linux-kernel - Re: [PATCH v5 2/2] PCI: xilinx-cpm: Add Versal CPM Root Port driver

lists.openwall.net		lists / announce owl-users owl-dev john-users john-dev passwdqc-users yescrypt popa3d-users / oss-security kernel-hardening musl sabotage tlsify passwords / crypt-dev xvendor / Bugtraq Full-Disclosure linux-kernel linux-netdev linux-ext4 linux-hardening linux-cve-announce PHC
Open Source and information security mailing list archives

Hash Suite: Windows password security audit tool. GUI, reports in PDF.

[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]

Message-ID: <20200228104442.GA2874@e121166-lin.cambridge.arm.com>
Date:   Fri, 28 Feb 2020 10:44:42 +0000
From:   Lorenzo Pieralisi <lorenzo.pieralisi@....com>
To:     Bharat Kumar Gogada <bharatku@...inx.com>
Cc:     "linux-pci@...r.kernel.org" <linux-pci@...r.kernel.org>,
        "linux-kernel@...r.kernel.org" <linux-kernel@...r.kernel.org>,
        "bhelgaas@...gle.com" <bhelgaas@...gle.com>,
        Ravikiran Gummaluri <rgummal@...inx.com>, maz@...nel.org
Subject: Re: [PATCH v5 2/2] PCI: xilinx-cpm: Add Versal CPM Root Port driver

[+MarcZ, FHI]

On Tue, Feb 25, 2020 at 02:39:56PM +0000, Bharat Kumar Gogada wrote:

[...]

> > > +/* ECAM definitions */
> > > +#define ECAM_BUS_NUM_SHIFT		20
> > > +#define ECAM_DEV_NUM_SHIFT		12
> > 
> > You don't need these ECAM_* defines, you can use pci_generic_ecam_ops.
> Does this need separate ranges region for ECAM space ? 
> We have ECAM and controller space in same region.

You can create an ECAM window with pci_ecam_create where *cfgres
represent the ECAM area, I don't get what you mean by "same region".

Do you mean "contiguous" ? Or something else ?

> > > +
> > > +/**
> > > + * struct xilinx_cpm_pcie_port - PCIe port information
> > > + * @reg_base: Bridge Register Base
> > > + * @cpm_base: CPM System Level Control and Status Register(SLCR) Base
> > > + * @irq: Interrupt number
> > > + * @root_busno: Root Bus number
> > > + * @dev: Device pointer
> > > + * @leg_domain: Legacy IRQ domain pointer
> > > + * @irq_misc: Legacy and error interrupt number  */ struct
> > > +xilinx_cpm_pcie_port {
> > > +	void __iomem *reg_base;
> > > +	void __iomem *cpm_base;
> > > +	u32 irq;
> > > +	u8 root_busno;
> > > +	struct device *dev;
> > > +	struct irq_domain *leg_domain;
> > > +	int irq_misc;
> > > +};
> > > +
> > > +static inline u32 pcie_read(struct xilinx_cpm_pcie_port *port, u32
> > > +reg) {
> > > +	return readl(port->reg_base + reg);
> > > +}
> > > +
> > > +static inline void pcie_write(struct xilinx_cpm_pcie_port *port,
> > > +			      u32 val, u32 reg)
> > > +{
> > > +	writel(val, port->reg_base + reg);
> > > +}
> > > +
> > > +static inline bool cpm_pcie_link_up(struct xilinx_cpm_pcie_port
> > > +*port) {
> > > +	return (pcie_read(port, XILINX_CPM_PCIE_REG_PSCR) &
> > > +		XILINX_CPM_PCIE_REG_PSCR_LNKUP) ? 1 : 0;
> > 
> > 	u32 val = pcie_read(port, XILINX_CPM_PCIE_REG_PSCR);
> > 
> > 	return val & XILINX_CPM_PCIE_REG_PSCR_LNKUP;
> > 
> > And this function call is not that informative anyway - it is used just to print a log
> > whose usefulness is questionable.
> We need this logging information customers are using this info in case
> of link down failure.

Out of curiosity, to do what ?

[...]

> > > +/**
> > > + * xilinx_cpm_pcie_intx_map - Set the handler for the INTx and mark
> > > +IRQ as valid
> > > + * @domain: IRQ domain
> > > + * @irq: Virtual IRQ number
> > > + * @hwirq: HW interrupt number
> > > + *
> > > + * Return: Always returns 0.
> > > + */
> > > +static int xilinx_cpm_pcie_intx_map(struct irq_domain *domain,
> > > +				    unsigned int irq, irq_hw_number_t hwirq) {
> > > +	irq_set_chip_and_handler(irq, &dummy_irq_chip, handle_simple_irq);
> > 
> > INTX are level IRQs, the flow handler must be handle_level_irq.
> Accepted will change.
> > 
> > > +	irq_set_chip_data(irq, domain->host_data);
> > > +	irq_set_status_flags(irq, IRQ_LEVEL);
> > 
> > The way INTX are handled in this patch is wrong. You must set-up a chained IRQ
> > with the appropriate flow handler, current code uses an IRQ action and that's an
> > IRQ layer violation and it goes without saying that it is almost certainly broken.
> In our controller we use same irq line for controller errors and
> legacy errors.  we have two cases here where error interrupts are
> self-consumed by controller, and legacy interrupts are flow handled.
> Its not INTX handling alone for this IRQ line .  So chained IRQ can be
> used for self consumed interrupts too ?

No. In this specific case both solutions are not satisfying, we need to
give it some thought, I will talk to Marc (CC'ed) to find the best
option here going forward.

Thanks,
Lorenzo