From mboxrd@z Thu Jan 1 00:00:00 1970 From: Avaneesh Kumar Dwivedi Subject: Re: [PATCH 2/5] remoteproc: Adding q6v55 specific regulator, clk, reset interface. Date: Fri, 4 Nov 2016 19:11:20 +0530 Message-ID: References: <1477324559-24752-1-git-send-email-akdwived@codeaurora.org> <1477324559-24752-3-git-send-email-akdwived@codeaurora.org> <20161025190512.GN7509@tuxbot> Mime-Version: 1.0 Content-Type: text/plain; charset=windows-1252; format=flowed Content-Transfer-Encoding: 7bit Return-path: Received: from smtp.codeaurora.org ([198.145.29.96]:44404 "EHLO smtp.codeaurora.org" rhost-flags-OK-OK-OK-OK) by vger.kernel.org with ESMTP id S933194AbcKDNl1 (ORCPT ); Fri, 4 Nov 2016 09:41:27 -0400 In-Reply-To: <20161025190512.GN7509@tuxbot> Sender: linux-arm-msm-owner@vger.kernel.org List-Id: linux-arm-msm@vger.kernel.org To: Bjorn Andersson Cc: linux-remoteproc@vger.kernel.org, linux-arm-msm@vger.kernel.org, spjoshi@codeaurora.org, kaushalk@codeaurora.org On 10/26/2016 12:35 AM, Bjorn Andersson wrote: > On Mon 24 Oct 08:55 PDT 2016, Avaneesh Kumar Dwivedi wrote: > >> q6v55 use different regulator and clock resource than q6v5, hence >> using separate resource handling for same. >> Also reset register >> programming in q6v55 is non secure so resource controller init- >> ilization is not required for q6v55. > The reset comes from GCC, so please use the fact that gcc already is a > reset-controller. Already in patch 1/5 have explained reason of not using the GCC to reset pasting same here again Infact i had done it the way suggested before i sent out initial patchset, but then when i discussed with internal clock team, they said they will not support MSS RESET to be done via GCC because "MSS_RESET is not a block reset, GCC reset controller is used only when we need a BCR to be reset, and which has a special purpose. MSS RESET doesn't fall under block resets, it is not a clock and cannot be associated with clock." Downstream also MSS RESET is programmed through ioremap > >> Signed-off-by: Avaneesh Kumar Dwivedi >> --- >> drivers/remoteproc/qcom_q6v5_pil.c | 271 +++++++++++++++++++++++++++++++++++++ >> 1 file changed, 271 insertions(+) >> >> diff --git a/drivers/remoteproc/qcom_q6v5_pil.c b/drivers/remoteproc/qcom_q6v5_pil.c >> index 8df95a0..c7dca40 100644 >> --- a/drivers/remoteproc/qcom_q6v5_pil.c >> +++ b/drivers/remoteproc/qcom_q6v5_pil.c >> @@ -215,6 +215,203 @@ static void q6v5_regulator_disable(struct q6v5 *qproc) >> regulator_set_voltage(mss, 0, 1150000); >> } >> >> +static int q6v55_regulator_init(struct q6v5 *qproc) >> +{ >> + int ret; >> + >> + qproc->supply[Q6V5_SUPPLY_CX].supply = "vdd_cx"; >> + qproc->supply[Q6V5_SUPPLY_MX].supply = "vdd_mx"; >> + qproc->supply[Q6V5_SUPPLY_PLL].supply = "vdd_pll"; >> + >> + ret = devm_regulator_bulk_get(qproc->dev, >> + (ARRAY_SIZE(qproc->supply) - 1), qproc->supply); >> + if (ret < 0) { >> + dev_err(qproc->dev, "failed to get supplies\n"); >> + return ret; >> + } >> + >> + >> + return 0; >> +} > This is very very similar to the existing q6v5_regulator_init, please > figure out a way to make the mss supply optional here instead of > creating a new function. OK, Have implemented common init and enable function for clock and regulator in patchset v2 > >> + >> +static int q6v55_clk_enable(struct q6v5 *qproc) >> +{ >> + int ret; >> + >> + ret = clk_prepare_enable(qproc->ahb_clk); >> + if (ret) >> + goto out; >> + >> + ret = clk_prepare_enable(qproc->axi_clk); >> + if (ret) >> + goto err_axi_clk; >> + >> + ret = clk_prepare_enable(qproc->rom_clk); >> + if (ret) >> + goto err_rom_clk; >> + >> + ret = clk_prepare_enable(qproc->gpll0_mss_clk); >> + if (ret) >> + goto err_gpll0_mss_clk; >> + >> + ret = clk_prepare_enable(qproc->snoc_axi_clk); >> + if (ret) >> + goto err_snoc_axi_clk; >> + >> + ret = clk_prepare_enable(qproc->mnoc_axi_clk); >> + if (ret) >> + goto err_mnoc_axi_clk; > I believe it would be better to turn the clocks into an array, like the > regulators, so that we can loop over them rather than listing them > explicitly. OK , Have used clock array to initialize and enable clocks for q6 in patchset v2 > >> + >> + return 0; >> +err_mnoc_axi_clk: >> + clk_disable_unprepare(qproc->snoc_axi_clk); >> +err_snoc_axi_clk: >> + clk_disable_unprepare(qproc->gpll0_mss_clk); >> +err_gpll0_mss_clk: >> + clk_disable_unprepare(qproc->rom_clk); >> +err_rom_clk: >> + clk_disable_unprepare(qproc->axi_clk); >> +err_axi_clk: >> + clk_disable_unprepare(qproc->ahb_clk); >> +out: >> + dev_err(qproc->dev, "Clk Vote Failed\n"); > I believe the clock framework will complain if above fails, so no need > to add another print here. OK, Have removed print also > >> + return ret; >> +} >> + >> +static void q6v55_clk_disable(struct q6v5 *qproc) >> +{ >> + >> + clk_disable_unprepare(qproc->mnoc_axi_clk); >> + clk_disable_unprepare(qproc->snoc_axi_clk); >> + clk_disable_unprepare(qproc->gpll0_mss_clk); >> + clk_disable_unprepare(qproc->rom_clk); >> + clk_disable_unprepare(qproc->axi_clk); >> + if (!qproc->ahb_clk_vote) > Why do you check ahb_clk_vote in the disable case but not in the enable > case? > > Also, please move all ahb_clk_vote handling into the same patch. OK, Infact check was not required for v56 applicable to 8996, so have removed them. > >> + clk_disable_unprepare(qproc->ahb_clk); >> +} >> + >> +static int q6v55_proxy_vote(struct q6v5 *qproc) >> +{ >> + struct regulator *mx = qproc->supply[Q6V5_SUPPLY_MX].consumer; >> + struct regulator *cx = qproc->supply[Q6V5_SUPPLY_CX].consumer; >> + struct regulator *vdd_pll = qproc->supply[Q6V5_SUPPLY_PLL].consumer; >> + int ret; >> + >> + ret = regulator_set_voltage(mx, INT_MAX, INT_MAX); > I'm not convinced that we should vote MAX as minimum voltage here, is it > not 1.05V as in the q6v5? OK > >> + if (ret) { >> + dev_err(qproc->dev, "Failed to set voltage for vreg_mx\n"); >> + return ret; >> + } >> + >> + ret = regulator_enable(mx); >> + if (ret) { >> + dev_err(qproc->dev, "Failed to enable vreg_mx\n"); >> + goto err_mx_enable; >> + } >> + >> + ret = regulator_set_voltage(cx, INT_MAX, INT_MAX); >> + if (ret) { >> + dev_err(qproc->dev, "Failed to request vdd_cx voltage.\n"); >> + goto err_cx_voltage; >> + } >> + >> + ret = regulator_set_load(cx, 100000); >> + if (ret < 0) { >> + dev_err(qproc->dev, "Failed to set vdd_cx mode.\n"); >> + goto err_cx_set_load; >> + } >> + >> + ret = regulator_enable(cx); >> + if (ret) { >> + dev_err(qproc->dev, "Failed to vote for vdd_cx\n"); >> + goto err_cx_enable; >> + } >> + >> + ret = regulator_set_voltage(vdd_pll, 1800000, 1800000); > PLL is fed from a fixed voltage regulator of 1.8V, so we do not need to > request a voltage (per Mark Brown's request). OK. > >> + if (ret) { >> + dev_err(qproc->dev, "Failed to set voltage for vdd_pll\n"); >> + goto err_vreg_pll_set_vol; >> + } >> + >> + ret = regulator_set_load(vdd_pll, 10000); >> + if (ret < 0) { >> + dev_err(qproc->dev, "Failed to set vdd_pll mode.\n"); >> + goto err_vreg_pll_load; >> + } >> + >> + ret = regulator_enable(vdd_pll); >> + if (ret) { >> + dev_err(qproc->dev, "Failed to vote for vdd_pll\n"); >> + goto err_vreg_pll_enable; >> + } >> + >> + ret = clk_prepare_enable(qproc->xo); >> + if (ret) >> + goto err_xo_vote; >> + >> + ret = clk_prepare_enable(qproc->pnoc_clk); >> + if (ret) >> + goto err_pnoc_vote; >> + >> + ret = clk_prepare_enable(qproc->qdss_clk); >> + if (ret) >> + goto err_qdss_vote; >> + qproc->unvote_flag = false; >> + return 0; >> + >> +err_qdss_vote: >> + clk_disable_unprepare(qproc->pnoc_clk); >> +err_pnoc_vote: >> + clk_disable_unprepare(qproc->xo); >> +err_xo_vote: >> + regulator_disable(vdd_pll); >> +err_vreg_pll_enable: >> + regulator_set_load(vdd_pll, 0); >> +err_vreg_pll_load: >> + regulator_set_voltage(vdd_pll, 0, INT_MAX); >> +err_vreg_pll_set_vol: >> + regulator_disable(cx); >> +err_cx_enable: >> + regulator_set_load(cx, 0); >> +err_cx_set_load: >> + regulator_set_voltage(cx, 0, INT_MAX); >> +err_cx_voltage: >> + regulator_disable(mx); >> +err_mx_enable: >> + regulator_set_voltage(mx, 0, INT_MAX); >> + >> + return ret; >> +} > As with the init function, this is very similar to the q6v5 case, so I > don't think we need to have separate functions for it. > > A side note on this is that clk_prepare_enable(NULL) is valid, so if you > distinguish between the two cases during initialisation and put all > clocks in an array you could just call clk_prepare_enable() on all of > them, which will skip the ones that isn't initialised. > >> + >> +static void q6v55_proxy_unvote(struct q6v5 *qproc) >> +{ >> + if (!qproc->unvote_flag) { > I don't think the "unvote_flag" is good design and don't see how you > would end up unvoting multiple times based on my original design. > > But again, the answer is probably in some other patch - i.e. please > group your patches so I can see each step in its entirety. code is reorganized yet i am using unvote flag but in transparent way. i am setting unvote flag during enable and making it false during disable. so that we unvote only if we have voted earlier. > >> + regulator_disable(qproc->supply[Q6V5_SUPPLY_PLL].consumer); >> + regulator_set_load(qproc->supply[Q6V5_SUPPLY_PLL].consumer, 0); >> + regulator_disable(qproc->supply[Q6V5_SUPPLY_CX].consumer); >> + regulator_set_load(qproc->supply[Q6V5_SUPPLY_CX].consumer, 0); >> + regulator_set_voltage(qproc->supply[Q6V5_SUPPLY_CX].consumer, >> + 0, INT_MAX); >> + >> + clk_disable_unprepare(qproc->qdss_clk); >> + clk_disable_unprepare(qproc->pnoc_clk); >> + clk_disable_unprepare(qproc->xo); >> + >> + regulator_disable(qproc->supply[Q6V5_SUPPLY_MX].consumer); >> + regulator_set_voltage(qproc->supply[Q6V5_SUPPLY_MX].consumer, >> + 0, INT_MAX); >> + } >> + qproc->unvote_flag = true; >> +} >> + >> +static void pil_mss_restart_reg(struct q6v5 *qproc, u32 mss_restart) >> +{ >> + if (qproc->restart_reg) { >> + writel_relaxed(mss_restart, qproc->restart_reg); >> + udelay(2); >> + } >> +} > Drop this can not drop due to reason explained already that i can not use GCC for MSS reset. > >> + >> static int q6v5_load(struct rproc *rproc, const struct firmware *fw) >> { >> struct q6v5 *qproc = rproc->priv; >> @@ -751,6 +948,65 @@ static int q6v5_init_clocks(struct q6v5 *qproc) >> return 0; >> } >> >> +static int q6v55_init_clocks(struct q6v5 *qproc) >> +{ >> + qproc->ahb_clk = devm_clk_get(qproc->dev, "iface"); >> + if (IS_ERR(qproc->ahb_clk)) { >> + dev_err(qproc->dev, "failed to get iface clock\n"); >> + return PTR_ERR(qproc->ahb_clk); >> + } >> + >> + qproc->axi_clk = devm_clk_get(qproc->dev, "bus"); >> + if (IS_ERR(qproc->axi_clk)) { >> + dev_err(qproc->dev, "failed to get bus clock\n"); >> + return PTR_ERR(qproc->axi_clk); >> + } >> + >> + qproc->rom_clk = devm_clk_get(qproc->dev, "mem"); >> + if (IS_ERR(qproc->rom_clk)) { >> + dev_err(qproc->dev, "failed to get mem clock\n"); >> + return PTR_ERR(qproc->rom_clk); >> + } > These three are the same as q6v5... > >> + >> + qproc->snoc_axi_clk = devm_clk_get(qproc->dev, "snoc_axi_clk"); >> + if (IS_ERR(qproc->snoc_axi_clk)) { >> + dev_err(qproc->dev, "failed to get snoc_axi_clk\n"); >> + return PTR_ERR(qproc->snoc_axi_clk); >> + } >> + >> + qproc->mnoc_axi_clk = devm_clk_get(qproc->dev, "mnoc_axi_clk"); >> + if (IS_ERR(qproc->mnoc_axi_clk)) { >> + dev_err(qproc->dev, "failed to get mnoc_axi_clk clock\n"); >> + return PTR_ERR(qproc->mnoc_axi_clk); >> + } >> + >> + qproc->gpll0_mss_clk = devm_clk_get(qproc->dev, "gpll0_mss_clk"); >> + if (IS_ERR(qproc->gpll0_mss_clk)) { >> + dev_err(qproc->dev, "failed to get gpll0_mss_clk clock\n"); >> + return PTR_ERR(qproc->gpll0_mss_clk); >> + } >> + >> + qproc->xo = devm_clk_get(qproc->dev, "xo"); >> + if (IS_ERR(qproc->xo)) { >> + dev_err(qproc->dev, "failed to get xo\n"); >> + return PTR_ERR(qproc->xo); >> + } > And q6v5 will need "xo" as well, I missed this one. OK. > >> + >> + qproc->pnoc_clk = devm_clk_get(qproc->dev, "pnoc"); >> + if (IS_ERR(qproc->pnoc_clk)) { >> + dev_err(qproc->dev, "failed to get pnoc_clk clock\n"); >> + return PTR_ERR(qproc->pnoc_clk); >> + } >> + >> + qproc->qdss_clk = devm_clk_get(qproc->dev, "qdss"); >> + if (IS_ERR(qproc->qdss_clk)) { >> + dev_err(qproc->dev, "failed to get qdss_clk clock\n"); >> + return PTR_ERR(qproc->qdss_clk); >> + } >> + > I would rather see that we have a single init_clocks function that based > on compatible will get the appropriate clocks. OK. > >> + return 0; >> +} >> + >> static int q6v5_init_reset(struct q6v5 *qproc) >> { >> qproc->mss_restart = devm_reset_control_get(qproc->dev, NULL); >> @@ -762,6 +1018,21 @@ static int q6v5_init_reset(struct q6v5 *qproc) >> return 0; >> } >> >> +static int q6v55_init_reset(struct q6v5 *qproc, struct platform_device *pdev) >> +{ >> + struct resource *res; >> + >> + res = platform_get_resource_byname(pdev, IORESOURCE_MEM, "restart_reg"); >> + qproc->restart_reg = devm_ioremap(qproc->dev, res->start, >> + resource_size(res)); >> + if (IS_ERR(qproc->restart_reg)) { >> + dev_err(qproc->dev, "failed to get restart_reg\n"); >> + return PTR_ERR(qproc->restart_reg); >> + } >> + >> + return 0; >> +} > Drop this Can not drop as can not use GCC for MSS_RESET due to reason explained above. > >> + >> static int q6v5_request_irq(struct q6v5 *qproc, >> struct platform_device *pdev, >> const char *name, >> -- >> The Qualcomm Innovation Center, Inc. is a member of the Code Aurora Forum, a Linux Foundation Collaborative Project >>