + + +Description of Contents: +------------------------ + +1. Scope for tuning of logic to various needs + 1.1 Tuning based on device or low level driver capabilities + - Per-queue parameters + - Highmem I/O support + - I/O scheduler modularization + 1.2 Tuning based on high level requirements/capabilities + 1.2.1 Request Priority/Latency + 1.3 Direct access/bypass to lower layers for diagnostics and special + device operations + 1.3.1 Pre-built commands +2. New flexible and generic but minimalist i/o structure or descriptor + (instead of using buffer heads at the i/o layer) + 2.1 Requirements/Goals addressed + 2.2 The bio struct in detail (multi-page io unit) + 2.3 Changes in the request structure +3. Using bios + 3.1 Setup/teardown (allocation, splitting) + 3.2 Generic bio helper routines + 3.2.1 Traversing segments and completion units in a request + 3.2.2 Setting up DMA scatterlists + 3.2.3 I/O completion + 3.2.4 Implications for drivers that do not interpret bios (don't handle + multiple segments) + 3.2.5 Request command tagging + 3.3 I/O submission +4. The I/O scheduler +5. Scalability related changes + 5.1 Granular locking: Removal of io_request_lock + 5.2 Prepare for transition to 64 bit sector_t +6. Other Changes/Implications + 6.1 Partition re-mapping handled by the generic block layer +7. A few tips on migration of older drivers +8. A list of prior/related/impacted patches/ideas +9. Other References/Discussion Threads + +--------------------------------------------------------------------------- + +Bio Notes +-------- + +Let us discuss the changes in the context of how some overall goals for the +block layer are addressed. + +1. Scope for tuning the generic logic to satisfy various requirements + +The block layer design supports adaptable abstractions to handle common +processing with the ability to tune the logic to an appropriate extent +depending on the nature of the device and the requirements of the caller. +One of the objectives of the rewrite was to increase the degree of tunability +and to enable higher level code to utilize underlying device/driver +capabilities to the maximum extent for better i/o performance. This is +important especially in the light of ever improving hardware capabilities +and application/middleware software designed to take advantage of these +capabilities. + +1.1 Tuning based on low level device / driver capabilities + +Sophisticated devices with large built-in caches, intelligent i/o scheduling +optimizations, high memory DMA support, etc may find some of the +generic processing an overhead, while for less capable devices the +generic functionality is essential for performance or correctness reasons. +Knowledge of some of the capabilities or parameters of the device should be +used at the generic block layer to take the right decisions on +behalf of the driver. + +How is this achieved ? + +Tuning at a per-queue level: + +i. Per-queue limits/values exported to the generic layer by the driver + +Various parameters that the generic i/o scheduler logic uses are set at +a per-queue level (e.g maximum request size, maximum number of segments in +a scatter-gather list, logical block size) + +Some parameters that were earlier available as global arrays indexed by +major/minor are now directly associated with the queue. Some of these may +move into the block device structure in the future. Some characteristics +have been incorporated into a queue flags field rather than separate fields +in themselves. There are blk_queue_xxx functions to set the parameters, +rather than update the fields directly + +Some new queue property settings: + + blk_queue_bounce_limit(q, u64 dma_address) + Enable I/O to highmem pages, dma_address being the + limit. No highmem default. + + blk_queue_max_sectors(q, max_sectors) + Sets two variables that limit the size of the request. + + - The request queue's max_sectors, which is a soft size in + units of 512 byte sectors, and could be dynamically varied + by the core kernel. + + - The request queue's max_hw_sectors, which is a hard limit + and reflects the maximum size request a driver can handle + in units of 512 byte sectors. + + The default for both max_sectors and max_hw_sectors is + 255. The upper limit of max_sectors is 1024. + + blk_queue_max_phys_segments(q, max_segments) + Maximum physical segments you can handle in a request. 128 + default (driver limit). (See 3.2.2) + + blk_queue_max_hw_segments(q, max_segments) + Maximum dma segments the hardware can handle in a request. 128 + default (host adapter limit, after dma remapping). + (See 3.2.2) + + blk_queue_max_segment_size(q, max_seg_size) + Maximum size of a clustered segment, 64kB default. + + blk_queue_logical_block_size(q, logical_block_size) + Lowest possible sector size that the hardware can operate + on, 512 bytes default. + +New queue flags: + + QUEUE_FLAG_CLUSTER (see 3.2.2) + QUEUE_FLAG_QUEUED (see 3.2.4) + + +ii. High-mem i/o capabilities are now considered the default + +The generic bounce buffer logic, present in 2.4, where the block layer would +by default copyin/out i/o requests on high-memory buffers to low-memory buffers +assuming that the driver wouldn't be able to handle it directly, has been +changed in 2.5. The bounce logic is now applied only for memory ranges +for which the device cannot handle i/o. A driver can specify this by +setting the queue bounce limit for the request queue for the device +(blk_queue_bounce_limit()). This avoids the inefficiencies of the copyin/out +where a device is capable of handling high memory i/o. + +In order to enable high-memory i/o where the device is capable of supporting +it, the pci dma mapping routines and associated data structures have now been +modified to accomplish a direct page -> bus translation, without requiring +a virtual address mapping (unlike the earlier scheme of virtual address +-> bus translation). So this works uniformly for high-memory pages (which +do not have a corresponding kernel virtual address space mapping) and +low-memory pages. + +Note: Please refer to Documentation/DMA-API-HOWTO.txt for a discussion +on PCI high mem DMA aspects and mapping of scatter gather lists, and support +for 64 bit PCI. + +Special handling is required only for cases where i/o needs to happen on +pages at physical memory addresses beyond what the device can support. In these +cases, a bounce bio representing a buffer from the supported memory range +is used for performing the i/o with copyin/copyout as needed depending on +the type of the operation. For example, in case of a read operation, the +data read has to be copied to the original buffer on i/o completion, so a +callback routine is set up to do this, while for write, the data is copied +from the original buffer to the bounce buffer prior to issuing the +operation. Since an original buffer may be in a high memory area that's not +mapped in kernel virtual addr, a kmap operation may be required for +performing the copy, and special care may be needed in the completion path +as it may not be in irq context. Special care is also required (by way of +GFP flags) when allocating bounce buffers, to avoid certain highmem +deadlock possibilities. + +It is also possible that a bounce buffer may be allocated from high-memory +area that's not mapped in kernel virtual addr, but within the range that the +device can use directly; so the bounce page may need to be kmapped during +copy operations. [Note: This does not hold in the current implementation, +though] + +There are some situations when pages from high memory may need to +be kmapped, even if bounce buffers are not necessary. For example a device +may need to abort DMA operations and revert to PIO for the transfer, in +which case a virtual mapping of the page is required. For SCSI it is also +done in some scenarios where the low level driver cannot be trusted to +handle a single sg entry correctly. The driver is expected to perform the +kmaps as needed on such occasions as appropriate. A driver could also use +the blk_queue_bounce() routine on its own to bounce highmem i/o to low +memory for specific requests if so desired. + +iii. The i/o scheduler algorithm itself can be replaced/set as appropriate + +As in 2.4, it is possible to plugin a brand new i/o scheduler for a particular +queue or pick from (copy) existing generic schedulers and replace/override +certain portions of it. The 2.5 rewrite provides improved modularization +of the i/o scheduler. There are more pluggable callbacks, e.g for init, +add request, extract request, which makes it possible to abstract specific +i/o scheduling algorithm aspects and details outside of the generic loop. +It also makes it possible to completely hide the implementation details of +the i/o scheduler from block drivers. + +I/O scheduler wrappers are to be used instead of accessing the queue directly. +See section 4. The I/O scheduler for details. + +1.2 Tuning Based on High level code capabilities + +i. Application capabilities for raw i/o + +This comes from some of the high-performance database/middleware +requirements where an application prefers to make its own i/o scheduling +decisions based on an understanding of the access patterns and i/o +characteristics + +ii. High performance filesystems or other higher level kernel code's +capabilities + +Kernel components like filesystems could also take their own i/o scheduling +decisions for optimizing performance. Journalling filesystems may need +some control over i/o ordering. + +What kind of support exists at the generic block layer for this ? + +The flags and rw fields in the bio structure can be used for some tuning +from above e.g indicating that an i/o is just a readahead request, or priority +settings (currently unused). As far as user applications are concerned they +would need an additional mechanism either via open flags or ioctls, or some +other upper level mechanism to communicate such settings to block. + +1.2.1 Request Priority/Latency + +Todo/Under discussion: +Arjan's proposed request priority scheme allows higher levels some broad + control (high/med/low) over the priority of an i/o request vs other pending + requests in the queue. For example it allows reads for bringing in an + executable page on demand to be given a higher priority over pending write + requests which haven't aged too much on the queue. Potentially this priority + could even be exposed to applications in some manner, providing higher level + tunability. Time based aging avoids starvation of lower priority + requests. Some bits in the bi_opf flags field in the bio structure are + intended to be used for this priority information. + + +1.3 Direct Access to Low level Device/Driver Capabilities (Bypass mode) + (e.g Diagnostics, Systems Management) + +There are situations where high-level code needs to have direct access to +the low level device capabilities or requires the ability to issue commands +to the device bypassing some of the intermediate i/o layers. +These could, for example, be special control commands issued through ioctl +interfaces, or could be raw read/write commands that stress the drive's +capabilities for certain kinds of fitness tests. Having direct interfaces at +multiple levels without having to pass through upper layers makes +it possible to perform bottom up validation of the i/o path, layer by +layer, starting from the media. + +The normal i/o submission interfaces, e.g submit_bio, could be bypassed +for specially crafted requests which such ioctl or diagnostics +interfaces would typically use, and the elevator add_request routine +can instead be used to directly insert such requests in the queue or preferably +the blk_do_rq routine can be used to place the request on the queue and +wait for completion. Alternatively, sometimes the caller might just +invoke a lower level driver specific interface with the request as a +parameter. + +If the request is a means for passing on special information associated with +the command, then such information is associated with the request->special +field (rather than misuse the request->buffer field which is meant for the +request data buffer's virtual mapping). + +For passing request data, the caller must build up a bio descriptor +representing the concerned memory buffer if the underlying driver interprets +bio segments or uses the block layer end*request* functions for i/o +completion. Alternatively one could directly use the request->buffer field to +specify the virtual address of the buffer, if the driver expects buffer +addresses passed in this way and ignores bio entries for the request type +involved. In the latter case, the driver would modify and manage the +request->buffer, request->sector and request->nr_sectors or +request->current_nr_sectors fields itself rather than using the block layer +end_request or end_that_request_first completion interfaces. +(See 2.3 or Documentation/block/request.txt for a brief explanation of +the request structure fields) + +[TBD: end_that_request_last should be usable even in this case; +Perhaps an end_that_direct_request_first routine could be implemented to make +handling direct requests easier for such drivers; Also for drivers that +expect bios, a helper function could be provided for setting up a bio +corresponding to a data buffer] + +<JENS: I dont understand the above, why is end_that_request_first() not +usable? Or _last for that matter. I must be missing something> +<SUP: What I meant here was that if the request doesn't have a bio, then + end_that_request_first doesn't modify nr_sectors or current_nr_sectors, + and hence can't be used for advancing request state settings on the + completion of partial transfers. The driver has to modify these fields + directly by hand. + This is because end_that_request_first only iterates over the bio list, + and always returns 0 if there are none associated with the request. + _last works OK in this case, and is not a problem, as I mentioned earlier +> + +1.3.1 Pre-built Commands + +A request can be created with a pre-built custom command to be sent directly +to the device. The cmd block in the request structure has room for filling +in the command bytes. (i.e rq->cmd is now 16 bytes in size, and meant for +command pre-building, and the type of the request is now indicated +through rq->flags instead of via rq->cmd) + +The request structure flags can be set up to indicate the type of request +in such cases (REQ_PC: direct packet command passed to driver, REQ_BLOCK_PC: +packet command issued via blk_do_rq, REQ_SPECIAL: special request). + +It can help to pre-build device commands for requests in advance. +Drivers can now specify a request prepare function (q->prep_rq_fn) that the +block layer would invoke to pre-build device commands for a given request, +or perform other preparatory processing for the request. This is routine is +called by elv_next_request(), i.e. typically just before servicing a request. +(The prepare function would not be called for requests that have RQF_DONTPREP +enabled) + +Aside: + Pre-building could possibly even be done early, i.e before placing the + request on the queue, rather than construct the command on the fly in the + driver while servicing the request queue when it may affect latencies in + interrupt context or responsiveness in general. One way to add early + pre-building would be to do it whenever we fail to merge on a request. + Now REQ_NOMERGE is set in the request flags to skip this one in the future, + which means that it will not change before we feed it to the device. So + the pre-builder hook can be invoked there. + + +2. Flexible and generic but minimalist i/o structure/descriptor. + +2.1 Reason for a new structure and requirements addressed + +Prior to 2.5, buffer heads were used as the unit of i/o at the generic block +layer, and the low level request structure was associated with a chain of +buffer heads for a contiguous i/o request. This led to certain inefficiencies +when it came to large i/o requests and readv/writev style operations, as it +forced such requests to be broken up into small chunks before being passed +on to the generic block layer, only to be merged by the i/o scheduler +when the underlying device was capable of handling the i/o in one shot. +Also, using the buffer head as an i/o structure for i/os that didn't originate +from the buffer cache unnecessarily added to the weight of the descriptors +which were generated for each such chunk. + +The following were some of the goals and expectations considered in the +redesign of the block i/o data structure in 2.5. + +i. Should be appropriate as a descriptor for both raw and buffered i/o - + avoid cache related fields which are irrelevant in the direct/page i/o path, + or filesystem block size alignment restrictions which may not be relevant + for raw i/o. +ii. Ability to represent high-memory buffers (which do not have a virtual + address mapping in kernel address space). +iii.Ability to represent large i/os w/o unnecessarily breaking them up (i.e + greater than PAGE_SIZE chunks in one shot) +iv. At the same time, ability to retain independent identity of i/os from + different sources or i/o units requiring individual completion (e.g. for + latency reasons) +v. Ability to represent an i/o involving multiple physical memory segments + (including non-page aligned page fragments, as specified via readv/writev) + without unnecessarily breaking it up, if the underlying device is capable of + handling it. +vi. Preferably should be based on a memory descriptor structure that can be + passed around different types of subsystems or layers, maybe even + networking, without duplication or extra copies of data/descriptor fields + themselves in the process +vii.Ability to handle the possibility of splits/merges as the structure passes + through layered drivers (lvm, md, evms), with minimal overhead. + +The solution was to define a new structure (bio) for the block layer, +instead of using the buffer head structure (bh) directly, the idea being +avoidance of some associated baggage and limitations. The bio structure +is uniformly used for all i/o at the block layer ; it forms a part of the +bh structure for buffered i/o, and in the case of raw/direct i/o kiobufs are +mapped to bio structures. + +2.2 The bio struct + +The bio structure uses a vector representation pointing to an array of tuples +of <page, offset, len> to describe the i/o buffer, and has various other +fields describing i/o parameters and state that needs to be maintained for +performing the i/o. + +Notice that this representation means that a bio has no virtual address +mapping at all (unlike buffer heads). + +struct bio_vec { + struct page *bv_page; + unsigned short bv_len; + unsigned short bv_offset; +}; + +/* + * main unit of I/O for the block layer and lower layers (ie drivers) + */ +struct bio { + struct bio *bi_next; /* request queue link */ + struct block_device *bi_bdev; /* target device */ + unsigned long bi_flags; /* status, command, etc */ + unsigned long bi_opf; /* low bits: r/w, high: priority */ + + unsigned int bi_vcnt; /* how may bio_vec's */ + struct bvec_iter bi_iter; /* current index into bio_vec array */ + + unsigned int bi_size; /* total size in bytes */ + unsigned short bi_phys_segments; /* segments after physaddr coalesce*/ + unsigned short bi_hw_segments; /* segments after DMA remapping */ + unsigned int bi_max; /* max bio_vecs we can hold + used as index into pool */ + struct bio_vec *bi_io_vec; /* the actual vec list */ + bio_end_io_t *bi_end_io; /* bi_end_io (bio) */ + atomic_t bi_cnt; /* pin count: free when it hits zero */ + void *bi_private; +}; + +With this multipage bio design: + +- Large i/os can be sent down in one go using a bio_vec list consisting + of an array of <page, offset, len> fragments (similar to the way fragments + are represented in the zero-copy network code) +- Splitting of an i/o request across multiple devices (as in the case of + lvm or raid) is achieved by cloning the bio (where the clone points to + the same bi_io_vec array, but with the index and size accordingly modified) +- A linked list of bios is used as before for unrelated merges (*) - this + avoids reallocs and makes independent completions easier to handle. +- Code that traverses the req list can find all the segments of a bio + by using rq_for_each_segment. This handles the fact that a request + has multiple bios, each of which can have multiple segments. +- Drivers which can't process a large bio in one shot can use the bi_iter + field to keep track of the next bio_vec entry to process. + (e.g a 1MB bio_vec needs to be handled in max 128kB chunks for IDE) + [TBD: Should preferably also have a bi_voffset and bi_vlen to avoid modifying + bi_offset an len fields] + +(*) unrelated merges -- a request ends up containing two or more bios that + didn't originate from the same place. + +bi_end_io() i/o callback gets called on i/o completion of the entire bio. + +At a lower level, drivers build a scatter gather list from the merged bios. +The scatter gather list is in the form of an array of <page, offset, len> +entries with their corresponding dma address mappings filled in at the +appropriate time. As an optimization, contiguous physical pages can be +covered by a single entry where <page> refers to the first page and <len> +covers the range of pages (up to 16 contiguous pages could be covered this +way). There is a helper routine (blk_rq_map_sg) which drivers can use to build +the sg list. + +Note: Right now the only user of bios with more than one page is ll_rw_kio, +which in turn means that only raw I/O uses it (direct i/o may not work +right now). The intent however is to enable clustering of pages etc to +become possible. The pagebuf abstraction layer from SGI also uses multi-page +bios, but that is currently not included in the stock development kernels. +The same is true of Andrew Morton's work-in-progress multipage bio writeout +and readahead patches. + +2.3 Changes in the Request Structure + +The request structure is the structure that gets passed down to low level +drivers. The block layer make_request function builds up a request structure, +places it on the queue and invokes the drivers request_fn. The driver makes +use of block layer helper routine elv_next_request to pull the next request +off the queue. Control or diagnostic functions might bypass block and directly +invoke underlying driver entry points passing in a specially constructed +request structure. + +Only some relevant fields (mainly those which changed or may be referred +to in some of the discussion here) are listed below, not necessarily in +the order in which they occur in the structure (see include/linux/blkdev.h) +Refer to Documentation/block/request.txt for details about all the request +structure fields and a quick reference about the layers which are +supposed to use or modify those fields. + +struct request { + struct list_head queuelist; /* Not meant to be directly accessed by + the driver. + Used by q->elv_next_request_fn + rq->queue is gone + */ + . + . + unsigned char cmd[16]; /* prebuilt command data block */ + unsigned long flags; /* also includes earlier rq->cmd settings */ + . + . + sector_t sector; /* this field is now of type sector_t instead of int + preparation for 64 bit sectors */ + . + . + + /* Number of scatter-gather DMA addr+len pairs after + * physical address coalescing is performed. + */ + unsigned short nr_phys_segments; + + /* Number of scatter-gather addr+len pairs after + * physical and DMA remapping hardware coalescing is performed. + * This is the number of scatter-gather entries the driver + * will actually have to deal with after DMA mapping is done. + */ + unsigned short nr_hw_segments; + + /* Various sector counts */ + unsigned long nr_sectors; /* no. of sectors left: driver modifiable */ + unsigned long hard_nr_sectors; /* block internal copy of above */ + unsigned int current_nr_sectors; /* no. of sectors left in the + current segment:driver modifiable */ + unsigned long hard_cur_sectors; /* block internal copy of the above */ + . + . + int tag; /* command tag associated with request */ + void *special; /* same as before */ + char *buffer; /* valid only for low memory buffers up to + current_nr_sectors */ + . + . + struct bio *bio, *biotail; /* bio list instead of bh */ + struct request_list *rl; +} + +See the req_ops and req_flag_bits definitions for an explanation of the various +flags available. Some bits are used by the block layer or i/o scheduler. + +The behaviour of the various sector counts are almost the same as before, +except that since we have multi-segment bios, current_nr_sectors refers +to the numbers of sectors in the current segment being processed which could +be one of the many segments in the current bio (i.e i/o completion unit). +The nr_sectors value refers to the total number of sectors in the whole +request that remain to be transferred (no change). The purpose of the +hard_xxx values is for block to remember these counts every time it hands +over the request to the driver. These values are updated by block on +end_that_request_first, i.e. every time the driver completes a part of the +transfer and invokes block end*request helpers to mark this. The +driver should not modify these values. The block layer sets up the +nr_sectors and current_nr_sectors fields (based on the corresponding +hard_xxx values and the number of bytes transferred) and updates it on +every transfer that invokes end_that_request_first. It does the same for the +buffer, bio, bio->bi_iter fields too. + +The buffer field is just a virtual address mapping of the current segment +of the i/o buffer in cases where the buffer resides in low-memory. For high +memory i/o, this field is not valid and must not be used by drivers. + +Code that sets up its own request structures and passes them down to +a driver needs to be careful about interoperation with the block layer helper +functions which the driver uses. (Section 1.3) + +3. Using bios + +3.1 Setup/Teardown + +There are routines for managing the allocation, and reference counting, and +freeing of bios (bio_alloc, bio_get, bio_put). + +This makes use of Ingo Molnar's mempool implementation, which enables +subsystems like bio to maintain their own reserve memory pools for guaranteed +deadlock-free allocations during extreme VM load. For example, the VM +subsystem makes use of the block layer to writeout dirty pages in order to be +able to free up memory space, a case which needs careful handling. The +allocation logic draws from the preallocated emergency reserve in situations +where it cannot allocate through normal means. If the pool is empty and it +can wait, then it would trigger action that would help free up memory or +replenish the pool (without deadlocking) and wait for availability in the pool. +If it is in IRQ context, and hence not in a position to do this, allocation +could fail if the pool is empty. In general mempool always first tries to +perform allocation without having to wait, even if it means digging into the +pool as long it is not less that 50% full. + +On a free, memory is released to the pool or directly freed depending on +the current availability in the pool. The mempool interface lets the +subsystem specify the routines to be used for normal alloc and free. In the +case of bio, these routines make use of the standard slab allocator. + +The caller of bio_alloc is expected to taken certain steps to avoid +deadlocks, e.g. avoid trying to allocate more memory from the pool while +already holding memory obtained from the pool. +[TBD: This is a potential issue, though a rare possibility + in the bounce bio allocation that happens in the current code, since + it ends up allocating a second bio from the same pool while + holding the original bio ] + +Memory allocated from the pool should be released back within a limited +amount of time (in the case of bio, that would be after the i/o is completed). +This ensures that if part of the pool has been used up, some work (in this +case i/o) must already be in progress and memory would be available when it +is over. If allocating from multiple pools in the same code path, the order +or hierarchy of allocation needs to be consistent, just the way one deals +with multiple locks. + +The bio_alloc routine also needs to allocate the bio_vec_list (bvec_alloc()) +for a non-clone bio. There are the 6 pools setup for different size biovecs, +so bio_alloc(gfp_mask, nr_iovecs) will allocate a vec_list of the +given size from these slabs. + +The bio_get() routine may be used to hold an extra reference on a bio prior +to i/o submission, if the bio fields are likely to be accessed after the +i/o is issued (since the bio may otherwise get freed in case i/o completion +happens in the meantime). + +The bio_clone_fast() routine may be used to duplicate a bio, where the clone +shares the bio_vec_list with the original bio (i.e. both point to the +same bio_vec_list). This would typically be used for splitting i/o requests +in lvm or md. + +3.2 Generic bio helper Routines + +3.2.1 Traversing segments and completion units in a request + +The macro rq_for_each_segment() should be used for traversing the bios +in the request list (drivers should avoid directly trying to do it +themselves). Using these helpers should also make it easier to cope +with block changes in the future. + + struct req_iterator iter; + rq_for_each_segment(bio_vec, rq, iter) + /* bio_vec is now current segment */ + +I/O completion callbacks are per-bio rather than per-segment, so drivers +that traverse bio chains on completion need to keep that in mind. Drivers +which don't make a distinction between segments and completion units would +need to be reorganized to support multi-segment bios. + +3.2.2 Setting up DMA scatterlists + +The blk_rq_map_sg() helper routine would be used for setting up scatter +gather lists from a request, so a driver need not do it on its own. + + nr_segments = blk_rq_map_sg(q, rq, scatterlist); + +The helper routine provides a level of abstraction which makes it easier +to modify the internals of request to scatterlist conversion down the line +without breaking drivers. The blk_rq_map_sg routine takes care of several +things like collapsing physically contiguous segments (if QUEUE_FLAG_CLUSTER +is set) and correct segment accounting to avoid exceeding the limits which +the i/o hardware can handle, based on various queue properties. + +- Prevents a clustered segment from crossing a 4GB mem boundary +- Avoids building segments that would exceed the number of physical + memory segments that the driver can handle (phys_segments) and the + number that the underlying hardware can handle at once, accounting for + DMA remapping (hw_segments) (i.e. IOMMU aware limits). + +Routines which the low level driver can use to set up the segment limits: + +blk_queue_max_hw_segments() : Sets an upper limit of the maximum number of +hw data segments in a request (i.e. the maximum number of address/length +pairs the host adapter can actually hand to the device at once) + +blk_queue_max_phys_segments() : Sets an upper limit on the maximum number +of physical data segments in a request (i.e. the largest sized scatter list +a driver could handle) + +3.2.3 I/O completion + +The existing generic block layer helper routines end_request, +end_that_request_first and end_that_request_last can be used for i/o +completion (and setting things up so the rest of the i/o or the next +request can be kicked of) as before. With the introduction of multi-page +bio support, end_that_request_first requires an additional argument indicating +the number of sectors completed. + +3.2.4 Implications for drivers that do not interpret bios (don't handle + multiple segments) + +Drivers that do not interpret bios e.g those which do not handle multiple +segments and do not support i/o into high memory addresses (require bounce +buffers) and expect only virtually mapped buffers, can access the rq->buffer +field. As before the driver should use current_nr_sectors to determine the +size of remaining data in the current segment (that is the maximum it can +transfer in one go unless it interprets segments), and rely on the block layer +end_request, or end_that_request_first/last to take care of all accounting +and transparent mapping of the next bio segment when a segment boundary +is crossed on completion of a transfer. (The end*request* functions should +be used if only if the request has come down from block/bio path, not for +direct access requests which only specify rq->buffer without a valid rq->bio) + +3.2.5 Generic request command tagging + +3.2.5.1 Tag helpers + +Block now offers some simple generic functionality to help support command +queueing (typically known as tagged command queueing), ie manage more than +one outstanding command on a queue at any given time. + + blk_queue_init_tags(struct request_queue *q, int depth) + + Initialize internal command tagging structures for a maximum + depth of 'depth'. + + blk_queue_free_tags((struct request_queue *q) + + Teardown tag info associated with the queue. This will be done + automatically by block if blk_queue_cleanup() is called on a queue + that is using tagging. + +The above are initialization and exit management, the main helpers during +normal operations are: + + blk_queue_start_tag(struct request_queue *q, struct request *rq) + + Start tagged operation for this request. A free tag number between + 0 and 'depth' is assigned to the request (rq->tag holds this number), + and 'rq' is added to the internal tag management. If the maximum depth + for this queue is already achieved (or if the tag wasn't started for + some other reason), 1 is returned. Otherwise 0 is returned. + + blk_queue_end_tag(struct request_queue *q, struct request *rq) + + End tagged operation on this request. 'rq' is removed from the internal + book keeping structures. + +To minimize struct request and queue overhead, the tag helpers utilize some +of the same request members that are used for normal request queue management. +This means that a request cannot both be an active tag and be on the queue +list at the same time. blk_queue_start_tag() will remove the request, but +the driver must remember to call blk_queue_end_tag() before signalling +completion of the request to the block layer. This means ending tag +operations before calling end_that_request_last()! For an example of a user +of these helpers, see the IDE tagged command queueing support. + +3.2.5.2 Tag info + +Some block functions exist to query current tag status or to go from a +tag number to the associated request. These are, in no particular order: + + blk_queue_tagged(q) + + Returns 1 if the queue 'q' is using tagging, 0 if not. + + blk_queue_tag_request(q, tag) + + Returns a pointer to the request associated with tag 'tag'. + + blk_queue_tag_depth(q) + + Return current queue depth. + + blk_queue_tag_queue(q) + + Returns 1 if the queue can accept a new queued command, 0 if we are + at the maximum depth already. + + blk_queue_rq_tagged(rq) + + Returns 1 if the request 'rq' is tagged. + +3.2.5.2 Internal structure + +Internally, block manages tags in the blk_queue_tag structure: + + struct blk_queue_tag { + struct request **tag_index; /* array or pointers to rq */ + unsigned long *tag_map; /* bitmap of free tags */ + struct list_head busy_list; /* fifo list of busy tags */ + int busy; /* queue depth */ + int max_depth; /* max queue depth */ + }; + +Most of the above is simple and straight forward, however busy_list may need +a bit of explaining. Normally we don't care too much about request ordering, +but in the event of any barrier requests in the tag queue we need to ensure +that requests are restarted in the order they were queue. + +3.3 I/O Submission + +The routine submit_bio() is used to submit a single io. Higher level i/o +routines make use of this: + +(a) Buffered i/o: +The routine submit_bh() invokes submit_bio() on a bio corresponding to the +bh, allocating the bio if required. ll_rw_block() uses submit_bh() as before. + +(b) Kiobuf i/o (for raw/direct i/o): +The ll_rw_kio() routine breaks up the kiobuf into page sized chunks and +maps the array to one or more multi-page bios, issuing submit_bio() to +perform the i/o on each of these. + +The embedded bh array in the kiobuf structure has been removed and no +preallocation of bios is done for kiobufs. [The intent is to remove the +blocks array as well, but it's currently in there to kludge around direct i/o.] +Thus kiobuf allocation has switched back to using kmalloc rather than vmalloc. + +Todo/Observation: + + A single kiobuf structure is assumed to correspond to a contiguous range + of data, so brw_kiovec() invokes ll_rw_kio for each kiobuf in a kiovec. + So right now it wouldn't work for direct i/o on non-contiguous blocks. + This is to be resolved. The eventual direction is to replace kiobuf + by kvec's. + + Badari Pulavarty has a patch to implement direct i/o correctly using + bio and kvec. + + +(c) Page i/o: +Todo/Under discussion: + + Andrew Morton's multi-page bio patches attempt to issue multi-page + writeouts (and reads) from the page cache, by directly building up + large bios for submission completely bypassing the usage of buffer + heads. This work is still in progress. + + Christoph Hellwig had some code that uses bios for page-io (rather than + bh). This isn't included in bio as yet. Christoph was also working on a + design for representing virtual/real extents as an entity and modifying + some of the address space ops interfaces to utilize this abstraction rather + than buffer_heads. (This is somewhat along the lines of the SGI XFS pagebuf + abstraction, but intended to be as lightweight as possible). + +(d) Direct access i/o: +Direct access requests that do not contain bios would be submitted differently +as discussed earlier in section 1.3. + +Aside: + + Kvec i/o: + + Ben LaHaise's aio code uses a slightly different structure instead + of kiobufs, called a kvec_cb. This contains an array of <page, offset, len> + tuples (very much like the networking code), together with a callback function + and data pointer. This is embedded into a brw_cb structure when passed + to brw_kvec_async(). + + Now it should be possible to directly map these kvecs to a bio. Just as while + cloning, in this case rather than PRE_BUILT bio_vecs, we set the bi_io_vec + array pointer to point to the veclet array in kvecs. + + TBD: In order for this to work, some changes are needed in the way multi-page + bios are handled today. The values of the tuples in such a vector passed in + from higher level code should not be modified by the block layer in the course + of its request processing, since that would make it hard for the higher layer + to continue to use the vector descriptor (kvec) after i/o completes. Instead, + all such transient state should either be maintained in the request structure, + and passed on in some way to the endio completion routine. + + +4. The I/O scheduler +I/O scheduler, a.k.a. elevator, is implemented in two layers. Generic dispatch +queue and specific I/O schedulers. Unless stated otherwise, elevator is used +to refer to both parts and I/O scheduler to specific I/O schedulers. + +Block layer implements generic dispatch queue in block/*.c. +The generic dispatch queue is responsible for requeueing, handling non-fs +requests and all other subtleties. + +Specific I/O schedulers are responsible for ordering normal filesystem +requests. They can also choose to delay certain requests to improve +throughput or whatever purpose. As the plural form indicates, there are +multiple I/O schedulers. They can be built as modules but at least one should +be built inside the kernel. Each queue can choose different one and can also +change to another one dynamically. + +A block layer call to the i/o scheduler follows the convention elv_xxx(). This +calls elevator_xxx_fn in the elevator switch (block/elevator.c). Oh, xxx +and xxx might not match exactly, but use your imagination. If an elevator +doesn't implement a function, the switch does nothing or some minimal house +keeping work. + +4.1. I/O scheduler API + +The functions an elevator may implement are: (* are mandatory) +elevator_merge_fn called to query requests for merge with a bio + +elevator_merge_req_fn called when two requests get merged. the one + which gets merged into the other one will be + never seen by I/O scheduler again. IOW, after + being merged, the request is gone. + +elevator_merged_fn called when a request in the scheduler has been + involved in a merge. It is used in the deadline + scheduler for example, to reposition the request + if its sorting order has changed. + +elevator_allow_merge_fn called whenever the block layer determines + that a bio can be merged into an existing + request safely. The io scheduler may still + want to stop a merge at this point if it + results in some sort of conflict internally, + this hook allows it to do that. Note however + that two *requests* can still be merged at later + time. Currently the io scheduler has no way to + prevent that. It can only learn about the fact + from elevator_merge_req_fn callback. + +elevator_dispatch_fn* fills the dispatch queue with ready requests. + I/O schedulers are free to postpone requests by + not filling the dispatch queue unless @force + is non-zero. Once dispatched, I/O schedulers + are not allowed to manipulate the requests - + they belong to generic dispatch queue. + +elevator_add_req_fn* called to add a new request into the scheduler + +elevator_former_req_fn +elevator_latter_req_fn These return the request before or after the + one specified in disk sort order. Used by the + block layer to find merge possibilities. + +elevator_completed_req_fn called when a request is completed. + +elevator_may_queue_fn returns true if the scheduler wants to allow the + current context to queue a new request even if + it is over the queue limit. This must be used + very carefully!! + +elevator_set_req_fn +elevator_put_req_fn Must be used to allocate and free any elevator + specific storage for a request. + +elevator_activate_req_fn Called when device driver first sees a request. + I/O schedulers can use this callback to + determine when actual execution of a request + starts. +elevator_deactivate_req_fn Called when device driver decides to delay + a request by requeueing it. + +elevator_init_fn* +elevator_exit_fn Allocate and free any elevator specific storage + for a queue. + +4.2 Request flows seen by I/O schedulers +All requests seen by I/O schedulers strictly follow one of the following three +flows. + + set_req_fn -> + + i. add_req_fn -> (merged_fn ->)* -> dispatch_fn -> activate_req_fn -> + (deactivate_req_fn -> activate_req_fn ->)* -> completed_req_fn + ii. add_req_fn -> (merged_fn ->)* -> merge_req_fn + iii. [none] + + -> put_req_fn + +4.3 I/O scheduler implementation +The generic i/o scheduler algorithm attempts to sort/merge/batch requests for +optimal disk scan and request servicing performance (based on generic +principles and device capabilities), optimized for: +i. improved throughput +ii. improved latency +iii. better utilization of h/w & CPU time + +Characteristics: + +i. Binary tree +AS and deadline i/o schedulers use red black binary trees for disk position +sorting and searching, and a fifo linked list for time-based searching. This +gives good scalability and good availability of information. Requests are +almost always dispatched in disk sort order, so a cache is kept of the next +request in sort order to prevent binary tree lookups. + +This arrangement is not a generic block layer characteristic however, so +elevators may implement queues as they please. + +ii. Merge hash +AS and deadline use a hash table indexed by the last sector of a request. This +enables merging code to quickly look up "back merge" candidates, even when +multiple I/O streams are being performed at once on one disk. + +"Front merges", a new request being merged at the front of an existing request, +are far less common than "back merges" due to the nature of most I/O patterns. +Front merges are handled by the binary trees in AS and deadline schedulers. + +iii. Plugging the queue to batch requests in anticipation of opportunities for + merge/sort optimizations + +Plugging is an approach that the current i/o scheduling algorithm resorts to so +that it collects up enough requests in the queue to be able to take +advantage of the sorting/merging logic in the elevator. If the +queue is empty when a request comes in, then it plugs the request queue +(sort of like plugging the bath tub of a vessel to get fluid to build up) +till it fills up with a few more requests, before starting to service +the requests. This provides an opportunity to merge/sort the requests before +passing them down to the device. There are various conditions when the queue is +unplugged (to open up the flow again), either through a scheduled task or +could be on demand. For example wait_on_buffer sets the unplugging going +through sync_buffer() running blk_run_address_space(mapping). Or the caller +can do it explicity through blk_unplug(bdev). So in the read case, +the queue gets explicitly unplugged as part of waiting for completion on that +buffer. + +Aside: + This is kind of controversial territory, as it's not clear if plugging is + always the right thing to do. Devices typically have their own queues, + and allowing a big queue to build up in software, while letting the device be + idle for a while may not always make sense. The trick is to handle the fine + balance between when to plug and when to open up. Also now that we have + multi-page bios being queued in one shot, we may not need to wait to merge + a big request from the broken up pieces coming by. + +4.4 I/O contexts +I/O contexts provide a dynamically allocated per process data area. They may +be used in I/O schedulers, and in the block layer (could be used for IO statis, +priorities for example). See *io_context in block/ll_rw_blk.c, and as-iosched.c +for an example of usage in an i/o scheduler. + + +5. Scalability related changes + +5.1 Granular Locking: io_request_lock replaced by a per-queue lock + +The global io_request_lock has been removed as of 2.5, to avoid +the scalability bottleneck it was causing, and has been replaced by more +granular locking. The request queue structure has a pointer to the +lock to be used for that queue. As a result, locking can now be +per-queue, with a provision for sharing a lock across queues if +necessary (e.g the scsi layer sets the queue lock pointers to the +corresponding adapter lock, which results in a per host locking +granularity). The locking semantics are the same, i.e. locking is +still imposed by the block layer, grabbing the lock before +request_fn execution which it means that lots of older drivers +should still be SMP safe. Drivers are free to drop the queue +lock themselves, if required. Drivers that explicitly used the +io_request_lock for serialization need to be modified accordingly. +Usually it's as easy as adding a global lock: + + static DEFINE_SPINLOCK(my_driver_lock); + +and passing the address to that lock to blk_init_queue(). + +5.2 64 bit sector numbers (sector_t prepares for 64 bit support) + +The sector number used in the bio structure has been changed to sector_t, +which could be defined as 64 bit in preparation for 64 bit sector support. + +6. Other Changes/Implications + +6.1 Partition re-mapping handled by the generic block layer + +In 2.5 some of the gendisk/partition related code has been reorganized. +Now the generic block layer performs partition-remapping early and thus +provides drivers with a sector number relative to whole device, rather than +having to take partition number into account in order to arrive at the true +sector number. The routine blk_partition_remap() is invoked by +generic_make_request even before invoking the queue specific make_request_fn, +so the i/o scheduler also gets to operate on whole disk sector numbers. This +should typically not require changes to block drivers, it just never gets +to invoke its own partition sector offset calculations since all bios +sent are offset from the beginning of the device. + + +7. A Few Tips on Migration of older drivers + +Old-style drivers that just use CURRENT and ignores clustered requests, +may not need much change. The generic layer will automatically handle +clustered requests, multi-page bios, etc for the driver. + +For a low performance driver or hardware that is PIO driven or just doesn't +support scatter-gather changes should be minimal too. + +The following are some points to keep in mind when converting old drivers +to bio. + +Drivers should use elv_next_request to pick up requests and are no longer +supposed to handle looping directly over the request list. +(struct request->queue has been removed) + +Now end_that_request_first takes an additional number_of_sectors argument. +It used to handle always just the first buffer_head in a request, now +it will loop and handle as many sectors (on a bio-segment granularity) +as specified. + +Now bh->b_end_io is replaced by bio->bi_end_io, but most of the time the +right thing to use is bio_endio(bio) instead. + +If the driver is dropping the io_request_lock from its request_fn strategy, +then it just needs to replace that with q->queue_lock instead. + +As described in Sec 1.1, drivers can set max sector size, max segment size +etc per queue now. Drivers that used to define their own merge functions i +to handle things like this can now just use the blk_queue_* functions at +blk_init_queue time. + +Drivers no longer have to map a {partition, sector offset} into the +correct absolute location anymore, this is done by the block layer, so +where a driver received a request ala this before: + + rq->rq_dev = mk_kdev(3, 5); /* /dev/hda5 */ + rq->sector = 0; /* first sector on hda5 */ + + it will now see + + rq->rq_dev = mk_kdev(3, 0); /* /dev/hda */ + rq->sector = 123128; /* offset from start of disk */ + +As mentioned, there is no virtual mapping of a bio. For DMA, this is +not a problem as the driver probably never will need a virtual mapping. +Instead it needs a bus mapping (dma_map_page for a single segment or +use dma_map_sg for scatter gather) to be able to ship it to the driver. For +PIO drivers (or drivers that need to revert to PIO transfer once in a +while (IDE for example)), where the CPU is doing the actual data +transfer a virtual mapping is needed. If the driver supports highmem I/O, +(Sec 1.1, (ii) ) it needs to use kmap_atomic or similar to temporarily map +a bio into the virtual address space. + + +8. Prior/Related/Impacted patches + +8.1. Earlier kiobuf patches (sct/axboe/chait/hch/mkp) +- orig kiobuf & raw i/o patches (now in 2.4 tree) +- direct kiobuf based i/o to devices (no intermediate bh's) +- page i/o using kiobuf +- kiobuf splitting for lvm (mkp) +- elevator support for kiobuf request merging (axboe) +8.2. Zero-copy networking (Dave Miller) +8.3. SGI XFS - pagebuf patches - use of kiobufs +8.4. Multi-page pioent patch for bio (Christoph Hellwig) +8.5. Direct i/o implementation (Andrea Arcangeli) since 2.4.10-pre11 +8.6. Async i/o implementation patch (Ben LaHaise) +8.7. EVMS layering design (IBM EVMS team) +8.8. Larger page cache size patch (Ben LaHaise) and + Large page size (Daniel Phillips) + => larger contiguous physical memory buffers +8.9. VM reservations patch (Ben LaHaise) +8.10. Write clustering patches ? (Marcelo/Quintela/Riel ?) +8.11. Block device in page cache patch (Andrea Archangeli) - now in 2.4.10+ +8.12. Multiple block-size transfers for faster raw i/o (Shailabh Nagar, + Badari) +8.13 Priority based i/o scheduler - prepatches (Arjan van de Ven) +8.14 IDE Taskfile i/o patch (Andre Hedrick) +8.15 Multi-page writeout and readahead patches (Andrew Morton) +8.16 Direct i/o patches for 2.5 using kvec and bio (Badari Pulavarthy) + +9. Other References: + +9.1 The Splice I/O Model - Larry McVoy (and subsequent discussions on lkml, +and Linus' comments - Jan 2001) +9.2 Discussions about kiobuf and bh design on lkml between sct, linus, alan +et al - Feb-March 2001 (many of the initial thoughts that led to bio were +brought up in this discussion thread) +9.3 Discussions on mempool on lkml - Dec 2001. + diff --git a/Documentation/block/biovecs.txt b/Documentation/block/biovecs.txt new file mode 100644 index 000000000..25689584e --- /dev/null +++ b/Documentation/block/biovecs.txt @@ -0,0 +1,119 @@ + +Immutable biovecs and biovec iterators: +======================================= + +Kent Overstreet <kmo@daterainc.com> + +As of 3.13, biovecs should never be modified after a bio has been submitted. +Instead, we have a new struct bvec_iter which represents a range of a biovec - +the iterator will be modified as the bio is completed, not the biovec. + +More specifically, old code that needed to partially complete a bio would +update bi_sector and bi_size, and advance bi_idx to the next biovec. If it +ended up partway through a biovec, it would increment bv_offset and decrement +bv_len by the number of bytes completed in that biovec. + +In the new scheme of things, everything that must be mutated in order to +partially complete a bio is segregated into struct bvec_iter: bi_sector, +bi_size and bi_idx have been moved there; and instead of modifying bv_offset +and bv_len, struct bvec_iter has bi_bvec_done, which represents the number of +bytes completed in the current bvec. + +There are a bunch of new helper macros for hiding the gory details - in +particular, presenting the illusion of partially completed biovecs so that +normal code doesn't have to deal with bi_bvec_done. + + * Driver code should no longer refer to biovecs directly; we now have + bio_iovec() and bio_iter_iovec() macros that return literal struct biovecs, + constructed from the raw biovecs but taking into account bi_bvec_done and + bi_size. + + bio_for_each_segment() has been updated to take a bvec_iter argument + instead of an integer (that corresponded to bi_idx); for a lot of code the + conversion just required changing the types of the arguments to + bio_for_each_segment(). + + * Advancing a bvec_iter is done with bio_advance_iter(); bio_advance() is a + wrapper around bio_advance_iter() that operates on bio->bi_iter, and also + advances the bio integrity's iter if present. + + There is a lower level advance function - bvec_iter_advance() - which takes + a pointer to a biovec, not a bio; this is used by the bio integrity code. + +What's all this get us? +======================= + +Having a real iterator, and making biovecs immutable, has a number of +advantages: + + * Before, iterating over bios was very awkward when you weren't processing + exactly one bvec at a time - for example, bio_copy_data() in fs/bio.c, + which copies the contents of one bio into another. Because the biovecs + wouldn't necessarily be the same size, the old code was tricky convoluted - + it had to walk two different bios at the same time, keeping both bi_idx and + and offset into the current biovec for each. + + The new code is much more straightforward - have a look. This sort of + pattern comes up in a lot of places; a lot of drivers were essentially open + coding bvec iterators before, and having common implementation considerably + simplifies a lot of code. + + * Before, any code that might need to use the biovec after the bio had been + completed (perhaps to copy the data somewhere else, or perhaps to resubmit + it somewhere else if there was an error) had to save the entire bvec array + - again, this was being done in a fair number of places. + + * Biovecs can be shared between multiple bios - a bvec iter can represent an + arbitrary range of an existing biovec, both starting and ending midway + through biovecs. This is what enables efficient splitting of arbitrary + bios. Note that this means we _only_ use bi_size to determine when we've + reached the end of a bio, not bi_vcnt - and the bio_iovec() macro takes + bi_size into account when constructing biovecs. + + * Splitting bios is now much simpler. The old bio_split() didn't even work on + bios with more than a single bvec! Now, we can efficiently split arbitrary + size bios - because the new bio can share the old bio's biovec. + + Care must be taken to ensure the biovec isn't freed while the split bio is + still using it, in case the original bio completes first, though. Using + bio_chain() when splitting bios helps with this. + + * Submitting partially completed bios is now perfectly fine - this comes up + occasionally in stacking block drivers and various code (e.g. md and + bcache) had some ugly workarounds for this. + + It used to be the case that submitting a partially completed bio would work + fine to _most_ devices, but since accessing the raw bvec array was the + norm, not all drivers would respect bi_idx and those would break. Now, + since all drivers _must_ go through the bvec iterator - and have been + audited to make sure they are - submitting partially completed bios is + perfectly fine. + +Other implications: +=================== + + * Almost all usage of bi_idx is now incorrect and has been removed; instead, + where previously you would have used bi_idx you'd now use a bvec_iter, + probably passing it to one of the helper macros. + + I.e. instead of using bio_iovec_idx() (or bio->bi_iovec[bio->bi_idx]), you + now use bio_iter_iovec(), which takes a bvec_iter and returns a + literal struct bio_vec - constructed on the fly from the raw biovec but + taking into account bi_bvec_done (and bi_size). + + * bi_vcnt can't be trusted or relied upon by driver code - i.e. anything that + doesn't actually own the bio. The reason is twofold: firstly, it's not + actually needed for iterating over the bio anymore - we only use bi_size. + Secondly, when cloning a bio and reusing (a portion of) the original bio's + biovec, in order to calculate bi_vcnt for the new bio we'd have to iterate + over all the biovecs in the new bio - which is silly as it's not needed. + + So, don't use bi_vcnt anymore. + + * The current interface allows the block layer to split bios as needed, so we + could eliminate a lot of complexity particularly in stacked drivers. Code + that creates bios can then create whatever size bios are convenient, and + more importantly stacked drivers don't have to deal with both their own bio + size limitations and the limitations of the underlying devices. Thus + there's no need to define ->merge_bvec_fn() callbacks for individual block + drivers. diff --git a/Documentation/block/capability.txt b/Documentation/block/capability.txt new file mode 100644 index 000000000..2f1729424 --- /dev/null +++ b/Documentation/block/capability.txt @@ -0,0 +1,15 @@ +Generic Block Device Capability +=============================================================================== +This file documents the sysfs file block/<disk>/capability + +capability is a hex word indicating which capabilities a specific disk +supports. For more information on bits not listed here, see +include/linux/genhd.h + +Capability Value +------------------------------------------------------------------------------- +GENHD_FL_MEDIA_CHANGE_NOTIFY 4 + When this bit is set, the disk supports Asynchronous Notification + of media change events. These events will be broadcast to user + space via kernel uevent. + diff --git a/Documentation/block/cfq-iosched.txt b/Documentation/block/cfq-iosched.txt new file mode 100644 index 000000000..895bd3813 --- /dev/null +++ b/Documentation/block/cfq-iosched.txt @@ -0,0 +1,291 @@ +CFQ (Complete Fairness Queueing) +=============================== + +The main aim of CFQ scheduler is to provide a fair allocation of the disk +I/O bandwidth for all the processes which requests an I/O operation. + +CFQ maintains the per process queue for the processes which request I/O +operation(synchronous requests). In case of asynchronous requests, all the +requests from all the processes are batched together according to their +process's I/O priority. + +CFQ ioscheduler tunables +======================== + +slice_idle +---------- +This specifies how long CFQ should idle for next request on certain cfq queues +(for sequential workloads) and service trees (for random workloads) before +queue is expired and CFQ selects next queue to dispatch from. + +By default slice_idle is a non-zero value. That means by default we idle on +queues/service trees. This can be very helpful on highly seeky media like +single spindle SATA/SAS disks where we can cut down on overall number of +seeks and see improved throughput. + +Setting slice_idle to 0 will remove all the idling on queues/service tree +level and one should see an overall improved throughput on faster storage +devices like multiple SATA/SAS disks in hardware RAID configuration. The down +side is that isolation provided from WRITES also goes down and notion of +IO priority becomes weaker. + +So depending on storage and workload, it might be useful to set slice_idle=0. +In general I think for SATA/SAS disks and software RAID of SATA/SAS disks +keeping slice_idle enabled should be useful. For any configurations where +there are multiple spindles behind single LUN (Host based hardware RAID +controller or for storage arrays), setting slice_idle=0 might end up in better +throughput and acceptable latencies. + +back_seek_max +------------- +This specifies, given in Kbytes, the maximum "distance" for backward seeking. +The distance is the amount of space from the current head location to the +sectors that are backward in terms of distance. + +This parameter allows the scheduler to anticipate requests in the "backward" +direction and consider them as being the "next" if they are within this +distance from the current head location. + +back_seek_penalty +----------------- +This parameter is used to compute the cost of backward seeking. If the +backward distance of request is just 1/back_seek_penalty from a "front" +request, then the seeking cost of two requests is considered equivalent. + +So scheduler will not bias toward one or the other request (otherwise scheduler +will bias toward front request). Default value of back_seek_penalty is 2. + +fifo_expire_async +----------------- +This parameter is used to set the timeout of asynchronous requests. Default +value of this is 248ms. + +fifo_expire_sync +---------------- +This parameter is used to set the timeout of synchronous requests. Default +value of this is 124ms. In case to favor synchronous requests over asynchronous +one, this value should be decreased relative to fifo_expire_async. + +group_idle +----------- +This parameter forces idling at the CFQ group level instead of CFQ +queue level. This was introduced after a bottleneck was observed +in higher end storage due to idle on sequential queue and allow dispatch +from a single queue. The idea with this parameter is that it can be run with +slice_idle=0 and group_idle=8, so that idling does not happen on individual +queues in the group but happens overall on the group and thus still keeps the +IO controller working. +Not idling on individual queues in the group will dispatch requests from +multiple queues in the group at the same time and achieve higher throughput +on higher end storage. + +Default value for this parameter is 8ms. + +low_latency +----------- +This parameter is used to enable/disable the low latency mode of the CFQ +scheduler. If enabled, CFQ tries to recompute the slice time for each process +based on the target_latency set for the system. This favors fairness over +throughput. Disabling low latency (setting it to 0) ignores target latency, +allowing each process in the system to get a full time slice. + +By default low latency mode is enabled. + +target_latency +-------------- +This parameter is used to calculate the time slice for a process if cfq's +latency mode is enabled. It will ensure that sync requests have an estimated +latency. But if sequential workload is higher(e.g. sequential read), +then to meet the latency constraints, throughput may decrease because of less +time for each process to issue I/O request before the cfq queue is switched. + +Though this can be overcome by disabling the latency_mode, it may increase +the read latency for some applications. This parameter allows for changing +target_latency through the sysfs interface which can provide the balanced +throughput and read latency. + +Default value for target_latency is 300ms. + +slice_async +----------- +This parameter is same as of slice_sync but for asynchronous queue. The +default value is 40ms. + +slice_async_rq +-------------- +This parameter is used to limit the dispatching of asynchronous request to +device request queue in queue's slice time. The maximum number of request that +are allowed to be dispatched also depends upon the io priority. Default value +for this is 2. + +slice_sync +---------- +When a queue is selected for execution, the queues IO requests are only +executed for a certain amount of time(time_slice) before switching to another +queue. This parameter is used to calculate the time slice of synchronous +queue. + +time_slice is computed using the below equation:- +time_slice = slice_sync + (slice_sync/5 * (4 - prio)). To increase the +time_slice of synchronous queue, increase the value of slice_sync. Default +value is 100ms. + +quantum +------- +This specifies the number of request dispatched to the device queue. In a +queue's time slice, a request will not be dispatched if the number of request +in the device exceeds this parameter. This parameter is used for synchronous +request. + +In case of storage with several disk, this setting can limit the parallel +processing of request. Therefore, increasing the value can improve the +performance although this can cause the latency of some I/O to increase due +to more number of requests. + +CFQ Group scheduling +==================== + +CFQ supports blkio cgroup and has "blkio." prefixed files in each +blkio cgroup directory. It is weight-based and there are four knobs +for configuration - weight[_device] and leaf_weight[_device]. +Internal cgroup nodes (the ones with children) can also have tasks in +them, so the former two configure how much proportion the cgroup as a +whole is entitled to at its parent's level while the latter two +configure how much proportion the tasks in the cgroup have compared to +its direct children. + +Another way to think about it is assuming that each internal node has +an implicit leaf child node which hosts all the tasks whose weight is +configured by leaf_weight[_device]. Let's assume a blkio hierarchy +composed of five cgroups - root, A, B, AA and AB - with the following +weights where the names represent the hierarchy. + + weight leaf_weight + root : 125 125 + A : 500 750 + B : 250 500 + AA : 500 500 + AB : 1000 500 + +root never has a parent making its weight is meaningless. For backward +compatibility, weight is always kept in sync with leaf_weight. B, AA +and AB have no child and thus its tasks have no children cgroup to +compete with. They always get 100% of what the cgroup won at the +parent level. Considering only the weights which matter, the hierarchy +looks like the following. + + root + / | \ + A B leaf + 500 250 125 + / | \ + AA AB leaf + 500 1000 750 + +If all cgroups have active IOs and competing with each other, disk +time will be distributed like the following. + +Distribution below root. The total active weight at this level is +A:500 + B:250 + C:125 = 875. + + root-leaf : 125 / 875 =~ 14% + A : 500 / 875 =~ 57% + B(-leaf) : 250 / 875 =~ 28% + +A has children and further distributes its 57% among the children and +the implicit leaf node. The total active weight at this level is +AA:500 + AB:1000 + A-leaf:750 = 2250. + + A-leaf : ( 750 / 2250) * A =~ 19% + AA(-leaf) : ( 500 / 2250) * A =~ 12% + AB(-leaf) : (1000 / 2250) * A =~ 25% + +CFQ IOPS Mode for group scheduling +=================================== +Basic CFQ design is to provide priority based time slices. Higher priority +process gets bigger time slice and lower priority process gets smaller time +slice. Measuring time becomes harder if storage is fast and supports NCQ and +it would be better to dispatch multiple requests from multiple cfq queues in +request queue at a time. In such scenario, it is not possible to measure time +consumed by single queue accurately. + +What is possible though is to measure number of requests dispatched from a +single queue and also allow dispatch from multiple cfq queue at the same time. +This effectively becomes the fairness in terms of IOPS (IO operations per +second). + +If one sets slice_idle=0 and if storage supports NCQ, CFQ internally switches +to IOPS mode and starts providing fairness in terms of number of requests +dispatched. Note that this mode switching takes effect only for group +scheduling. For non-cgroup users nothing should change. + +CFQ IO scheduler Idling Theory +=============================== +Idling on a queue is primarily about waiting for the next request to come +on same queue after completion of a request. In this process CFQ will not +dispatch requests from other cfq queues even if requests are pending there. + +The rationale behind idling is that it can cut down on number of seeks +on rotational media. For example, if a process is doing dependent +sequential reads (next read will come on only after completion of previous +one), then not dispatching request from other queue should help as we +did not move the disk head and kept on dispatching sequential IO from +one queue. + +CFQ has following service trees and various queues are put on these trees. + + sync-idle sync-noidle async + +All cfq queues doing synchronous sequential IO go on to sync-idle tree. +On this tree we idle on each queue individually. + +All synchronous non-sequential queues go on sync-noidle tree. Also any +synchronous write request which is not marked with REQ_IDLE goes on this +service tree. On this tree we do not idle on individual queues instead idle +on the whole group of queues or the tree. So if there are 4 queues waiting +for IO to dispatch we will idle only once last queue has dispatched the IO +and there is no more IO on this service tree. + +All async writes go on async service tree. There is no idling on async +queues. + +CFQ has some optimizations for SSDs and if it detects a non-rotational +media which can support higher queue depth (multiple requests at in +flight at a time), then it cuts down on idling of individual queues and +all the queues move to sync-noidle tree and only tree idle remains. This +tree idling provides isolation with buffered write queues on async tree. + +FAQ +=== +Q1. Why to idle at all on queues not marked with REQ_IDLE. + +A1. We only do tree idle (all queues on sync-noidle tree) on queues not marked + with REQ_IDLE. This helps in providing isolation with all the sync-idle + queues. Otherwise in presence of many sequential readers, other + synchronous IO might not get fair share of disk. + + For example, if there are 10 sequential readers doing IO and they get + 100ms each. If a !REQ_IDLE request comes in, it will be scheduled + roughly after 1 second. If after completion of !REQ_IDLE request we + do not idle, and after a couple of milli seconds a another !REQ_IDLE + request comes in, again it will be scheduled after 1second. Repeat it + and notice how a workload can lose its disk share and suffer due to + multiple sequential readers. + + fsync can generate dependent IO where bunch of data is written in the + context of fsync, and later some journaling data is written. Journaling + data comes in only after fsync has finished its IO (atleast for ext4 + that seemed to be the case). Now if one decides not to idle on fsync + thread due to !REQ_IDLE, then next journaling write will not get + scheduled for another second. A process doing small fsync, will suffer + badly in presence of multiple sequential readers. + + Hence doing tree idling on threads using !REQ_IDLE flag on requests + provides isolation from multiple sequential readers and at the same + time we do not idle on individual threads. + +Q2. When to specify REQ_IDLE +A2. I would think whenever one is doing synchronous write and expecting + more writes to be dispatched from same context soon, should be able + to specify REQ_IDLE on writes and that probably should work well for + most of the cases. diff --git a/Documentation/block/cmdline-partition.txt b/Documentation/block/cmdline-partition.txt new file mode 100644 index 000000000..760a3f7c3 --- /dev/null +++ b/Documentation/block/cmdline-partition.txt @@ -0,0 +1,46 @@ +Embedded device command line partition parsing +===================================================================== + +The "blkdevparts" command line option adds support for reading the +block device partition table from the kernel command line. + +It is typically used for fixed block (eMMC) embedded devices. +It has no MBR, so saves storage space. Bootloader can be easily accessed +by absolute address of data on the block device. +Users can easily change the partition. + +The format for the command line is just like mtdparts: + +blkdevparts=<blkdev-def>[;<blkdev-def>] + <blkdev-def> := <blkdev-id>:<partdef>[,<partdef>] + <partdef> := <size>[@<offset>](part-name) + +<blkdev-id> + block device disk name. Embedded device uses fixed block device. + Its disk name is also fixed, such as: mmcblk0, mmcblk1, mmcblk0boot0. + +<size> + partition size, in bytes, such as: 512, 1m, 1G. + size may contain an optional suffix of (upper or lower case): + K, M, G, T, P, E. + "-" is used to denote all remaining space. + +<offset> + partition start address, in bytes. + offset may contain an optional suffix of (upper or lower case): + K, M, G, T, P, E. + +(part-name) + partition name. Kernel sends uevent with "PARTNAME". Application can + create a link to block device partition with the name "PARTNAME". + User space application can access partition by partition name. + +Example: + eMMC disk names are "mmcblk0" and "mmcblk0boot0". + + bootargs: + 'blkdevparts=mmcblk0:1G(data0),1G(data1),-;mmcblk0boot0:1m(boot),-(kernel)' + + dmesg: + mmcblk0: p1(data0) p2(data1) p3() + mmcblk0boot0: p1(boot) p2(kernel) diff --git a/Documentation/block/data-integrity.txt b/Documentation/block/data-integrity.txt new file mode 100644 index 000000000..934c44ea0 --- /dev/null +++ b/Documentation/block/data-integrity.txt @@ -0,0 +1,281 @@ +---------------------------------------------------------------------- +1. INTRODUCTION + +Modern filesystems feature checksumming of data and metadata to +protect against data corruption. However, the detection of the +corruption is done at read time which could potentially be months +after the data was written. At that point the original data that the +application tried to write is most likely lost. + +The solution is to ensure that the disk is actually storing what the +application meant it to. Recent additions to both the SCSI family +protocols (SBC Data Integrity Field, SCC protection proposal) as well +as SATA/T13 (External Path Protection) try to remedy this by adding +support for appending integrity metadata to an I/O. The integrity +metadata (or protection information in SCSI terminology) includes a +checksum for each sector as well as an incrementing counter that +ensures the individual sectors are written in the right order. And +for some protection schemes also that the I/O is written to the right +place on disk. + +Current storage controllers and devices implement various protective +measures, for instance checksumming and scrubbing. But these +technologies are working in their own isolated domains or at best +between adjacent nodes in the I/O path. The interesting thing about +DIF and the other integrity extensions is that the protection format +is well defined and every node in the I/O path can verify the +integrity of the I/O and reject it if corruption is detected. This +allows not only corruption prevention but also isolation of the point +of failure. + +---------------------------------------------------------------------- +2. THE DATA INTEGRITY EXTENSIONS + +As written, the protocol extensions only protect the path between +controller and storage device. However, many controllers actually +allow the operating system to interact with the integrity metadata +(IMD). We have been working with several FC/SAS HBA vendors to enable +the protection information to be transferred to and from their +controllers. + +The SCSI Data Integrity Field works by appending 8 bytes of protection +information to each sector. The data + integrity metadata is stored +in 520 byte sectors on disk. Data + IMD are interleaved when +transferred between the controller and target. The T13 proposal is +similar. + +Because it is highly inconvenient for operating systems to deal with +520 (and 4104) byte sectors, we approached several HBA vendors and +encouraged them to allow separation of the data and integrity metadata +scatter-gather lists. + +The controller will interleave the buffers on write and split them on +read. This means that Linux can DMA the data buffers to and from +host memory without changes to the page cache. + +Also, the 16-bit CRC checksum mandated by both the SCSI and SATA specs +is somewhat heavy to compute in software. Benchmarks found that +calculating this checksum had a significant impact on system +performance for a number of workloads. Some controllers allow a +lighter-weight checksum to be used when interfacing with the operating +system. Emulex, for instance, supports the TCP/IP checksum instead. +The IP checksum received from the OS is converted to the 16-bit CRC +when writing and vice versa. This allows the integrity metadata to be +generated by Linux or the application at very low cost (comparable to +software RAID5). + +The IP checksum is weaker than the CRC in terms of detecting bit +errors. However, the strength is really in the separation of the data +buffers and the integrity metadata. These two distinct buffers must +match up for an I/O to complete. + +The separation of the data and integrity metadata buffers as well as +the choice in checksums is referred to as the Data Integrity +Extensions. As these extensions are outside the scope of the protocol +bodies (T10, T13), Oracle and its partners are trying to standardize +them within the Storage Networking Industry Association. + +---------------------------------------------------------------------- +3. KERNEL CHANGES + +The data integrity framework in Linux enables protection information +to be pinned to I/Os and sent to/received from controllers that +support it. + +The advantage to the integrity extensions in SCSI and SATA is that +they enable us to protect the entire path from application to storage +device. However, at the same time this is also the biggest +disadvantage. It means that the protection information must be in a +format that can be understood by the disk. + +Generally Linux/POSIX applications are agnostic to the intricacies of +the storage devices they are accessing. The virtual filesystem switch +and the block layer make things like hardware sector size and +transport protocols completely transparent to the application. + +However, this level of detail is required when preparing the +protection information to send to a disk. Consequently, the very +concept of an end-to-end protection scheme is a layering violation. +It is completely unreasonable for an application to be aware whether +it is accessing a SCSI or SATA disk. + +The data integrity support implemented in Linux attempts to hide this +from the application. As far as the application (and to some extent +the kernel) is concerned, the integrity metadata is opaque information +that's attached to the I/O. + +The current implementation allows the block layer to automatically +generate the protection information for any I/O. Eventually the +intent is to move the integrity metadata calculation to userspace for +user data. Metadata and other I/O that originates within the kernel +will still use the automatic generation interface. + +Some storage devices allow each hardware sector to be tagged with a +16-bit value. The owner of this tag space is the owner of the block +device. I.e. the filesystem in most cases. The filesystem can use +this extra space to tag sectors as they see fit. Because the tag +space is limited, the block interface allows tagging bigger chunks by +way of interleaving. This way, 8*16 bits of information can be +attached to a typical 4KB filesystem block. + +This also means that applications such as fsck and mkfs will need +access to manipulate the tags from user space. A passthrough +interface for this is being worked on. + + +---------------------------------------------------------------------- +4. BLOCK LAYER IMPLEMENTATION DETAILS + +4.1 BIO + +The data integrity patches add a new field to struct bio when +CONFIG_BLK_DEV_INTEGRITY is enabled. bio_integrity(bio) returns a +pointer to a struct bip which contains the bio integrity payload. +Essentially a bip is a trimmed down struct bio which holds a bio_vec +containing the integrity metadata and the required housekeeping +information (bvec pool, vector count, etc.) + +A kernel subsystem can enable data integrity protection on a bio by +calling bio_integrity_alloc(bio). This will allocate and attach the +bip to the bio. + +Individual pages containing integrity metadata can subsequently be +attached using bio_integrity_add_page(). + +bio_free() will automatically free the bip. + + +4.2 BLOCK DEVICE + +Because the format of the protection data is tied to the physical +disk, each block device has been extended with a block integrity +profile (struct blk_integrity). This optional profile is registered +with the block layer using blk_integrity_register(). + +The profile contains callback functions for generating and verifying +the protection data, as well as getting and setting application tags. +The profile also contains a few constants to aid in completing, +merging and splitting the integrity metadata. + +Layered block devices will need to pick a profile that's appropriate +for all subdevices. blk_integrity_compare() can help with that. DM +and MD linear, RAID0 and RAID1 are currently supported. RAID4/5/6 +will require extra work due to the application tag. + + +---------------------------------------------------------------------- +5.0 BLOCK LAYER INTEGRITY API + +5.1 NORMAL FILESYSTEM + + The normal filesystem is unaware that the underlying block device + is capable of sending/receiving integrity metadata. The IMD will + be automatically generated by the block layer at submit_bio() time + in case of a WRITE. A READ request will cause the I/O integrity + to be verified upon completion. + + IMD generation and verification can be toggled using the + + /sys/block/<bdev>/integrity/write_generate + + and + + /sys/block/<bdev>/integrity/read_verify + + flags. + + +5.2 INTEGRITY-AWARE FILESYSTEM + + A filesystem that is integrity-aware can prepare I/Os with IMD + attached. It can also use the application tag space if this is + supported by the block device. + + + bool bio_integrity_prep(bio); + + To generate IMD for WRITE and to set up buffers for READ, the + filesystem must call bio_integrity_prep(bio). + + Prior to calling this function, the bio data direction and start + sector must be set, and the bio should have all data pages + added. It is up to the caller to ensure that the bio does not + change while I/O is in progress. + Complete bio with error if prepare failed for some reson. + + +5.3 PASSING EXISTING INTEGRITY METADATA + + Filesystems that either generate their own integrity metadata or + are capable of transferring IMD from user space can use the + following calls: + + + struct bip * bio_integrity_alloc(bio, gfp_mask, nr_pages); + + Allocates the bio integrity payload and hangs it off of the bio. + nr_pages indicate how many pages of protection data need to be + stored in the integrity bio_vec list (similar to bio_alloc()). + + The integrity payload will be freed at bio_free() time. + + + int bio_integrity_add_page(bio, page, len, offset); + + Attaches a page containing integrity metadata to an existing + bio. The bio must have an existing bip, + i.e. bio_integrity_alloc() must have been called. For a WRITE, + the integrity metadata in the pages must be in a format + understood by the target device with the notable exception that + the sector numbers will be remapped as the request traverses the + I/O stack. This implies that the pages added using this call + will be modified during I/O! The first reference tag in the + integrity metadata must have a value of bip->bip_sector. + + Pages can be added using bio_integrity_add_page() as long as + there is room in the bip bio_vec array (nr_pages). + + Upon completion of a READ operation, the attached pages will + contain the integrity metadata received from the storage device. + It is up to the receiver to process them and verify data + integrity upon completion. + + +5.4 REGISTERING A BLOCK DEVICE AS CAPABLE OF EXCHANGING INTEGRITY + METADATA + + To enable integrity exchange on a block device the gendisk must be + registered as capable: + + int blk_integrity_register(gendisk, blk_integrity); + + The blk_integrity struct is a template and should contain the + following: + + static struct blk_integrity my_profile = { + .name = "STANDARDSBODY-TYPE-VARIANT-CSUM", + .generate_fn = my_generate_fn, + .verify_fn = my_verify_fn, + .tuple_size = sizeof(struct my_tuple_size), + .tag_size = <tag bytes per hw sector>, + }; + + 'name' is a text string which will be visible in sysfs. This is + part of the userland API so chose it carefully and never change + it. The format is standards body-type-variant. + E.g. T10-DIF-TYPE1-IP or T13-EPP-0-CRC. + + 'generate_fn' generates appropriate integrity metadata (for WRITE). + + 'verify_fn' verifies that the data buffer matches the integrity + metadata. + + 'tuple_size' must be set to match the size of the integrity + metadata per sector. I.e. 8 for DIF and EPP. + + 'tag_size' must be set to identify how many bytes of tag space + are available per hardware sector. For DIF this is either 2 or + 0 depending on the value of the Control Mode Page ATO bit. + +---------------------------------------------------------------------- +2007-12-24 Martin K. Petersen <martin.petersen@oracle.com> diff --git a/Documentation/block/deadline-iosched.txt b/Documentation/block/deadline-iosched.txt new file mode 100644 index 000000000..2d82c8032 --- /dev/null +++ b/Documentation/block/deadline-iosched.txt @@ -0,0 +1,75 @@ +Deadline IO scheduler tunables +============================== + +This little file attempts to document how the deadline io scheduler works. +In particular, it will clarify the meaning of the exposed tunables that may be +of interest to power users. + +Selecting IO schedulers +----------------------- +Refer to Documentation/block/switching-sched.txt for information on +selecting an io scheduler on a per-device basis. + + +******************************************************************************** + + +read_expire (in ms) +----------- + +The goal of the deadline io scheduler is to attempt to guarantee a start +service time for a request. As we focus mainly on read latencies, this is +tunable. When a read request first enters the io scheduler, it is assigned +a deadline that is the current time + the read_expire value in units of +milliseconds. + + +write_expire (in ms) +----------- + +Similar to read_expire mentioned above, but for writes. + + +fifo_batch (number of requests) +---------- + +Requests are grouped into ``batches'' of a particular data direction (read or +write) which are serviced in increasing sector order. To limit extra seeking, +deadline expiries are only checked between batches. fifo_batch controls the +maximum number of requests per batch. + +This parameter tunes the balance between per-request latency and aggregate +throughput. When low latency is the primary concern, smaller is better (where +a value of 1 yields first-come first-served behaviour). Increasing fifo_batch +generally improves throughput, at the cost of latency variation. + + +writes_starved (number of dispatches) +-------------- + +When we have to move requests from the io scheduler queue to the block +device dispatch queue, we always give a preference to reads. However, we +don't want to starve writes indefinitely either. So writes_starved controls +how many times we give preference to reads over writes. When that has been +done writes_starved number of times, we dispatch some writes based on the +same criteria as reads. + + +front_merges (bool) +------------ + +Sometimes it happens that a request enters the io scheduler that is contiguous +with a request that is already on the queue. Either it fits in the back of that +request, or it fits at the front. That is called either a back merge candidate +or a front merge candidate. Due to the way files are typically laid out, +back merges are much more common than front merges. For some work loads, you +may even know that it is a waste of time to spend any time attempting to +front merge requests. Setting front_merges to 0 disables this functionality. +Front merges may still occur due to the cached last_merge hint, but since +that comes at basically 0 cost we leave that on. We simply disable the +rbtree front sector lookup when the io scheduler merge function is called. + + +Nov 11 2002, Jens Axboe <jens.axboe@oracle.com> + + diff --git a/Documentation/block/ioprio.txt b/Documentation/block/ioprio.txt new file mode 100644 index 000000000..8ed8c5938 --- /dev/null +++ b/Documentation/block/ioprio.txt @@ -0,0 +1,183 @@ +Block io priorities +=================== + + +Intro +----- + +With the introduction of cfq v3 (aka cfq-ts or time sliced cfq), basic io +priorities are supported for reads on files. This enables users to io nice +processes or process groups, similar to what has been possible with cpu +scheduling for ages. This document mainly details the current possibilities +with cfq; other io schedulers do not support io priorities thus far. + +Scheduling classes +------------------ + +CFQ implements three generic scheduling classes that determine how io is +served for a process. + +IOPRIO_CLASS_RT: This is the realtime io class. This scheduling class is given +higher priority than any other in the system, processes from this class are +given first access to the disk every time. Thus it needs to be used with some +care, one io RT process can starve the entire system. Within the RT class, +there are 8 levels of class data that determine exactly how much time this +process needs the disk for on each service. In the future this might change +to be more directly mappable to performance, by passing in a wanted data +rate instead. + +IOPRIO_CLASS_BE: This is the best-effort scheduling class, which is the default +for any process that hasn't set a specific io priority. The class data +determines how much io bandwidth the process will get, it's directly mappable +to the cpu nice levels just more coarsely implemented. 0 is the highest +BE prio level, 7 is the lowest. The mapping between cpu nice level and io +nice level is determined as: io_nice = (cpu_nice + 20) / 5. + +IOPRIO_CLASS_IDLE: This is the idle scheduling class, processes running at this +level only get io time when no one else needs the disk. The idle class has no +class data, since it doesn't really apply here. + +Tools +----- + +See below for a sample ionice tool. Usage: + +# ionice -c<class> -n<level> -p<pid> + +If pid isn't given, the current process is assumed. IO priority settings +are inherited on fork, so you can use ionice to start the process at a given +level: + +# ionice -c2 -n0 /bin/ls + +will run ls at the best-effort scheduling class at the highest priority. +For a running process, you can give the pid instead: + +# ionice -c1 -n2 -p100 + +will change pid 100 to run at the realtime scheduling class, at priority 2. + +---> snip ionice.c tool <--- + +#include <stdio.h> +#include <stdlib.h> +#include <errno.h> +#include <getopt.h> +#include <unistd.h> +#include <sys/ptrace.h> +#include <asm/unistd.h> + +extern int sys_ioprio_set(int, int, int); +extern int sys_ioprio_get(int, int); + +#if defined(__i386__) +#define __NR_ioprio_set 289 +#define __NR_ioprio_get 290 +#elif defined(__ppc__) +#define __NR_ioprio_set 273 +#define __NR_ioprio_get 274 +#elif defined(__x86_64__) +#define __NR_ioprio_set 251 +#define __NR_ioprio_get 252 +#elif defined(__ia64__) +#define __NR_ioprio_set 1274 +#define __NR_ioprio_get 1275 +#else +#error "Unsupported arch" +#endif + +static inline int ioprio_set(int which, int who, int ioprio) +{ + return syscall(__NR_ioprio_set, which, who, ioprio); +} + +static inline int ioprio_get(int which, int who) +{ + return syscall(__NR_ioprio_get, which, who); +} + +enum { + IOPRIO_CLASS_NONE, + IOPRIO_CLASS_RT, + IOPRIO_CLASS_BE, + IOPRIO_CLASS_IDLE, +}; + +enum { + IOPRIO_WHO_PROCESS = 1, + IOPRIO_WHO_PGRP, + IOPRIO_WHO_USER, +}; + +#define IOPRIO_CLASS_SHIFT 13 + +const char *to_prio[] = { "none", "realtime", "best-effort", "idle", }; + +int main(int argc, char *argv[]) +{ + int ioprio = 4, set = 0, ioprio_class = IOPRIO_CLASS_BE; + int c, pid = 0; + + while ((c = getopt(argc, argv, "+n:c:p:")) != EOF) { + switch (c) { + case 'n': + ioprio = strtol(optarg, NULL, 10); + set = 1; + break; + case 'c': + ioprio_class = strtol(optarg, NULL, 10); + set = 1; + break; + case 'p': + pid = strtol(optarg, NULL, 10); + break; + } + } + + switch (ioprio_class) { + case IOPRIO_CLASS_NONE: + ioprio_class = IOPRIO_CLASS_BE; + break; + case IOPRIO_CLASS_RT: + case IOPRIO_CLASS_BE: + break; + case IOPRIO_CLASS_IDLE: + ioprio = 7; + break; + default: + printf("bad prio class %d\n", ioprio_class); + return 1; + } + + if (!set) { + if (!pid && argv[optind]) + pid = strtol(argv[optind], NULL, 10); + + ioprio = ioprio_get(IOPRIO_WHO_PROCESS, pid); + + printf("pid=%d, %d\n", pid, ioprio); + + if (ioprio == -1) + perror("ioprio_get"); + else { + ioprio_class = ioprio >> IOPRIO_CLASS_SHIFT; + ioprio = ioprio & 0xff; + printf("%s: prio %d\n", to_prio[ioprio_class], ioprio); + } + } else { + if (ioprio_set(IOPRIO_WHO_PROCESS, pid, ioprio | ioprio_class << IOPRIO_CLASS_SHIFT) == -1) { + perror("ioprio_set"); + return 1; + } + + if (argv[optind]) + execvp(argv[optind], &argv[optind]); + } + + return 0; +} + +---> snip ionice.c tool <--- + + +March 11 2005, Jens Axboe <jens.axboe@oracle.com> diff --git a/Documentation/block/kyber-iosched.txt b/Documentation/block/kyber-iosched.txt new file mode 100644 index 000000000..e94feacd7 --- /dev/null +++ b/Documentation/block/kyber-iosched.txt @@ -0,0 +1,14 @@ +Kyber I/O scheduler tunables +=========================== + +The only two tunables for the Kyber scheduler are the target latencies for +reads and synchronous writes. Kyber will throttle requests in order to meet +these target latencies. + +read_lat_nsec +------------- +Target latency for reads (in nanoseconds). + +write_lat_nsec +-------------- +Target latency for synchronous writes (in nanoseconds). diff --git a/Documentation/block/null_blk.txt b/Documentation/block/null_blk.txt new file mode 100644 index 000000000..ea2dafe49 --- /dev/null +++ b/Documentation/block/null_blk.txt @@ -0,0 +1,94 @@ +Null block device driver +================================================================================ + +I. Overview + +The null block device (/dev/nullb*) is used for benchmarking the various +block-layer implementations. It emulates a block device of X gigabytes in size. +The following instances are possible: + + Single-queue block-layer + - Request-based. + - Single submission queue per device. + - Implements IO scheduling algorithms (CFQ, Deadline, noop). + Multi-queue block-layer + - Request-based. + - Configurable submission queues per device. + No block-layer (Known as bio-based) + - Bio-based. IO requests are submitted directly to the device driver. + - Directly accepts bio data structure and returns them. + +All of them have a completion queue for each core in the system. + +II. Module parameters applicable for all instances: + +queue_mode=[0-2]: Default: 2-Multi-queue + Selects which block-layer the module should instantiate with. + + 0: Bio-based. + 1: Single-queue. + 2: Multi-queue. + +home_node=[0--nr_nodes]: Default: NUMA_NO_NODE + Selects what CPU node the data structures are allocated from. + +gb=[Size in GB]: Default: 250GB + The size of the device reported to the system. + +bs=[Block size (in bytes)]: Default: 512 bytes + The block size reported to the system. + +nr_devices=[Number of devices]: Default: 1 + Number of block devices instantiated. They are instantiated as /dev/nullb0, + etc. + +irqmode=[0-2]: Default: 1-Soft-irq + The completion mode used for completing IOs to the block-layer. + + 0: None. + 1: Soft-irq. Uses IPI to complete IOs across CPU nodes. Simulates the overhead + when IOs are issued from another CPU node than the home the device is + connected to. + 2: Timer: Waits a specific period (completion_nsec) for each IO before + completion. + +completion_nsec=[ns]: Default: 10,000ns + Combined with irqmode=2 (timer). The time each completion event must wait. + +submit_queues=[1..nr_cpus]: + The number of submission queues attached to the device driver. If unset, it + defaults to 1. For multi-queue, it is ignored when use_per_node_hctx module + parameter is 1. + +hw_queue_depth=[0..qdepth]: Default: 64 + The hardware queue depth of the device. + +III: Multi-queue specific parameters + +use_per_node_hctx=[0/1]: Default: 0 + 0: The number of submit queues are set to the value of the submit_queues + parameter. + 1: The multi-queue block layer is instantiated with a hardware dispatch + queue for each CPU node in the system. + +no_sched=[0/1]: Default: 0 + 0: nullb* use default blk-mq io scheduler. + 1: nullb* doesn't use io scheduler. + +blocking=[0/1]: Default: 0 + 0: Register as a non-blocking blk-mq driver device. + 1: Register as a blocking blk-mq driver device, null_blk will set + the BLK_MQ_F_BLOCKING flag, indicating that it sometimes/always + needs to block in its ->queue_rq() function. + +shared_tags=[0/1]: Default: 0 + 0: Tag set is not shared. + 1: Tag set shared between devices for blk-mq. Only makes sense with + nr_devices > 1, otherwise there's no tag set to share. + +zoned=[0/1]: Default: 0 + 0: Block device is exposed as a random-access block device. + 1: Block device is exposed as a host-managed zoned block device. + +zone_size=[MB]: Default: 256 + Per zone size when exposed as a zoned block device. Must be a power of two. diff --git a/Documentation/block/pr.txt b/Documentation/block/pr.txt new file mode 100644 index 000000000..ac9b8e70e --- /dev/null +++ b/Documentation/block/pr.txt @@ -0,0 +1,119 @@ + +Block layer support for Persistent Reservations +=============================================== + +The Linux kernel supports a user space interface for simplified +Persistent Reservations which map to block devices that support +these (like SCSI). Persistent Reservations allow restricting +access to block devices to specific initiators in a shared storage +setup. + +This document gives a general overview of the support ioctl commands. +For a more detailed reference please refer the the SCSI Primary +Commands standard, specifically the section on Reservations and the +"PERSISTENT RESERVE IN" and "PERSISTENT RESERVE OUT" commands. + +All implementations are expected to ensure the reservations survive +a power loss and cover all connections in a multi path environment. +These behaviors are optional in SPC but will be automatically applied +by Linux. + + +The following types of reservations are supported: +-------------------------------------------------- + + - PR_WRITE_EXCLUSIVE + + Only the initiator that owns the reservation can write to the + device. Any initiator can read from the device. + + - PR_EXCLUSIVE_ACCESS + + Only the initiator that owns the reservation can access the + device. + + - PR_WRITE_EXCLUSIVE_REG_ONLY + + Only initiators with a registered key can write to the device, + Any initiator can read from the device. + + - PR_EXCLUSIVE_ACCESS_REG_ONLY + + Only initiators with a registered key can access the device. + + - PR_WRITE_EXCLUSIVE_ALL_REGS + + Only initiators with a registered key can write to the device, + Any initiator can read from the device. + All initiators with a registered key are considered reservation + holders. + Please reference the SPC spec on the meaning of a reservation + holder if you want to use this type. + + - PR_EXCLUSIVE_ACCESS_ALL_REGS + + Only initiators with a registered key can access the device. + All initiators with a registered key are considered reservation + holders. + Please reference the SPC spec on the meaning of a reservation + holder if you want to use this type. + + +The following ioctl are supported: +---------------------------------- + +1. IOC_PR_REGISTER + +This ioctl command registers a new reservation if the new_key argument +is non-null. If no existing reservation exists old_key must be zero, +if an existing reservation should be replaced old_key must contain +the old reservation key. + +If the new_key argument is 0 it unregisters the existing reservation passed +in old_key. + + +2. IOC_PR_RESERVE + +This ioctl command reserves the device and thus restricts access for other +devices based on the type argument. The key argument must be the existing +reservation key for the device as acquired by the IOC_PR_REGISTER, +IOC_PR_REGISTER_IGNORE, IOC_PR_PREEMPT or IOC_PR_PREEMPT_ABORT commands. + + +3. IOC_PR_RELEASE + +This ioctl command releases the reservation specified by key and flags +and thus removes any access restriction implied by it. + + +4. IOC_PR_PREEMPT + +This ioctl command releases the existing reservation referred to by +old_key and replaces it with a new reservation of type for the +reservation key new_key. + + +5. IOC_PR_PREEMPT_ABORT + +This ioctl command works like IOC_PR_PREEMPT except that it also aborts +any outstanding command sent over a connection identified by old_key. + +6. IOC_PR_CLEAR + +This ioctl command unregisters both key and any other reservation key +registered with the device and drops any existing reservation. + + +Flags +----- + +All the ioctls have a flag field. Currently only one flag is supported: + + - PR_FL_IGNORE_KEY + + Ignore the existing reservation key. This is commonly supported for + IOC_PR_REGISTER, and some implementation may support the flag for + IOC_PR_RESERVE. + +For all unknown flags the kernel will return -EOPNOTSUPP. diff --git a/Documentation/block/queue-sysfs.txt b/Documentation/block/queue-sysfs.txt new file mode 100644 index 000000000..2c1e67058 --- /dev/null +++ b/Documentation/block/queue-sysfs.txt @@ -0,0 +1,197 @@ +Queue sysfs files +================= + +This text file will detail the queue files that are located in the sysfs tree +for each block device. Note that stacked devices typically do not export +any settings, since their queue merely functions are a remapping target. +These files are the ones found in the /sys/block/xxx/queue/ directory. + +Files denoted with a RO postfix are readonly and the RW postfix means +read-write. + +add_random (RW) +---------------- +This file allows to turn off the disk entropy contribution. Default +value of this file is '1'(on). + +dax (RO) +-------- +This file indicates whether the device supports Direct Access (DAX), +used by CPU-addressable storage to bypass the pagecache. It shows '1' +if true, '0' if not. + +discard_granularity (RO) +----------------------- +This shows the size of internal allocation of the device in bytes, if +reported by the device. A value of '0' means device does not support +the discard functionality. + +discard_max_hw_bytes (RO) +---------------------- +Devices that support discard functionality may have internal limits on +the number of bytes that can be trimmed or unmapped in a single operation. +The discard_max_bytes parameter is set by the device driver to the maximum +number of bytes that can be discarded in a single operation. Discard +requests issued to the device must not exceed this limit. A discard_max_bytes +value of 0 means that the device does not support discard functionality. + +discard_max_bytes (RW) +---------------------- +While discard_max_hw_bytes is the hardware limit for the device, this +setting is the software limit. Some devices exhibit large latencies when +large discards are issued, setting this value lower will make Linux issue +smaller discards and potentially help reduce latencies induced by large +discard operations. + +hw_sector_size (RO) +------------------- +This is the hardware sector size of the device, in bytes. + +io_poll (RW) +------------ +When read, this file shows whether polling is enabled (1) or disabled +(0). Writing '0' to this file will disable polling for this device. +Writing any non-zero value will enable this feature. + +io_poll_delay (RW) +------------------ +If polling is enabled, this controls what kind of polling will be +performed. It defaults to -1, which is classic polling. In this mode, +the CPU will repeatedly ask for completions without giving up any time. +If set to 0, a hybrid polling mode is used, where the kernel will attempt +to make an educated guess at when the IO will complete. Based on this +guess, the kernel will put the process issuing IO to sleep for an amount +of time, before entering a classic poll loop. This mode might be a +little slower than pure classic polling, but it will be more efficient. +If set to a value larger than 0, the kernel will put the process issuing +IO to sleep for this amont of microseconds before entering classic +polling. + +iostats (RW) +------------- +This file is used to control (on/off) the iostats accounting of the +disk. + +logical_block_size (RO) +----------------------- +This is the logical block size of the device, in bytes. + +max_hw_sectors_kb (RO) +---------------------- +This is the maximum number of kilobytes supported in a single data transfer. + +max_integrity_segments (RO) +--------------------------- +When read, this file shows the max limit of integrity segments as +set by block layer which a hardware controller can handle. + +max_sectors_kb (RW) +------------------- +This is the maximum number of kilobytes that the block layer will allow +for a filesystem request. Must be smaller than or equal to the maximum +size allowed by the hardware. + +max_segments (RO) +----------------- +Maximum number of segments of the device. + +max_segment_size (RO) +--------------------- +Maximum segment size of the device. + +minimum_io_size (RO) +-------------------- +This is the smallest preferred IO size reported by the device. + +nomerges (RW) +------------- +This enables the user to disable the lookup logic involved with IO +merging requests in the block layer. By default (0) all merges are +enabled. When set to 1 only simple one-hit merges will be tried. When +set to 2 no merge algorithms will be tried (including one-hit or more +complex tree/hash lookups). + +nr_requests (RW) +---------------- +This controls how many requests may be allocated in the block layer for +read or write requests. Note that the total allocated number may be twice +this amount, since it applies only to reads or writes (not the accumulated +sum). + +To avoid priority inversion through request starvation, a request +queue maintains a separate request pool per each cgroup when +CONFIG_BLK_CGROUP is enabled, and this parameter applies to each such +per-block-cgroup request pool. IOW, if there are N block cgroups, +each request queue may have up to N request pools, each independently +regulated by nr_requests. + +optimal_io_size (RO) +-------------------- +This is the optimal IO size reported by the device. + +physical_block_size (RO) +------------------------ +This is the physical block size of device, in bytes. + +read_ahead_kb (RW) +------------------ +Maximum number of kilobytes to read-ahead for filesystems on this block +device. + +rotational (RW) +--------------- +This file is used to stat if the device is of rotational type or +non-rotational type. + +rq_affinity (RW) +---------------- +If this option is '1', the block layer will migrate request completions to the +cpu "group" that originally submitted the request. For some workloads this +provides a significant reduction in CPU cycles due to caching effects. + +For storage configurations that need to maximize distribution of completion +processing setting this option to '2' forces the completion to run on the +requesting cpu (bypassing the "group" aggregation logic). + +scheduler (RW) +-------------- +When read, this file will display the current and available IO schedulers +for this block device. The currently active IO scheduler will be enclosed +in [] brackets. Writing an IO scheduler name to this file will switch +control of this block device to that new IO scheduler. Note that writing +an IO scheduler name to this file will attempt to load that IO scheduler +module, if it isn't already present in the system. + +write_cache (RW) +---------------- +When read, this file will display whether the device has write back +caching enabled or not. It will return "write back" for the former +case, and "write through" for the latter. Writing to this file can +change the kernels view of the device, but it doesn't alter the +device state. This means that it might not be safe to toggle the +setting from "write back" to "write through", since that will also +eliminate cache flushes issued by the kernel. + +write_same_max_bytes (RO) +------------------------- +This is the number of bytes the device can write in a single write-same +command. A value of '0' means write-same is not supported by this +device. + +wb_lat_usec (RW) +---------------- +If the device is registered for writeback throttling, then this file shows +the target minimum read latency. If this latency is exceeded in a given +window of time (see wb_window_usec), then the writeback throttling will start +scaling back writes. Writing a value of '0' to this file disables the +feature. Writing a value of '-1' to this file resets the value to the +default setting. + +throttle_sample_time (RW) +------------------------- +This is the time window that blk-throttle samples data, in millisecond. +blk-throttle makes decision based on the samplings. Lower time means cgroups +have more smooth throughput, but higher CPU overhead. This exists only when +CONFIG_BLK_DEV_THROTTLING_LOW is enabled. + +Jens Axboe <jens.axboe@oracle.com>, February 2009 diff --git a/Documentation/block/request.txt b/Documentation/block/request.txt new file mode 100644 index 000000000..754e104ed --- /dev/null +++ b/Documentation/block/request.txt @@ -0,0 +1,88 @@ + +struct request documentation + +Jens Axboe <jens.axboe@oracle.com> 27/05/02 + +1.0 +Index + +2.0 Struct request members classification + + 2.1 struct request members explanation + +3.0 + + +2.0 +Short explanation of request members + +Classification flags: + + D driver member + B block layer member + I I/O scheduler member + +Unless an entry contains a D classification, a device driver must not access +this member. Some members may contain D classifications, but should only be +access through certain macros or functions (eg ->flags). + +<linux/blkdev.h> + +2.1 +Member Flag Comment +------ ---- ------- + +struct list_head queuelist BI Organization on various internal + queues + +void *elevator_private I I/O scheduler private data + +unsigned char cmd[16] D Driver can use this for setting up + a cdb before execution, see + blk_queue_prep_rq + +unsigned long flags DBI Contains info about data direction, + request type, etc. + +int rq_status D Request status bits + +kdev_t rq_dev DBI Target device + +int errors DB Error counts + +sector_t sector DBI Target location + +unsigned long hard_nr_sectors B Used to keep sector sane + +unsigned long nr_sectors DBI Total number of sectors in request + +unsigned long hard_nr_sectors B Used to keep nr_sectors sane + +unsigned short nr_phys_segments DB Number of physical scatter gather + segments in a request + +unsigned short nr_hw_segments DB Number of hardware scatter gather + segments in a request + +unsigned int current_nr_sectors DB Number of sectors in first segment + of request + +unsigned int hard_cur_sectors B Used to keep current_nr_sectors sane + +int tag DB TCQ tag, if assigned + +void *special D Free to be used by driver + +char *buffer D Map of first segment, also see + section on bouncing SECTION + +struct completion *waiting D Can be used by driver to get signalled + on request completion + +struct bio *bio DBI First bio in request + +struct bio *biotail DBI Last bio in request + +struct request_queue *q DB Request queue this request belongs to + +struct request_list *rl B Request list this request came from diff --git a/Documentation/block/stat.txt b/Documentation/block/stat.txt new file mode 100644 index 000000000..0aace9cc5 --- /dev/null +++ b/Documentation/block/stat.txt @@ -0,0 +1,86 @@ +Block layer statistics in /sys/block/<dev>/stat +=============================================== + +This file documents the contents of the /sys/block/<dev>/stat file. + +The stat file provides several statistics about the state of block +device <dev>. + +Q. Why are there multiple statistics in a single file? Doesn't sysfs + normally contain a single value per file? +A. By having a single file, the kernel can guarantee that the statistics + represent a consistent snapshot of the state of the device. If the + statistics were exported as multiple files containing one statistic + each, it would be impossible to guarantee that a set of readings + represent a single point in time. + +The stat file consists of a single line of text containing 11 decimal +values separated by whitespace. The fields are summarized in the +following table, and described in more detail below. + +Name units description +---- ----- ----------- +read I/Os requests number of read I/Os processed +read merges requests number of read I/Os merged with in-queue I/O +read sectors sectors number of sectors read +read ticks milliseconds total wait time for read requests +write I/Os requests number of write I/Os processed +write merges requests number of write I/Os merged with in-queue I/O +write sectors sectors number of sectors written +write ticks milliseconds total wait time for write requests +in_flight requests number of I/Os currently in flight +io_ticks milliseconds total time this block device has been active +time_in_queue milliseconds total wait time for all requests +discard I/Os requests number of discard I/Os processed +discard merges requests number of discard I/Os merged with in-queue I/O +discard sectors sectors number of sectors discarded +discard ticks milliseconds total wait time for discard requests + +read I/Os, write I/Os, discard I/0s +=================================== + +These values increment when an I/O request completes. + +read merges, write merges, discard merges +========================================= + +These values increment when an I/O request is merged with an +already-queued I/O request. + +read sectors, write sectors, discard_sectors +============================================ + +These values count the number of sectors read from, written to, or +discarded from this block device. The "sectors" in question are the +standard UNIX 512-byte sectors, not any device- or filesystem-specific +block size. The counters are incremented when the I/O completes. + +read ticks, write ticks, discard ticks +====================================== + +These values count the number of milliseconds that I/O requests have +waited on this block device. If there are multiple I/O requests waiting, +these values will increase at a rate greater than 1000/second; for +example, if 60 read requests wait for an average of 30 ms, the read_ticks +field will increase by 60*30 = 1800. + +in_flight +========= + +This value counts the number of I/O requests that have been issued to +the device driver but have not yet completed. It does not include I/O +requests that are in the queue but not yet issued to the device driver. + +io_ticks +======== + +This value counts the number of milliseconds during which the device has +had I/O requests queued. + +time_in_queue +============= + +This value counts the number of milliseconds that I/O requests have waited +on this block device. If there are multiple I/O requests waiting, this +value will increase as the product of the number of milliseconds times the +number of requests waiting (see "read ticks" above for an example). diff --git a/Documentation/block/switching-sched.txt b/Documentation/block/switching-sched.txt new file mode 100644 index 000000000..3b2612e34 --- /dev/null +++ b/Documentation/block/switching-sched.txt @@ -0,0 +1,37 @@ +To choose IO schedulers at boot time, use the argument 'elevator=deadline'. +'noop' and 'cfq' (the default) are also available. IO schedulers are assigned +globally at boot time only presently. + +Each io queue has a set of io scheduler tunables associated with it. These +tunables control how the io scheduler works. You can find these entries +in: + +/sys/block/<device>/queue/iosched + +assuming that you have sysfs mounted on /sys. If you don't have sysfs mounted, +you can do so by typing: + +# mount none /sys -t sysfs + +As of the Linux 2.6.10 kernel, it is now possible to change the +IO scheduler for a given block device on the fly (thus making it possible, +for instance, to set the CFQ scheduler for the system default, but +set a specific device to use the deadline or noop schedulers - which +can improve that device's throughput). + +To set a specific scheduler, simply do this: + +echo SCHEDNAME > /sys/block/DEV/queue/scheduler + +where SCHEDNAME is the name of a defined IO scheduler, and DEV is the +device name (hda, hdb, sga, or whatever you happen to have). + +The list of defined schedulers can be found by simply doing +a "cat /sys/block/DEV/queue/scheduler" - the list of valid names +will be displayed, with the currently selected scheduler in brackets: + +# cat /sys/block/hda/queue/scheduler +noop deadline [cfq] +# echo deadline > /sys/block/hda/queue/scheduler +# cat /sys/block/hda/queue/scheduler +noop [deadline] cfq diff --git a/Documentation/block/writeback_cache_control.txt b/Documentation/block/writeback_cache_control.txt new file mode 100644 index 000000000..8a6bdada5 --- /dev/null +++ b/Documentation/block/writeback_cache_control.txt @@ -0,0 +1,86 @@ + +Explicit volatile write back cache control +===================================== + +Introduction +------------ + +Many storage devices, especially in the consumer market, come with volatile +write back caches. That means the devices signal I/O completion to the +operating system before data actually has hit the non-volatile storage. This +behavior obviously speeds up various workloads, but it means the operating +system needs to force data out to the non-volatile storage when it performs +a data integrity operation like fsync, sync or an unmount. + +The Linux block layer provides two simple mechanisms that let filesystems +control the caching behavior of the storage device. These mechanisms are +a forced cache flush, and the Force Unit Access (FUA) flag for requests. + + +Explicit cache flushes +---------------------- + +The REQ_PREFLUSH flag can be OR ed into the r/w flags of a bio submitted from +the filesystem and will make sure the volatile cache of the storage device +has been flushed before the actual I/O operation is started. This explicitly +guarantees that previously completed write requests are on non-volatile +storage before the flagged bio starts. In addition the REQ_PREFLUSH flag can be +set on an otherwise empty bio structure, which causes only an explicit cache +flush without any dependent I/O. It is recommend to use +the blkdev_issue_flush() helper for a pure cache flush. + + +Forced Unit Access +----------------- + +The REQ_FUA flag can be OR ed into the r/w flags of a bio submitted from the +filesystem and will make sure that I/O completion for this request is only +signaled after the data has been committed to non-volatile storage. + + +Implementation details for filesystems +-------------------------------------- + +Filesystems can simply set the REQ_PREFLUSH and REQ_FUA bits and do not have to +worry if the underlying devices need any explicit cache flushing and how +the Forced Unit Access is implemented. The REQ_PREFLUSH and REQ_FUA flags +may both be set on a single bio. + + +Implementation details for make_request_fn based block drivers +-------------------------------------------------------------- + +These drivers will always see the REQ_PREFLUSH and REQ_FUA bits as they sit +directly below the submit_bio interface. For remapping drivers the REQ_FUA +bits need to be propagated to underlying devices, and a global flush needs +to be implemented for bios with the REQ_PREFLUSH bit set. For real device +drivers that do not have a volatile cache the REQ_PREFLUSH and REQ_FUA bits +on non-empty bios can simply be ignored, and REQ_PREFLUSH requests without +data can be completed successfully without doing any work. Drivers for +devices with volatile caches need to implement the support for these +flags themselves without any help from the block layer. + + +Implementation details for request_fn based block drivers +-------------------------------------------------------------- + +For devices that do not support volatile write caches there is no driver +support required, the block layer completes empty REQ_PREFLUSH requests before +entering the driver and strips off the REQ_PREFLUSH and REQ_FUA bits from +requests that have a payload. For devices with volatile write caches the +driver needs to tell the block layer that it supports flushing caches by +doing: + + blk_queue_write_cache(sdkp->disk->queue, true, false); + +and handle empty REQ_OP_FLUSH requests in its prep_fn/request_fn. Note that +REQ_PREFLUSH requests with a payload are automatically turned into a sequence +of an empty REQ_OP_FLUSH request followed by the actual write by the block +layer. For devices that also support the FUA bit the block layer needs +to be told to pass through the REQ_FUA bit using: + + blk_queue_write_cache(sdkp->disk->queue, true, true); + +and the driver must handle write requests that have the REQ_FUA bit set +in prep_fn/request_fn. If the FUA bit is not natively supported the block +layer turns it into an empty REQ_OP_FLUSH request after the actual write. |