From: Anshuman Khandual <anshuman.khandual@arm.com>
To: Suzuki K Poulose <suzuki.poulose@arm.com>,
linux-arm-kernel@lists.infradead.org
Cc: linux-kernel@vger.kernel.org, maz@kernel.org,
catalin.marinas@arm.com, mark.rutland@arm.com,
james.morse@arm.com, leo.yan@linaro.org, mike.leach@linaro.org,
mathieu.poirier@linaro.org, will@kernel.org,
lcherian@marvell.com, coresight@lists.linaro.org
Subject: Re: [PATCH v2 09/17] coresight: trbe: Workaround TRBE errata overwrite in FILL mode
Date: Thu, 23 Sep 2021 11:43:11 +0530 [thread overview]
Message-ID: <7ae6b3a6-1fa2-a655-9347-2263a1701c91@arm.com> (raw)
In-Reply-To: <20210921134121.2423546-10-suzuki.poulose@arm.com>
On 9/21/21 7:11 PM, Suzuki K Poulose wrote:
> ARM Neoverse-N2 (#2139208) and Cortex-A710(##2119858) suffers from
> an erratum, which when triggered, might cause the TRBE to overwrite
> the trace data already collected in FILL mode, in the event of a WRAP.
> i.e, the TRBE doesn't stop writing the data, instead wraps to the base
> and could write upto 3 cache line size worth trace. Thus, this could
> corrupt the trace at the "BASE" pointer.
>
> The workaround is to program the write pointer 256bytes from the
3 cache lines = 256 bytes on all implementation which might have TRBE ?
OR this skid bytes should be derived from the platform cache line size
instead.
> base, such that if the erratum is triggered, it doesn't overwrite
> the trace data that was captured. This skipped region could be
> padded with ignore packets at the end of the session, so that
> the decoder sees a continuous buffer with some padding at the
> beginning. The trace data written at the base is considered
> lost as the limit could have been in the middle of the perf
> ring buffer, and jumping to the "base" is not acceptable.
> We set the flags already to indicate that some amount of trace
> was lost during the FILL event IRQ. So this is fine.
Via PERF_AUX_FLAG_TRUNCATED ? Should be specified here to be clear.
>
> One important change with the work around is, we program the
> TRBBASER_EL1 to current page where we are allowed to write.
> Otherwise, it could overwrite a region that may be consumed
> by the perf. Towards this, we always make sure that the
> "handle->head" and thus the trbe_write is PAGE_SIZE aligned,
> so that we can set the BASE to the PAGE base and move the
> TRBPTR to the 256bytes offset.
>
> Cc: Mike Leach <mike.leach@linaro.org>
> Cc: Mathieu Poirier <mathieu.poirier@linaro.org>
> Cc: Anshuman Khandual <anshuman.khandual@arm.com>
> Cc: Leo Yan <leo.yan@linaro.org>
> Signed-off-by: Suzuki K Poulose <suzuki.poulose@arm.com>
> ---
> Change since v1:
> - Updated comment with ASCII art
> - Add _BYTES suffix for the space to skip for the work around.
> ---
> drivers/hwtracing/coresight/coresight-trbe.c | 144 +++++++++++++++++--
> 1 file changed, 132 insertions(+), 12 deletions(-)
>
> diff --git a/drivers/hwtracing/coresight/coresight-trbe.c b/drivers/hwtracing/coresight/coresight-trbe.c
> index f569010c672b..983dd5039e52 100644
> --- a/drivers/hwtracing/coresight/coresight-trbe.c
> +++ b/drivers/hwtracing/coresight/coresight-trbe.c
> @@ -16,6 +16,7 @@
> #define pr_fmt(fmt) DRVNAME ": " fmt
>
> #include <asm/barrier.h>
> +#include <asm/cpufeature.h>
> #include <asm/cputype.h>
>
> #include "coresight-self-hosted-trace.h"
> @@ -84,9 +85,17 @@ struct trbe_buf {
> * per TRBE instance, we keep track of the list of errata that
> * affects the given instance of the TRBE.
> */
> -#define TRBE_ERRATA_MAX 0
> +#define TRBE_WORKAROUND_OVERWRITE_FILL_MODE 0
> +#define TRBE_ERRATA_MAX 1
> +
> +/*
> + * Safe limit for the number of bytes that may be overwritten
> + * when the erratum is triggered.
> + */
> +#define TRBE_WORKAROUND_OVERWRITE_FILL_MODE_SKIP_BYTES 256
As mentioned earlier, does it depend on the platform cache line size ?
Otherwise if the skip bytes is something platform independent, should
be mentioned here in a comment.
>
> static unsigned long trbe_errata_cpucaps[TRBE_ERRATA_MAX] = {
> + [TRBE_WORKAROUND_OVERWRITE_FILL_MODE] = ARM64_WORKAROUND_TRBE_OVERWRITE_FILL_MODE,
> };
>
> /*
> @@ -519,10 +528,13 @@ static void trbe_enable_hw(struct trbe_buf *buf)
> set_trbe_limit_pointer_enabled(buf->trbe_limit);
> }
>
> -static enum trbe_fault_action trbe_get_fault_act(u64 trbsr)
> +static enum trbe_fault_action trbe_get_fault_act(struct perf_output_handle *handle,
> + u64 trbsr)
> {
> int ec = get_trbe_ec(trbsr);
> int bsc = get_trbe_bsc(trbsr);
> + struct trbe_buf *buf = etm_perf_sink_config(handle);
> + struct trbe_cpudata *cpudata = buf->cpudata;
Passing down the perf handle to derive trbe_cpudata seems to be right.
>
> WARN_ON(is_trbe_running(trbsr));
> if (is_trbe_trg(trbsr) || is_trbe_abort(trbsr))
> @@ -531,10 +543,16 @@ static enum trbe_fault_action trbe_get_fault_act(u64 trbsr)
> if ((ec == TRBE_EC_STAGE1_ABORT) || (ec == TRBE_EC_STAGE2_ABORT))
> return TRBE_FAULT_ACT_FATAL;
>
> - if (is_trbe_wrap(trbsr) && (ec == TRBE_EC_OTHERS) && (bsc == TRBE_BSC_FILLED)) {
> - if (get_trbe_write_pointer() == get_trbe_base_pointer())
> - return TRBE_FAULT_ACT_WRAP;
> - }
> + /*
> + * If the trbe is affected by TRBE_WORKAROUND_OVERWRITE_FILL_MODE,
> + * it might write data after a WRAP event in the fill mode.
> + * Thus the check TRBPTR == TRBBASER will not be honored.
> + */
Needs bit formatting/alignment cleanup.
> + if ((is_trbe_wrap(trbsr) && (ec == TRBE_EC_OTHERS) && (bsc == TRBE_BSC_FILLED)) &&
> + (trbe_has_erratum(cpudata, TRBE_WORKAROUND_OVERWRITE_FILL_MODE) ||
> + get_trbe_write_pointer() == get_trbe_base_pointer()))
> + return TRBE_FAULT_ACT_WRAP;
> +
Right, TRBE without the errata should continue to have the write
pointer = base pointer check. Could all TRBE errata checks like
the following be shortened (without the workaround index) for
better readability ? But not something very important.
trbe_has_erratum(cpudata, TRBE_WORKAROUND_OVERWRITE_FILL_MODE)
> return TRBE_FAULT_ACT_SPURIOUS;
> }
>
> @@ -544,6 +562,8 @@ static unsigned long trbe_get_trace_size(struct perf_output_handle *handle,
> {
> u64 write;
> u64 start_off, end_off;
> + u64 size;
> + u64 overwrite_skip = TRBE_WORKAROUND_OVERWRITE_FILL_MODE_SKIP_BYTES;
>
> /*
> * If the TRBE has wrapped around the write pointer has
> @@ -559,7 +579,18 @@ static unsigned long trbe_get_trace_size(struct perf_output_handle *handle,
>
> if (WARN_ON_ONCE(end_off < start_off))
> return 0;
> - return (end_off - start_off);
> +
> + size = end_off - start_off;
> + /*
> + * If the TRBE is affected by the following erratum, we must fill
> + * the space we skipped with IGNORE packets. And we are always
> + * guaranteed to have at least a PAGE_SIZE space in the buffer.
> + */
> + if (trbe_has_erratum(buf->cpudata, TRBE_WORKAROUND_OVERWRITE_FILL_MODE) &&
> + !WARN_ON(size < overwrite_skip))
> + __trbe_pad_buf(buf, start_off, overwrite_skip);
> +
> + return size;
> }
>
> static void *arm_trbe_alloc_buffer(struct coresight_device *csdev,
> @@ -678,7 +709,7 @@ static unsigned long arm_trbe_update_buffer(struct coresight_device *csdev,
> clr_trbe_irq();
> isb();
>
> - act = trbe_get_fault_act(status);
> + act = trbe_get_fault_act(handle, status);
> /*
> * If this was not due to a WRAP event, we have some
> * errors and as such buffer is empty.
> @@ -702,21 +733,95 @@ static unsigned long arm_trbe_update_buffer(struct coresight_device *csdev,
> return size;
> }
>
> +
> +static int trbe_apply_work_around_before_enable(struct trbe_buf *buf)
> +{
> + /*
> + * TRBE_WORKAROUND_OVERWRITE_FILL_MODE causes the TRBE to overwrite a few cache
few cache lines = 3 cache lines ?
> + * line size from the "TRBBASER_EL1" in the event of a "FILL".
> + * Thus, we could loose some amount of the trace at the base.
> + *
> + * Before Fix:
> + *
> + * normal-BASE head normal-PTR tail normal-LIMIT
> + * | \/ /
> + * -------------------------------------------------------------
> + * | | |xyzdefghij..|... tuvw| |
> + * -------------------------------------------------------------
> + * / | \
> + * After Fix-> TRBBASER TRBPTR TRBLIMITR.LIMIT
> + *
> + * In the normal course of action, we would set the TRBBASER to the
> + * beginning of the ring-buffer (normal-BASE). But with the erratum,
> + * the TRBE could overwrite the contents at the "normal-BASE", after
> + * hitting the "normal-LIMIT", since it doesn't stop as expected. And
> + * this is wrong. So we must always make sure that the TRBBASER is
> + * within the region [head, head+size].
> + *
> + * Also, we would set the TRBPTR to head (after adjusting for
> + * alignment) at normal-PTR. This would mean that the last few bytes
> + * of the trace (say, "xyz") might overwrite the first few bytes of
> + * trace written ("abc"). More importantly they will appear in what\
> + * userspace sees as the beginning of the trace, which is wrong. We may
> + * not always have space to move the latest trace "xyz" to the correct
> + * order as it must appear beyond the LIMIT. (i.e, [head..head+size].
> + * Thus it is easier to ignore those bytes than to complicate the
> + * driver to move it, assuming that the erratum was triggered and doing
> + * additional checks to see if there is indeed allowed space at
> + * TRBLIMITR.LIMIT.
> + *
> + * To summarize, with the work around:
> + *
> + * - We always align the offset for the next session to PAGE_SIZE
> + * (This is to ensure we can program the TRBBASER to this offset
> + * within the region [head...head+size]).
> + *
> + * - At TRBE enable:
> + * - Set the TRBBASER to the page aligned offset of the current
> + * proposed write offset. (which is guaranteed to be aligned
> + * as above)
> + * - Move the TRBPTR to skip first 256bytes (that might be
> + * overwritten with the erratum). This ensures that the trace
> + * generated in the session is not re-written.
> + *
> + * - At trace collection:
> + * - Pad the 256bytes skipped above again with IGNORE packets.
> + */
> + if (trbe_has_erratum(buf->cpudata, TRBE_WORKAROUND_OVERWRITE_FILL_MODE)) {
> + if (WARN_ON(!IS_ALIGNED(buf->trbe_write, PAGE_SIZE)))
> + return -EINVAL;
> + buf->trbe_hw_base = buf->trbe_write;
> + buf->trbe_write += TRBE_WORKAROUND_OVERWRITE_FILL_MODE_SKIP_BYTES;
> + }
> +
> + return 0;
> +}
> +
> static int __arm_trbe_enable(struct trbe_buf *buf,
> struct perf_output_handle *handle)
> {
> + int ret = 0;
> +
> perf_aux_output_flag(handle, PERF_AUX_FLAG_CORESIGHT_FORMAT_RAW);
> buf->trbe_limit = compute_trbe_buffer_limit(handle);
> buf->trbe_write = buf->trbe_base + PERF_IDX2OFF(handle->head, buf);
> if (buf->trbe_limit == buf->trbe_base) {
> - trbe_stop_and_truncate_event(handle);
> - return -ENOSPC;
> + ret = -ENOSPC;
> + goto err;
> }
> /* Set the base of the TRBE to the buffer base */
> buf->trbe_hw_base = buf->trbe_base;
> +
> + ret = trbe_apply_work_around_before_enable(buf);
> + if (ret)
> + goto err;
> +
> *this_cpu_ptr(buf->cpudata->drvdata->handle) = handle;
> trbe_enable_hw(buf);
> return 0;
> +err:
> + trbe_stop_and_truncate_event(handle);
> + return ret;
> }
>
> static int arm_trbe_enable(struct coresight_device *csdev, u32 mode, void *data)
> @@ -860,7 +965,7 @@ static irqreturn_t arm_trbe_irq_handler(int irq, void *dev)
> if (!is_perf_trbe(handle))
> return IRQ_NONE;
>
> - act = trbe_get_fault_act(status);
> + act = trbe_get_fault_act(handle, status);
> switch (act) {
> case TRBE_FAULT_ACT_WRAP:
> truncated = !!trbe_handle_overflow(handle);
> @@ -1000,7 +1105,22 @@ static void arm_trbe_probe_cpu(void *info)
> }
>
> trbe_check_errata(cpudata);
> - cpudata->trbe_align = cpudata->trbe_hw_align;
> + /*
> + * If the TRBE is affected by erratum TRBE_WORKAROUND_OVERWRITE_FILL_MODE,
> + * we must always program the TBRPTR_EL1, 256bytes from a page
> + * boundary, with TRBBASER_EL1 set to the page, to prevent
> + * TRBE over-writing 256bytes at TRBBASER_EL1 on FILL event.
> + *
> + * Thus make sure we always align our write pointer to a PAGE_SIZE,
> + * which also guarantees that we have at least a PAGE_SIZE space in
> + * the buffer (TRBLIMITR is PAGE aligned) and thus we can skip
> + * the required bytes at the base.
> + */
> + if (trbe_has_erratum(cpudata, TRBE_WORKAROUND_OVERWRITE_FILL_MODE))
> + cpudata->trbe_align = PAGE_SIZE;
> + else
> + cpudata->trbe_align = cpudata->trbe_hw_align;
> +
But like trbe_apply_work_around_before_enable(), trbe_align assignment
should also be wrapped inside a new helper which should contain these
comments and conditional block. Because it makes sense to have errata
work arounds in the leaf level helper functions, rather than TRBE core
operations.
> cpudata->trbe_flag = get_trbe_flag_update(trbidr);
> cpudata->cpu = cpu;
> cpudata->drvdata = drvdata;
>
next prev parent reply other threads:[~2021-09-23 6:12 UTC|newest]
Thread overview: 62+ messages / expand[flat|nested] mbox.gz Atom feed top
2021-09-21 13:41 [PATCH v2 00/17] arm64: Self-hosted trace related errata workarounds Suzuki K Poulose
2021-09-21 13:41 ` [PATCH v2 01/17] coresight: trbe: Fix incorrect access of the sink specific data Suzuki K Poulose
2021-09-22 5:41 ` Anshuman Khandual
2021-09-30 17:57 ` Mathieu Poirier
2021-09-21 13:41 ` [PATCH v2 02/17] coresight: trbe: Add infrastructure for Errata handling Suzuki K Poulose
2021-09-22 6:47 ` Anshuman Khandual
2021-10-05 16:46 ` Mathieu Poirier
2021-09-21 13:41 ` [PATCH v2 03/17] coresight: trbe: Add a helper to calculate the trace generated Suzuki K Poulose
2021-09-30 17:54 ` Mathieu Poirier
2021-10-01 8:36 ` Suzuki K Poulose
2021-10-01 15:15 ` Mathieu Poirier
2021-10-01 15:22 ` Suzuki K Poulose
2021-09-21 13:41 ` [PATCH v2 04/17] coresight: trbe: Add a helper to pad a given buffer area Suzuki K Poulose
2021-09-21 13:41 ` [PATCH v2 05/17] coresight: trbe: Decouple buffer base from the hardware base Suzuki K Poulose
2021-09-21 13:41 ` [PATCH v2 06/17] coresight: trbe: Allow driver to choose a different alignment Suzuki K Poulose
2021-09-21 13:41 ` [PATCH v2 07/17] arm64: Add Neoverse-N2, Cortex-A710 CPU part definition Suzuki K Poulose
2021-09-22 6:57 ` Anshuman Khandual
2021-09-21 13:41 ` [PATCH v2 08/17] arm64: Add erratum detection for TRBE overwrite in FILL mode Suzuki K Poulose
2021-09-21 13:41 ` [PATCH v2 09/17] coresight: trbe: Workaround TRBE errata " Suzuki K Poulose
2021-09-23 6:13 ` Anshuman Khandual [this message]
2021-09-28 10:40 ` Suzuki K Poulose
2021-10-01 4:21 ` Anshuman Khandual
2021-10-01 17:15 ` Mathieu Poirier
2021-10-04 8:46 ` Suzuki K Poulose
2021-10-04 16:47 ` Mathieu Poirier
2021-09-21 13:41 ` [PATCH v2 10/17] arm64: Enable workaround for TRBE " Suzuki K Poulose
2021-09-22 7:23 ` Anshuman Khandual
2021-09-22 8:11 ` Suzuki K Poulose
2021-10-01 4:35 ` Anshuman Khandual
2021-10-07 16:09 ` Catalin Marinas
2021-09-21 13:41 ` [PATCH v2 11/17] arm64: errata: Add workaround for TSB flush failures Suzuki K Poulose
2021-09-22 7:39 ` Anshuman Khandual
2021-09-22 12:03 ` Suzuki K Poulose
2021-10-01 4:38 ` Anshuman Khandual
2021-10-07 16:10 ` Catalin Marinas
2021-09-21 13:41 ` [PATCH v2 12/17] coresight: trbe: Add a helper to fetch cpudata from perf handle Suzuki K Poulose
2021-09-22 7:59 ` Anshuman Khandual
2021-10-04 17:42 ` Mathieu Poirier
2021-10-05 22:35 ` Suzuki K Poulose
2021-10-06 17:15 ` Mathieu Poirier
2021-10-07 9:18 ` Suzuki K Poulose
2021-09-21 13:41 ` [PATCH v2 13/17] coresight: trbe: Add a helper to determine the minimum buffer size Suzuki K Poulose
2021-09-22 9:51 ` Anshuman Khandual
2021-09-21 13:41 ` [PATCH v2 14/17] coresight: trbe: Make sure we have enough space Suzuki K Poulose
2021-09-22 9:58 ` Anshuman Khandual
2021-09-22 10:16 ` Suzuki K Poulose
2021-10-01 4:40 ` Anshuman Khandual
2021-09-21 13:41 ` [PATCH v2 15/17] arm64: Add erratum detection for TRBE write to out-of-range Suzuki K Poulose
2021-09-22 10:59 ` Anshuman Khandual
2021-10-07 16:10 ` Catalin Marinas
2021-09-21 13:41 ` [PATCH v2 16/17] coresight: trbe: Work around write to out of range Suzuki K Poulose
2021-09-23 3:15 ` Anshuman Khandual
2021-09-28 10:32 ` Suzuki K Poulose
2021-10-01 4:56 ` Anshuman Khandual
2021-09-21 13:41 ` [PATCH v2 17/17] arm64: Advertise TRBE erratum workaround for write to out-of-range address Suzuki K Poulose
2021-09-22 11:03 ` Anshuman Khandual
2021-10-07 16:11 ` Catalin Marinas
2021-10-05 17:04 ` [PATCH v2 00/17] arm64: Self-hosted trace related errata workarounds Mathieu Poirier
2021-10-08 7:32 ` Will Deacon
2021-10-08 9:25 ` Suzuki K Poulose
2021-10-08 9:52 ` Will Deacon
2021-10-08 9:57 ` Suzuki K Poulose
Reply instructions:
You may reply publicly to this message via plain-text email
using any one of the following methods:
* Save the following mbox file, import it into your mail client,
and reply-to-all from there: mbox
Avoid top-posting and favor interleaved quoting:
https://en.wikipedia.org/wiki/Posting_style#Interleaved_style
* Reply using the --to, --cc, and --in-reply-to
switches of git-send-email(1):
git send-email \
--in-reply-to=7ae6b3a6-1fa2-a655-9347-2263a1701c91@arm.com \
--to=anshuman.khandual@arm.com \
--cc=catalin.marinas@arm.com \
--cc=coresight@lists.linaro.org \
--cc=james.morse@arm.com \
--cc=lcherian@marvell.com \
--cc=leo.yan@linaro.org \
--cc=linux-arm-kernel@lists.infradead.org \
--cc=linux-kernel@vger.kernel.org \
--cc=mark.rutland@arm.com \
--cc=mathieu.poirier@linaro.org \
--cc=maz@kernel.org \
--cc=mike.leach@linaro.org \
--cc=suzuki.poulose@arm.com \
--cc=will@kernel.org \
/path/to/YOUR_REPLY
https://kernel.org/pub/software/scm/git/docs/git-send-email.html
* If your mail client supports setting the In-Reply-To header
via mailto: links, try the mailto: link
Be sure your reply has a Subject: header at the top and a blank line
before the message body.
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox;
as well as URLs for NNTP newsgroup(s).