CoreSight August 2018

coresight@lists.linaro.org

11 participants
9 discussions

by Ajith Kurian Issac

Hi, We are trying to us the Open CSD for decoding a onchip trace in our ETB. The trace was enabled and is captured in the ETB. We read the trace back and dumped it into a text file. I am attaching it. How can I use the CSD tool to decode it? Thanks Ajith

6 years, 2 months

[PATCH v3 0/9] coresight: Update device tree bindings

by Suzuki K Poulose

Coresight uses DT graph bindings to describe the connections of the components. However we have some undocumented usage of the bindings to describe some of the properties of the connections. The coresight driver needs to know the hardware ports invovled in the connection and the direction of data flow to effectively manage the trace sessions. So far we have relied on the "port" address (as described by the generic graph bindings) to represent the hardware port of the component for a connection. The hardware uses separate numbering scheme for input and output ports, which implies, we could have two different (input and output) ports with the same port number. This could create problems in the graph bindings where the label of the port wouldn't match the address. e.g, with the existing bindings we get : port@0{ // Output port 0 reg = <0>; ... }; port@1{ reg = <0>; // Input port 0 endpoint { slave-mode; ... }; }; With the new enforcement in the DT rules, mismatches in label and address are not allowed (as see in the case for port@1). So, we need a new mechanism to describe the hardware port number reliably. Also, we relied on an undocumented "slave-mode" property (see the above example) to indicate if the port is an input port. Let us formalise and switch to a new property to describe the direction of data flow. There were three options considered for the hardware port number scheme: 1) Use natural ordering in the DT to infer the hardware port number. i.e, Mandate that the all ports are listed in the DT and in the ascending order for each class (input and output respectively). Pros : - We don't need new properties and if the existing DTS list them in order (which most of them do), they work out of the box. Cons : - We must list all the ports even if the system cannot/shouldn't use it. - It is prone to human errors (if the order is not kept). 2) Use an explicit property to list both the direction and the hw port number and direction. Define "coresight,hwid" as 2 member array of u32, where the members are port number and the direction respectively. e.g port@0{ reg = <0>; endpoint { coresight,hwid = <0 1>; // Port # 0, Output } }; port@1{ reg = <1>; endpoint { coresight,hwid = <0 0>; // Port # 0, Input }; }; Pros: - The bindings are formal but not so reader friendly and could potentially lead to human errors. Cons: - Backward compatiblity is lost. 3) Use explicit properties (implemented in the series) for the hardware port id and direction. We define a new property "coresight,hwid" for each endpoint in coresight devices to specify the hardware port number explicitly. Also use a separate property "direction" to specify the direction of the data flow. e.g, port@0{ reg = <0>; endpoint { direction = <1>; // Output coresight,hwid = <0>; // Port # 0 } }; port@1{ reg = <1>; endpoint { direction = <0>; // Input coresight,hwid = <0>; // Port # 0 }; }; Pros: - The bindings are formal and reader friendly, and less prone to errors. Cons: - Backward compatibility is lost. After a round of discussions [1], the following option (4) is adopted : 4) Group ports based on the directions under a dedicated node. This has been checked with the upstream DTC tool to resolve the "address mismatch" issue. e.g, out-ports { // Output ports for this component port@0 { // Outport 0 reg = 0; endpoint { ... }; }; port@1 { // Outport 1 reg = 1; endpoint { ... }; }; }; in-ports { // Input ports for this component port@0 { // Inport 0 reg = 0; endpoint { ... }; }; port@1 { // Inport 1 reg = 1; endpoint { ... }; }; }; This series implements Option (4) listed above and falls back to the old bindings if the new bindings are not available. This allows the systems with old bindings work with the new driver. The driver now issues a warning (once) when it encounters the old bindings. The series contains DT update for Juno platform. The remaining in-kernel sources could be updated once we are fine with the proposal. It also cleans up the platform parsing code to reduce the memory usage by reusing the platform description. Applies on coresight/next Changes since V2: - Clean of_coresight_parse_endpoint() to return 1 to indicate a connection record was updated. - Drop documentation for old bindings Changes since V1: - Implement the proposal by Rob. - Drop the DTS updates for all platforms except Juno - Drop the incorrect fix in coresight_register. Instead document the code to prevent people trying to un-fix it again. - Add a patch to drop remote device references in DT graph parsing - Split of_node refcount fixing patch, fix a typo in the comment. - Add Reviewed-by tags from Mathieu. - Drop patches picked up for 4.18-rc series Changes since RFC: - Fixed style issues - Fix an existing memory leak coresight_register (Found in code update) - Fix missing of_node_put() in the existing driver (Reported-by Mathieu) - Update the existing dts in kernel tree. Suzuki K Poulose (9): coresight: Document error handling in coresight_register coresight: platform: Refactor graph endpoint parsing coresight: platform: Fix refcounting for graph nodes coresight: platform: Fix leaking device reference coresight: Fix remote endpoint parsing coresight: Add helper to check if the endpoint is input coresight: platform: Cleanup coresight connection handling coresight: Cleanup coresight DT bindings dts: juno: Update coresight bindings .../devicetree/bindings/arm/coresight.txt | 95 +++++--- arch/arm64/boot/dts/arm/juno-base.dtsi | 161 ++++++------ arch/arm64/boot/dts/arm/juno-cs-r1r2.dtsi | 52 ++-- arch/arm64/boot/dts/arm/juno.dts | 13 +- drivers/hwtracing/coresight/coresight.c | 35 +-- drivers/hwtracing/coresight/of_coresight.c | 269 ++++++++++++++------- include/linux/coresight.h | 9 +- 7 files changed, 359 insertions(+), 275 deletions(-) -- 2.7.4

6 years, 4 months

Re: ThunderX2 and Coresight bring up

by Mathieu Poirier

On Wed, 8 Aug 2018 at 01:59, Tomasz Nowicki <tnowicki(a)caviumnetworks.com> wrote: > > Hi Mathieu, > > It's been a while but I am back to Coresight. > > Let me remind my setup and the issue I am struggling with now. > > Kernel baseline: > https://github.com/Linaro/perf-opencsd (perf-opencsd-v4.16) > OpenCSD: > https://github.com/Linaro/OpenCSD.git (master) > > The simplest Coresight components path I used as a start point: > ETMv4.1 -> TDR -> FUNNEL -> ETF > > As I mentioned TDR is built by Cavium and it was added to aggregate 128 > inputs into one output rather than cascading funnels. TDR has its own > driver just to keep path connected in Linux Coresight framework. > > Here is how I catch some trace data: > sudo perf record -C 0 -e cs_etm/@etf0/ --per-thread test_app The above command line tells perf to trace everything that is happening on CPU0 for as long as "test_app" is executing. In this case the "--per-thread" option is ignored. This is called a CPU-wide trace scenario and is currently not supported for CS (I am currently working on it). If you want to make sure "test_app" executes on CPU0 and that you trace just that you will need to use the "taskset" utility: sudo perf record -e cs_etm/@etf0/ --per-thread taskset 0x1 test_app An alternative to the above would be to CPU-hotplug out CPU128-255 while you are testing. Let's start with that before going further. Thanks, Mathieu > > I need to use -C because my machines has 2 nodes, 32 cores (128 threads) > each and each node has different ETF. So I have to specify which CPU is > the source and for specified ETF sink (EFT0 can be a sink for > CPU0-CPU127, ETF1 can be a sink for CPU128-CPU255). Otherwise Linux > cannot find path for ETMs related to CPU128-CPU255 if I specify ETF0 as > a sink. > > Overall, I can see some data using: > # sudo perf report --stdio --dump > [...] > . ... CoreSight ETM Trace data: size 16384 bytes > Frame deformatter: Found 4 FSYNCS > ID:12 RESET operation on trace decode path > Idx:108; ID:12; I_NOT_SYNC : I Stream not synchronised > Idx:455; ID:12; I_ASYNC : Alignment Synchronisation. > Idx:468; ID:12; I_TRACE_INFO : Trace Info.; INFO=0x0 > Idx:470; ID:12; I_TRACE_ON : Trace On. > Idx:471; ID:12; I_CTXT : Context Packet.; Ctxt: AArch64,EL0, NS; > Idx:473; ID:12; I_ADDR_L_64IS0 : Address, Long, 64 bit, IS0.; > Addr=0x0000AAABE0B09584; > Idx:483; ID:12; I_ATOM_F1 : Atom format 1.; N > Idx:484; ID:12; I_TIMESTAMP : Timestamp.; Updated val = > 0x1b6a5d937cc1 > Idx:492; ID:12; I_ATOM_F3 : Atom format 3.; NNE > Idx:493; ID:12; I_ADDR_L_64IS0 : Address, Long, 64 bit, IS0.; > Addr=0x0000AAABE0B0D210; > Idx:504; ID:12; I_ATOM_F3 : Atom format 3.; NEE > Idx:505; ID:12; I_ATOM_F3 : Atom format 3.; NEN > Idx:506; ID:12; I_ATOM_F6 : Atom format 6.; EEEN > Idx:507; ID:12; I_ATOM_F3 : Atom format 3.; NNE > Idx:508; ID:12; I_ATOM_F1 : Atom format 1.; N > Idx:509; ID:12; I_ATOM_F3 : Atom format 3.; NNN > Idx:510; ID:12; I_ATOM_F3 : Atom format 3.; EEN > Idx:512; ID:12; I_ATOM_F1 : Atom format 1.; E > [...] > > However, I still see errors while using: > # sudo perf report --stdio > 0x1e8 [0x60]: failed to process type: 1 > Error: > failed to process sample > # To display the perf.data header info, please use > --header/--header-only options. > > The reason is that cs_etm__process_event() is failing on: > if (!etm->timeless_decoding) > return -EINVAL; > > and etm->timeless_decoding is setup in cs_etm__is_timeless_decoding(). > For some events time bit set and so far I failed to figure out what is > going on. Have you met similar issue so far? Any pointers or hints are > very appreciated. > > One more comment below. > > On 10.01.2018 21:10, Mathieu Poirier wrote: > > On 10 January 2018 at 06:57, Tomasz Nowicki <tnowicki(a)caviumnetworks.com> wrote: > >> Hello Mathieu, > >> > >> Thank you for your response. Please see comments below. > >> > >> On 08.01.2018 17:53, Mathieu Poirier wrote: > >>> > >>> Good day Tomasz, > >>> > >>> > >>> On 5 January 2018 at 05:51, tn <Tomasz.Nowicki(a)caviumnetworks.com> wrote: > >>>> > >>>> Hi Mathieu, > >>>> > >>>> I am bringing up Coresight functiproject zeroonality on ThunderX2. While > >>>> ramping up I > >>>> come across your Connect session: > >>>> > >>>> which I found very helpful. > >>> > >>> > >>> Perfect - a few things have changed since then, see below. > >>> > >>>> > >>>> During my research I had to create new Coresight component driver for > >>>> Linux, > >>>> here is the story. For ThunderX2, we aggregate data trace from all 128 > >>>> ETMs > >>>> into one funnel inport using so called TDR (Trace Data Ring) component. > >>>> This > >>>> should be transparent to software and does not require configuration at > >>>> all. > >>>> However, Linux Coresight framework requires components to be connected > >>>> each > >>>> other so we cannot leave funnel and ETMs disconnected in DT. I decided to > >>>> create pure software component i.e. TDR which is meant to connect chain > >>>> only, no actions on registers. > >>> > >>> > >>> Is this TDR an ARM IP or built in-house by Cavium? > >> > >> > >> This is Cavium specific component which I am going to upstream once I test > >> the whole functionality. > >> > >> And I suppose it > >>> > >>> was added there to aggregate 128 input into one output rather than > >>> cascading funnels? > >> > >> > >> Correct. > >> > >>>> > >>>> Now I am able to enable ETF sink and path from ETM via TDR via FUNNEL up > >>>> to > >>>> ETF and gather some data. To be sure things work properly I want to > >>>> decode > >>>> data using Linaro OpenCSD library following instructions from here: > >>>> > >>>> https://community.arm.com/tools/b/blog/posts/do-a-coresight-trace-on-linux-… > >>> > >>> > >>> Thanks for pointing this out, I didn't know about it. > >>> > >>>> but still got error while doing 'perf report' step. Kernel perf tool > >>>> support > >>>> for OpenCSD is out of tree for now so I may miss some patches. > >>> > >>> > >>> Can you get me a pastebin of the errors you're getting? > >> > >> > >> Sure, see: > >> https://pastebin.com/6YDq8KfC > >> As you see there is not much info about error cause. > >> > >>> > >>>> > >>>> Here is my setup: > >>>> https://github.com/Linaro/perf-opencsd/commits/upstream-v1 (+ ThunderX2 > >>>> specific patches) > >>> > >>> > >>> Oh boy... I wasn't expecting people to use that but I suppose it is > >>> the right thing to do. Keep going with that code. > >>> > >>>> https://github.com/Linaro/OpenCSD/commits/master > >>> > >>> > >>> This, in combination with the upstream-v1 branch should work properly. > >>> That's how I test things on my Juno and Dragon board. > >>> > >>>> > >>>> # echo 1 > etf0/enable_sink > >>>> # perf record -C 0 -e cs_etm// sleep 2 > >>> > >>> > >>> Ok, that won't work as the -C option is currently not supported (I am > >>> working on it). I also suggest to make sure you have the very latest > >>> TIP [1] on branch [2] and to carefully read the README.md. We > >>> recently updated the instructions to fit the newest development. > >>> Lastly we have deprecated enabling the sink from the sysFS interface - > >>> it can still work but no guarantees are provided. It is better to > >>> specify the sink as part of the perf record command line, as shown in > >>> the most recent HOWTO.md. > >> > >> > >> I am able to specify sink as part of the perf record command line only for > >> Linux Perf master branch: > >> https://github.com/Linaro/perf-opencsd/commits/master > >> > >> For upstream-v1 branch I am getting: > >> $ perf record -vvv -e cs_etm/@etf0/ --per-thread uname > >> Using CPUID 0x00000000420f5160 > >> perf: util/evsel.c:783: apply_config_terms: Assertion `!(1)' failed. > >> Aborted (core dumped) > > > > > > Ok, I've uploaded upstream-v2. With that branch everything works fine > > on my side, no changes needed. I added a fix for a regression in the > > perf tip tree and the code required to use the ETR from the perf > > interface. > > > > One thing about the above: "@etf0". Is this really the name you gave > > to the device in the DT? Look under /sys/bus/coresight/devices/ for > > an etf entry. What is listed there should is the name of the ETF as > > it is known to the system. > > Indeed, the name is different but for perf command clarity I use shortcut. > > Thanks, > Tomasz

6 years, 4 months

[PATCH v2] perf: Support for Arm A32/T32 instruction sets in CoreSight trace

by Robert Walker

This patch adds support for generating instruction samples from trace of AArch32 programs using the A32 and T32 instruction sets. T32 has variable 2 or 4 byte instruction size, so the conversion between addresses and instruction counts requires extra information from the trace decoder, requiring version 0.9.1 of OpenCSD. A check for the new struct member has been added to the feature check for OpenCSD. Signed-off-by: Robert Walker <robert.walker(a)arm.com> --- v2: Minor fixes following review comments from Mathieu Rebased on v4.19-rc1 tools/build/feature/test-libopencsd.c | 7 +++ tools/perf/util/cs-etm-decoder/cs-etm-decoder.c | 27 ++++++++++ tools/perf/util/cs-etm-decoder/cs-etm-decoder.h | 10 ++++ tools/perf/util/cs-etm.c | 71 +++++++++++-------------- 4 files changed, 75 insertions(+), 40 deletions(-) diff --git a/tools/build/feature/test-libopencsd.c b/tools/build/feature/test-libopencsd.c index 5ff1246..d96b2df 100644 --- a/tools/build/feature/test-libopencsd.c +++ b/tools/build/feature/test-libopencsd.c @@ -3,6 +3,13 @@ int main(void) { + /* + * Requires ocsd_generic_trace_elem.num_instr_range introduced in + * OpenCSD 0.9 + */ + ocsd_generic_trace_elem elem; + (void)elem.num_instr_range; + (void)ocsd_get_version(); return 0; } diff --git a/tools/perf/util/cs-etm-decoder/cs-etm-decoder.c b/tools/perf/util/cs-etm-decoder/cs-etm-decoder.c index 938def6..73d8384 100644 --- a/tools/perf/util/cs-etm-decoder/cs-etm-decoder.c +++ b/tools/perf/util/cs-etm-decoder/cs-etm-decoder.c @@ -263,9 +263,12 @@ static void cs_etm_decoder__clear_buffer(struct cs_etm_decoder *decoder) decoder->tail = 0; decoder->packet_count = 0; for (i = 0; i < MAX_BUFFER; i++) { + decoder->packet_buffer[i].isa = CS_ETM_ISA_UNKNOWN; decoder->packet_buffer[i].start_addr = CS_ETM_INVAL_ADDR; decoder->packet_buffer[i].end_addr = CS_ETM_INVAL_ADDR; + decoder->packet_buffer[i].instr_count = 0; decoder->packet_buffer[i].last_instr_taken_branch = false; + decoder->packet_buffer[i].last_instr_size = 0; decoder->packet_buffer[i].exc = false; decoder->packet_buffer[i].exc_ret = false; decoder->packet_buffer[i].cpu = INT_MIN; @@ -294,11 +297,13 @@ cs_etm_decoder__buffer_packet(struct cs_etm_decoder *decoder, decoder->packet_count++; decoder->packet_buffer[et].sample_type = sample_type; + decoder->packet_buffer[et].isa = CS_ETM_ISA_UNKNOWN; decoder->packet_buffer[et].exc = false; decoder->packet_buffer[et].exc_ret = false; decoder->packet_buffer[et].cpu = *((int *)inode->priv); decoder->packet_buffer[et].start_addr = CS_ETM_INVAL_ADDR; decoder->packet_buffer[et].end_addr = CS_ETM_INVAL_ADDR; + decoder->packet_buffer[et].instr_count = 0; if (decoder->packet_count == MAX_BUFFER - 1) return OCSD_RESP_WAIT; @@ -321,8 +326,28 @@ cs_etm_decoder__buffer_range(struct cs_etm_decoder *decoder, packet = &decoder->packet_buffer[decoder->tail]; + switch (elem->isa) { + case ocsd_isa_aarch64: + packet->isa = CS_ETM_ISA_A64; + break; + case ocsd_isa_arm: + packet->isa = CS_ETM_ISA_A32; + break; + case ocsd_isa_thumb2: + packet->isa = CS_ETM_ISA_T32; + break; + case ocsd_isa_tee: + case ocsd_isa_jazelle: + case ocsd_isa_custom: + case ocsd_isa_unknown: + default: + packet->isa = CS_ETM_ISA_UNKNOWN; + } + packet->start_addr = elem->st_addr; packet->end_addr = elem->en_addr; + packet->instr_count = elem->num_instr_range; + switch (elem->last_i_type) { case OCSD_INSTR_BR: case OCSD_INSTR_BR_INDIRECT: @@ -336,6 +361,8 @@ cs_etm_decoder__buffer_range(struct cs_etm_decoder *decoder, break; } + packet->last_instr_size = elem->last_instr_sz; + return ret; } diff --git a/tools/perf/util/cs-etm-decoder/cs-etm-decoder.h b/tools/perf/util/cs-etm-decoder/cs-etm-decoder.h index 612b575..9351bd1 100644 --- a/tools/perf/util/cs-etm-decoder/cs-etm-decoder.h +++ b/tools/perf/util/cs-etm-decoder/cs-etm-decoder.h @@ -28,11 +28,21 @@ enum cs_etm_sample_type { CS_ETM_TRACE_ON = 1 << 1, }; +enum cs_etm_isa { + CS_ETM_ISA_UNKNOWN, + CS_ETM_ISA_A64, + CS_ETM_ISA_A32, + CS_ETM_ISA_T32, +}; + struct cs_etm_packet { enum cs_etm_sample_type sample_type; + enum cs_etm_isa isa; u64 start_addr; u64 end_addr; + u32 instr_count; u8 last_instr_taken_branch; + u8 last_instr_size; u8 exc; u8 exc_ret; int cpu; diff --git a/tools/perf/util/cs-etm.c b/tools/perf/util/cs-etm.c index 2ae6402..fcaa73f 100644 --- a/tools/perf/util/cs-etm.c +++ b/tools/perf/util/cs-etm.c @@ -31,14 +31,6 @@ #define MAX_TIMESTAMP (~0ULL) -/* - * A64 instructions are always 4 bytes - * - * Only A64 is supported, so can use this constant for converting between - * addresses and instruction counts, calculting offsets etc - */ -#define A64_INSTR_SIZE 4 - struct cs_etm_auxtrace { struct auxtrace auxtrace; struct auxtrace_queues queues; @@ -492,21 +484,16 @@ static inline void cs_etm__reset_last_branch_rb(struct cs_etm_queue *etmq) etmq->last_branch_rb->nr = 0; } -static inline u64 cs_etm__last_executed_instr(struct cs_etm_packet *packet) -{ - /* Returns 0 for the CS_ETM_TRACE_ON packet */ - if (packet->sample_type == CS_ETM_TRACE_ON) - return 0; +static inline int cs_etm__t32_instr_size(struct cs_etm_queue *etmq, + u64 addr) { + u8 instrBytes[2]; - /* - * The packet records the execution range with an exclusive end address - * - * A64 instructions are constant size, so the last executed - * instruction is A64_INSTR_SIZE before the end address - * Will need to do instruction level decode for T32 instructions as - * they can be variable size (not yet supported). + cs_etm__mem_access(etmq, addr, ARRAY_SIZE(instrBytes), instrBytes); + /* T32 instruction size is indicated by bits[15:11] of the first + * 16-bit word of the instruction: 0b11101, 0b11110 and 0b11111 + * denote a 32-bit instruction. */ - return packet->end_addr - A64_INSTR_SIZE; + return ((instrBytes[1] & 0xF8) >= 0xE8) ? 4 : 2; } static inline u64 cs_etm__first_executed_instr(struct cs_etm_packet *packet) @@ -518,27 +505,32 @@ static inline u64 cs_etm__first_executed_instr(struct cs_etm_packet *packet) return packet->start_addr; } -static inline u64 cs_etm__instr_count(const struct cs_etm_packet *packet) +static inline +u64 cs_etm__last_executed_instr(const struct cs_etm_packet *packet) { - /* - * Only A64 instructions are currently supported, so can get - * instruction count by dividing. - * Will need to do instruction level decode for T32 instructions as - * they can be variable size (not yet supported). - */ - return (packet->end_addr - packet->start_addr) / A64_INSTR_SIZE; + /* Returns 0 for the CS_ETM_TRACE_ON packet */ + if (packet->sample_type == CS_ETM_TRACE_ON) + return 0; + + return packet->end_addr - packet->last_instr_size; } -static inline u64 cs_etm__instr_addr(const struct cs_etm_packet *packet, +static inline u64 cs_etm__instr_addr(struct cs_etm_queue *etmq, + const struct cs_etm_packet *packet, u64 offset) { - /* - * Only A64 instructions are currently supported, so can get - * instruction address by muliplying. - * Will need to do instruction level decode for T32 instructions as - * they can be variable size (not yet supported). - */ - return packet->start_addr + offset * A64_INSTR_SIZE; + if (packet->isa == CS_ETM_ISA_T32) { + u64 addr = packet->start_addr; + + while (offset > 0) { + addr += cs_etm__t32_instr_size(etmq, addr); + offset--; + } + return addr; + } + + /* Assume a 4 byte instruction size (A32/A64) */ + return packet->start_addr + offset * 4; } static void cs_etm__update_last_branch_rb(struct cs_etm_queue *etmq) @@ -867,9 +859,8 @@ static int cs_etm__sample(struct cs_etm_queue *etmq) struct cs_etm_auxtrace *etm = etmq->etm; struct cs_etm_packet *tmp; int ret; - u64 instrs_executed; + u64 instrs_executed = etmq->packet->instr_count; - instrs_executed = cs_etm__instr_count(etmq->packet); etmq->period_instructions += instrs_executed; /* @@ -899,7 +890,7 @@ static int cs_etm__sample(struct cs_etm_queue *etmq) * executed, but PC has not advanced to next instruction) */ u64 offset = (instrs_executed - instrs_over - 1); - u64 addr = cs_etm__instr_addr(etmq->packet, offset); + u64 addr = cs_etm__instr_addr(etmq, etmq->packet, offset); ret = cs_etm__synth_instruction_sample( etmq, addr, etm->instructions_sample_period); -- 2.7.4

6 years, 4 months

Enabling Coresight in atomic context.

by Mike Bazov

Greetings, I'm trying to enable the Coresight ETMv4 trace from kernel mode. I saw there is no documentation on how to do this, except using the sysfs user mode interface and perf. To overcome this i looked a little bit in the coresight.h header file, and came to these APIs: > extern int coresight_enable(struct coresight_device *csdev); > > extern void coresight_disable(struct coresight_device *csdev); And the sysfs implementation uses these APIs when enabing/disabling the trace code, so i thought this could suit my needs. The next problem was actually getting the coresight devices data structures, which aren't exported and are actually provided internally to perf and sysfs. So i exported the coresight bus type: struct bus_type coresight_bustype = { > > .name = "coresight", > > }; > > EXPORT_SYMBOL(coresight_bustype); And enumerated the bus and just looked for the type CORESIGHT_DEV_TYPE_SOURCE, since the only source on my board is ETMv4, there isn't any conflicts so i should get only ETMv4s. This worked, and indeed i collected a coresight device for every CPU, and enabled the trace successfully while being in *non-atomic context*. The problem is, i must enable the trace in *atomic context synchronously *on the current thread's CPU(i can't issue a work-queue to enable the trace for me). So.. i got many BUG() errors because of non-atomic API usage, for example, the allocation of the TMC-ETR buffer: > dma_alloc_coherent(drvdata->dev, etr_buf->size, > &flat_buf->daddr, *GFP_KERNEL*); So my questions are: 1) Is there a more documented way of enabling coresight from kernel mode? i believe i achieved this using cheats. 2) I see there is no exported kernel API to config the coresight trace attributes(for example, filter EL0). I can only do so from sysfs.. am i missing something? 3) Are there any plans to make the Coresight infrastructure atomic-context friendly? If there are, is the development in progress? if not.. how would you suggest tackling the issues i've described in this message? Thank you!

6 years, 4 months

Decoding STM traces with OpenCSD test programs

by Mathieu Poirier

Good day Leo, Please meet my new friend Christophe. He is from ST-Micro and is currently working on integrating the CS framework in their next platform. I would be grateful if you could share with him the instructions you have used to get STM doing on the dragon board and how you got the snapshot decoder to decode the traces. Christophe is also looking to get a trace snapshot that is known to be working properly to test in his environment. Many thanks, Mathieu

6 years, 4 months

[PATCH] perf: Support for Arm A32/T32 instruction sets in CoreSight trace

by Robert Walker

This patch adds support for generating instruction samples from trace of AArch32 programs using the A32 and T32 instruction sets. T32 has variable 2 or 4 byte instruction size, so the conversion between addresses and instruction counts requires extra information from the trace decoder, requiring version 0.9.1 of OpenCSD. A check for the new struct member has been added to the feature check for OpenCSD. Signed-off-by: Robert Walker <robert.walker(a)arm.com> --- tools/build/feature/test-libopencsd.c | 7 +++ tools/perf/util/cs-etm-decoder/cs-etm-decoder.c | 27 ++++++++++ tools/perf/util/cs-etm-decoder/cs-etm-decoder.h | 10 ++++ tools/perf/util/cs-etm.c | 71 +++++++++++-------------- 4 files changed, 75 insertions(+), 40 deletions(-) diff --git a/tools/build/feature/test-libopencsd.c b/tools/build/feature/test-libopencsd.c index 5ff1246..d96b2df 100644 --- a/tools/build/feature/test-libopencsd.c +++ b/tools/build/feature/test-libopencsd.c @@ -3,6 +3,13 @@ int main(void) { + /* + * Requires ocsd_generic_trace_elem.num_instr_range introduced in + * OpenCSD 0.9 + */ + ocsd_generic_trace_elem elem; + (void)elem.num_instr_range; + (void)ocsd_get_version(); return 0; } diff --git a/tools/perf/util/cs-etm-decoder/cs-etm-decoder.c b/tools/perf/util/cs-etm-decoder/cs-etm-decoder.c index 938def6..73d8384 100644 --- a/tools/perf/util/cs-etm-decoder/cs-etm-decoder.c +++ b/tools/perf/util/cs-etm-decoder/cs-etm-decoder.c @@ -263,9 +263,12 @@ static void cs_etm_decoder__clear_buffer(struct cs_etm_decoder *decoder) decoder->tail = 0; decoder->packet_count = 0; for (i = 0; i < MAX_BUFFER; i++) { + decoder->packet_buffer[i].isa = CS_ETM_ISA_UNKNOWN; decoder->packet_buffer[i].start_addr = CS_ETM_INVAL_ADDR; decoder->packet_buffer[i].end_addr = CS_ETM_INVAL_ADDR; + decoder->packet_buffer[i].instr_count = 0; decoder->packet_buffer[i].last_instr_taken_branch = false; + decoder->packet_buffer[i].last_instr_size = 0; decoder->packet_buffer[i].exc = false; decoder->packet_buffer[i].exc_ret = false; decoder->packet_buffer[i].cpu = INT_MIN; @@ -294,11 +297,13 @@ cs_etm_decoder__buffer_packet(struct cs_etm_decoder *decoder, decoder->packet_count++; decoder->packet_buffer[et].sample_type = sample_type; + decoder->packet_buffer[et].isa = CS_ETM_ISA_UNKNOWN; decoder->packet_buffer[et].exc = false; decoder->packet_buffer[et].exc_ret = false; decoder->packet_buffer[et].cpu = *((int *)inode->priv); decoder->packet_buffer[et].start_addr = CS_ETM_INVAL_ADDR; decoder->packet_buffer[et].end_addr = CS_ETM_INVAL_ADDR; + decoder->packet_buffer[et].instr_count = 0; if (decoder->packet_count == MAX_BUFFER - 1) return OCSD_RESP_WAIT; @@ -321,8 +326,28 @@ cs_etm_decoder__buffer_range(struct cs_etm_decoder *decoder, packet = &decoder->packet_buffer[decoder->tail]; + switch (elem->isa) { + case ocsd_isa_aarch64: + packet->isa = CS_ETM_ISA_A64; + break; + case ocsd_isa_arm: + packet->isa = CS_ETM_ISA_A32; + break; + case ocsd_isa_thumb2: + packet->isa = CS_ETM_ISA_T32; + break; + case ocsd_isa_tee: + case ocsd_isa_jazelle: + case ocsd_isa_custom: + case ocsd_isa_unknown: + default: + packet->isa = CS_ETM_ISA_UNKNOWN; + } + packet->start_addr = elem->st_addr; packet->end_addr = elem->en_addr; + packet->instr_count = elem->num_instr_range; + switch (elem->last_i_type) { case OCSD_INSTR_BR: case OCSD_INSTR_BR_INDIRECT: @@ -336,6 +361,8 @@ cs_etm_decoder__buffer_range(struct cs_etm_decoder *decoder, break; } + packet->last_instr_size = elem->last_instr_sz; + return ret; } diff --git a/tools/perf/util/cs-etm-decoder/cs-etm-decoder.h b/tools/perf/util/cs-etm-decoder/cs-etm-decoder.h index 612b575..9a10eda 100644 --- a/tools/perf/util/cs-etm-decoder/cs-etm-decoder.h +++ b/tools/perf/util/cs-etm-decoder/cs-etm-decoder.h @@ -28,11 +28,21 @@ enum cs_etm_sample_type { CS_ETM_TRACE_ON = 1 << 1, }; +enum cs_etm_isa { + CS_ETM_ISA_UNKNOWN, + CS_ETM_ISA_A64, + CS_ETM_ISA_A32, + CS_ETM_ISA_T32, +}; + struct cs_etm_packet { enum cs_etm_sample_type sample_type; + enum cs_etm_isa isa; u64 start_addr; u64 end_addr; + u64 instr_count; u8 last_instr_taken_branch; + u8 last_instr_size; u8 exc; u8 exc_ret; int cpu; diff --git a/tools/perf/util/cs-etm.c b/tools/perf/util/cs-etm.c index 2ae6402..579f52a 100644 --- a/tools/perf/util/cs-etm.c +++ b/tools/perf/util/cs-etm.c @@ -31,14 +31,6 @@ #define MAX_TIMESTAMP (~0ULL) -/* - * A64 instructions are always 4 bytes - * - * Only A64 is supported, so can use this constant for converting between - * addresses and instruction counts, calculting offsets etc - */ -#define A64_INSTR_SIZE 4 - struct cs_etm_auxtrace { struct auxtrace auxtrace; struct auxtrace_queues queues; @@ -492,21 +484,16 @@ static inline void cs_etm__reset_last_branch_rb(struct cs_etm_queue *etmq) etmq->last_branch_rb->nr = 0; } -static inline u64 cs_etm__last_executed_instr(struct cs_etm_packet *packet) -{ - /* Returns 0 for the CS_ETM_TRACE_ON packet */ - if (packet->sample_type == CS_ETM_TRACE_ON) - return 0; +static inline int cs_etm__t32_instr_size(struct cs_etm_queue *etmq, + u64 addr) { + u8 instrBytes[2]; - /* - * The packet records the execution range with an exclusive end address - * - * A64 instructions are constant size, so the last executed - * instruction is A64_INSTR_SIZE before the end address - * Will need to do instruction level decode for T32 instructions as - * they can be variable size (not yet supported). + cs_etm__mem_access(etmq, addr, 2, instrBytes); + /* T32 instruction size is indicated by bits[15:11] of the first + * 16-bit word of the instruction: 0b11101, 0b11110 and 0b11111 + * denote a 32-bit instruction. */ - return packet->end_addr - A64_INSTR_SIZE; + return ((instrBytes[1] & 0xF8) >= 0xE8) ? 4 : 2; } static inline u64 cs_etm__first_executed_instr(struct cs_etm_packet *packet) @@ -518,27 +505,32 @@ static inline u64 cs_etm__first_executed_instr(struct cs_etm_packet *packet) return packet->start_addr; } -static inline u64 cs_etm__instr_count(const struct cs_etm_packet *packet) +static inline +u64 cs_etm__last_executed_instr(const struct cs_etm_packet *packet) { - /* - * Only A64 instructions are currently supported, so can get - * instruction count by dividing. - * Will need to do instruction level decode for T32 instructions as - * they can be variable size (not yet supported). - */ - return (packet->end_addr - packet->start_addr) / A64_INSTR_SIZE; + /* Returns 0 for the CS_ETM_TRACE_ON packet */ + if (packet->sample_type == CS_ETM_TRACE_ON) + return 0; + + return packet->end_addr - packet->last_instr_size; } -static inline u64 cs_etm__instr_addr(const struct cs_etm_packet *packet, +static inline u64 cs_etm__instr_addr(struct cs_etm_queue *etmq, + const struct cs_etm_packet *packet, u64 offset) { - /* - * Only A64 instructions are currently supported, so can get - * instruction address by muliplying. - * Will need to do instruction level decode for T32 instructions as - * they can be variable size (not yet supported). - */ - return packet->start_addr + offset * A64_INSTR_SIZE; + if (packet->isa == CS_ETM_ISA_T32) { + u64 addr = packet->start_addr; + + while (offset > 0) { + addr += cs_etm__t32_instr_size(etmq, addr); + offset--; + } + return addr; + } + + /* Assume a 4 byte instruction size (A32/A64) */ + return packet->start_addr + offset * 4; } static void cs_etm__update_last_branch_rb(struct cs_etm_queue *etmq) @@ -867,9 +859,8 @@ static int cs_etm__sample(struct cs_etm_queue *etmq) struct cs_etm_auxtrace *etm = etmq->etm; struct cs_etm_packet *tmp; int ret; - u64 instrs_executed; + u64 instrs_executed = etmq->packet->instr_count; - instrs_executed = cs_etm__instr_count(etmq->packet); etmq->period_instructions += instrs_executed; /* @@ -899,7 +890,7 @@ static int cs_etm__sample(struct cs_etm_queue *etmq) * executed, but PC has not advanced to next instruction) */ u64 offset = (instrs_executed - instrs_over - 1); - u64 addr = cs_etm__instr_addr(etmq->packet, offset); + u64 addr = cs_etm__instr_addr(etmq, etmq->packet, offset); ret = cs_etm__synth_instruction_sample( etmq, addr, etm->instructions_sample_period); -- 2.7.4

6 years, 4 months

хпфн

by всэ

цхэ гхжщ мюма

6 years, 5 months

[PATCH v3 00/13] coresight: perf: Support for TMC ETR backend

by Suzuki K Poulose

This series adds the support for using the tmc-etr in perf mode for storing the trace to system RAM. The ETR uses a separate buffer (double buffering) for storing the trace. This is copied back to the ring buffer when the event is stopped. We try to match the ETR buffer to the larger of perf ring buffer or the size configured for the ETR via sysfs. This allows tuning the buffer size to prevent overflows and loosing trace data, as we don't have overflow interrupt support (yet). Applies on coresight/next Changes since v2: - Split hotplug lock cleanup from per-cpu path management - Add support for tracing on hotplugged CPUs (based on patch by Mathieu) - Fix handling of perf mode in etb10 driver - Fix CS_UNLOCK() typo in ETR perf driver - Rename call backs from tmc_etr_<op>_perf_buffer => tmc_<op>_etr_buffer to be consistent with ETF driver - Modify the commit description for removing set_buffer call back. Changes since v1 : - Fix path for each CPU where the event might be recorded. - Handle errors in enabling source - Remove set_buffer callback to avoid complicating the error handling. - Fix a bug in handling sysfs mode specific buffers Changes since [0] : - Drop buffer rotation logic for etr-buf - Do not use perf ring buffer. (Add support later) - Fix handling of sink, preventing mixed modes of operation. [0] - TMC ETR perf support - http://lists.infradead.org/pipermail/linux-arm-kernel/2018-May/574875.html Suzuki K Poulose (13): coresight: Fix handling of sinks coresight: etb10: Fix handling of perf mode coresight: perf: Fix per cpu path management coresight: perf: Avoid unncessary CPU hotplug read lock coresight: perf: Allow tracing on hotplugged CPUs coresight: perf: Disable trace path upon source error coresight: tmc-etr: Handle driver mode specific ETR buffers coresight: tmc-etr: Relax collection of trace from sysfs mode coresight: Convert driver messages to dev_dbg coresight: perf: Remove reset_buffer call back for sinks coresight: perf: Add helper to retrieve sink configuration coresight: perf: Remove set_buffer call back coresight: etm-perf: Add support for ETR backend .../coresight/coresight-dynamic-replicator.c | 4 +- drivers/hwtracing/coresight/coresight-etb10.c | 100 +++---- drivers/hwtracing/coresight/coresight-etm-perf.c | 134 +++++---- drivers/hwtracing/coresight/coresight-etm-perf.h | 26 ++ drivers/hwtracing/coresight/coresight-etm3x.c | 4 +- drivers/hwtracing/coresight/coresight-etm4x.c | 4 +- drivers/hwtracing/coresight/coresight-funnel.c | 4 +- drivers/hwtracing/coresight/coresight-priv.h | 2 +- drivers/hwtracing/coresight/coresight-replicator.c | 4 +- drivers/hwtracing/coresight/coresight-stm.c | 4 +- drivers/hwtracing/coresight/coresight-tmc-etf.c | 94 +++--- drivers/hwtracing/coresight/coresight-tmc-etr.c | 327 ++++++++++++++++++--- drivers/hwtracing/coresight/coresight-tmc.c | 4 +- drivers/hwtracing/coresight/coresight-tmc.h | 4 + drivers/hwtracing/coresight/coresight-tpiu.c | 6 +- drivers/hwtracing/coresight/coresight.c | 31 +- include/linux/coresight.h | 12 +- 17 files changed, 517 insertions(+), 247 deletions(-) -- 2.7.4

6 years, 5 months

2025

2024

2023

2022

2021

2020

2019

2018

2017

2016

2015

CoreSight August 2018