[PATCH 12/12] drm/i915/xe3lpd: Power request asserting/deasserting

Kahola, Mika mika.kahola at intel.com
Mon Oct 28 09:11:59 UTC 2024


> -----Original Message-----
> From: Sousa, Gustavo <gustavo.sousa at intel.com>
> Sent: Wednesday, 23 October 2024 23.40
> To: Atwood, Matthew S <matthew.s.atwood at intel.com>; intel-
> gfx at lists.freedesktop.org; intel-xe at lists.freedesktop.org
> Cc: Kahola, Mika <mika.kahola at intel.com>; Atwood, Matthew S
> <matthew.s.atwood at intel.com>; Taylor, Clinton A <clinton.a.taylor at intel.com>
> Subject: Re: [PATCH 12/12] drm/i915/xe3lpd: Power request
> asserting/deasserting
> 
> Quoting Matt Atwood (2024-10-18 17:49:41-03:00)
> >From: Mika Kahola <mika.kahola at intel.com>
> >
> >There is a HW issue that arises when there are race conditions between
> >TCSS entering/exiting TC7 or TC10 states while the driver is
> >asserting/deasserting TCSS power request. As a workaround, Display
> >driver will implement a mailbox sequence to ensure that the TCSS is in
> >TC0 when TCSS power request is asserted/deasserted.
> >
> >The sequence is the following
> >
> >1. Read mailbox command status and wait until run/busy bit is
> >   clear
> >2. Write mailbox data value '1' for power request asserting
> >   and '0' for power request deasserting 3. Write mailbox command
> >run/busy bit and command value with 0x1 4. Read mailbox command and
> >wait until run/busy bit is clear
> >   before continuing power request.
> >
> >Signed-off-by: Mika Kahola <mika.kahola at intel.com>
> >Signed-off-by: Matt Atwood <matthew.s.atwood at intel.com>
> >---
> > drivers/gpu/drm/i915/display/intel_tc.c | 40 +++++++++++++++++++++++++
> > drivers/gpu/drm/i915/i915_reg.h         |  7 +++++
> > 2 files changed, 47 insertions(+)
> >
> >diff --git a/drivers/gpu/drm/i915/display/intel_tc.c
> >b/drivers/gpu/drm/i915/display/intel_tc.c
> >index 6f2ee7dbc43b..7d9f87db381c 100644
> >--- a/drivers/gpu/drm/i915/display/intel_tc.c
> >+++ b/drivers/gpu/drm/i915/display/intel_tc.c
> >@@ -1013,6 +1013,39 @@ xelpdp_tc_phy_wait_for_tcss_power(struct
> intel_tc_port *tc, bool enabled)
> >         return true;
> > }
> >
> >+static bool xelpdp_tc_phy_wait_for_tcss_ready(struct drm_i915_private
> >+*i915,
> 
> I think xelpdp_ is not right here as this does not apply to Xe_LPD+. I think we
> could simply use the workaround lineage number for the name of this function.
> Something like wa_14020908590().

I couldn't find any workarounds with this 14020908590 number. Maybe we could rename the function wa_tcss_power_assert()?

> 
> >+                                              bool enable) {
> >+        if (DISPLAY_VER(i915) < 30)
> 
> The description of the internal ticket that resulted in this workaround makes me
> wonder if this is actually an issue associated to the SoC instead of display or PICA
> IP. However the ticket metadata indicates the PICA IP as the one affected. It
> would be good to confirm the correct association here.
> 
> In any case, this seems not really related to the display IP, so checking
> DISPLAY_VER(i915) seems not very precise here.
> 
> If it turns out that this is a SoC-related issue, it would be better to check if the
> platform is PTL.
> 
> Now, if this is indeed an issue associated to the PICA IP, then I see the following
> alternatives:
> 
>  - add an earlier patch to detect the PICA IP and add that info to
>    intel_display_runtime_info. Then, here we use that info in the
>    condition for this workaround;
> 
>  - at least add a comment here that we are checking the display version
>    because we do not have PICA IP detection in the driver yet. In this
>    case.
> 
> I tend to think that checking version equality would make more sense (assuming
> the issue would not be seen in a future platform).

I'm assuming this is more related to PICA IP than platform but I cannot confirm that yet. In the meantime, I could add a comment and check display version only for the PTL platform.

> 
> >+                return true;
> >+
> >+        /* check if mailbox is running busy */
> >+        if (intel_de_wait_for_clear(i915, TCSS_DISP_MAILBOX_IN_CMD,
> >+                                    TCSS_DISP_MAILBOX_IN_CMD_RUN_BUSY, 10)) {
> >+                drm_dbg_kms(&i915->drm,
> >+                            "timeout waiting for TCSS mailbox run/busy bit to clear\n");
> >+                return false;
> >+        }
> >+
> >+        if (enable)
> >+                intel_de_write(i915, TCSS_DISP_MAILBOX_IN_DATA, 1);
> >+        else
> >+                intel_de_write(i915, TCSS_DISP_MAILBOX_IN_DATA, 0);
> >+
> >+        intel_de_write(i915, TCSS_DISP_MAILBOX_IN_CMD,
> >+                       TCSS_DISP_MAILBOX_IN_CMD_DATA(1));
> 
> Nitpick: I would prefer a more explicit version of this. Something like:
> 
>     intel_de_write(i915, TCSS_DISP_MAILBOX_IN_CMD,
>                    TCSS_DISP_MAILBOX_IN_CMD_RUN_BUSY |
>                    TCSS_DISP_MAILBOX_IN_CMD_CMD(0x1));
> 
> With the current version, I had to go and check that
> TCSS_DISP_MAILBOX_IN_CMD_DATA() also includes the run/busy bit.
> 
> >+
> >+        /* wait to clear mailbox running busy bit before continuing */
> >+        if (intel_de_wait_for_clear(i915, TCSS_DISP_MAILBOX_IN_CMD,
> >+                                    TCSS_DISP_MAILBOX_IN_CMD_RUN_BUSY, 10)) {
> >+                drm_dbg_kms(&i915->drm,
> >+                            "timeout waiting for TCSS mailbox run/busy
> >+ bit to clear\n");
> 
> I think would be good to have different timeout messages so that it is easy to
> differentiate whether we timed out while waiting for our turn to use the mailbox
> or while waiting for our command to be handled.

I'll rephrase the wording here.

> 
> >+                return false;
> >+        }
> >+
> >+        return true;
> >+}
> >+
> > static void __xelpdp_tc_phy_enable_tcss_power(struct intel_tc_port
> >*tc, bool enable)  {
> >         struct drm_i915_private *i915 = tc_to_i915(tc); @@ -1022,6
> >+1055,13 @@ static void __xelpdp_tc_phy_enable_tcss_power(struct
> >intel_tc_port *tc, bool ena
> >
> >         assert_tc_cold_blocked(tc);
> >
> >+        /*
> >+         * Gfx driver workaround for PTL tcss_rxdetect_clkswb_req/ack handshake
> >+         * violation when pwwreq= 0->1 during TC7/10 entry
> >+         */
> >+        drm_WARN_ON(&i915->drm,
> >+                    !xelpdp_tc_phy_wait_for_tcss_ready(i915, enable));
> >+
> >         val = intel_de_read(i915, reg);
> >         if (enable)
> >                 val |= XELPDP_TCSS_POWER_REQUEST; diff --git
> >a/drivers/gpu/drm/i915/i915_reg.h b/drivers/gpu/drm/i915/i915_reg.h
> >index 2743a2dd0a3d..d2775a32bf18 100644
> >--- a/drivers/gpu/drm/i915/i915_reg.h
> >+++ b/drivers/gpu/drm/i915/i915_reg.h
> 
> Maybe intel_cx0_phy_regs.h would be a better home for the mailbox registers,
> since it is where XELPDP_PORT_BUF_CTL1 and
> XELPDP_TCSS_POWER_{REQUEST,STATE} are defined?
> 
> Not the perfect place, but at least we would not add new definitions to
> i915_reg.h and add to the work of separating display code from i915.

Ok, I will move these defs to intel_cx0_phy_regs.h

> 
> >@@ -4539,6 +4539,13 @@ enum skl_power_gate {
> > #define  TCSS_DDI_STATUS_HPD_LIVE_STATUS_TBT        REG_BIT(1)
> > #define  TCSS_DDI_STATUS_HPD_LIVE_STATUS_ALT        REG_BIT(0)
> >
> >+#define TCSS_DISP_MAILBOX_IN_CMD                _MMIO(0x161300)
> >+#define   TCSS_DISP_MAILBOX_IN_CMD_RUN_BUSY        REG_BIT(31)
> >+#define   TCSS_DISP_MAILBOX_IN_CMD_CMD_MASK        REG_GENMASK(7, 0)
> >+#define   TCSS_DISP_MAILBOX_IN_CMD_DATA(x)
> TCSS_DISP_MAILBOX_IN_CMD_RUN_BUSY | \
> >+
> >+REG_FIELD_PREP(TCSS_DISP_MAILBOX_IN_CMD_CMD_MASK, (x))
> 
> Missing a blank line here.
Ok

I will make few adjustments to the patch. Thanks for the comments and a review!

-Mika-

> 
> --
> Gustavo Sousa
> 
> >+#define TCSS_DISP_MAILBOX_IN_DATA                _MMIO(0x161304)
> >+
> > #define PRIMARY_SPI_TRIGGER                        _MMIO(0x102040)
> > #define PRIMARY_SPI_ADDRESS                        _MMIO(0x102080)
> > #define PRIMARY_SPI_REGIONID                        _MMIO(0x102084)
> >--
> >2.45.0
> >


More information about the Intel-xe mailing list