Donate to e Foundation | Murena handsets with /e/OS | Own a part of Murena! Learn more

Commit 792e8f6f authored by qctecmdr Service's avatar qctecmdr Service Committed by Gerrit - the friendly Code Review server
Browse files

Merge "msm: gsi: Change user_data to support OOO completions"

parents 64dd490f 06ae5b7b

drivers/platform/msm/gsi/gsi.c

+123 −17
Original line number Diff line number Diff line
@@ -466,7 +466,8 @@ static void gsi_process_chan(struct gsi_xfer_compl_evt *evt, 
	}



	ch_ctx = &gsi_ctx->chan[ch_id];

	if (WARN_ON(ch_ctx->props.prot != GSI_CHAN_PROT_GPI))

	if (WARN_ON(ch_ctx->props.prot != GSI_CHAN_PROT_GPI &&

		ch_ctx->props.prot != GSI_CHAN_PROT_GCI))

		return;



	if (evt->type != GSI_XFER_COMPL_TYPE_GCI) {
@@ -493,7 +494,10 @@ static void gsi_process_chan(struct gsi_xfer_compl_evt *evt, 


	ch_ctx->stats.completed++;



	notify->xfer_user_data = ch_ctx->user_data[rp_idx];

	WARN_ON(!ch_ctx->user_data[rp_idx].valid);

	notify->xfer_user_data = ch_ctx->user_data[rp_idx].p;

	ch_ctx->user_data[rp_idx].valid = false;



	notify->chan_user_data = ch_ctx->props.chan_user_data;

	notify->evt_id = evt->code;

	notify->bytes_xfered = evt->len;
@@ -525,6 +529,7 @@ static void gsi_ring_evt_doorbell(struct gsi_evt_ctx *ctx) 
{

	uint32_t val;



	ctx->ring.wp = ctx->ring.wp_local;

	val = (ctx->ring.wp_local &

			GSI_EE_n_EV_CH_k_DOORBELL_0_WRITE_PTR_LSB_BMSK) <<

			GSI_EE_n_EV_CH_k_DOORBELL_0_WRITE_PTR_LSB_SHFT;
@@ -1485,6 +1490,55 @@ static int gsi_validate_evt_ring_props(struct gsi_evt_ring_props *props) 
	return GSI_STATUS_SUCCESS;

}



/**

 * gsi_cleanup_xfer_user_data: cleanup the user data array using callback passed

 *	by IPA driver. Need to do this in GSI since only GSI knows which TRE

 *	are being used or not. However, IPA is the one that does cleaning,

 *	therefore we pass a callback from IPA and call it using params from GSI

 *

 * @chan_hdl: hdl of the gsi channel user data array to be cleaned

 * @cleanup_cb: callback used to clean the user data array. takes 2 inputs

 *	@chan_user_data: ipa_sys_context of the gsi_channel

 *	@xfer_uder_data: user data array element (rx_pkt wrapper)

 *

 * Returns: 0 on success, negative on failure

 */

static int gsi_cleanup_xfer_user_data(unsigned long chan_hdl,

	void (*cleanup_cb)(void *chan_user_data, void *xfer_user_data))

{

	struct gsi_chan_ctx *ctx;

	uint64_t i;

	uint16_t rp_idx;



	ctx = &gsi_ctx->chan[chan_hdl];

	if (ctx->state != GSI_CHAN_STATE_ALLOCATED) {

		GSIERR("bad state %d\n", ctx->state);

		return -GSI_STATUS_UNSUPPORTED_OP;

	}



	/* for coalescing, traverse the whole array */

	if (ctx->props.prot == GSI_CHAN_PROT_GCI) {

		size_t user_data_size =

			ctx->ring.max_num_elem + 1 + GSI_VEID_MAX;

		for (i = 0; i < user_data_size; i++) {

			if (ctx->user_data[i].valid)

				cleanup_cb(ctx->props.chan_user_data,

					ctx->user_data[i].p);

		}

	} else {

		/* for non-coalescing, clean between RP and WP */

		while (ctx->ring.rp_local != ctx->ring.wp_local) {

			rp_idx = gsi_find_idx_from_addr(&ctx->ring,

				ctx->ring.rp_local);

			WARN_ON(!ctx->user_data[rp_idx].valid);

			cleanup_cb(ctx->props.chan_user_data,

				ctx->user_data[rp_idx].p);

			gsi_incr_ring_rp(&ctx->ring);

		}

	}

	return 0;

}



int gsi_alloc_evt_ring(struct gsi_evt_ring_props *props, unsigned long dev_hdl,

		unsigned long *evt_ring_hdl)

{
@@ -2163,7 +2217,8 @@ int gsi_alloc_channel(struct gsi_chan_props *props, unsigned long dev_hdl, 
	int ee;

	enum gsi_ch_cmd_opcode op = GSI_CH_ALLOCATE;

	uint8_t erindex;

	void **user_data;

	struct gsi_user_data *user_data;

	size_t user_data_size;



	if (!gsi_ctx) {

		pr_err("%s:%d gsi context not allocated\n", __func__, __LINE__);
@@ -2187,10 +2242,11 @@ int gsi_alloc_channel(struct gsi_chan_props *props, unsigned long dev_hdl, 
			return -GSI_STATUS_INVALID_PARAMS;

		}



		if (props->prot != GSI_CHAN_PROT_GCI &&

			atomic_read(

		if (atomic_read(

			&gsi_ctx->evtr[props->evt_ring_hdl].chan_ref_cnt) &&

			gsi_ctx->evtr[props->evt_ring_hdl].props.exclusive) {

			gsi_ctx->evtr[props->evt_ring_hdl].props.exclusive &&

			gsi_ctx->evtr[props->evt_ring_hdl].chan->props.prot !=

			GSI_CHAN_PROT_GCI) {

			GSIERR("evt ring=%lu exclusively used by ch_hdl=%pK\n",

				props->evt_ring_hdl, chan_hdl);

			return -GSI_STATUS_UNSUPPORTED_OP;
@@ -2202,10 +2258,18 @@ int gsi_alloc_channel(struct gsi_chan_props *props, unsigned long dev_hdl, 
		GSIERR("chan %d already allocated\n", props->ch_id);

		return -GSI_STATUS_NODEV;

	}



	memset(ctx, 0, sizeof(*ctx));

	user_data_size = props->ring_len / props->re_size;

	/*

	 * GCI channels might have OOO event completions up to GSI_VEID_MAX.

	 * user_data needs to be large enough to accommodate those.

	 * TODO: increase user data size if GSI_VEID_MAX is not enough

	 */

	if (props->prot == GSI_CHAN_PROT_GCI)

		user_data_size += GSI_VEID_MAX;



	user_data = devm_kzalloc(gsi_ctx->dev,

		(props->ring_len / props->re_size) * sizeof(void *),

		user_data_size * sizeof(*user_data),

		GFP_KERNEL);

	if (user_data == NULL) {

		GSIERR("context not allocated\n");
@@ -2256,7 +2320,9 @@ int gsi_alloc_channel(struct gsi_chan_props *props, unsigned long dev_hdl, 
	}

	ctx->evtr = &gsi_ctx->evtr[erindex];

	atomic_inc(&ctx->evtr->chan_ref_cnt);

	if (props->prot != GSI_CHAN_PROT_GCI && ctx->evtr->props.exclusive)

	if (props->prot != GSI_CHAN_PROT_GCI &&

		ctx->evtr->props.exclusive &&

		atomic_read(&ctx->evtr->chan_ref_cnt) == 1)

		ctx->evtr->chan = ctx;



	gsi_program_chan_ctx(props, gsi_ctx->per.ee, erindex);
@@ -2882,6 +2948,9 @@ int gsi_reset_channel(unsigned long chan_hdl) 
		goto reset;

	}



	if (ctx->props.cleanup_cb)

		gsi_cleanup_xfer_user_data(chan_hdl, ctx->props.cleanup_cb);



	gsi_program_chan_ctx(&ctx->props, gsi_ctx->per.ee,

			ctx->evtr ? ctx->evtr->id : GSI_NO_EVT_ERINDEX);

	gsi_init_chan_ring(&ctx->props, &ctx->ring);
@@ -3111,7 +3180,8 @@ int gsi_is_channel_empty(unsigned long chan_hdl, bool *is_empty) 
	ctx = &gsi_ctx->chan[chan_hdl];

	ee = gsi_ctx->per.ee;



	if (ctx->props.prot != GSI_CHAN_PROT_GPI) {

	if (ctx->props.prot != GSI_CHAN_PROT_GPI &&

		ctx->props.prot != GSI_CHAN_PROT_GCI) {

		GSIERR("op not supported for protocol %u\n", ctx->props.prot);

		return -GSI_STATUS_UNSUPPORTED_OP;

	}
@@ -3167,7 +3237,37 @@ int gsi_is_channel_empty(unsigned long chan_hdl, bool *is_empty) 
}

EXPORT_SYMBOL(gsi_is_channel_empty);



int __gsi_populate_gci_tre(struct gsi_chan_ctx *ctx, struct gsi_xfer_elem *xfer)

int __gsi_get_gci_cookie(struct gsi_chan_ctx *ctx, uint16_t idx)

{

	int i;

	int end;



	if (!ctx->user_data[idx].valid) {

		ctx->user_data[idx].valid = true;

		return idx;

	}



	/*

	 * at this point we need to find an "escape buffer" for the cookie

	 * as the userdata in this spot is in use. This happens if the TRE at

	 * idx is not completed yet and it is getting reused by a new TRE.

	 */

	ctx->stats.userdata_in_use++;

	for (i = 0; i < GSI_VEID_MAX; i++) {

		end = ctx->ring.max_num_elem + 1;

		if (!ctx->user_data[end + i].valid) {

			ctx->user_data[end + i].valid = true;

			return end + i;

		}

	}



	/* TODO: Increase escape buffer size if we hit this */

	GSIERR("user_data is full\n");

	return -EPERM;

}



int __gsi_populate_gci_tre(struct gsi_chan_ctx *ctx,

	struct gsi_xfer_elem *xfer)

{

	struct gsi_gci_tre gci_tre;

	struct gsi_gci_tre *tre_gci_ptr;
@@ -3193,11 +3293,13 @@ int __gsi_populate_gci_tre(struct gsi_chan_ctx *ctx, struct gsi_xfer_elem *xfer) 
	gci_tre.buffer_ptr = xfer->addr;

	gci_tre.buf_len = xfer->len;

	gci_tre.re_type = GSI_RE_COAL;

	gci_tre.cookie = idx;

	gci_tre.cookie = __gsi_get_gci_cookie(ctx, idx);

	if (gci_tre.cookie < 0)

		return -EPERM;



	/* write the TRE to ring */

	*tre_gci_ptr = gci_tre;

	ctx->user_data[idx] = xfer->xfer_user_data;

	ctx->user_data[idx].p = xfer->xfer_user_data;



	return 0;

}
@@ -3235,7 +3337,8 @@ int __gsi_populate_tre(struct gsi_chan_ctx *ctx, 


	/* write the TRE to ring */

	*tre_ptr = tre;

	ctx->user_data[idx] = xfer->xfer_user_data;

	ctx->user_data[idx].valid = true;

	ctx->user_data[idx].p = xfer->xfer_user_data;



	return 0;

}
@@ -3338,7 +3441,8 @@ int gsi_start_xfer(unsigned long chan_hdl) 


	ctx = &gsi_ctx->chan[chan_hdl];



	if (ctx->props.prot != GSI_CHAN_PROT_GPI) {

	if (ctx->props.prot != GSI_CHAN_PROT_GPI &&

		ctx->props.prot != GSI_CHAN_PROT_GCI) {

		GSIERR("op not supported for protocol %u\n", ctx->props.prot);

		return -GSI_STATUS_UNSUPPORTED_OP;

	}
@@ -3393,7 +3497,8 @@ int gsi_poll_n_channel(unsigned long chan_hdl, 
	ctx = &gsi_ctx->chan[chan_hdl];

	ee = gsi_ctx->per.ee;



	if (ctx->props.prot != GSI_CHAN_PROT_GPI) {

	if (ctx->props.prot != GSI_CHAN_PROT_GPI &&

		ctx->props.prot != GSI_CHAN_PROT_GCI) {

		GSIERR("op not supported for protocol %u\n", ctx->props.prot);

		return -GSI_STATUS_UNSUPPORTED_OP;

	}
@@ -3455,7 +3560,8 @@ int gsi_config_channel_mode(unsigned long chan_hdl, enum gsi_chan_mode mode) 


	ctx = &gsi_ctx->chan[chan_hdl];



	if (ctx->props.prot != GSI_CHAN_PROT_GPI) {

	if (ctx->props.prot != GSI_CHAN_PROT_GPI &&

		ctx->props.prot != GSI_CHAN_PROT_GCI) {

		GSIERR("op not supported for protocol %u\n", ctx->props.prot);

		return -GSI_STATUS_UNSUPPORTED_OP;

	}

drivers/platform/msm/gsi/gsi.h

+13 −1
Original line number Diff line number Diff line
@@ -121,14 +121,26 @@ struct gsi_chan_stats { 
	unsigned long invalid_tre_error;

	unsigned long poll_ok;

	unsigned long poll_empty;

	unsigned long userdata_in_use;

	struct gsi_chan_dp_stats dp;

};



/**

 * struct gsi_user_data - user_data element pointed by the TRE

 * @valid: valid to be cleaned. if its true that means it is being used.

 *	false means its free to overwrite

 * @p: pointer to the user data array element

 */

struct gsi_user_data {

	bool valid;

	void *p;

};



struct gsi_chan_ctx {

	struct gsi_chan_props props;

	enum gsi_chan_state state;

	struct gsi_ring_ctx ring;

	void **user_data;

	struct gsi_user_data *user_data;

	struct gsi_evt_ctx *evtr;

	struct mutex mlock;

	struct completion compl;

drivers/platform/msm/gsi/gsi_dbg.c

+1 −0
Original line number Diff line number Diff line
@@ -278,6 +278,7 @@ static void gsi_dump_ch_stats(struct gsi_chan_ctx *ctx) 
	if (ctx->evtr)

		PRT_STAT("compl_evt=%lu\n",

			ctx->evtr->stats.completed);

	PRT_STAT("userdata_in_use=%lu\n", ctx->stats.userdata_in_use);



	PRT_STAT("ch_below_lo=%lu\n", ctx->stats.dp.ch_below_lo);

	PRT_STAT("ch_below_hi=%lu\n", ctx->stats.dp.ch_below_hi);

drivers/platform/msm/ipa/ipa_v3/ipa_dma.c

+1 −3
Original line number Diff line number Diff line 
// SPDX-License-Identifier: GPL-2.0

/*

 * Copyright (c) 2015-2018, The Linux Foundation. All rights reserved.

 * Copyright (c) 2015-2019, The Linux Foundation. All rights reserved.

 */
@@ -1089,11 +1089,9 @@ void ipa3_dma_async_memcpy_notify_cb(void *priv 
	struct ipa3_dma_xfer_wrapper *xfer_descr_expected;

	struct ipa3_sys_context *sys;

	unsigned long flags;

	struct ipa_mem_buffer *mem_info;



	IPADMA_FUNC_ENTRY();



	mem_info = (struct ipa_mem_buffer *)data;

	ep_idx = ipa3_get_ep_mapping(IPA_CLIENT_MEMCPY_DMA_ASYNC_CONS);

	if (ep_idx < 0) {

		IPADMA_ERR("IPA Client mapping failed\n");

drivers/platform/msm/ipa/ipa_v3/ipa_dp.c

+223 −160

File changed.

Preview size limit exceeded, changes collapsed.