2024-01-17 03:13:06

by Erico Nunes

[permalink] [raw]
Subject: [PATCH v1 3/6] drm/lima: set bus_stop bit before hard reset

This is required for reliable hard resets. Otherwise, doing a hard reset
while a task is still running (such as a task which is being stopped by
the drm_sched timeout handler) may result in random mmu write timeouts
or lockups which cause the entire gpu to hang.

Signed-off-by: Erico Nunes <[email protected]>
---
drivers/gpu/drm/lima/lima_pp.c | 13 +++++++++++++
1 file changed, 13 insertions(+)

diff --git a/drivers/gpu/drm/lima/lima_pp.c b/drivers/gpu/drm/lima/lima_pp.c
index a8f8f63b8295..ac097dd75072 100644
--- a/drivers/gpu/drm/lima/lima_pp.c
+++ b/drivers/gpu/drm/lima/lima_pp.c
@@ -168,6 +168,11 @@ static void lima_pp_write_frame(struct lima_ip *ip, u32 *frame, u32 *wb)
}
}

+static int lima_pp_bus_stop_poll(struct lima_ip *ip)
+{
+ return !!(pp_read(LIMA_PP_STATUS) & LIMA_PP_STATUS_BUS_STOPPED);
+}
+
static int lima_pp_hard_reset_poll(struct lima_ip *ip)
{
pp_write(LIMA_PP_PERF_CNT_0_LIMIT, 0xC01A0000);
@@ -181,6 +186,14 @@ static int lima_pp_hard_reset(struct lima_ip *ip)

pp_write(LIMA_PP_PERF_CNT_0_LIMIT, 0xC0FFE000);
pp_write(LIMA_PP_INT_MASK, 0);
+
+ pp_write(LIMA_PP_CTRL, LIMA_PP_CTRL_STOP_BUS);
+ ret = lima_poll_timeout(ip, lima_pp_bus_stop_poll, 10, 100);
+ if (ret) {
+ dev_err(dev->dev, "pp %s bus stop timeout\n", lima_ip_name(ip));
+ return ret;
+ }
+
pp_write(LIMA_PP_CTRL, LIMA_PP_CTRL_FORCE_RESET);
ret = lima_poll_timeout(ip, lima_pp_hard_reset_poll, 10, 100);
if (ret) {
--
2.43.0



2024-01-17 18:24:24

by Vasily Khoruzhick

[permalink] [raw]
Subject: Re: [PATCH v1 3/6] drm/lima: set bus_stop bit before hard reset

On Tue, Jan 16, 2024 at 7:12 PM Erico Nunes <[email protected]> wrote:
>
> This is required for reliable hard resets. Otherwise, doing a hard reset
> while a task is still running (such as a task which is being stopped by
> the drm_sched timeout handler) may result in random mmu write timeouts
> or lockups which cause the entire gpu to hang.

It looks like Mali driver is doing the same, so it totally makes sense.

> Signed-off-by: Erico Nunes <[email protected]>
Reviewed-by: Vasily Khoruzhick <[email protected]>

> ---
> drivers/gpu/drm/lima/lima_pp.c | 13 +++++++++++++
> 1 file changed, 13 insertions(+)
>
> diff --git a/drivers/gpu/drm/lima/lima_pp.c b/drivers/gpu/drm/lima/lima_pp.c
> index a8f8f63b8295..ac097dd75072 100644
> --- a/drivers/gpu/drm/lima/lima_pp.c
> +++ b/drivers/gpu/drm/lima/lima_pp.c
> @@ -168,6 +168,11 @@ static void lima_pp_write_frame(struct lima_ip *ip, u32 *frame, u32 *wb)
> }
> }
>
> +static int lima_pp_bus_stop_poll(struct lima_ip *ip)
> +{
> + return !!(pp_read(LIMA_PP_STATUS) & LIMA_PP_STATUS_BUS_STOPPED);
> +}
> +
> static int lima_pp_hard_reset_poll(struct lima_ip *ip)
> {
> pp_write(LIMA_PP_PERF_CNT_0_LIMIT, 0xC01A0000);
> @@ -181,6 +186,14 @@ static int lima_pp_hard_reset(struct lima_ip *ip)
>
> pp_write(LIMA_PP_PERF_CNT_0_LIMIT, 0xC0FFE000);
> pp_write(LIMA_PP_INT_MASK, 0);
> +
> + pp_write(LIMA_PP_CTRL, LIMA_PP_CTRL_STOP_BUS);
> + ret = lima_poll_timeout(ip, lima_pp_bus_stop_poll, 10, 100);
> + if (ret) {
> + dev_err(dev->dev, "pp %s bus stop timeout\n", lima_ip_name(ip));
> + return ret;
> + }
> +
> pp_write(LIMA_PP_CTRL, LIMA_PP_CTRL_FORCE_RESET);
> ret = lima_poll_timeout(ip, lima_pp_hard_reset_poll, 10, 100);
> if (ret) {
> --
> 2.43.0
>

2024-01-18 02:01:40

by Qiang Yu

[permalink] [raw]
Subject: Re: [PATCH v1 3/6] drm/lima: set bus_stop bit before hard reset

Do we need same for GP?

Regards,
Qiang

On Wed, Jan 17, 2024 at 11:12 AM Erico Nunes <[email protected]> wrote:
>
> This is required for reliable hard resets. Otherwise, doing a hard reset
> while a task is still running (such as a task which is being stopped by
> the drm_sched timeout handler) may result in random mmu write timeouts
> or lockups which cause the entire gpu to hang.
>
> Signed-off-by: Erico Nunes <[email protected]>
> ---
> drivers/gpu/drm/lima/lima_pp.c | 13 +++++++++++++
> 1 file changed, 13 insertions(+)
>
> diff --git a/drivers/gpu/drm/lima/lima_pp.c b/drivers/gpu/drm/lima/lima_pp.c
> index a8f8f63b8295..ac097dd75072 100644
> --- a/drivers/gpu/drm/lima/lima_pp.c
> +++ b/drivers/gpu/drm/lima/lima_pp.c
> @@ -168,6 +168,11 @@ static void lima_pp_write_frame(struct lima_ip *ip, u32 *frame, u32 *wb)
> }
> }
>
> +static int lima_pp_bus_stop_poll(struct lima_ip *ip)
> +{
> + return !!(pp_read(LIMA_PP_STATUS) & LIMA_PP_STATUS_BUS_STOPPED);
> +}
> +
> static int lima_pp_hard_reset_poll(struct lima_ip *ip)
> {
> pp_write(LIMA_PP_PERF_CNT_0_LIMIT, 0xC01A0000);
> @@ -181,6 +186,14 @@ static int lima_pp_hard_reset(struct lima_ip *ip)
>
> pp_write(LIMA_PP_PERF_CNT_0_LIMIT, 0xC0FFE000);
> pp_write(LIMA_PP_INT_MASK, 0);
> +
> + pp_write(LIMA_PP_CTRL, LIMA_PP_CTRL_STOP_BUS);
> + ret = lima_poll_timeout(ip, lima_pp_bus_stop_poll, 10, 100);
> + if (ret) {
> + dev_err(dev->dev, "pp %s bus stop timeout\n", lima_ip_name(ip));
> + return ret;
> + }
> +
> pp_write(LIMA_PP_CTRL, LIMA_PP_CTRL_FORCE_RESET);
> ret = lima_poll_timeout(ip, lima_pp_hard_reset_poll, 10, 100);
> if (ret) {
> --
> 2.43.0
>

2024-01-18 11:44:13

by Erico Nunes

[permalink] [raw]
Subject: Re: [PATCH v1 3/6] drm/lima: set bus_stop bit before hard reset

On Thu, Jan 18, 2024 at 3:01 AM Qiang Yu <[email protected]> wrote:
>
> Do we need same for GP?

I don't have an issue reproducer for gp so far, but the hardware does
have the same bit and the mali driver does it for both gp and pp, so I
think we can also add it to gp.