| Message ID | 20241121101845.1815660-1-Jonathan.Cameron@huawei.com (mailing list archive) |
|---|---|
| Headers | show |
| Series | CXL Hotness Monitoring Unit perf driver | expand |
On Thu, 21 Nov 2024 10:18:41 +0000 Jonathan Cameron <Jonathan.Cameron@huawei.com> wrote: > The CXL specification release 3.2 is now available under a click through at > https://computeexpresslink.org/cxl-specification/ and it brings new > shiny toys. > > RFC reason > - Whilst trace capture with a particular configuration is potentially useful > the intent is that CXL HMU units will be used to drive various forms of > hotpage migration for memory tiering setups. This driver doesn't do this > (yet), but rather provides data capture etc for experimentation and > for working out how to mostly put the allocations in the right place to > start with by tuning applications. > > CXL r3.2 introduces a CXL Hotness Monitoring Unit definition. The intent > of this is to provide a way to establish which units of memory (typically > pages or larger) in CXL attached memory are hot. The implementation details > and algorithm are all implementation defined. The specification simply > describes the 'interface' which takes the form of ring buffer of hotness > records in a PCI BAR and defined capability, configuration and status > registers. > > The hardware may have constraints on what it can track, granularity etc > and on how accurately it tracks (e.g. counter exhaustion, inaccurate > trackers). Some of these constraints are discoverable from the hardware > registers, others such as loss of accuracy have no universally accepted > measures as they are typically access pattern dependent. Sadly it is > very unlikely any hardware will implement a truly precise tracker given > the large resource requirements for tracking at a useful granularity. > > There are two fundamental operation modes: > > * Epoch based. Counters are checked after a period of time (Epoch) and > if over a threshold added to the hotlist. > * Always on. Counters run until a threshold is reached, after that the > hot unit is added to the hotlist and the counter released. > > Counting can be filtered on: > > * Region of CXL DPA space (256MiB per bit in a bitmap). > * Type of access - Trusted and non trusted or non trusted only, R/W/RW > > Sampling can be modified by: > > * Downsampling including potentially randomized downsampling. > > The driver presented here is intended to be useful in its own right but > also to act as the first step of a possible path towards hotness monitoring > based hot page migration. Those steps might look like. > > 1. Gather data - drivers provide telemetry like solutions to get that > data. May be enhanced, for example in this driver by providing the > HPA address rather than DPA Unit Address. Userspace can access enough > information to do this so maybe not. > 2. Userspace algorithm development, possibly combined with userspace > triggered migration by PA. Working out how to use different levels > of constrained hardware resources will be challenging. > 3. Move those algorithms in kernel. Will require generalization across > different hotpage trackers etc. > > So far this driver just gives access to the raw data. I will probably kick > of a longer discussion on how to do adaptive sampling needed to actually > use these units for tiering etc, sometime soon (if no one one else beats > me too it). There is a follow up topic of how to virtualize this stuff > for memory stranding cases (VM gets a fixed mixture of fast and slow > memory and should do it's own tiering). > > More details in the Documentation patch but typical commands are: > > $perf record -a -e cxl_hmu_mem0.0.0/epoch_type=0,access_type=6,\ > hotness_threshold=1024,epoch_multiplier=4,epoch_scale=4,range_base=0,\ > range_size=1024,randomized_downsampling=0,downsampling_factor=32,\ > hotness_granual=12 > > $perf report --dump-raw-traces > > Example output. With a counter_width of 16 (0x10) the least significant > 4 bytes are the counter value and the unit index is bits 16-63. > Here all units are over the threshold and the indexes are 0,1,2 etc. > > . ... CXL_HMU data: size 33512 bytes > Header 0: units: 29c counter_width 10 > Header 1 : deadbeef > 0000000000000283 > 0000000000010364 > 0000000000020366 > 000000000003033c > 0000000000040343 > 00000000000502ff > 000000000006030d > 000000000007031a > > Which will produce a list of hotness entries. > Bits[N-1:0] counter value > Bits[63:N] Unit ID (combine with unit size and DPA base + HDM decoder > config to get to a Host Physical Address) > > Specific RFC questions. > - What should be in the header added to the aux buffer. > Currently just the minimum is provided. Number of records > and the counter width needed to decode them. > - Should we reset the counters when doing sampling "-F X" > If the frequency is higher than the epoch we never see any hot units. > If so, when should we reset them? > > Note testing has been light and on emulation only + as perf tool is > a pain to build on a striped back VM, build testing has all be on > arm64 so far. The driver loads though on both arm64 and x86 so > any problems are likely in the perf tool arch specific code > which is build tested (on wrong machine) FWIW, runs on x86. However, it triggers a lockdep warning in both start and stop due to the spin lock. Something to tidy up for RFCv2. J > > The QEMU emulation needs some cleanup, but I should be able to post > that shortly to let people actually play with this. There are lots > of open questions there on how 'right' we want the emulation to be > and what counting uarch to emulate. > > Jonathan Cameron (4): > cxl: Register devices for CXL Hotness Monitoring Units (CHMU) > cxl: Hotness Monitoring Unit via a Perf AUX Buffer. > perf: Add support for CXL Hotness Monitoring Units (CHMU) > hwtrace: Document CXL Hotness Monitoring Unit driver > > Documentation/trace/cxl-hmu.rst | 197 +++++++ > Documentation/trace/index.rst | 1 + > drivers/cxl/Kconfig | 6 + > drivers/cxl/Makefile | 3 + > drivers/cxl/core/Makefile | 1 + > drivers/cxl/core/core.h | 1 + > drivers/cxl/core/hmu.c | 64 ++ > drivers/cxl/core/port.c | 2 + > drivers/cxl/core/regs.c | 14 + > drivers/cxl/cxl.h | 5 + > drivers/cxl/cxlpci.h | 1 + > drivers/cxl/hmu.c | 880 ++++++++++++++++++++++++++++ > drivers/cxl/hmu.h | 23 + > drivers/cxl/pci.c | 26 +- > tools/perf/arch/arm/util/auxtrace.c | 58 ++ > tools/perf/arch/x86/util/auxtrace.c | 76 +++ > tools/perf/util/Build | 1 + > tools/perf/util/auxtrace.c | 4 + > tools/perf/util/auxtrace.h | 1 + > tools/perf/util/cxl-hmu.c | 367 ++++++++++++ > tools/perf/util/cxl-hmu.h | 18 + > 21 files changed, 1748 insertions(+), 1 deletion(-) > create mode 100644 Documentation/trace/cxl-hmu.rst > create mode 100644 drivers/cxl/core/hmu.c > create mode 100644 drivers/cxl/hmu.c > create mode 100644 drivers/cxl/hmu.h > create mode 100644 tools/perf/util/cxl-hmu.c > create mode 100644 tools/perf/util/cxl-hmu.h >
On Thu, Nov 21, 2024 at 10:18:41AM +0000, Jonathan Cameron wrote: > The CXL specification release 3.2 is now available under a click through at > https://computeexpresslink.org/cxl-specification/ and it brings new > shiny toys. > > RFC reason > - Whilst trace capture with a particular configuration is potentially useful > the intent is that CXL HMU units will be used to drive various forms of > hotpage migration for memory tiering setups. This driver doesn't do this > (yet), but rather provides data capture etc for experimentation and > for working out how to mostly put the allocations in the right place to > start with by tuning applications. > > CXL r3.2 introduces a CXL Hotness Monitoring Unit definition. The intent > of this is to provide a way to establish which units of memory (typically > pages or larger) in CXL attached memory are hot. The implementation details > and algorithm are all implementation defined. The specification simply > describes the 'interface' which takes the form of ring buffer of hotness > records in a PCI BAR and defined capability, configuration and status > registers. > > The hardware may have constraints on what it can track, granularity etc > and on how accurately it tracks (e.g. counter exhaustion, inaccurate > trackers). Some of these constraints are discoverable from the hardware > registers, others such as loss of accuracy have no universally accepted > measures as they are typically access pattern dependent. Sadly it is > very unlikely any hardware will implement a truly precise tracker given > the large resource requirements for tracking at a useful granularity. > > There are two fundamental operation modes: > > * Epoch based. Counters are checked after a period of time (Epoch) and > if over a threshold added to the hotlist. > * Always on. Counters run until a threshold is reached, after that the > hot unit is added to the hotlist and the counter released. > > Counting can be filtered on: > > * Region of CXL DPA space (256MiB per bit in a bitmap). > * Type of access - Trusted and non trusted or non trusted only, R/W/RW > > Sampling can be modified by: > > * Downsampling including potentially randomized downsampling. > > The driver presented here is intended to be useful in its own right but > also to act as the first step of a possible path towards hotness monitoring > based hot page migration. Those steps might look like. > > 1. Gather data - drivers provide telemetry like solutions to get that > data. May be enhanced, for example in this driver by providing the > HPA address rather than DPA Unit Address. Userspace can access enough > information to do this so maybe not. > 2. Userspace algorithm development, possibly combined with userspace > triggered migration by PA. Working out how to use different levels > of constrained hardware resources will be challenging. FWIW this is what i was thinking about for this extension: https://lore.kernel.org/all/20240319172609.332900-1-gregory.price@memverge.com/ At least for testing CHMU stuff. So if anyone is poking at testing such things, they can feel free to use that for prototyping. However, I think there is general discomfort around userspace handling HPA/DPA. So it might look more like echo nr_pages > /sys/.../tiering/nodeN/promote_pages rather than handling the raw data from the CHMU to make decisions. > 3. Move those algorithms in kernel. Will require generalization across > different hotpage trackers etc. > In a longer discussion with Dan, we considered something a little more abstract - like a system that monitors bandwidth and memory access stalls and decide to promote X pages from Y device. This carries a pretty tall generalization cost, but it's pretty exciting to say the least. Definitely worth a discussion for later. > > So far this driver just gives access to the raw data. I will probably kick > of a longer discussion on how to do adaptive sampling needed to actually > use these units for tiering etc, sometime soon (if no one one else beats > me too it). There is a follow up topic of how to virtualize this stuff > for memory stranding cases (VM gets a fixed mixture of fast and slow > memory and should do it's own tiering). > Without having looked at the patches yet, I would presume this interface is at least gated to admin/root? (raw data is physical address info) ~Gregory
On Thu, 21 Nov 2024 09:24:43 -0500 Gregory Price <gourry@gourry.net> wrote: > On Thu, Nov 21, 2024 at 10:18:41AM +0000, Jonathan Cameron wrote: > > The CXL specification release 3.2 is now available under a click through at > > https://computeexpresslink.org/cxl-specification/ and it brings new > > shiny toys. > > > > RFC reason > > - Whilst trace capture with a particular configuration is potentially useful > > the intent is that CXL HMU units will be used to drive various forms of > > hotpage migration for memory tiering setups. This driver doesn't do this > > (yet), but rather provides data capture etc for experimentation and > > for working out how to mostly put the allocations in the right place to > > start with by tuning applications. > > > > CXL r3.2 introduces a CXL Hotness Monitoring Unit definition. The intent > > of this is to provide a way to establish which units of memory (typically > > pages or larger) in CXL attached memory are hot. The implementation details > > and algorithm are all implementation defined. The specification simply > > describes the 'interface' which takes the form of ring buffer of hotness > > records in a PCI BAR and defined capability, configuration and status > > registers. > > > > The hardware may have constraints on what it can track, granularity etc > > and on how accurately it tracks (e.g. counter exhaustion, inaccurate > > trackers). Some of these constraints are discoverable from the hardware > > registers, others such as loss of accuracy have no universally accepted > > measures as they are typically access pattern dependent. Sadly it is > > very unlikely any hardware will implement a truly precise tracker given > > the large resource requirements for tracking at a useful granularity. > > > > There are two fundamental operation modes: > > > > * Epoch based. Counters are checked after a period of time (Epoch) and > > if over a threshold added to the hotlist. > > * Always on. Counters run until a threshold is reached, after that the > > hot unit is added to the hotlist and the counter released. > > > > Counting can be filtered on: > > > > * Region of CXL DPA space (256MiB per bit in a bitmap). > > * Type of access - Trusted and non trusted or non trusted only, R/W/RW > > > > Sampling can be modified by: > > > > * Downsampling including potentially randomized downsampling. > > > > The driver presented here is intended to be useful in its own right but > > also to act as the first step of a possible path towards hotness monitoring > > based hot page migration. Those steps might look like. > > > > 1. Gather data - drivers provide telemetry like solutions to get that > > data. May be enhanced, for example in this driver by providing the > > HPA address rather than DPA Unit Address. Userspace can access enough > > information to do this so maybe not. > > 2. Userspace algorithm development, possibly combined with userspace > > triggered migration by PA. Working out how to use different levels > > of constrained hardware resources will be challenging. > > FWIW this is what i was thinking about for this extension: > > https://lore.kernel.org/all/20240319172609.332900-1-gregory.price@memverge.com/ Yup. I had that in mind. Forgot to actually add a link. > > At least for testing CHMU stuff. So if anyone is poking at testing such > things, they can feel free to use that for prototyping. However, I think > there is general discomfort around userspace handling HPA/DPA. > > So it might look more like > > echo nr_pages > /sys/.../tiering/nodeN/promote_pages > > rather than handling the raw data from the CHMU to make decisions. Agreed, but I think we are far away from a point where we can implement that. Just working out how to tune the hardware to grab useful data is going to take a while to figure out, let alone doing anything much with it. Without care you won't get a meaningful signal for what is actually hot out of the box. Lots of reasons why including: a) Exhaustion of tracking resources, due to looking at too large a window or for too long. Will probably need some form of auto updating of what is being scanning (coarse to fine might work though I'm doubtful, scanning across small regions maybe). b) Threshold too high, no detections. c) Threshold too low, everything hot. d) Wrong timescales. Hot is not a well defined thing. e) Hardware that won't do tracking at fine enough granularity. > > > > 3. Move those algorithms in kernel. Will require generalization across > > different hotpage trackers etc. > > > > In a longer discussion with Dan, we considered something a little more > abstract - like a system that monitors bandwidth and memory access stalls > and decide to promote X pages from Y device. This carries a pretty tall > generalization cost, but it's pretty exciting to say the least. Agreed that ultimately we'll end up somewhere like that. These units are just a small part of what is needed in total. > > Definitely worth a discussion for later. > > > > > So far this driver just gives access to the raw data. I will probably kick > > of a longer discussion on how to do adaptive sampling needed to actually > > use these units for tiering etc, sometime soon (if no one one else beats > > me too it). There is a follow up topic of how to virtualize this stuff > > for memory stranding cases (VM gets a fixed mixture of fast and slow > > memory and should do it's own tiering). > > > > Without having looked at the patches yet, I would presume this interface > is at least gated to admin/root? (raw data is physical address info) That's certainly the intent. It's not going upstream in this form so I haven't actually checked yet :) Uses similar infrastructure to ARM SPE which can also give physical address info + a lot more than that. Jonathan > > ~Gregory >
On Thu, Nov 21, 2024 at 02:58:52PM +0000, Jonathan Cameron wrote: > On Thu, 21 Nov 2024 09:24:43 -0500 > Gregory Price <gourry@gourry.net> wrote: > > > On Thu, Nov 21, 2024 at 10:18:41AM +0000, Jonathan Cameron wrote: > > > The CXL specification release 3.2 is now available under a click through at > > > https://computeexpresslink.org/cxl-specification/ and it brings new > > > shiny toys. > > > > > > RFC reason > > > - Whilst trace capture with a particular configuration is potentially useful > > > the intent is that CXL HMU units will be used to drive various forms of > > > hotpage migration for memory tiering setups. This driver doesn't do this > > > (yet), but rather provides data capture etc for experimentation and > > > for working out how to mostly put the allocations in the right place to > > > start with by tuning applications. > > > > > > CXL r3.2 introduces a CXL Hotness Monitoring Unit definition. The intent > > > of this is to provide a way to establish which units of memory (typically > > > pages or larger) in CXL attached memory are hot. The implementation details > > > and algorithm are all implementation defined. The specification simply > > > describes the 'interface' which takes the form of ring buffer of hotness > > > records in a PCI BAR and defined capability, configuration and status > > > registers. > > > > > > The hardware may have constraints on what it can track, granularity etc > > > and on how accurately it tracks (e.g. counter exhaustion, inaccurate > > > trackers). Some of these constraints are discoverable from the hardware > > > registers, others such as loss of accuracy have no universally accepted > > > measures as they are typically access pattern dependent. Sadly it is > > > very unlikely any hardware will implement a truly precise tracker given > > > the large resource requirements for tracking at a useful granularity. > > > > > > There are two fundamental operation modes: > > > > > > * Epoch based. Counters are checked after a period of time (Epoch) and > > > if over a threshold added to the hotlist. > > > * Always on. Counters run until a threshold is reached, after that the > > > hot unit is added to the hotlist and the counter released. > > > > > > Counting can be filtered on: > > > > > > * Region of CXL DPA space (256MiB per bit in a bitmap). > > > * Type of access - Trusted and non trusted or non trusted only, R/W/RW > > > > > > Sampling can be modified by: > > > > > > * Downsampling including potentially randomized downsampling. > > > > > > The driver presented here is intended to be useful in its own right but > > > also to act as the first step of a possible path towards hotness monitoring > > > based hot page migration. Those steps might look like. > > > > > > 1. Gather data - drivers provide telemetry like solutions to get that > > > data. May be enhanced, for example in this driver by providing the > > > HPA address rather than DPA Unit Address. Userspace can access enough > > > information to do this so maybe not. > > > 2. Userspace algorithm development, possibly combined with userspace > > > triggered migration by PA. Working out how to use different levels > > > of constrained hardware resources will be challenging. > > > > FWIW this is what i was thinking about for this extension: > > > > https://lore.kernel.org/all/20240319172609.332900-1-gregory.price@memverge.com/ > > Yup. I had that in mind. Forgot to actually add a link. > > > > > At least for testing CHMU stuff. So if anyone is poking at testing such > > things, they can feel free to use that for prototyping. However, I think > > there is general discomfort around userspace handling HPA/DPA. > > > > So it might look more like > > > > echo nr_pages > /sys/.../tiering/nodeN/promote_pages > > > > rather than handling the raw data from the CHMU to make decisions. > > Agreed, but I think we are far away from a point where we can implement that. > > Just working out how to tune the hardware to grab useful data is going > to take a while to figure out, let alone doing anything much with it. > > Without care you won't get a meaningful signal for what is actually > hot out of the box. Lots of reasons why including: > a) Exhaustion of tracking resources, due to looking at too large a window > or for too long. Will probably need some form of auto updating of > what is being scanning (coarse to fine might work though I'm doubtful, > scanning across small regions maybe). > b) Threshold too high, no detections. > c) Threshold too low, everything hot. > d) Wrong timescales. Hot is not a well defined thing. > e) Hardware that won't do tracking at fine enough granularity. > f) How does this even work with interleaving on larger pools :B It's pretend-addressing all the way down :D Lots of conceptually complex and fun questions here. ~Gregory
On Thu, 21 Nov 2024 14:58:52 +0000 Jonathan Cameron <Jonathan.Cameron@huawei.com> wrote: > On Thu, 21 Nov 2024 09:24:43 -0500 > Gregory Price <gourry@gourry.net> wrote: > > > On Thu, Nov 21, 2024 at 10:18:41AM +0000, Jonathan Cameron wrote: [...] > Just working out how to tune the hardware to grab useful data is going > to take a while to figure out, let alone doing anything much with it. > > Without care you won't get a meaningful signal for what is actually > hot out of the box. Lots of reasons why including: > a) Exhaustion of tracking resources, due to looking at too large a window > or for too long. Will probably need some form of auto updating of > what is being scanning (coarse to fine might work though I'm doubtful, > scanning across small regions maybe). > b) Threshold too high, no detections. > c) Threshold too low, everything hot. > d) Wrong timescales. Hot is not a well defined thing. > e) Hardware that won't do tracking at fine enough granularity. Similar questions can be raised to general hotness monitoring including that for DAMON. I'm trying to summarize[1] rules of thumbs for DAMON tuning based on my humble experiences. Once it is done, I will further try automations of tunings. In future, hopefully DAMON can be extended to utilize CXL hotness monitoring unit as low level primitive for access check. Then, the guidance and automation of DAMON tuning could be just applied. Note that I'm not saying DAMON should be the only way to utilize CXL hotness monitoring unit. I'm saying DAMON could be one of the ways :) [1] https://lore.kernel.org/20241108232536.73843-1-sj@kernel.org Thanks, SJ [...]