[Intel-gfx] [PATCH v2] drm/i915/bxt: Broxton decoupled MMIO
Tvrtko Ursulin
tvrtko.ursulin at linux.intel.com
Fri Sep 23 09:49:38 UTC 2016
Hi,
On 19/09/2016 18:15, Praveen Paneri wrote:
> Decoupled MMIO is an alternative way to access forcewake domain
> registers, which requires less cycles and avoids frequent software
> forcewake.
I would like to see a sentence or two on how it works here.
> v2:
> - Moved platform check out of the function and got rid of duplicate
> functions to find out decoupled power domain.
> - Added a check for forcewake already held and skipped decoupled access
> - Skipped writing 64 bit register through decoupled MMIO
>
> Signed-off-by: Zhe Wang <zhe1.wang at intel.com>
> Signed-off-by: Damien Lespiau <damien.lespiau at intel.com>
> Signed-off-by: Ankitprasad Sharma <ankitprasad.r.sharma at intel.com>
> Signed-off-by: Praveen Paneri <praveen.paneri at intel.com>
How come this list of SoBs?
> ---
> drivers/gpu/drm/i915/i915_drv.h | 11 ++++
> drivers/gpu/drm/i915/i915_reg.h | 7 +++
> drivers/gpu/drm/i915/intel_uncore.c | 118 +++++++++++++++++++++++++++++++++++-
> 3 files changed, 134 insertions(+), 2 deletions(-)
>
> diff --git a/drivers/gpu/drm/i915/i915_drv.h b/drivers/gpu/drm/i915/i915_drv.h
> index 4dd307e..065247b 100644
> --- a/drivers/gpu/drm/i915/i915_drv.h
> +++ b/drivers/gpu/drm/i915/i915_drv.h
> @@ -558,6 +558,16 @@ enum forcewake_domains {
> #define FW_REG_READ (1)
> #define FW_REG_WRITE (2)
>
> +enum power_domains {
If this is specific to decouples mmio maybe call it decoupled_power_domains?
> + GEN9_DECOUPLED_PD_BLITTER = 0,
> + GEN9_DECOUPLED_PD_RENDER,
> + GEN9_DECOUPLED_PD_MEDIA,
> + GEN9_DECOUPLED_PD_ALL
> +};
> +
> +#define GEN9_DECOUPLED_OP_WRITE (0)
> +#define GEN9_DECOUPLED_OP_READ (1)
Would it be more consistent to make these ones enums as well?
> +
> enum forcewake_domains
> intel_uncore_forcewake_for_reg(struct drm_i915_private *dev_priv,
> i915_reg_t reg, unsigned int op);
> @@ -2854,6 +2864,7 @@ struct drm_i915_cmd_table {
> #define GT_FREQUENCY_MULTIPLIER 50
> #define GEN9_FREQ_SCALER 3
>
> +#define HAS_DECOUPLED_MMIO(dev_priv) (IS_BROXTON(dev_priv) && IS_BXT_REVID(dev_priv, BXT_REVID_C0, REVID_FOREVER))
There is a recent patch series from Carlos Santa which moved all these
type of things to device info. So I think you have to make this
compliant with that new style.
> #include "i915_trace.h"
>
> static inline bool intel_scanout_needs_vtd_wa(struct drm_i915_private *dev_priv)
> diff --git a/drivers/gpu/drm/i915/i915_reg.h b/drivers/gpu/drm/i915/i915_reg.h
> index 70d9616..be59803 100644
> --- a/drivers/gpu/drm/i915/i915_reg.h
> +++ b/drivers/gpu/drm/i915/i915_reg.h
> @@ -7394,6 +7394,13 @@ enum {
> #define SKL_FUSE_PG1_DIST_STATUS (1<<26)
> #define SKL_FUSE_PG2_DIST_STATUS (1<<25)
>
> +/* Decoupled MMIO register pair for kernel driver */
> +#define GEN9_DECOUPLED_REG0_DW0 _MMIO(0xF00)
> +#define GEN9_DECOUPLED_REG0_DW1 _MMIO(0xF04)
> +#define GEN9_DECOUPLED_DW1_GO (1<<31)
> +#define GEN9_DECOUPLED_PD_SHIFT 28
> +#define GEN9_DECOUPLED_OP_SHIFT 24
> +
> /* Per-pipe DDI Function Control */
> #define _TRANS_DDI_FUNC_CTL_A 0x60400
> #define _TRANS_DDI_FUNC_CTL_B 0x61400
> diff --git a/drivers/gpu/drm/i915/intel_uncore.c b/drivers/gpu/drm/i915/intel_uncore.c
> index a9b6c93..06fffba 100644
> --- a/drivers/gpu/drm/i915/intel_uncore.c
> +++ b/drivers/gpu/drm/i915/intel_uncore.c
> @@ -816,6 +816,42 @@ unclaimed_reg_debug(struct drm_i915_private *dev_priv,
> __unclaimed_reg_debug(dev_priv, reg, read, before);
> }
>
> +/*
> + * Decoupled MMIO access for only 1 DWORD
> + */
> +static void __gen9_decoupled_mmio_access(struct drm_i915_private *dev_priv,
> + uint32_t reg, u32 *ptr_data,
Seeing uint32_t and u32 on the same line looks a bit untidy to me. I
think you should only use one and that in kernel u32 is preferred.
> + enum forcewake_domains fw_engine, int operation)
> +{
> + enum power_domains pd_engine;
> + u32 ctrl_reg_data = 0;
> +
> + if (operation == GEN9_DECOUPLED_OP_WRITE)
> + __raw_i915_write32(dev_priv,
> + GEN9_DECOUPLED_REG0_DW0,
> + *ptr_data);
> +
> + pd_engine = (fw_engine == (FORCEWAKE_RENDER || FORCEWAKE_BLITTER)) ? \
> + !(fw_engine >> 1) : (fw_engine >> 1); \
Feels that look up table would be better.
> +
> + ctrl_reg_data |= reg;
> + ctrl_reg_data |= (operation << GEN9_DECOUPLED_OP_SHIFT);
> + ctrl_reg_data |= (pd_engine << GEN9_DECOUPLED_PD_SHIFT);
> + __raw_i915_write32(dev_priv, GEN9_DECOUPLED_REG0_DW1, ctrl_reg_data);
> +
> + ctrl_reg_data |= GEN9_DECOUPLED_DW1_GO;
> + __raw_i915_write32(dev_priv, GEN9_DECOUPLED_REG0_DW1, ctrl_reg_data);
> +
> + if (wait_for_atomic((__raw_i915_read32(dev_priv,
> + GEN9_DECOUPLED_REG0_DW1) & GEN9_DECOUPLED_DW1_GO) == 0,
> + FORCEWAKE_ACK_TIMEOUT_MS))
> + DRM_ERROR("Decoupled MMIO wait timed out\n");
> +
Is FORCEWAKE_ACK_TIMEOUT_MS correct/relevant for decoupled MMIO? It is
potentially quite a long atomic wait.
Would it be worth returning some fixed value in the case of a timeout?
Might be better than random stuff? (applies in the 64-bit read case)
> + if (operation == GEN9_DECOUPLED_OP_READ)
> + *ptr_data = __raw_i915_read32(dev_priv,
> + GEN9_DECOUPLED_REG0_DW0);
> +}
> +
> #define GEN2_READ_HEADER(x) \
> u##x val = 0; \
> assert_rpm_wakelock_held(dev_priv);
> @@ -892,6 +928,20 @@ static inline void __force_wake_auto(struct drm_i915_private *dev_priv,
> dev_priv->uncore.funcs.force_wake_get(dev_priv, fw_domains);
> }
>
> +static inline bool __is_forcewake_active(struct drm_i915_private *dev_priv,
> + enum forcewake_domains fw_domains)
> +{
> + struct intel_uncore_forcewake_domain *domain;
> +
> + /* Ideally GCC would be constant-fold and eliminate this loop */
> + for_each_fw_domain_masked(domain, fw_domains, dev_priv) {
> + if (domain->wake_count)
> + fw_domains &= ~domain->mask;
> + }
> +
> + return fw_domains ? 0 : 1;
> +}
> +
> #define __gen6_read(x) \
> static u##x \
> gen6_read##x(struct drm_i915_private *dev_priv, i915_reg_t reg, bool trace) { \
> @@ -940,6 +990,37 @@ gen9_read##x(struct drm_i915_private *dev_priv, i915_reg_t reg, bool trace) { \
> GEN6_READ_FOOTER; \
> }
>
> +#define __gen9_decoupled_read(x) \
> +static u##x \
> +gen9_decoupled_read##x(struct drm_i915_private *dev_priv, i915_reg_t reg, bool trace) { \
> + enum forcewake_domains fw_engine; \
fw_engines
> + GEN6_READ_HEADER(x); \
> + fw_engine = __gen9_reg_read_fw_domains(offset); \
> + if (fw_engine && x%32 == 0) { \
> + if (__is_forcewake_active(dev_priv, fw_engine)) \
> + __raw_i915_write##x(dev_priv, reg, val); \
Write in the read macro, I don't understand!?
Also, would a single mmio read call be possible, something like below?
if (x % 32 || !fw_engines || __is_forcewake_active()) {
if (fw_engines)
__force_wake_auto
__raw_i915_read
} else {
... decoupled mmio loop ...
}
I might have made an oversight, no guarantees that I haven't. :)
> + else { \
> + unsigned i; \
> + u32 *ptr_data = (u32 *) &val; \
Hm, curios if taking the address of val causes some issues for code
generation. I am not set up yet to try it out myself, but would be
interesting to see the difference between your approach and the one
where reads are done via return value and not pointer writes. Even if it
means refactoring the common accessor a bit.
> + for (i = 0; i < x/32; i++) \
Could you check if for the 64-bit read compiler manages to unroll this loop?
> + __gen9_decoupled_mmio_access(dev_priv, \
> + (offset + i*4), \
Hopefully compiler manages to optimize, but please check and in case it
doesn't consider offsett += sizeof(u32) in the for statement as an
alternative.
> + ptr_data + i, \
> + fw_engine, \
> + GEN9_DECOUPLED_OP_READ); \
> + } \
> + } else { \
> + if (fw_engine) \
> + __force_wake_auto(dev_priv, fw_engine); \
> + val = __raw_i915_read##x(dev_priv, reg); \
> + } \
> + GEN6_READ_FOOTER; \
> +}
> +
> +__gen9_decoupled_read(8)
> +__gen9_decoupled_read(16)
What is the value of defining 8 and 16-bit version of those when they
won't use the decoupled mmio path at all? Wouldn't it be better to just
assign a mix of gen9 mmio functions in that case?
> +__gen9_decoupled_read(32)
> +__gen9_decoupled_read(64)
> __gen9_read(8)
> __gen9_read(16)
> __gen9_read(32)
> @@ -1107,6 +1188,34 @@ gen9_write##x(struct drm_i915_private *dev_priv, i915_reg_t reg, u##x val, \
> GEN6_WRITE_FOOTER; \
> }
>
> +#define __gen9_decoupled_write(x) \
> +static void \
> +gen9_decoupled_write##x(struct drm_i915_private *dev_priv, i915_reg_t reg, u##x val, \
> + bool trace) { \
> + enum forcewake_domains fw_engine; \
> + GEN6_WRITE_HEADER; \
> + fw_engine = __gen9_reg_write_fw_domains(offset); \
> + if (fw_engine && x == 32) { \
> + u32 *ptr_data = (u32 *) &val; \
> + if (__is_forcewake_active(dev_priv, fw_engine)) \
> + __raw_i915_write##x(dev_priv, reg, val); \
> + else \
> + __gen9_decoupled_mmio_access(dev_priv, \
> + offset, \
> + ptr_data, \
> + fw_engine, \
> + GEN9_DECOUPLED_OP_WRITE); \
> + } else { \
> + if (fw_engine) \
> + __force_wake_auto(dev_priv, fw_engine); \
> + __raw_i915_write##x(dev_priv, reg, val); \
> + } \
> + GEN6_WRITE_FOOTER; \
> +}
> +
> +__gen9_decoupled_write(8)
> +__gen9_decoupled_write(16)
All same comments as for the read block.
> +__gen9_decoupled_write(32)
> __gen9_write(8)
> __gen9_write(16)
> __gen9_write(32)
> @@ -1328,8 +1437,13 @@ void intel_uncore_init(struct drm_i915_private *dev_priv)
> switch (INTEL_INFO(dev_priv)->gen) {
> default:
> case 9:
> - ASSIGN_WRITE_MMIO_VFUNCS(gen9);
> - ASSIGN_READ_MMIO_VFUNCS(gen9);
> + if (HAS_DECOUPLED_MMIO(dev_priv)) {
> + ASSIGN_WRITE_MMIO_VFUNCS(gen9_decoupled);
> + ASSIGN_READ_MMIO_VFUNCS(gen9_decoupled);
> + } else {
> + ASSIGN_WRITE_MMIO_VFUNCS(gen9);
> + ASSIGN_READ_MMIO_VFUNCS(gen9);
> + }
So with the suggestion from above this would have to be open coded to
only override 32 and 64 bit ops with decoupled versions, but I think
that is better than having duplicated code compiled in the module.
> break;
> case 8:
> if (IS_CHERRYVIEW(dev_priv)) {
Regards,
Tvrtko
More information about the Intel-gfx
mailing list