From: Andrew Halaney <ahalaney@redhat.com>
To: Akhil P Oommen <quic_akhilpo@quicinc.com>
Cc: Konrad Dybcio <konrad.dybcio@linaro.org>,
Rob Clark <robdclark@gmail.com>, Sean Paul <sean@poorly.run>,
Abhinav Kumar <quic_abhinavk@quicinc.com>,
Dmitry Baryshkov <dmitry.baryshkov@linaro.org>,
Marijn Suijten <marijn.suijten@somainline.org>,
David Airlie <airlied@gmail.com>,
Daniel Vetter <daniel@ffwll.ch>,
Rob Clark <robdclark@chromium.org>,
linux-arm-msm@vger.kernel.org, dri-devel@lists.freedesktop.org,
freedreno@lists.freedesktop.org, linux-kernel@vger.kernel.org
Subject: Re: [PATCH] drm/msm/adreno: De-spaghettify the use of memory barriers
Date: Thu, 16 May 2024 08:15:34 -0500 [thread overview]
Message-ID: <ae4a77wt3kc73ejshptldqx6ugzrqguyq7etbbu54y4avhbdlt@qyt4r6gma7ev> (raw)
In-Reply-To: <20240514183849.6lpyplifero5u35r@hu-akhilpo-hyd.qualcomm.com>
On Wed, May 15, 2024 at 12:08:49AM GMT, Akhil P Oommen wrote:
> On Wed, May 08, 2024 at 07:46:31PM +0200, Konrad Dybcio wrote:
> > Memory barriers help ensure instruction ordering, NOT time and order
> > of actual write arrival at other observers (e.g. memory-mapped IP).
> > On architectures employing weak memory ordering, the latter can be a
> > giant pain point, and it has been as part of this driver.
> >
> > Moreover, the gpu_/gmu_ accessors already use non-relaxed versions of
> > readl/writel, which include r/w (respectively) barriers.
> >
> > Replace the barriers with a readback that ensures the previous writes
> > have exited the write buffer (as the CPU must flush the write to the
> > register it's trying to read back) and subsequently remove the hack
> > introduced in commit b77532803d11 ("drm/msm/a6xx: Poll for GBIF unhalt
> > status in hw_init").
For what its worth, I've been eyeing (but haven't tested) sending some
patches to clean up dsi_phy_write_udelay/ndelay(). There's no ordering
guarantee between a writel() and a delay(), so the expected "write then
delay" sequence might not be happening.. you need to write, read, delay.
memory-barriers.txt:
5. A readX() by a CPU thread from the peripheral will complete before
any subsequent delay() loop can begin execution on the same thread.
This ensures that two MMIO register writes by the CPU to a peripheral
will arrive at least 1us apart if the first write is immediately read
back with readX() and udelay(1) is called prior to the second
writeX():
writel(42, DEVICE_REGISTER_0); // Arrives at the device...
readl(DEVICE_REGISTER_0);
udelay(1);
writel(42, DEVICE_REGISTER_1); // ...at least 1us before this.
> >
> > Fixes: b77532803d11 ("drm/msm/a6xx: Poll for GBIF unhalt status in hw_init")
> > Signed-off-by: Konrad Dybcio <konrad.dybcio@linaro.org>
> > ---
> > drivers/gpu/drm/msm/adreno/a6xx_gmu.c | 5 ++---
> > drivers/gpu/drm/msm/adreno/a6xx_gpu.c | 14 ++++----------
> > 2 files changed, 6 insertions(+), 13 deletions(-)
>
> I prefer this version compared to the v2. A helper routine is
> unnecessary here because:
> 1. there are very few scenarios where we have to read back the same
> register.
> 2. we may accidently readback a write only register.
>
> >
> > diff --git a/drivers/gpu/drm/msm/adreno/a6xx_gmu.c b/drivers/gpu/drm/msm/adreno/a6xx_gmu.c
> > index 0e3dfd4c2bc8..4135a53b55a7 100644
> > --- a/drivers/gpu/drm/msm/adreno/a6xx_gmu.c
> > +++ b/drivers/gpu/drm/msm/adreno/a6xx_gmu.c
> > @@ -466,9 +466,8 @@ static int a6xx_rpmh_start(struct a6xx_gmu *gmu)
> > int ret;
> > u32 val;
> >
> > - gmu_write(gmu, REG_A6XX_GMU_RSCC_CONTROL_REQ, 1 << 1);
> > - /* Wait for the register to finish posting */
> > - wmb();
> > + gmu_write(gmu, REG_A6XX_GMU_RSCC_CONTROL_REQ, BIT(1));
> > + gmu_read(gmu, REG_A6XX_GMU_RSCC_CONTROL_REQ);
>
> This is unnecessary because we are polling on a register on the same port below. But I think we
> can replace "wmb()" above with "mb()" to avoid reordering between read
> and write IO instructions.
If I understand correctly, you don't need any memory barrier.
writel()/readl()'s are ordered to the same endpoint. That goes for all
the reordering/barrier comments mentioned below too.
device-io.rst:
The read and write functions are defined to be ordered. That is the
compiler is not permitted to reorder the I/O sequence. When the ordering
can be compiler optimised, you can use __readb() and friends to
indicate the relaxed ordering. Use this with care.
memory-barriers.txt:
(*) readX(), writeX():
The readX() and writeX() MMIO accessors take a pointer to the
peripheral being accessed as an __iomem * parameter. For pointers
mapped with the default I/O attributes (e.g. those returned by
ioremap()), the ordering guarantees are as follows:
1. All readX() and writeX() accesses to the same peripheral are ordered
with respect to each other. This ensures that MMIO register accesses
by the same CPU thread to a particular device will arrive in program
order.
>
> >
> > ret = gmu_poll_timeout(gmu, REG_A6XX_GMU_RSCC_CONTROL_ACK, val,
> > val & (1 << 1), 100, 10000);
> > diff --git a/drivers/gpu/drm/msm/adreno/a6xx_gpu.c b/drivers/gpu/drm/msm/adreno/a6xx_gpu.c
> > index 973872ad0474..0acbc38b8e70 100644
> > --- a/drivers/gpu/drm/msm/adreno/a6xx_gpu.c
> > +++ b/drivers/gpu/drm/msm/adreno/a6xx_gpu.c
> > @@ -1713,22 +1713,16 @@ static int hw_init(struct msm_gpu *gpu)
> > }
> >
> > /* Clear GBIF halt in case GX domain was not collapsed */
> > + gpu_write(gpu, REG_A6XX_GBIF_HALT, 0);
>
> We need a full barrier here to avoid reordering. Also, lets add a
> comment about why we are doing this odd looking sequence.
>
> > + gpu_read(gpu, REG_A6XX_GBIF_HALT);
> > if (adreno_is_a619_holi(adreno_gpu)) {
> > - gpu_write(gpu, REG_A6XX_GBIF_HALT, 0);
> > gpu_write(gpu, REG_A6XX_RBBM_GPR0_CNTL, 0);
> > - /* Let's make extra sure that the GPU can access the memory.. */
> > - mb();
>
> We need a full barrier here.
>
> > + gpu_read(gpu, REG_A6XX_RBBM_GPR0_CNTL);
> > } else if (a6xx_has_gbif(adreno_gpu)) {
> > - gpu_write(gpu, REG_A6XX_GBIF_HALT, 0);
> > gpu_write(gpu, REG_A6XX_RBBM_GBIF_HALT, 0);
> > - /* Let's make extra sure that the GPU can access the memory.. */
> > - mb();
>
> We need a full barrier here.
>
> > + gpu_read(gpu, REG_A6XX_RBBM_GBIF_HALT);
> > }
> >
> > - /* Some GPUs are stubborn and take their sweet time to unhalt GBIF! */
> > - if (adreno_is_a7xx(adreno_gpu) && a6xx_has_gbif(adreno_gpu))
> > - spin_until(!gpu_read(gpu, REG_A6XX_GBIF_HALT_ACK));
> > -
>
> Why is this removed?
>
> -Akhil
>
> > gpu_write(gpu, REG_A6XX_RBBM_SECVID_TSB_CNTL, 0);
> >
> > if (adreno_is_a619_holi(adreno_gpu))
> >
> > ---
> > base-commit: 93a39e4766083050ca0ecd6a3548093a3b9eb60c
> > change-id: 20240508-topic-adreno-a2d199cd4152
> >
> > Best regards,
> > --
> > Konrad Dybcio <konrad.dybcio@linaro.org>
> >
>
next prev parent reply other threads:[~2024-05-16 13:15 UTC|newest]
Thread overview: 8+ messages / expand[flat|nested] mbox.gz Atom feed top
2024-05-08 17:46 [PATCH] drm/msm/adreno: De-spaghettify the use of memory barriers Konrad Dybcio
2024-05-14 18:38 ` Akhil P Oommen
2024-05-16 13:15 ` Andrew Halaney [this message]
2024-05-16 14:50 ` Akhil P Oommen
2024-05-16 18:55 ` Andrew Halaney
2024-06-04 14:40 ` Will Deacon
2024-06-06 12:03 ` Konrad Dybcio
2024-06-04 17:35 ` Konrad Dybcio
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=ae4a77wt3kc73ejshptldqx6ugzrqguyq7etbbu54y4avhbdlt@qyt4r6gma7ev \
--to=ahalaney@redhat.com \
--cc=airlied@gmail.com \
--cc=daniel@ffwll.ch \
--cc=dmitry.baryshkov@linaro.org \
--cc=dri-devel@lists.freedesktop.org \
--cc=freedreno@lists.freedesktop.org \
--cc=konrad.dybcio@linaro.org \
--cc=linux-arm-msm@vger.kernel.org \
--cc=linux-kernel@vger.kernel.org \
--cc=marijn.suijten@somainline.org \
--cc=quic_abhinavk@quicinc.com \
--cc=quic_akhilpo@quicinc.com \
--cc=robdclark@chromium.org \
--cc=robdclark@gmail.com \
--cc=sean@poorly.run \
/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 read-only IMAP folder(s) and NNTP newsgroup(s).