Re: [PATCH] drivers: irqchip: Allocate alignment addr by ITS_BASER.Page_size

[Marc Zyngier <maz@kernel.org>]

On Mon, 18 Jul 2022 08:35:47 +0100,
wangwudi <wangwudi@hisilicon.com> wrote:
> 
> Hi Marc,
> 
> 在 2022/7/16 17:30, Marc Zyngier 写道:
> > On Sat, 16 Jul 2022 08:05:36 +0100,
> > wangwudi <wangwudi@hisilicon.com> wrote:
> >>
> >> The description of the ITS_BASER.Physical_Address field in the ARM GIC spec is as 
> >> follows:
> >> "The address must be aligned to the size specified in the Page Size field."
> >> The Page_Size field in ITS_BASER might be RO.
> >>
> >> Currently, the address is aligned based on the system page_size, not the HW 
> >> Page_Size field. In some case, this is in contradiction with the spec.
> >>
> >> For example:
> >> ITS_BASER.Page_Size indicate 16K, and kernel page size is 4K.
> >> If HW need 4K-size memory, the driver may alloc a 4K aligned address.
> >> This has been proven in hardware.
> > 
> > Ah, interesting bug. Thanks for bringing this up. Can you describe how
> > this occurs? I suspect you are using indirect tables.
> > 
> 
> Sure. In the system, kernel page size is 4K, and ITS_BASER.Page_Size is 16K.
> 
> As you suspected, HW used indirect VPE table and indication supports a small
> number of vpeids, like 2-bits vpeid. So that HW requires only less than 4K-
> size memory, and 16K aligned base address. But driver alloctes 4K aligend base
> address.

Well, that I dispute, see below.

> 
> >>
> >> Cc: Thomas Gleixner <tglx@linutronix.de>
> >> Cc: Marc Zyngier <maz@kernel.org>
> >> Signed-off-by: wangwudi <wangwudi@hisilicon.com>
> >> ---
> >>  drivers/irqchip/irq-gic-v3-its.c | 3 +++
> >>  1 file changed, 3 insertions(+)
> >>
> >> diff --git a/drivers/irqchip/irq-gic-v3-its.c b/drivers/irqchip/irq-gic-v3-its.c
> >> index 5ff09de6c48f..0e25e887d45c 100644
> >> --- a/drivers/irqchip/irq-gic-v3-its.c
> >> +++ b/drivers/irqchip/irq-gic-v3-its.c
> >> @@ -2310,6 +2310,9 @@ static int its_setup_baser(struct its_node *its, struct its_baser *baser,
> >>  		order = get_order(GITS_BASER_PAGES_MAX * psz);
> >>  	}
> >>  
> >> +	if ((psz > PAGE_SIZE) && (PAGE_ORDER_TO_SIZE(order) < psz)) {
> >> +		order = get_order(psz);
> >> +	}
> >>  	page = alloc_pages_node(its->numa_node, GFP_KERNEL | __GFP_ZERO, order);
> >>  	if (!page)
> >>  		return -ENOMEM;
> > 
> > However, I don't see how you end-up with the incorrect value here.
> > 
> > * No indirect table:
> > 	alloc_its_tables():
> > 		order = get_order(baser->psz);
> > 
> > * Indirect tables:
> > 	alloc_its_tables():
> > 		order = get_order(baser->psz);
> > 	its_parse_indirect_baser():
> > 		new_order = *order;
> > 		new_order = max_t(u32, get_order(esz << ids), new_order);
> > 
> > So in both cases, we should end-up with order >= get_order(psz).
> Yes, totally agree.

OK. So what does your patch actually fixes?

> 
> > 
> > Clearly, I'm missing a path, but your commit message doesn't make it
> > obvious. Can you please enlighten me?
> > 
> My commit is based on the premise:
> "alloc_pages_node gives a size-aligend address".
> For example, if HW apply for 4K-size memory, then allocated address is 4K
> aligned.

Right. And if baser->psz is 16K, the memory returned will be 16K
aligned.

The only thing I can imagine is that there is a code path that doesn't
use baser->psz as the minimum value when allocating memory programmed
into a ITS_BASER register. But I can't see that path. Can you?

	M.

-- 
Without deviation from the norm, progress is not possible.