On Fri, Jan 17, 2025 at 10:09:03AM +0100, Christian Bruel wrote:
> Fixes an oops found while testing the stm32_pcie ep driver with handling
> of PERST# deassertion:
>
> [ 92.154549] ------------[ cut here ]------------
> [ 92.159093] Trying to vunmap() nonexistent vm area (0000000031e0f06f)
> ...
> [ 92.288763] vunmap+0x58/0x60 (P)
> [ 92.292096] dma_direct_free+0x88/0x18c
> [ 92.295932] dma_free_attrs+0x84/0xf8
> [ 92.299664] pci_epf_free_space+0x48/0x78
> [ 92.303698] pci_epf_test_epc_init+0x184/0x3c0 [pci_epf_test]
> [ 92.309446] pci_epc_init_notify+0x70/0xb4
> [ 92.313578] stm32_pcie_ep_perst_irq_thread+0xf8/0x24c
> ...
Personally, I would omit the stack trace.
>
> During EP initialization, pci_epf_test_alloc_space allocates all BARs,
> which are further freed if epc_set_bar fails (for instance, due to
> no free inbound window).
>
> However, when pci_epc_set_bar fails, the error path:
> pci_epc_set_bar -> pci_epf_free_space
> does not reset epf_test->reg[bar].
>
> Then, if the host reboots, PERST# deassertion restarts the BAR allocation
> sequence with the same allocation failure (no free inbound window).
>
> So, two subsequent calls to the sequence:
>
> if (!epf_test->reg[bar])
> continue;
>
> ret = pci_epc_set_bar(epc, epf->func_no, epf->vfunc_no, &epf->bar[bar]);
> if (ret) {
> pci_epf_free_space(epf, epf_test->reg[bar], bar, PRIMARY_INTERFACE);
> }
Personally, I would omit the copy pasted code.
>
> create a double free situation since epf_test->reg[bar] was deallocated
> and is still non-NULL.
>
> This patch makes pci_epf_alloc_space/pci_epf_free_space symmetric
> by resetting epf_test->reg[bar] when memory is deallocated.
>
> Signed-off-by: Christian Bruel <christian.bruel@xxxxxxxxxxx>
> ---
Reviewed-by: Niklas Cassel <cassel@xxxxxxxxxx>
