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: <afd8ed16-638d-c176-b84e-503ee0ece668@synopsys.com>
Date:   Fri, 25 Aug 2017 10:13:37 -0700
From:   Vineet Gupta <Vineet.Gupta1@...opsys.com>
To:     Alexey Brodkin <Alexey.Brodkin@...opsys.com>,
        "linux-snps-arc@...ts.infradead.org" 
        <linux-snps-arc@...ts.infradead.org>
CC:     "linux-kernel@...r.kernel.org" <linux-kernel@...r.kernel.org>,
        "Alexey Brodkin" <Alexey.Brodkin@...opsys.com>,
        Eugeniy Paltsev <Eugeniy.Paltsev@...opsys.com>
Subject: Re: [PATCHi v2] arc: Mask individual IRQ lines during core INTC init

On 08/25/2017 07:22 AM, Alexey Brodkin wrote:
> From: Alexey Brodkin <Alexey.Brodkin@...opsys.com>
>
> ARC cores on reset have all interrupt lines of built-in INTC enabled.
> Which means once we globally enable interrupts (very early on boot)
> faulty hardware blocks may trigger an interrupt that Linux kernel
> cannot handle yet as corresponding handler is not yet installed.
>
> In that case system falls in "interrupt storm" and basically never
> does anything useful except entering and exiting generic IRQ handling
> code.
>
> One real example of that kind of problematic hardware is DW GMAC which
> also has interrupts enabled on reset and if Ethernet PHY informs GMAC
> about link state, GMAC immediately reports that upstream to ARC core
> and here we are.
>
> Now with that change we mask all individual IRQ lines making entire
> system more fool-proof.
>
> [This patch was motivated by Adaptrum platform support]
>
> Signed-off-by: Alexey Brodkin <abrodkin@...opsys.com>
> Cc: Eugeniy Paltsev <paltsev@...opsys.com>
> Tested-by: Alexandru Gagniuc <alex.g@...ptrum.com>
> Signed-off-by: Vineet Gupta <vgupta@...opsys.com>

v1 was merged in mainline this week - please provide a fixup patch on top of 
mainline / for-curr !

> ---
>
> Changes v1 -> v2:
>    Fixed arcv2 SMP case.
>    1. IDU simply routes signals from its inputs to a particular IRQ lines of
>       ARC core in SMP cluster. Moreover with IRQ affinity IDU could be set such
>       that it routes intrrupt signals every time to a different core
>       (round-robin) or always to a dedicated core. So it is not practical to
>       manage common interrupts on ARC cores but instead set IDU properly.
>       That said on ARC cores we unconditinally enable all "common" interrupt
>       lines expecting IDU to only routes interrupts expectedly (i.e. not before
>       a corresponding handler was installed).
>    2. All priveta-per-core interrupt lines are not covered by IDU so we have to
>       manage them right in the ARC's INTC and so we do now
>       (hwirq < FIRST_EXT_IRQ).
>
>   arch/arc/kernel/intc-arcv2.c   |  4 ++++
>   arch/arc/kernel/intc-compact.c | 14 +++++++++++++-
>   arch/arc/kernel/mcip.c         | 15 +++++++++++++++
>   3 files changed, 32 insertions(+), 1 deletion(-)
>
> diff --git a/arch/arc/kernel/intc-arcv2.c b/arch/arc/kernel/intc-arcv2.c
> index f928795fd07a..2db639874849 100644
> --- a/arch/arc/kernel/intc-arcv2.c
> +++ b/arch/arc/kernel/intc-arcv2.c
> @@ -75,10 +75,14 @@ void arc_init_IRQ(void)
>   	 * Set a default priority for all available interrupts to prevent
>   	 * switching of register banks if Fast IRQ and multiple register banks
>   	 * are supported by CPU.
> +	 * Also disable private-per-core IRQ lines so faulty external HW won't
> +	 * trigger interrupt that kernel is not ready to handle.
>   	 */
>   	for (i = NR_EXCEPTIONS; i < irq_bcr.irqs + NR_EXCEPTIONS; i++) {
>   		write_aux_reg(AUX_IRQ_SELECT, i);
>   		write_aux_reg(AUX_IRQ_PRIORITY, ARCV2_IRQ_DEF_PRIO);
> +		if (i < FIRST_EXT_IRQ)
> +			write_aux_reg(AUX_IRQ_ENABLE, 0);
>   	}
>   
>   	/* setup status32, don't enable intr yet as kernel doesn't want */
> diff --git a/arch/arc/kernel/intc-compact.c b/arch/arc/kernel/intc-compact.c
> index 7e608c6b0a01..cef388025adf 100644
> --- a/arch/arc/kernel/intc-compact.c
> +++ b/arch/arc/kernel/intc-compact.c
> @@ -27,7 +27,7 @@
>    */
>   void arc_init_IRQ(void)
>   {
> -	int level_mask = 0;
> +	int level_mask = 0, i;
>   
>          /* Is timer high priority Interrupt (Level2 in ARCompact jargon) */
>   	level_mask |= IS_ENABLED(CONFIG_ARC_COMPACT_IRQ_LEVELS) << TIMER0_IRQ;
> @@ -40,6 +40,18 @@ void arc_init_IRQ(void)
>   
>   	if (level_mask)
>   		pr_info("Level-2 interrupts bitset %x\n", level_mask);
> +
> +	/*
> +	 * Disable all IRQ lines so faulty external hardware won't
> +	 * trigger interrupt that kernel is not ready to handle.
> +	 */
> +	for (i = TIMER0_IRQ; i < NR_CPU_IRQS; i++) {
> +		unsigned int ienb;
> +
> +		ienb = read_aux_reg(AUX_IENABLE);
> +		ienb &= ~(1 << i);
> +		write_aux_reg(AUX_IENABLE, ienb);
> +	}
>   }
>   
>   /*
> diff --git a/arch/arc/kernel/mcip.c b/arch/arc/kernel/mcip.c
> index f61a52b01625..6a889828db4f 100644
> --- a/arch/arc/kernel/mcip.c
> +++ b/arch/arc/kernel/mcip.c
> @@ -303,6 +303,21 @@ idu_of_init(struct device_node *intc, struct device_node *parent)
>   		virq = irq_create_mapping(NULL, i + FIRST_EXT_IRQ);
>   		BUG_ON(!virq);
>   		irq_set_chained_handler_and_data(virq, idu_cascade_isr, domain);
> +		/*
> +		 * irq_set_chained_handler_and_data() silently executes
> +		 * idu_irq_enable() which leaves us with all common interrupts
> +		 * enabled right on start which allows faulty HW to generate
> +		 * an interrupt which we are not ready to support yet.
> +		 * So explicitly mask freshly set IRQ.
> +		 * Note we have to do all that magic because IDU is not a real
> +		 * cascaded INTC with N inputs and one upstream output.
> +		 * What IDU really does it just mirrors its N inputs to
> +		 * N*core_num outputs so we either need to selectively disable
> +		 * or enable IRQ lines on upstream INTCs of ARC cores (and do
> +		 * it for each and every ARC core in the cluster) or simply mask
> +		 * inputs of the IDU just in one place here.
> +		 */
> +		idu_irq_mask_raw(i);
>   	}
>   
>   	__mcip_cmd(CMD_IDU_ENABLE, 0);

Powered by blists - more mailing lists

Powered by Openwall GNU/*/Linux Powered by OpenVZ