Re: [PATCH igt v2 3/3] msm: Add recovery tests

[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

 



On Thu, Nov 11, 2021 at 09:35:16AM -0800, Rob Clark wrote:
> On Thu, Nov 11, 2021 at 4:13 AM Petri Latvala <petri.latvala@xxxxxxxxx> wrote:
> >
> > On Wed, Nov 10, 2021 at 11:00:41AM -0800, Rob Clark wrote:
> > > On Wed, Nov 10, 2021 at 10:37 AM Rob Clark <robdclark@xxxxxxxxx> wrote:
> > > >
> > > > From: Rob Clark <robdclark@xxxxxxxxxxxx>
> > > >
> > > > Add tests to exercise:
> > > >
> > > > 1. sw hangcheck timeout
> > > > 2. gpu fault (hang) recovery
> > > > 3. iova fault recovery
> > > >
> > > > Signed-off-by: Rob Clark <robdclark@xxxxxxxxxxxx>
> > > > ---
> > > >  lib/igt_msm.h        |   3 +
> > > >  tests/meson.build    |   1 +
> > > >  tests/msm_recovery.c | 172 +++++++++++++++++++++++++++++++++++++++++++
> > > >  3 files changed, 176 insertions(+)
> > > >  create mode 100644 tests/msm_recovery.c
> > > >
> > > > diff --git a/lib/igt_msm.h b/lib/igt_msm.h
> > > > index 1a66c806..421d23ed 100644
> > > > --- a/lib/igt_msm.h
> > > > +++ b/lib/igt_msm.h
> > > > @@ -97,6 +97,9 @@ enum adreno_pm4_packet_type {
> > > >
> > > >  enum adreno_pm4_type3_packets {
> > > >         CP_NOP = 16,
> > > > +       CP_WAIT_MEM_GTE = 20,
> > > > +       CP_WAIT_REG_MEM = 60,
> > > > +       CP_MEM_WRITE = 61,
> > > >  };
> > > >
> > > >  static inline unsigned
> > > > diff --git a/tests/meson.build b/tests/meson.build
> > > > index 0af3e03a..166e3494 100644
> > > > --- a/tests/meson.build
> > > > +++ b/tests/meson.build
> > > > @@ -60,6 +60,7 @@ test_progs = [
> > > >         'kms_vrr',
> > > >         'kms_writeback',
> > > >         'meta_test',
> > > > +       'msm_recovery',
> > > >         'msm_submit',
> > > >         'panfrost_get_param',
> > > >         'panfrost_gem_new',
> > > > diff --git a/tests/msm_recovery.c b/tests/msm_recovery.c
> > > > new file mode 100644
> > > > index 00000000..b71326b8
> > > > --- /dev/null
> > > > +++ b/tests/msm_recovery.c
> > > > @@ -0,0 +1,172 @@
> > > > +/*
> > > > + * Copyright © 2021 Google, Inc.
> > > > + *
> > > > + * Permission is hereby granted, free of charge, to any person obtaining a
> > > > + * copy of this software and associated documentation files (the "Software"),
> > > > + * to deal in the Software without restriction, including without limitation
> > > > + * the rights to use, copy, modify, merge, publish, distribute, sublicense,
> > > > + * and/or sell copies of the Software, and to permit persons to whom the
> > > > + * Software is furnished to do so, subject to the following conditions:
> > > > + *
> > > > + * The above copyright notice and this permission notice (including the next
> > > > + * paragraph) shall be included in all copies or substantial portions of the
> > > > + * Software.
> > > > + *
> > > > + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
> > > > + * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
> > > > + * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
> > > > + * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
> > > > + * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING
> > > > + * FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS
> > > > + * IN THE SOFTWARE.
> > > > + */
> > > > +
> > > > +#include <sys/poll.h>
> > > > +
> > > > +#include "igt.h"
> > > > +#include "igt_msm.h"
> > > > +
> > > > +static struct msm_device *dev;
> > > > +static struct msm_bo *scratch_bo;
> > > > +static uint32_t *scratch;
> > > > +
> > > > +/*
> > > > + * Helpers for cmdstream packet building:
> > > > + */
> > > > +
> > > > +static void
> > > > +wait_mem_gte(struct msm_cmd *cmd, uint32_t offset_dwords, uint32_t ref)
> > > > +{
> > > > +       msm_cmd_pkt7(cmd, CP_WAIT_MEM_GTE, 4);
> > > > +       msm_cmd_emit(cmd, 0);                              /* RESERVED */
> > > > +       msm_cmd_bo  (cmd, scratch_bo, offset_dwords * 4);  /* POLL_ADDR_LO/HI */
> > > > +       msm_cmd_emit(cmd, ref);                            /* REF */
> > > > +}
> > > > +
> > > > +static void
> > > > +mem_write(struct msm_cmd *cmd, uint32_t offset_dwords, uint32_t val)
> > > > +{
> > > > +       msm_cmd_pkt7(cmd, CP_MEM_WRITE, 3);
> > > > +       msm_cmd_bo  (cmd, scratch_bo, offset_dwords * 4);  /* ADDR_LO/HI */
> > > > +       msm_cmd_emit(cmd, val);                            /* VAL */
> > > > +}
> > > > +
> > > > +/*
> > > > + * Helper to wait on a fence-fd:
> > > > + */
> > > > +static void
> > > > +wait_and_close(int fence_fd)
> > > > +{
> > > > +       poll(&(struct pollfd){fence_fd, POLLIN}, 1, -1);
> > > > +       close(fence_fd);
> > > > +}
> > > > +
> > > > +/*
> > > > + * Helper for hang tests.  Emits multiple submits, with one in the middle
> > > > + * that triggers a fault, and confirms that the submits before and after
> > > > + * the faulting one execute properly, ie. that the driver properly manages
> > > > + * to recover and re-queue the submits after the faulting submit;
> > > > + */
> > > > +static void
> > > > +do_hang_test(struct msm_pipe *pipe)
> > > > +{
> > > > +       struct msm_cmd *cmds[16];
> > > > +       int fence_fds[ARRAY_SIZE(cmds)];
> > > > +
> > > > +       memset(scratch, 0, 0x1000);
> > > > +
> > > > +       for (unsigned i = 0; i < ARRAY_SIZE(cmds); i++) {
> > > > +               struct msm_cmd *cmd = igt_msm_cmd_new(pipe, 0x1000);
> > > > +
> > > > +               cmds[i] = cmd;
> > > > +
> > > > +               /*
> > > > +                * Emit a packet to wait for scratch[0] to be >= 1
> > > > +                *
> > > > +                * This lets us force the GPU to wait until all the cmdstream is
> > > > +                * queued up.
> > > > +                */
> > > > +               wait_mem_gte(cmd, 0, 1);
> > > > +
> > > > +               if (i == 10) {
> > > > +                       msm_cmd_emit(cmd, 0xdeaddead);
> > > > +               }
> > > > +
> > > > +               /* Emit a packet to write scratch[1+i] = 2+i: */
> > > > +               mem_write(cmd, 1+i, 2+i);
> > > > +       }
> > > > +
> > > > +       for (unsigned i = 0; i < ARRAY_SIZE(cmds); i++) {
> > > > +               fence_fds[i] = igt_msm_cmd_submit(cmds[i]);
> > > > +       }
> > > > +
> > > > +       usleep(10000);
> > > > +
> > > > +       /* Let the WAIT_MEM_GTE complete: */
> > > > +       scratch[0] = 1;
> > > > +
> > > > +       for (unsigned i = 0; i < ARRAY_SIZE(cmds); i++) {
> > > > +               wait_and_close(fence_fds[i]);
> > > > +               igt_msm_cmd_free(cmds[i]);
> > > > +               if (i == 10)
> > > > +                       continue;
> > > > +               igt_assert_eq(scratch[1+i], 2+i);
> > > > +       }
> > > > +}
> > > > +
> > > > +/*
> > > > + * Tests for drm/msm hangcheck, recovery, and fault handling
> > > > + */
> > > > +
> > > > +igt_main
> > > > +{
> > > > +       static struct msm_pipe *pipe = NULL;
> > > > +
> > > > +       igt_fixture {
> > > > +               dev = igt_msm_dev_open();
> > > > +               pipe = igt_msm_pipe_open(dev, 0);
> > > > +               scratch_bo = igt_msm_bo_new(dev, 0x1000, MSM_BO_WC);
> > > > +               scratch = igt_msm_bo_map(scratch_bo);
> > > > +       }
> > > > +
> > > > +       igt_describe("Test sw hangcheck handling");
> > > > +       igt_subtest("hangcheck") {
> > > > +               igt_require(dev->gen >= 6);
> > > > +
> > > > +               /* Disable hw hang detection to force fallback to sw hangcheck: */
> > > > +               igt_debugfs_write(dev->fd, "disable_err_irq", "Y");
> > >
> > > note that this depends on [1].. not sure if there is any constraint
> > > about landing igt tests before some debugfs they depend on lands
> > > upstream on the kernel side?
> > >
> > > [1] https://patchwork.freedesktop.org/patch/462625/?series=96725&rev=1
> >
> > The usual ordering dependency of kernel vs userspace, where both sides
> > are reviewed and ready to go before anything lands. But then the
> > actual merging with IGT is not so strict, IGT can go in first as long
> > as there's a consensus that the required kernel changes are going in
> > in finite time.
> >
> > That said, IGT is supposed to do the right thing for kernels some time
> > back, especially LTS kernels, so this test needs to properly handle
> > the debugfs file missing regardless of merging order.
> >
> > Some kind of igt_require check for the debugfs file takes care of
> > that.
> >
> 
> Hmm, unless I'm overlooking something, I guess we need a helper to
> check for the existence of a debugfs file?

Yeah =(

Unless the file supports reading.


-- 
Petri Latvala


> 
> BR,
> -R
> 
> >
> > --
> > Petri Latvala
> >
> >
> >
> >
> > >
> > > BR,
> > > -R
> > >
> > > > +
> > > > +               do_hang_test(pipe);
> > > > +
> > > > +               igt_debugfs_write(dev->fd, "disable_err_irq", "N");
> > > > +       }
> > > > +
> > > > +       igt_describe("Test hw fault handling");
> > > > +       igt_subtest("gpu-fault") {
> > > > +               igt_require(dev->gen >= 6);
> > > > +
> > > > +               do_hang_test(pipe);
> > > > +       }
> > > > +
> > > > +       igt_describe("Test iova fault handling");
> > > > +       igt_subtest("iova-fault") {
> > > > +               struct msm_cmd *cmd;
> > > > +
> > > > +               igt_require(dev->gen >= 6);
> > > > +
> > > > +               cmd = igt_msm_cmd_new(pipe, 0x1000);
> > > > +
> > > > +               msm_cmd_pkt7(cmd, CP_MEM_WRITE, 3);
> > > > +               msm_cmd_emit(cmd, 0xdeaddead);           /* ADDR_LO */
> > > > +               msm_cmd_emit(cmd, 0x1);                  /* ADDR_HI */
> > > > +               msm_cmd_emit(cmd, 0x123);                /* VAL */
> > > > +
> > > > +               wait_and_close(igt_msm_cmd_submit(cmd));
> > > > +       }
> > > > +
> > > > +       igt_fixture {
> > > > +               igt_msm_bo_free(scratch_bo);
> > > > +               igt_msm_pipe_close(pipe);
> > > > +               igt_msm_dev_close(dev);
> > > > +       }
> > > > +}
> > > > --
> > > > 2.31.1
> > > >



[Index of Archives]     [Linux ARM Kernel]     [Linux ARM]     [Linux Omap]     [Fedora ARM]     [Linux for Sparc]     [IETF Annouce]     [Security]     [Bugtraq]     [Linux MIPS]     [ECOS]     [Asterisk Internet PBX]     [Linux API]

  Powered by Linux