Re: [PATCH v1] hw/usb/hcd-ehci: Handle get_dwords() failures in async writeback

On 14/8/26 11:06, Peter Maydell wrote:
> On Thu, 13 Aug 2026 at 08:24, Jamin Lin <[email protected]> wrote:
>>
>> Coverity reports that ehci_writeback_async_complete_packet() ignores
>> the return value of get_dwords() when reading the QH and qTD.
>>
>> Handle read failures in the same way as QH and qTD verification
>> failures by freeing the packet and returning early.
>>
>> Signed-off-by: Jamin Lin <[email protected]>
>> ---
>>   hw/usb/hcd-ehci.c | 10 +++++-----
>>   1 file changed, 5 insertions(+), 5 deletions(-)
>>
>> diff --git a/hw/usb/hcd-ehci.c b/hw/usb/hcd-ehci.c
>> index 451a918e9f..d43951975a 100644
>> --- a/hw/usb/hcd-ehci.c
>> +++ b/hw/usb/hcd-ehci.c
>> @@ -533,11 +533,11 @@ static void ehci_writeback_async_complete_packet(EHCIPacket *p)
>>       /* Verify the qh + qtd, like we do when going through fetchqh & fetchqtd */
>>       memset(&qh, 0, sizeof(qh));
>>       memset(&qtd, 0, sizeof(qtd));
>> -    get_dwords(q->ehci, NLPTR_GET(q->qhaddr),
>> -               (uint32_t *) &qh, ehci_qh_dwords(q->ehci));
>> -    get_dwords(q->ehci, NLPTR_GET(q->qtdaddr),
>> -               (uint32_t *) &qtd, ehci_qtd_dwords(q->ehci));
>> -    if (!ehci_verify_qh(q, &qh) || !ehci_verify_qtd(p, &qtd)) {
>> +    if (get_dwords(q->ehci, NLPTR_GET(q->qhaddr),
>> +                   (uint32_t *) &qh, ehci_qh_dwords(q->ehci)) < 0 ||
>> +        get_dwords(q->ehci, NLPTR_GET(q->qtdaddr),
>> +                   (uint32_t *) &qtd, ehci_qtd_dwords(q->ehci)) < 0 ||
>> +        !ehci_verify_qh(q, &qh) || !ehci_verify_qtd(p, &qtd)) {
>>           p->async = EHCI_ASYNC_INITIALIZED;
>>           ehci_free_packet(p);
>>           return;
> 
> Looking at the coverity report, my question is whether echi->as
> can ever actually be NULL. This is the address space we use to
> do DMA, so it feels like every EHCI device must set that up
> somehow. ehci_sysbus_init() does. So does usb_ehci_pci_realize().
> usb_ehci_pci_write_config() can change it, but never to NULL.
> 
> If echi->as is always non-NULL then we could change get_dwords()
> and put_dwords() to return "void".
> 
> Alternatively, maybe get_dwords() and put_dwords() should be
> checking the return value from dma_memory_write() and
> dma_memory_read() so that they fail if the DMA fails...

Oh good point, I missed that. We really should qualify
dma_memory_write() & co with G_GNUC_WARN_UNUSED_RESULT, that'd
help us preventing such mistakes.