|
Okay, I will incorporate Shaoyun's input.
Yong
From: Liu, Shaoyun <Shaoyun.Liu@xxxxxxx>
Sent: Monday, September 23, 2019 10:27 AM To: Zhao, Yong <Yong.Zhao@xxxxxxx>; amd-gfx@xxxxxxxxxxxxxxxxxxxxx <amd-gfx@xxxxxxxxxxxxxxxxxxxxx> Subject: Re: [PATCH 2/2] drm/amdkfd: Use better name for sdma queue non HWS path Probably rename to sdma_rlc to avoid the confusion of rlc used in
other amdgpu driver . Regards shaoyun.liu On 2019-09-22 11:56 p.m., Zhao, Yong wrote: > The old name is prone to confusion. The register offset is for a RLC queue > rather than a SDMA engine. The value is not a base address, but a > register offset. > > Change-Id: I55fb835f2105392344b1c17323bb55c03f927836 > Signed-off-by: Yong Zhao <Yong.Zhao@xxxxxxx> > --- > .../drm/amd/amdgpu/amdgpu_amdkfd_arcturus.c | 85 +++++++++--------- > .../drm/amd/amdgpu/amdgpu_amdkfd_gfx_v10.c | 90 +++++++++---------- > .../gpu/drm/amd/amdgpu/amdgpu_amdkfd_gfx_v7.c | 10 +-- > .../gpu/drm/amd/amdgpu/amdgpu_amdkfd_gfx_v8.c | 10 +-- > .../gpu/drm/amd/amdgpu/amdgpu_amdkfd_gfx_v9.c | 86 +++++++++--------- > 5 files changed, 137 insertions(+), 144 deletions(-) > > diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_arcturus.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_arcturus.c > index c9ce1516956e..d2c0666c2798 100644 > --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_arcturus.c > +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_arcturus.c > @@ -70,11 +70,11 @@ static inline struct v9_sdma_mqd *get_sdma_mqd(void *mqd) > return (struct v9_sdma_mqd *)mqd; > } > > -static uint32_t get_sdma_base_addr(struct amdgpu_device *adev, > +static uint32_t get_rlc_reg_offset(struct amdgpu_device *adev, > unsigned int engine_id, > unsigned int queue_id) > { > - uint32_t base[8] = { > + uint32_t sdma_engine_reg_base[8] = { > SOC15_REG_OFFSET(SDMA0, 0, > mmSDMA0_RLC0_RB_CNTL) - mmSDMA0_RLC0_RB_CNTL, > SOC15_REG_OFFSET(SDMA1, 0, > @@ -92,12 +92,11 @@ static uint32_t get_sdma_base_addr(struct amdgpu_device *adev, > SOC15_REG_OFFSET(SDMA7, 0, > mmSDMA7_RLC0_RB_CNTL) - mmSDMA7_RLC0_RB_CNTL > }; > - uint32_t retval; > > - retval = base[engine_id] + queue_id * (mmSDMA0_RLC1_RB_CNTL - > - mmSDMA0_RLC0_RB_CNTL); > + uint32_t retval = sdma_engine_reg_base[engine_id] > + + queue_id * (mmSDMA0_RLC1_RB_CNTL - mmSDMA0_RLC0_RB_CNTL); > > - pr_debug("sdma base address: 0x%x\n", retval); > + pr_debug("RLC register offset: 0x%x\n", retval); > > return retval; > } > @@ -107,22 +106,22 @@ static int kgd_hqd_sdma_load(struct kgd_dev *kgd, void *mqd, > { > struct amdgpu_device *adev = get_amdgpu_device(kgd); > struct v9_sdma_mqd *m; > - uint32_t sdma_base_addr; > + uint32_t rlc_reg_offset; > unsigned long end_jiffies; > uint32_t data; > uint64_t data64; > uint64_t __user *wptr64 = (uint64_t __user *)wptr; > > m = get_sdma_mqd(mqd); > - sdma_base_addr = get_sdma_base_addr(adev, m->sdma_engine_id, > + rlc_reg_offset = get_rlc_reg_offset(adev, m->sdma_engine_id, > m->sdma_queue_id); > > - WREG32(sdma_base_addr + mmSDMA0_RLC0_RB_CNTL, > + WREG32(rlc_reg_offset + mmSDMA0_RLC0_RB_CNTL, > m->sdmax_rlcx_rb_cntl & (~SDMA0_RLC0_RB_CNTL__RB_ENABLE_MASK)); > > end_jiffies = msecs_to_jiffies(2000) + jiffies; > while (true) { > - data = "" + mmSDMA0_RLC0_CONTEXT_STATUS); > + data = "" + mmSDMA0_RLC0_CONTEXT_STATUS); > if (data & SDMA0_RLC0_CONTEXT_STATUS__IDLE_MASK) > break; > if (time_after(jiffies, end_jiffies)) > @@ -130,41 +129,41 @@ static int kgd_hqd_sdma_load(struct kgd_dev *kgd, void *mqd, > usleep_range(500, 1000); > } > > - WREG32(sdma_base_addr + mmSDMA0_RLC0_DOORBELL_OFFSET, > + WREG32(rlc_reg_offset + mmSDMA0_RLC0_DOORBELL_OFFSET, > m->sdmax_rlcx_doorbell_offset); > > data = "" SDMA0_RLC0_DOORBELL, > ENABLE, 1); > - WREG32(sdma_base_addr + mmSDMA0_RLC0_DOORBELL, data); > - WREG32(sdma_base_addr + mmSDMA0_RLC0_RB_RPTR, m->sdmax_rlcx_rb_rptr); > - WREG32(sdma_base_addr + mmSDMA0_RLC0_RB_RPTR_HI, > + WREG32(rlc_reg_offset + mmSDMA0_RLC0_DOORBELL, data); > + WREG32(rlc_reg_offset + mmSDMA0_RLC0_RB_RPTR, m->sdmax_rlcx_rb_rptr); > + WREG32(rlc_reg_offset + mmSDMA0_RLC0_RB_RPTR_HI, > m->sdmax_rlcx_rb_rptr_hi); > > - WREG32(sdma_base_addr + mmSDMA0_RLC0_MINOR_PTR_UPDATE, 1); > + WREG32(rlc_reg_offset + mmSDMA0_RLC0_MINOR_PTR_UPDATE, 1); > if (read_user_wptr(mm, wptr64, data64)) { > - WREG32(sdma_base_addr + mmSDMA0_RLC0_RB_WPTR, > + WREG32(rlc_reg_offset + mmSDMA0_RLC0_RB_WPTR, > lower_32_bits(data64)); > - WREG32(sdma_base_addr + mmSDMA0_RLC0_RB_WPTR_HI, > + WREG32(rlc_reg_offset + mmSDMA0_RLC0_RB_WPTR_HI, > upper_32_bits(data64)); > } else { > - WREG32(sdma_base_addr + mmSDMA0_RLC0_RB_WPTR, > + WREG32(rlc_reg_offset + mmSDMA0_RLC0_RB_WPTR, > m->sdmax_rlcx_rb_rptr); > - WREG32(sdma_base_addr + mmSDMA0_RLC0_RB_WPTR_HI, > + WREG32(rlc_reg_offset + mmSDMA0_RLC0_RB_WPTR_HI, > m->sdmax_rlcx_rb_rptr_hi); > } > - WREG32(sdma_base_addr + mmSDMA0_RLC0_MINOR_PTR_UPDATE, 0); > + WREG32(rlc_reg_offset + mmSDMA0_RLC0_MINOR_PTR_UPDATE, 0); > > - WREG32(sdma_base_addr + mmSDMA0_RLC0_RB_BASE, m->sdmax_rlcx_rb_base); > - WREG32(sdma_base_addr + mmSDMA0_RLC0_RB_BASE_HI, > + WREG32(rlc_reg_offset + mmSDMA0_RLC0_RB_BASE, m->sdmax_rlcx_rb_base); > + WREG32(rlc_reg_offset + mmSDMA0_RLC0_RB_BASE_HI, > m->sdmax_rlcx_rb_base_hi); > - WREG32(sdma_base_addr + mmSDMA0_RLC0_RB_RPTR_ADDR_LO, > + WREG32(rlc_reg_offset + mmSDMA0_RLC0_RB_RPTR_ADDR_LO, > m->sdmax_rlcx_rb_rptr_addr_lo); > - WREG32(sdma_base_addr + mmSDMA0_RLC0_RB_RPTR_ADDR_HI, > + WREG32(rlc_reg_offset + mmSDMA0_RLC0_RB_RPTR_ADDR_HI, > m->sdmax_rlcx_rb_rptr_addr_hi); > > data = "" SDMA0_RLC0_RB_CNTL, > RB_ENABLE, 1); > - WREG32(sdma_base_addr + mmSDMA0_RLC0_RB_CNTL, data); > + WREG32(rlc_reg_offset + mmSDMA0_RLC0_RB_CNTL, data); > > return 0; > } > @@ -174,7 +173,7 @@ static int kgd_hqd_sdma_dump(struct kgd_dev *kgd, > uint32_t (**dump)[2], uint32_t *n_regs) > { > struct amdgpu_device *adev = get_amdgpu_device(kgd); > - uint32_t sdma_base_addr = get_sdma_base_addr(adev, engine_id, queue_id); > + uint32_t rlc_reg_offset = get_rlc_reg_offset(adev, engine_id, queue_id); > uint32_t i = 0, reg; > #undef HQD_N_REGS > #define HQD_N_REGS (19+6+7+10) > @@ -184,15 +183,15 @@ static int kgd_hqd_sdma_dump(struct kgd_dev *kgd, > return -ENOMEM; > > for (reg = mmSDMA0_RLC0_RB_CNTL; reg <= mmSDMA0_RLC0_DOORBELL; reg++) > - DUMP_REG(sdma_base_addr + reg); > + DUMP_REG(rlc_reg_offset + reg); > for (reg = mmSDMA0_RLC0_STATUS; reg <= mmSDMA0_RLC0_CSA_ADDR_HI; reg++) > - DUMP_REG(sdma_base_addr + reg); > + DUMP_REG(rlc_reg_offset + reg); > for (reg = mmSDMA0_RLC0_IB_SUB_REMAIN; > reg <= mmSDMA0_RLC0_MINOR_PTR_UPDATE; reg++) > - DUMP_REG(sdma_base_addr + reg); > + DUMP_REG(rlc_reg_offset + reg); > for (reg = mmSDMA0_RLC0_MIDCMD_DATA0; > reg <= mmSDMA0_RLC0_MIDCMD_CNTL; reg++) > - DUMP_REG(sdma_base_addr + reg); > + DUMP_REG(rlc_reg_offset + reg); > > WARN_ON_ONCE(i != HQD_N_REGS); > *n_regs = i; > @@ -204,14 +203,14 @@ static bool kgd_hqd_sdma_is_occupied(struct kgd_dev *kgd, void *mqd) > { > struct amdgpu_device *adev = get_amdgpu_device(kgd); > struct v9_sdma_mqd *m; > - uint32_t sdma_base_addr; > + uint32_t rlc_reg_offset; > uint32_t sdma_rlc_rb_cntl; > > m = get_sdma_mqd(mqd); > - sdma_base_addr = get_sdma_base_addr(adev, m->sdma_engine_id, > + rlc_reg_offset = get_rlc_reg_offset(adev, m->sdma_engine_id, > m->sdma_queue_id); > > - sdma_rlc_rb_cntl = RREG32(sdma_base_addr + mmSDMA0_RLC0_RB_CNTL); > + sdma_rlc_rb_cntl = RREG32(rlc_reg_offset + mmSDMA0_RLC0_RB_CNTL); > > if (sdma_rlc_rb_cntl & SDMA0_RLC0_RB_CNTL__RB_ENABLE_MASK) > return true; > @@ -224,20 +223,20 @@ static int kgd_hqd_sdma_destroy(struct kgd_dev *kgd, void *mqd, > { > struct amdgpu_device *adev = get_amdgpu_device(kgd); > struct v9_sdma_mqd *m; > - uint32_t sdma_base_addr; > + uint32_t rlc_reg_offset; > uint32_t temp; > unsigned long end_jiffies = (utimeout * HZ / 1000) + jiffies; > > m = get_sdma_mqd(mqd); > - sdma_base_addr = get_sdma_base_addr(adev, m->sdma_engine_id, > + rlc_reg_offset = get_rlc_reg_offset(adev, m->sdma_engine_id, > m->sdma_queue_id); > > - temp = RREG32(sdma_base_addr + mmSDMA0_RLC0_RB_CNTL); > + temp = RREG32(rlc_reg_offset + mmSDMA0_RLC0_RB_CNTL); > temp = temp & ~SDMA0_RLC0_RB_CNTL__RB_ENABLE_MASK; > - WREG32(sdma_base_addr + mmSDMA0_RLC0_RB_CNTL, temp); > + WREG32(rlc_reg_offset + mmSDMA0_RLC0_RB_CNTL, temp); > > while (true) { > - temp = RREG32(sdma_base_addr + mmSDMA0_RLC0_CONTEXT_STATUS); > + temp = RREG32(rlc_reg_offset + mmSDMA0_RLC0_CONTEXT_STATUS); > if (temp & SDMA0_RLC0_CONTEXT_STATUS__IDLE_MASK) > break; > if (time_after(jiffies, end_jiffies)) > @@ -245,14 +244,14 @@ static int kgd_hqd_sdma_destroy(struct kgd_dev *kgd, void *mqd, > usleep_range(500, 1000); > } > > - WREG32(sdma_base_addr + mmSDMA0_RLC0_DOORBELL, 0); > - WREG32(sdma_base_addr + mmSDMA0_RLC0_RB_CNTL, > - RREG32(sdma_base_addr + mmSDMA0_RLC0_RB_CNTL) | > + WREG32(rlc_reg_offset + mmSDMA0_RLC0_DOORBELL, 0); > + WREG32(rlc_reg_offset + mmSDMA0_RLC0_RB_CNTL, > + RREG32(rlc_reg_offset + mmSDMA0_RLC0_RB_CNTL) | > SDMA0_RLC0_RB_CNTL__RB_ENABLE_MASK); > > - m->sdmax_rlcx_rb_rptr = RREG32(sdma_base_addr + mmSDMA0_RLC0_RB_RPTR); > + m->sdmax_rlcx_rb_rptr = RREG32(rlc_reg_offset + mmSDMA0_RLC0_RB_RPTR); > m->sdmax_rlcx_rb_rptr_hi = > - RREG32(sdma_base_addr + mmSDMA0_RLC0_RB_RPTR_HI); > + RREG32(rlc_reg_offset + mmSDMA0_RLC0_RB_RPTR_HI); > > return 0; > } > diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_gfx_v10.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_gfx_v10.c > index a4325db8d093..ee520ad90717 100644 > --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_gfx_v10.c > +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_gfx_v10.c > @@ -307,11 +307,11 @@ static int kgd_init_interrupts(struct kgd_dev *kgd, uint32_t pipe_id) > return 0; > } > > -static uint32_t get_sdma_base_addr(struct amdgpu_device *adev, > +static uint32_t get_rlc_reg_offset(struct amdgpu_device *adev, > unsigned int engine_id, > unsigned int queue_id) > { > - uint32_t base[2] = { > + uint32_t sdma_engine_reg_base[2] = { > SOC15_REG_OFFSET(SDMA0, 0, > mmSDMA0_RLC0_RB_CNTL) - mmSDMA0_RLC0_RB_CNTL, > /* On gfx10, mmSDMA1_xxx registers are defined NOT based > @@ -323,12 +323,11 @@ static uint32_t get_sdma_base_addr(struct amdgpu_device *adev, > SOC15_REG_OFFSET(SDMA1, 0, > mmSDMA1_RLC0_RB_CNTL) - mmSDMA0_RLC0_RB_CNTL > }; > - uint32_t retval; > > - retval = base[engine_id] + queue_id * (mmSDMA0_RLC1_RB_CNTL - > - mmSDMA0_RLC0_RB_CNTL); > + uint32_t retval = sdma_engine_reg_base[engine_id] > + + queue_id * (mmSDMA0_RLC1_RB_CNTL - mmSDMA0_RLC0_RB_CNTL); > > - pr_debug("sdma base address: 0x%x\n", retval); > + pr_debug("RLC register offset: 0x%x\n", retval); > > return retval; > } > @@ -489,23 +488,23 @@ static int kgd_hqd_sdma_load(struct kgd_dev *kgd, void *mqd, > { > struct amdgpu_device *adev = get_amdgpu_device(kgd); > struct v10_sdma_mqd *m; > - uint32_t sdma_base_addr; > + uint32_t rlc_reg_offset; > unsigned long end_jiffies; > uint32_t data; > uint64_t data64; > uint64_t __user *wptr64 = (uint64_t __user *)wptr; > > m = get_sdma_mqd(mqd); > - sdma_base_addr = get_sdma_base_addr(adev, m->sdma_engine_id, > + rlc_reg_offset = get_rlc_reg_offset(adev, m->sdma_engine_id, > m->sdma_queue_id); > - pr_debug("sdma load base addr %x for engine %d, queue %d\n", sdma_base_addr, m->sdma_engine_id, m->sdma_queue_id); > + pr_debug("sdma load base addr 0x%x for engine %d, queue %d\n", rlc_reg_offset, m->sdma_engine_id, m->sdma_queue_id); > > - WREG32(sdma_base_addr + mmSDMA0_RLC0_RB_CNTL, > + WREG32(rlc_reg_offset + mmSDMA0_RLC0_RB_CNTL, > m->sdmax_rlcx_rb_cntl & (~SDMA0_RLC0_RB_CNTL__RB_ENABLE_MASK)); > > end_jiffies = msecs_to_jiffies(2000) + jiffies; > while (true) { > - data = "" + mmSDMA0_RLC0_CONTEXT_STATUS); > + data = "" + mmSDMA0_RLC0_CONTEXT_STATUS); > if (data & SDMA0_RLC0_CONTEXT_STATUS__IDLE_MASK) > break; > if (time_after(jiffies, end_jiffies)) > @@ -513,41 +512,41 @@ static int kgd_hqd_sdma_load(struct kgd_dev *kgd, void *mqd, > usleep_range(500, 1000); > } > > - WREG32(sdma_base_addr + mmSDMA0_RLC0_DOORBELL_OFFSET, > + WREG32(rlc_reg_offset + mmSDMA0_RLC0_DOORBELL_OFFSET, > m->sdmax_rlcx_doorbell_offset); > > data = "" SDMA0_RLC0_DOORBELL, > ENABLE, 1); > - WREG32(sdma_base_addr + mmSDMA0_RLC0_DOORBELL, data); > - WREG32(sdma_base_addr + mmSDMA0_RLC0_RB_RPTR, m->sdmax_rlcx_rb_rptr); > - WREG32(sdma_base_addr + mmSDMA0_RLC0_RB_RPTR_HI, > + WREG32(rlc_reg_offset + mmSDMA0_RLC0_DOORBELL, data); > + WREG32(rlc_reg_offset + mmSDMA0_RLC0_RB_RPTR, m->sdmax_rlcx_rb_rptr); > + WREG32(rlc_reg_offset + mmSDMA0_RLC0_RB_RPTR_HI, > m->sdmax_rlcx_rb_rptr_hi); > > - WREG32(sdma_base_addr + mmSDMA0_RLC0_MINOR_PTR_UPDATE, 1); > + WREG32(rlc_reg_offset + mmSDMA0_RLC0_MINOR_PTR_UPDATE, 1); > if (read_user_wptr(mm, wptr64, data64)) { > - WREG32(sdma_base_addr + mmSDMA0_RLC0_RB_WPTR, > + WREG32(rlc_reg_offset + mmSDMA0_RLC0_RB_WPTR, > lower_32_bits(data64)); > - WREG32(sdma_base_addr + mmSDMA0_RLC0_RB_WPTR_HI, > + WREG32(rlc_reg_offset + mmSDMA0_RLC0_RB_WPTR_HI, > upper_32_bits(data64)); > } else { > - WREG32(sdma_base_addr + mmSDMA0_RLC0_RB_WPTR, > + WREG32(rlc_reg_offset + mmSDMA0_RLC0_RB_WPTR, > m->sdmax_rlcx_rb_rptr); > - WREG32(sdma_base_addr + mmSDMA0_RLC0_RB_WPTR_HI, > + WREG32(rlc_reg_offset + mmSDMA0_RLC0_RB_WPTR_HI, > m->sdmax_rlcx_rb_rptr_hi); > } > - WREG32(sdma_base_addr + mmSDMA0_RLC0_MINOR_PTR_UPDATE, 0); > + WREG32(rlc_reg_offset + mmSDMA0_RLC0_MINOR_PTR_UPDATE, 0); > > - WREG32(sdma_base_addr + mmSDMA0_RLC0_RB_BASE, m->sdmax_rlcx_rb_base); > - WREG32(sdma_base_addr + mmSDMA0_RLC0_RB_BASE_HI, > + WREG32(rlc_reg_offset + mmSDMA0_RLC0_RB_BASE, m->sdmax_rlcx_rb_base); > + WREG32(rlc_reg_offset + mmSDMA0_RLC0_RB_BASE_HI, > m->sdmax_rlcx_rb_base_hi); > - WREG32(sdma_base_addr + mmSDMA0_RLC0_RB_RPTR_ADDR_LO, > + WREG32(rlc_reg_offset + mmSDMA0_RLC0_RB_RPTR_ADDR_LO, > m->sdmax_rlcx_rb_rptr_addr_lo); > - WREG32(sdma_base_addr + mmSDMA0_RLC0_RB_RPTR_ADDR_HI, > + WREG32(rlc_reg_offset + mmSDMA0_RLC0_RB_RPTR_ADDR_HI, > m->sdmax_rlcx_rb_rptr_addr_hi); > > data = "" SDMA0_RLC0_RB_CNTL, > RB_ENABLE, 1); > - WREG32(sdma_base_addr + mmSDMA0_RLC0_RB_CNTL, data); > + WREG32(rlc_reg_offset + mmSDMA0_RLC0_RB_CNTL, data); > > return 0; > } > @@ -557,28 +556,25 @@ static int kgd_hqd_sdma_dump(struct kgd_dev *kgd, > uint32_t (**dump)[2], uint32_t *n_regs) > { > struct amdgpu_device *adev = get_amdgpu_device(kgd); > - uint32_t sdma_base_addr = get_sdma_base_addr(adev, engine_id, queue_id); > + uint32_t rlc_reg_offset = get_rlc_reg_offset(adev, engine_id, queue_id); > uint32_t i = 0, reg; > #undef HQD_N_REGS > #define HQD_N_REGS (19+6+7+10) > > - pr_debug("sdma dump engine id %d queue_id %d\n", engine_id, queue_id); > - pr_debug("sdma base addr %x\n", sdma_base_addr); > - > *dump = kmalloc(HQD_N_REGS*2*sizeof(uint32_t), GFP_KERNEL); > if (*dump == NULL) > return -ENOMEM; > > for (reg = mmSDMA0_RLC0_RB_CNTL; reg <= mmSDMA0_RLC0_DOORBELL; reg++) > - DUMP_REG(sdma_base_addr + reg); > + DUMP_REG(rlc_reg_offset + reg); > for (reg = mmSDMA0_RLC0_STATUS; reg <= mmSDMA0_RLC0_CSA_ADDR_HI; reg++) > - DUMP_REG(sdma_base_addr + reg); > + DUMP_REG(rlc_reg_offset + reg); > for (reg = mmSDMA0_RLC0_IB_SUB_REMAIN; > reg <= mmSDMA0_RLC0_MINOR_PTR_UPDATE; reg++) > - DUMP_REG(sdma_base_addr + reg); > + DUMP_REG(rlc_reg_offset + reg); > for (reg = mmSDMA0_RLC0_MIDCMD_DATA0; > reg <= mmSDMA0_RLC0_MIDCMD_CNTL; reg++) > - DUMP_REG(sdma_base_addr + reg); > + DUMP_REG(rlc_reg_offset + reg); > > WARN_ON_ONCE(i != HQD_N_REGS); > *n_regs = i; > @@ -612,14 +608,14 @@ static bool kgd_hqd_sdma_is_occupied(struct kgd_dev *kgd, void *mqd) > { > struct amdgpu_device *adev = get_amdgpu_device(kgd); > struct v10_sdma_mqd *m; > - uint32_t sdma_base_addr; > + uint32_t rlc_reg_offset; > uint32_t sdma_rlc_rb_cntl; > > m = get_sdma_mqd(mqd); > - sdma_base_addr = get_sdma_base_addr(adev, m->sdma_engine_id, > + rlc_reg_offset = get_rlc_reg_offset(adev, m->sdma_engine_id, > m->sdma_queue_id); > > - sdma_rlc_rb_cntl = RREG32(sdma_base_addr + mmSDMA0_RLC0_RB_CNTL); > + sdma_rlc_rb_cntl = RREG32(rlc_reg_offset + mmSDMA0_RLC0_RB_CNTL); > > if (sdma_rlc_rb_cntl & SDMA0_RLC0_RB_CNTL__RB_ENABLE_MASK) > return true; > @@ -740,20 +736,20 @@ static int kgd_hqd_sdma_destroy(struct kgd_dev *kgd, void *mqd, > { > struct amdgpu_device *adev = get_amdgpu_device(kgd); > struct v10_sdma_mqd *m; > - uint32_t sdma_base_addr; > + uint32_t rlc_reg_offset; > uint32_t temp; > unsigned long end_jiffies = (utimeout * HZ / 1000) + jiffies; > > m = get_sdma_mqd(mqd); > - sdma_base_addr = get_sdma_base_addr(adev, m->sdma_engine_id, > + rlc_reg_offset = get_rlc_reg_offset(adev, m->sdma_engine_id, > m->sdma_queue_id); > > - temp = RREG32(sdma_base_addr + mmSDMA0_RLC0_RB_CNTL); > + temp = RREG32(rlc_reg_offset + mmSDMA0_RLC0_RB_CNTL); > temp = temp & ~SDMA0_RLC0_RB_CNTL__RB_ENABLE_MASK; > - WREG32(sdma_base_addr + mmSDMA0_RLC0_RB_CNTL, temp); > + WREG32(rlc_reg_offset + mmSDMA0_RLC0_RB_CNTL, temp); > > while (true) { > - temp = RREG32(sdma_base_addr + mmSDMA0_RLC0_CONTEXT_STATUS); > + temp = RREG32(rlc_reg_offset + mmSDMA0_RLC0_CONTEXT_STATUS); > if (temp & SDMA0_RLC0_CONTEXT_STATUS__IDLE_MASK) > break; > if (time_after(jiffies, end_jiffies)) > @@ -761,14 +757,14 @@ static int kgd_hqd_sdma_destroy(struct kgd_dev *kgd, void *mqd, > usleep_range(500, 1000); > } > > - WREG32(sdma_base_addr + mmSDMA0_RLC0_DOORBELL, 0); > - WREG32(sdma_base_addr + mmSDMA0_RLC0_RB_CNTL, > - RREG32(sdma_base_addr + mmSDMA0_RLC0_RB_CNTL) | > + WREG32(rlc_reg_offset + mmSDMA0_RLC0_DOORBELL, 0); > + WREG32(rlc_reg_offset + mmSDMA0_RLC0_RB_CNTL, > + RREG32(rlc_reg_offset + mmSDMA0_RLC0_RB_CNTL) | > SDMA0_RLC0_RB_CNTL__RB_ENABLE_MASK); > > - m->sdmax_rlcx_rb_rptr = RREG32(sdma_base_addr + mmSDMA0_RLC0_RB_RPTR); > + m->sdmax_rlcx_rb_rptr = RREG32(rlc_reg_offset + mmSDMA0_RLC0_RB_RPTR); > m->sdmax_rlcx_rb_rptr_hi = > - RREG32(sdma_base_addr + mmSDMA0_RLC0_RB_RPTR_HI); > + RREG32(rlc_reg_offset + mmSDMA0_RLC0_RB_RPTR_HI); > > return 0; > } > diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_gfx_v7.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_gfx_v7.c > index c6abcf72e822..d0517b7ae089 100644 > --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_gfx_v7.c > +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_gfx_v7.c > @@ -303,14 +303,14 @@ static int kgd_init_interrupts(struct kgd_dev *kgd, uint32_t pipe_id) > return 0; > } > > -static inline uint32_t get_sdma_base_addr(struct cik_sdma_rlc_registers *m) > +static inline uint32_t get_rlc_reg_offset(struct cik_sdma_rlc_registers *m) > { > uint32_t retval; > > retval = m->sdma_engine_id * SDMA1_REGISTER_OFFSET + > m->sdma_queue_id * KFD_CIK_SDMA_QUEUE_OFFSET; > > - pr_debug("sdma base address: 0x%x\n", retval); > + pr_debug("RLC register offset: 0x%x\n", retval); > > return retval; > } > @@ -417,7 +417,7 @@ static int kgd_hqd_sdma_load(struct kgd_dev *kgd, void *mqd, > uint32_t data; > > m = get_sdma_mqd(mqd); > - sdma_base_addr = get_sdma_base_addr(m); > + sdma_base_addr = get_rlc_reg_offset(m); > > WREG32(sdma_base_addr + mmSDMA0_RLC0_RB_CNTL, > m->sdma_rlc_rb_cntl & (~SDMA0_RLC0_RB_CNTL__RB_ENABLE_MASK)); > @@ -528,7 +528,7 @@ static bool kgd_hqd_sdma_is_occupied(struct kgd_dev *kgd, void *mqd) > uint32_t sdma_rlc_rb_cntl; > > m = get_sdma_mqd(mqd); > - sdma_base_addr = get_sdma_base_addr(m); > + sdma_base_addr = get_rlc_reg_offset(m); > > sdma_rlc_rb_cntl = RREG32(sdma_base_addr + mmSDMA0_RLC0_RB_CNTL); > > @@ -650,7 +650,7 @@ static int kgd_hqd_sdma_destroy(struct kgd_dev *kgd, void *mqd, > unsigned long end_jiffies = (utimeout * HZ / 1000) + jiffies; > > m = get_sdma_mqd(mqd); > - sdma_base_addr = get_sdma_base_addr(m); > + sdma_base_addr = get_rlc_reg_offset(m); > > temp = RREG32(sdma_base_addr + mmSDMA0_RLC0_RB_CNTL); > temp = temp & ~SDMA0_RLC0_RB_CNTL__RB_ENABLE_MASK; > diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_gfx_v8.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_gfx_v8.c > index 4e8b4e949926..373501abdb6b 100644 > --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_gfx_v8.c > +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_gfx_v8.c > @@ -260,13 +260,13 @@ static int kgd_init_interrupts(struct kgd_dev *kgd, uint32_t pipe_id) > return 0; > } > > -static inline uint32_t get_sdma_base_addr(struct vi_sdma_mqd *m) > +static inline uint32_t get_rlc_reg_offset(struct vi_sdma_mqd *m) > { > uint32_t retval; > > retval = m->sdma_engine_id * SDMA1_REGISTER_OFFSET + > m->sdma_queue_id * KFD_VI_SDMA_QUEUE_OFFSET; > - pr_debug("sdma base address: 0x%x\n", retval); > + pr_debug("RLC register offset: 0x%x\n", retval); > > return retval; > } > @@ -402,7 +402,7 @@ static int kgd_hqd_sdma_load(struct kgd_dev *kgd, void *mqd, > uint32_t data; > > m = get_sdma_mqd(mqd); > - sdma_base_addr = get_sdma_base_addr(m); > + sdma_base_addr = get_rlc_reg_offset(m); > WREG32(sdma_base_addr + mmSDMA0_RLC0_RB_CNTL, > m->sdmax_rlcx_rb_cntl & (~SDMA0_RLC0_RB_CNTL__RB_ENABLE_MASK)); > > @@ -521,7 +521,7 @@ static bool kgd_hqd_sdma_is_occupied(struct kgd_dev *kgd, void *mqd) > uint32_t sdma_rlc_rb_cntl; > > m = get_sdma_mqd(mqd); > - sdma_base_addr = get_sdma_base_addr(m); > + sdma_base_addr = get_rlc_reg_offset(m); > > sdma_rlc_rb_cntl = RREG32(sdma_base_addr + mmSDMA0_RLC0_RB_CNTL); > > @@ -646,7 +646,7 @@ static int kgd_hqd_sdma_destroy(struct kgd_dev *kgd, void *mqd, > unsigned long end_jiffies = (utimeout * HZ / 1000) + jiffies; > > m = get_sdma_mqd(mqd); > - sdma_base_addr = get_sdma_base_addr(m); > + sdma_base_addr = get_rlc_reg_offset(m); > > temp = RREG32(sdma_base_addr + mmSDMA0_RLC0_RB_CNTL); > temp = temp & ~SDMA0_RLC0_RB_CNTL__RB_ENABLE_MASK; > diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_gfx_v9.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_gfx_v9.c > index 55437f160a72..2dd5bc676029 100644 > --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_gfx_v9.c > +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_gfx_v9.c > @@ -226,22 +226,20 @@ int kgd_gfx_v9_init_interrupts(struct kgd_dev *kgd, uint32_t pipe_id) > return 0; > } > > -static uint32_t get_sdma_base_addr(struct amdgpu_device *adev, > +static uint32_t get_rlc_reg_offset(struct amdgpu_device *adev, > unsigned int engine_id, > unsigned int queue_id) > { > - uint32_t base[2] = { > + uint32_t sdma_engine_reg_base[2] = { > SOC15_REG_OFFSET(SDMA0, 0, > mmSDMA0_RLC0_RB_CNTL) - mmSDMA0_RLC0_RB_CNTL, > SOC15_REG_OFFSET(SDMA1, 0, > mmSDMA1_RLC0_RB_CNTL) - mmSDMA1_RLC0_RB_CNTL > }; > - uint32_t retval; > + uint32_t retval = sdma_engine_reg_base[engine_id] > + + queue_id * (mmSDMA0_RLC1_RB_CNTL - mmSDMA0_RLC0_RB_CNTL); > > - retval = base[engine_id] + queue_id * (mmSDMA0_RLC1_RB_CNTL - > - mmSDMA0_RLC0_RB_CNTL); > - > - pr_debug("sdma base address: 0x%x\n", retval); > + pr_debug("RLC register offset: 0x%x\n", retval); > > return retval; > } > @@ -388,22 +386,22 @@ static int kgd_hqd_sdma_load(struct kgd_dev *kgd, void *mqd, > { > struct amdgpu_device *adev = get_amdgpu_device(kgd); > struct v9_sdma_mqd *m; > - uint32_t sdma_base_addr; > + uint32_t rlc_reg_offset; > unsigned long end_jiffies; > uint32_t data; > uint64_t data64; > uint64_t __user *wptr64 = (uint64_t __user *)wptr; > > m = get_sdma_mqd(mqd); > - sdma_base_addr = get_sdma_base_addr(adev, m->sdma_engine_id, > + rlc_reg_offset = get_rlc_reg_offset(adev, m->sdma_engine_id, > m->sdma_queue_id); > > - WREG32(sdma_base_addr + mmSDMA0_RLC0_RB_CNTL, > + WREG32(rlc_reg_offset + mmSDMA0_RLC0_RB_CNTL, > m->sdmax_rlcx_rb_cntl & (~SDMA0_RLC0_RB_CNTL__RB_ENABLE_MASK)); > > end_jiffies = msecs_to_jiffies(2000) + jiffies; > while (true) { > - data = "" + mmSDMA0_RLC0_CONTEXT_STATUS); > + data = "" + mmSDMA0_RLC0_CONTEXT_STATUS); > if (data & SDMA0_RLC0_CONTEXT_STATUS__IDLE_MASK) > break; > if (time_after(jiffies, end_jiffies)) > @@ -411,41 +409,41 @@ static int kgd_hqd_sdma_load(struct kgd_dev *kgd, void *mqd, > usleep_range(500, 1000); > } > > - WREG32(sdma_base_addr + mmSDMA0_RLC0_DOORBELL_OFFSET, > + WREG32(rlc_reg_offset + mmSDMA0_RLC0_DOORBELL_OFFSET, > m->sdmax_rlcx_doorbell_offset); > > data = "" SDMA0_RLC0_DOORBELL, > ENABLE, 1); > - WREG32(sdma_base_addr + mmSDMA0_RLC0_DOORBELL, data); > - WREG32(sdma_base_addr + mmSDMA0_RLC0_RB_RPTR, m->sdmax_rlcx_rb_rptr); > - WREG32(sdma_base_addr + mmSDMA0_RLC0_RB_RPTR_HI, > + WREG32(rlc_reg_offset + mmSDMA0_RLC0_DOORBELL, data); > + WREG32(rlc_reg_offset + mmSDMA0_RLC0_RB_RPTR, m->sdmax_rlcx_rb_rptr); > + WREG32(rlc_reg_offset + mmSDMA0_RLC0_RB_RPTR_HI, > m->sdmax_rlcx_rb_rptr_hi); > > - WREG32(sdma_base_addr + mmSDMA0_RLC0_MINOR_PTR_UPDATE, 1); > + WREG32(rlc_reg_offset + mmSDMA0_RLC0_MINOR_PTR_UPDATE, 1); > if (read_user_wptr(mm, wptr64, data64)) { > - WREG32(sdma_base_addr + mmSDMA0_RLC0_RB_WPTR, > + WREG32(rlc_reg_offset + mmSDMA0_RLC0_RB_WPTR, > lower_32_bits(data64)); > - WREG32(sdma_base_addr + mmSDMA0_RLC0_RB_WPTR_HI, > + WREG32(rlc_reg_offset + mmSDMA0_RLC0_RB_WPTR_HI, > upper_32_bits(data64)); > } else { > - WREG32(sdma_base_addr + mmSDMA0_RLC0_RB_WPTR, > + WREG32(rlc_reg_offset + mmSDMA0_RLC0_RB_WPTR, > m->sdmax_rlcx_rb_rptr); > - WREG32(sdma_base_addr + mmSDMA0_RLC0_RB_WPTR_HI, > + WREG32(rlc_reg_offset + mmSDMA0_RLC0_RB_WPTR_HI, > m->sdmax_rlcx_rb_rptr_hi); > } > - WREG32(sdma_base_addr + mmSDMA0_RLC0_MINOR_PTR_UPDATE, 0); > + WREG32(rlc_reg_offset + mmSDMA0_RLC0_MINOR_PTR_UPDATE, 0); > > - WREG32(sdma_base_addr + mmSDMA0_RLC0_RB_BASE, m->sdmax_rlcx_rb_base); > - WREG32(sdma_base_addr + mmSDMA0_RLC0_RB_BASE_HI, > + WREG32(rlc_reg_offset + mmSDMA0_RLC0_RB_BASE, m->sdmax_rlcx_rb_base); > + WREG32(rlc_reg_offset + mmSDMA0_RLC0_RB_BASE_HI, > m->sdmax_rlcx_rb_base_hi); > - WREG32(sdma_base_addr + mmSDMA0_RLC0_RB_RPTR_ADDR_LO, > + WREG32(rlc_reg_offset + mmSDMA0_RLC0_RB_RPTR_ADDR_LO, > m->sdmax_rlcx_rb_rptr_addr_lo); > - WREG32(sdma_base_addr + mmSDMA0_RLC0_RB_RPTR_ADDR_HI, > + WREG32(rlc_reg_offset + mmSDMA0_RLC0_RB_RPTR_ADDR_HI, > m->sdmax_rlcx_rb_rptr_addr_hi); > > data = "" SDMA0_RLC0_RB_CNTL, > RB_ENABLE, 1); > - WREG32(sdma_base_addr + mmSDMA0_RLC0_RB_CNTL, data); > + WREG32(rlc_reg_offset + mmSDMA0_RLC0_RB_CNTL, data); > > return 0; > } > @@ -455,7 +453,7 @@ static int kgd_hqd_sdma_dump(struct kgd_dev *kgd, > uint32_t (**dump)[2], uint32_t *n_regs) > { > struct amdgpu_device *adev = get_amdgpu_device(kgd); > - uint32_t sdma_base_addr = get_sdma_base_addr(adev, engine_id, queue_id); > + uint32_t rlc_reg_offset = get_rlc_reg_offset(adev, engine_id, queue_id); > uint32_t i = 0, reg; > #undef HQD_N_REGS > #define HQD_N_REGS (19+6+7+10) > @@ -465,15 +463,15 @@ static int kgd_hqd_sdma_dump(struct kgd_dev *kgd, > return -ENOMEM; > > for (reg = mmSDMA0_RLC0_RB_CNTL; reg <= mmSDMA0_RLC0_DOORBELL; reg++) > - DUMP_REG(sdma_base_addr + reg); > + DUMP_REG(rlc_reg_offset + reg); > for (reg = mmSDMA0_RLC0_STATUS; reg <= mmSDMA0_RLC0_CSA_ADDR_HI; reg++) > - DUMP_REG(sdma_base_addr + reg); > + DUMP_REG(rlc_reg_offset + reg); > for (reg = mmSDMA0_RLC0_IB_SUB_REMAIN; > reg <= mmSDMA0_RLC0_MINOR_PTR_UPDATE; reg++) > - DUMP_REG(sdma_base_addr + reg); > + DUMP_REG(rlc_reg_offset + reg); > for (reg = mmSDMA0_RLC0_MIDCMD_DATA0; > reg <= mmSDMA0_RLC0_MIDCMD_CNTL; reg++) > - DUMP_REG(sdma_base_addr + reg); > + DUMP_REG(rlc_reg_offset + reg); > > WARN_ON_ONCE(i != HQD_N_REGS); > *n_regs = i; > @@ -507,14 +505,14 @@ static bool kgd_hqd_sdma_is_occupied(struct kgd_dev *kgd, void *mqd) > { > struct amdgpu_device *adev = get_amdgpu_device(kgd); > struct v9_sdma_mqd *m; > - uint32_t sdma_base_addr; > + uint32_t rlc_reg_offset; > uint32_t sdma_rlc_rb_cntl; > > m = get_sdma_mqd(mqd); > - sdma_base_addr = get_sdma_base_addr(adev, m->sdma_engine_id, > + rlc_reg_offset = get_rlc_reg_offset(adev, m->sdma_engine_id, > m->sdma_queue_id); > > - sdma_rlc_rb_cntl = RREG32(sdma_base_addr + mmSDMA0_RLC0_RB_CNTL); > + sdma_rlc_rb_cntl = RREG32(rlc_reg_offset + mmSDMA0_RLC0_RB_CNTL); > > if (sdma_rlc_rb_cntl & SDMA0_RLC0_RB_CNTL__RB_ENABLE_MASK) > return true; > @@ -577,20 +575,20 @@ static int kgd_hqd_sdma_destroy(struct kgd_dev *kgd, void *mqd, > { > struct amdgpu_device *adev = get_amdgpu_device(kgd); > struct v9_sdma_mqd *m; > - uint32_t sdma_base_addr; > + uint32_t rlc_reg_offset; > uint32_t temp; > unsigned long end_jiffies = (utimeout * HZ / 1000) + jiffies; > > m = get_sdma_mqd(mqd); > - sdma_base_addr = get_sdma_base_addr(adev, m->sdma_engine_id, > + rlc_reg_offset = get_rlc_reg_offset(adev, m->sdma_engine_id, > m->sdma_queue_id); > > - temp = RREG32(sdma_base_addr + mmSDMA0_RLC0_RB_CNTL); > + temp = RREG32(rlc_reg_offset + mmSDMA0_RLC0_RB_CNTL); > temp = temp & ~SDMA0_RLC0_RB_CNTL__RB_ENABLE_MASK; > - WREG32(sdma_base_addr + mmSDMA0_RLC0_RB_CNTL, temp); > + WREG32(rlc_reg_offset + mmSDMA0_RLC0_RB_CNTL, temp); > > while (true) { > - temp = RREG32(sdma_base_addr + mmSDMA0_RLC0_CONTEXT_STATUS); > + temp = RREG32(rlc_reg_offset + mmSDMA0_RLC0_CONTEXT_STATUS); > if (temp & SDMA0_RLC0_CONTEXT_STATUS__IDLE_MASK) > break; > if (time_after(jiffies, end_jiffies)) > @@ -598,14 +596,14 @@ static int kgd_hqd_sdma_destroy(struct kgd_dev *kgd, void *mqd, > usleep_range(500, 1000); > } > > - WREG32(sdma_base_addr + mmSDMA0_RLC0_DOORBELL, 0); > - WREG32(sdma_base_addr + mmSDMA0_RLC0_RB_CNTL, > - RREG32(sdma_base_addr + mmSDMA0_RLC0_RB_CNTL) | > + WREG32(rlc_reg_offset + mmSDMA0_RLC0_DOORBELL, 0); > + WREG32(rlc_reg_offset + mmSDMA0_RLC0_RB_CNTL, > + RREG32(rlc_reg_offset + mmSDMA0_RLC0_RB_CNTL) | > SDMA0_RLC0_RB_CNTL__RB_ENABLE_MASK); > > - m->sdmax_rlcx_rb_rptr = RREG32(sdma_base_addr + mmSDMA0_RLC0_RB_RPTR); > + m->sdmax_rlcx_rb_rptr = RREG32(rlc_reg_offset + mmSDMA0_RLC0_RB_RPTR); > m->sdmax_rlcx_rb_rptr_hi = > - RREG32(sdma_base_addr + mmSDMA0_RLC0_RB_RPTR_HI); > + RREG32(rlc_reg_offset + mmSDMA0_RLC0_RB_RPTR_HI); > > return 0; > } |
_______________________________________________ amd-gfx mailing list amd-gfx@xxxxxxxxxxxxxxxxxxxxx https://lists.freedesktop.org/mailman/listinfo/amd-gfx
