Received-SPF: pass (google.com: best guess record for domain of linux-kernel-owner@vger.kernel.org designates 209.132.180.67 as permitted sender) client-ip=209.132.180.67;
MIME-Version: 1.0
References: <1570240177-8934-1-git-send-email-alan.mikhak@sifive.com>
In-Reply-To: <1570240177-8934-1-git-send-email-alan.mikhak@sifive.com>
From:   Alan Mikhak <alan.mikhak@sifive.com>
Date:   Mon, 7 Oct 2019 10:44:21 -0700
Message-ID: <CABEDWGx5MzsdcKzNzCtt3DxXAEWK69Bm-QBK0248rGAvWaU22w@mail.gmail.com>
Subject: Re: [PATCH] PCI: endpoint: cast the page number to phys_addr_t
To:     linux-kernel@vger.kernel.org, linux-pci@vger.kernel.org,
        Kishon Vijay Abraham I <kishon@ti.com>,
        lorenzo.pieralisi@arm.com, Bjorn Helgaas <bhelgaas@google.com>,
        Palmer Dabbelt <palmer@sifive.com>,
        Paul Walmsley <paul.walmsley@sifive.com>
Content-Type: text/plain; charset="UTF-8"
Sender: linux-kernel-owner@vger.kernel.org
Precedence: bulk

On Fri, Oct 4, 2019 at 6:49 PM Alan Mikhak <alan.mikhak@sifive.com> wrote:
>
> From: Alan Mikhak <alan.mikhak@sifive.com>
>
> Modify pci_epc_mem_alloc_addr() to cast the variable 'pageno'
> from type 'int' to 'phys_addr_t' before shifting left. This
> cast is needed to avoid treating bit 31 of 'pageno' as the
> sign bit which would otherwise get sign-extended to produce
> a negative value. When added to the base address of PCI memory
> space, the negative value would produce an invalid physical
> address which falls before the start of the PCI memory space.
>
> Signed-off-by: Alan Mikhak <alan.mikhak@sifive.com>
> ---
>  drivers/pci/endpoint/pci-epc-mem.c | 2 +-
>  1 file changed, 1 insertion(+), 1 deletion(-)
>
> diff --git a/drivers/pci/endpoint/pci-epc-mem.c b/drivers/pci/endpoint/pci-epc-mem.c
> index 2bf8bd1f0563..d2b174ce15de 100644
> --- a/drivers/pci/endpoint/pci-epc-mem.c
> +++ b/drivers/pci/endpoint/pci-epc-mem.c
> @@ -134,7 +134,7 @@ void __iomem *pci_epc_mem_alloc_addr(struct pci_epc *epc,
>         if (pageno < 0)
>                 return NULL;
>
> -       *phys_addr = mem->phys_base + (pageno << page_shift);
> +       *phys_addr = mem->phys_base + ((phys_addr_t)pageno << page_shift);
>         virt_addr = ioremap(*phys_addr, size);
>         if (!virt_addr)
>                 bitmap_release_region(mem->bitmap, pageno, order);
> --
> 2.7.4
>

Hi Kishon,

This issue was observed when requesting pci_epc_mem_alloc_addr()
to allocate a region of size 0x40010000ULL (1GB + 64KB) from a
128GB PCI address space with page sizes being 64KB. This resulted
in 'pageno' value of '0x8000' as the first available page in a
contiguous region for the requested size due to other smaller
regions having been allocated earlier. With 64KB page sizes,
the variable 'page_shift' holds a value of 0x10. Shifting 'pageno'
16 bits to the left results in an 'int' value whose bit 31 is set.

[   10.565256] __pci_epc_mem_init: mem size 0x2000000000 page_size 0x10000
[   10.571613] __pci_epc_mem_init: mem pages 0x200000 bitmap_size
0x40000 page_shift 0x10

PCI memory base 0x2000000000
PCI memory size 128M 0x2000000000
page_size 64K 0x10000
page_shift  16 0x10
pages 2M 0x200000
bitmap_size 256K 0x40000

[  702.050299] pci_epc_mem_alloc_addr: size 0x10000 order 0x0 pageno
0x4 virt_add 0xffffffd0047b0000 phys_addr 0x2000040000
[  702.061424] pci_epc_mem_alloc_addr: size 0x10000 order 0x0 pageno
0x5 virt_add 0xffffffd0047d0000 phys_addr 0x2000050000
[  702.203933] pci_epc_mem_alloc_addr: size 0x40010000 order 0xf
pageno 0x8000 virt_add 0xffffffd004800000 phys_addr 0x1f80000000
[  702.216547] Oops - store (or AMO) access fault [#1]
:::
[  702.310198] sstatus: 0000000200000120 sbadaddr: ffffffd004804000
scause: 0000000000000007

Regards,
Alan