Message-ID: <20240319042519.GA52500@thinkpad>
Date: Tue, 19 Mar 2024 09:55:19 +0530
From: Manivannan Sadhasivam <manivannan.sadhasivam@...aro.org>
To: Frank Li <Frank.li@....com>
Cc: Serge Semin <fancer.lancer@...il.com>,
	Bjorn Helgaas <bhelgaas@...gle.com>,
	Lorenzo Pieralisi <lpieralisi@...nel.org>,
	Krzysztof Wilczyński <kw@...ux.com>,
	Rob Herring <robh@...nel.org>, Jingoo Han <jingoohan1@...il.com>,
	Gustavo Pimentel <gustavo.pimentel@...opsys.com>,
	Krzysztof Kozlowski <krzysztof.kozlowski+dt@...aro.org>,
	Conor Dooley <conor+dt@...nel.org>, imx@...ts.linux.dev,
	linux-pci@...r.kernel.org, linux-kernel@...r.kernel.org,
	devicetree@...r.kernel.org
Subject: Re: [PATCH v4 5/5] PCI: dwc: Add common send PME_Turn_Off message
 method

On Mon, Feb 26, 2024 at 05:43:29PM -0500, Frank Li wrote:
> On Tue, Feb 27, 2024 at 12:52:45AM +0300, Serge Semin wrote:
> > On Tue, Feb 13, 2024 at 04:50:26PM -0500, Frank Li wrote:
> > > Reserve space at end of first IORESOURCE_MEM window as message TLP MMIO
> > > window. This space's size is 'region_align'.
> > > 
> > > Set outbound ATU map memory write to send PCI message. So one MMIO write
> > > can trigger a PCI message, such as PME_Turn_Off.
> > > 
> > > Add common dwc_pme_turn_off() function.
> > > 
> > > Call dwc_pme_turn_off() to send out PME_Turn_Off message in general
> > > dw_pcie_suspend_noirq() if there are not platform callback pme_turn_off()
> > > exist.
> > > 
> > > Signed-off-by: Frank Li <Frank.Li@....com>
> > > ---
> > >  drivers/pci/controller/dwc/pcie-designware-host.c | 93 ++++++++++++++++++++++-
> > >  drivers/pci/controller/dwc/pcie-designware.h      |  2 +
> > >  2 files changed, 91 insertions(+), 4 deletions(-)
> > > 
> > > diff --git a/drivers/pci/controller/dwc/pcie-designware-host.c b/drivers/pci/controller/dwc/pcie-designware-host.c
> > > index 267687ab33cbc..5e83756492462 100644
> > > --- a/drivers/pci/controller/dwc/pcie-designware-host.c
> > > +++ b/drivers/pci/controller/dwc/pcie-designware-host.c
> > > @@ -393,6 +393,31 @@ static int dw_pcie_msi_host_init(struct dw_pcie_rp *pp)
> > >  	return 0;
> > >  }
> > >  
> > > +static void dw_pcie_host_request_msg_tlp_res(struct dw_pcie_rp *pp)
> > > +{
> > > +	struct dw_pcie *pci = to_dw_pcie_from_pp(pp);
> > > +	struct resource_entry *win;
> > > +	struct resource *res;
> > > +
> > > +	win = resource_list_first_type(&pp->bridge->windows, IORESOURCE_MEM);
> > > +	if (win) {
> > > +		res = devm_kzalloc(pci->dev, sizeof(*res), GFP_KERNEL);
> > > +		if (!res)
> > > +			return;
> > > +
> > > +		/* Reserve last region_align block as message TLP space */
> > > +		res->start = win->res->end - pci->region_align + 1;
> > > +		res->end = win->res->end;
> > > +		res->name = "msg";
> > > +		res->flags = win->res->flags | IORESOURCE_BUSY;
> > > +
> > > +		if (!request_resource(win->res, res))
> > > +			pp->msg_res = res;
> > > +		else
> > > +			devm_kfree(pci->dev, res);
> > > +	}
> > > +}
> > > +
> > >  int dw_pcie_host_init(struct dw_pcie_rp *pp)
> > >  {
> > >  	struct dw_pcie *pci = to_dw_pcie_from_pp(pp);
> > > @@ -479,6 +504,9 @@ int dw_pcie_host_init(struct dw_pcie_rp *pp)
> > >  
> > >  	dw_pcie_iatu_detect(pci);
> > >  
> > 
> > > +	/* Need call after dw_pcie_iatu_detect() before dw_pcie_setup_rc() */
> > > +	dw_pcie_host_request_msg_tlp_res(pp);
> > 
> > This may cause regressions for instance if the outbound memory window
> > is small and is fully dedicated for some device like VGA/GPU/etc.
> 
> There are host memory map windows, which are quite big. It will be too
> small if only because one page size windows less.
> 
> Look like it is impossible to dedicated to one device, all pcie devices
> (VGA/GPU) should go through standard pcie probe flow, the bar will be
> allocated from bridge windows space.
> 

No, Sergey's concern is still valid. You are allocating the memory for TLPs at
the end of the existing MEM range. But there is a chance that it could cause
memory hungry devices like GPU to run out of memory in an existing setup.

That being said, I also do not want to hold off merging this series. So let's
make this region opt-in for drivers making use of this feature. This way, if we
migrate to a dedicated 'ranges' region in the future, we can remove the
conditional check and base the check on the existence of the new region.

- Mani

> > 
> > Why not to use a new ranges-based mapping as we discussed earlier:
> > https://lore.kernel.org/linux-pci/20240214061412.GB4618@thinkpad/
> > ?
> 
> If driver can auto alloc from known range, why need static defined in dts
> files.
> 
> static alloc can't resolve small outbound memory windows problem. It may
> disable this feature. 
> 
> > 
> > I know it might be troublesome but still it would be much better
> > and more portable across various platforms.
> > 
> > Bjorn, Lorenzo, Krzysztofm Rob could you please follow the link above
> > and give your opinion about the solution suggested there?
> > 
> > -Serge(y)
> > 
> > > +
> > >  	ret = dw_pcie_edma_detect(pci);
> > >  	if (ret)
> > >  		goto err_free_msi;
> > > @@ -536,6 +564,11 @@ void dw_pcie_host_deinit(struct dw_pcie_rp *pp)
> > >  
> > >  	dw_pcie_edma_remove(pci);
> > >  
> > > +	if (pp->msg_res) {
> > > +		release_resource(pp->msg_res);
> > > +		devm_kfree(pci->dev, pp->msg_res);
> > > +	}
> > > +
> > >  	if (pp->has_msi_ctrl)
> > >  		dw_pcie_free_msi(pp);
> > >  
> > > @@ -697,6 +730,10 @@ static int dw_pcie_iatu_setup(struct dw_pcie_rp *pp)
> > >  		atu.pci_addr = entry->res->start - entry->offset;
> > >  		atu.size = resource_size(entry->res);
> > >  
> > > +		/* MSG TLB window resource reserve at one of end of IORESOURCE_MEM resource */
> > > +		if (pp->msg_res && pp->msg_res->parent == entry->res)
> > > +			atu.size -= resource_size(pp->msg_res);
> > > +
> > >  		ret = dw_pcie_prog_outbound_atu(pci, &atu);
> > >  		if (ret) {
> > >  			dev_err(pci->dev, "Failed to set MEM range %pr\n",
> > > @@ -728,6 +765,8 @@ static int dw_pcie_iatu_setup(struct dw_pcie_rp *pp)
> > >  		dev_warn(pci->dev, "Ranges exceed outbound iATU size (%d)\n",
> > >  			 pci->num_ob_windows);
> > >  
> > > +	pp->msg_atu_index = i;
> > > +
> > >  	i = 0;
> > >  	resource_list_for_each_entry(entry, &pp->bridge->dma_ranges) {
> > >  		if (resource_type(entry->res) != IORESOURCE_MEM)
> > > @@ -833,11 +872,54 @@ int dw_pcie_setup_rc(struct dw_pcie_rp *pp)
> > >  }
> > >  EXPORT_SYMBOL_GPL(dw_pcie_setup_rc);
> > >  
> > > +/* Using message outbound ATU to send out PME_Turn_Off message for dwc PCIe controller */
> > > +static int dw_pcie_pme_turn_off(struct dw_pcie *pci)
> > > +{
> > > +	struct dw_pcie_ob_atu_cfg atu = { 0 };
> > > +	void __iomem *m;
> > > +	int ret;
> > > +
> > > +	if (pci->num_ob_windows <= pci->pp.msg_atu_index)
> > > +		return -EINVAL;
> > > +
> > > +	if (!pci->pp.msg_res)
> > > +		return -EINVAL;
> > > +
> > > +	atu.code = PCIE_MSG_CODE_PME_TURN_OFF;
> > > +	atu.routing = PCIE_MSG_TYPE_R_BC;
> > > +	atu.type = PCIE_ATU_TYPE_MSG;
> > > +	atu.size = resource_size(pci->pp.msg_res);
> > > +	atu.index = pci->pp.msg_atu_index;
> > > +
> > > +	if (!atu.size) {
> > > +		dev_dbg(pci->dev,
> > > +			"atu memory map windows is zero, please check 'msg' reg in dts\n");
> > > +		return -ENOMEM;
> > > +	}
> > > +
> > > +	atu.cpu_addr = pci->pp.msg_res->start;
> > > +
> > > +	ret = dw_pcie_prog_outbound_atu(pci, &atu);
> > > +	if (ret)
> > > +		return ret;
> > > +
> > > +	m = ioremap(atu.cpu_addr, pci->region_align);
> > > +	if (!m)
> > > +		return -ENOMEM;
> > > +
> > > +	/* A dummy write is converted to a Msg TLP */
> > > +	writel(0, m);
> > > +
> > > +	iounmap(m);
> > > +
> > > +	return 0;
> > > +}
> > > +
> > >  int dw_pcie_suspend_noirq(struct dw_pcie *pci)
> > >  {
> > >  	u8 offset = dw_pcie_find_capability(pci, PCI_CAP_ID_EXP);
> > >  	u32 val;
> > > -	int ret;
> > > +	int ret = 0;
> > >  
> > >  	/*
> > >  	 * If L1SS is supported, then do not put the link into L2 as some
> > > @@ -849,10 +931,13 @@ int dw_pcie_suspend_noirq(struct dw_pcie *pci)
> > >  	if (dw_pcie_get_ltssm(pci) <= DW_PCIE_LTSSM_DETECT_ACT)
> > >  		return 0;
> > >  
> > > -	if (!pci->pp.ops->pme_turn_off)
> > > -		return 0;
> > > +	if (pci->pp.ops->pme_turn_off)
> > > +		pci->pp.ops->pme_turn_off(&pci->pp);
> > > +	else
> > > +		ret = dw_pcie_pme_turn_off(pci);
> > >  
> > > -	pci->pp.ops->pme_turn_off(&pci->pp);
> > > +	if (ret)
> > > +		return ret;
> > >  
> > >  	ret = read_poll_timeout(dw_pcie_get_ltssm, val, val == DW_PCIE_LTSSM_L2_IDLE,
> > >  				PCIE_PME_TO_L2_TIMEOUT_US/10,
> > > diff --git a/drivers/pci/controller/dwc/pcie-designware.h b/drivers/pci/controller/dwc/pcie-designware.h
> > > index 703b50bc5e0f1..9e6076aa4c850 100644
> > > --- a/drivers/pci/controller/dwc/pcie-designware.h
> > > +++ b/drivers/pci/controller/dwc/pcie-designware.h
> > > @@ -341,6 +341,8 @@ struct dw_pcie_rp {
> > >  	struct pci_host_bridge  *bridge;
> > >  	raw_spinlock_t		lock;
> > >  	DECLARE_BITMAP(msi_irq_in_use, MAX_MSI_IRQS);
> > > +	int			msg_atu_index;
> > > +	struct resource		*msg_res;
> > >  };
> > >  
> > >  struct dw_pcie_ep_ops {
> > > 
> > > -- 
> > > 2.34.1
> > > 
> > > 

-- 
மணிவண்ணன் சதாசிவம்

Powered by blists - more mailing lists