| Message ID | 1381408907-11724-1-git-send-email-wangyijing@huawei.com |
|---|---|
| State | Superseded |
| Headers | show |
On Thu, 2013-10-10 at 20:41 +0800, Yijing Wang wrote: > Currently, if device power state != PCI_D0, we still initialize > device MSI/MSIX, but we won't write the MSI message to device > MSI/MSIX registers. It's weird, we don't configure MSI/MSIX > registers properly, but pci_enable_msi() or pci_enable_msix() > return success, and even these registers will never be updated later. > So I think it should return error if device power state != PCI_D0. > > --- > > v1->v2: keep __write_msi_msg() function device power state statement, > because Ben Hutchings point out this function may be called > during device in D3 state to set irq affinity. We should not > touch that. > > Signed-off-by: Yijing Wang <wangyijing@huawei.com> S-o-b needs to be above the first '---' but perhaps Bjorn can fix that up before applying. Acked-by: Ben Hutchings <bhutchings@solarflare.com> > --- > drivers/pci/msi.c | 6 +++--- > 1 files changed, 3 insertions(+), 3 deletions(-) > > diff --git a/drivers/pci/msi.c b/drivers/pci/msi.c > index d5f90d6..604265c 100644 > --- a/drivers/pci/msi.c > +++ b/drivers/pci/msi.c > @@ -831,7 +831,7 @@ int pci_enable_msi_block(struct pci_dev *dev, unsigned int nvec) > int status, maxvec; > u16 msgctl; > > - if (!dev->msi_cap) > + if (!dev->msi_cap || dev->current_state != PCI_D0) > return -EINVAL; > > pci_read_config_word(dev, dev->msi_cap + PCI_MSI_FLAGS, &msgctl); > @@ -862,7 +862,7 @@ int pci_enable_msi_block_auto(struct pci_dev *dev, unsigned int *maxvec) > int ret, nvec; > u16 msgctl; > > - if (!dev->msi_cap) > + if (!dev->msi_cap || dev->current_state != PCI_D0) > return -EINVAL; > > pci_read_config_word(dev, dev->msi_cap + PCI_MSI_FLAGS, &msgctl); > @@ -955,7 +955,7 @@ int pci_enable_msix(struct pci_dev *dev, struct msix_entry *entries, int nvec) > int status, nr_entries; > int i, j; > > - if (!entries || !dev->msix_cap) > + if (!entries || !dev->msix_cap || dev->current_state != PCI_D0) > return -EINVAL; > > status = pci_msi_check_device(dev, nvec, PCI_CAP_ID_MSIX);
On 2013/10/10 20:45, Ben Hutchings wrote: > On Thu, 2013-10-10 at 20:41 +0800, Yijing Wang wrote: >> Currently, if device power state != PCI_D0, we still initialize >> device MSI/MSIX, but we won't write the MSI message to device >> MSI/MSIX registers. It's weird, we don't configure MSI/MSIX >> registers properly, but pci_enable_msi() or pci_enable_msix() >> return success, and even these registers will never be updated later. >> So I think it should return error if device power state != PCI_D0. >> >> --- >> >> v1->v2: keep __write_msi_msg() function device power state statement, >> because Ben Hutchings point out this function may be called >> during device in D3 state to set irq affinity. We should not >> touch that. >> >> Signed-off-by: Yijing Wang <wangyijing@huawei.com> > > S-o-b needs to be above the first '---' but perhaps Bjorn can fix that > up before applying. > > Acked-by: Ben Hutchings <bhutchings@solarflare.com> Thanks very much, I will update it, so Bjorn can apply it easily. > >> --- >> drivers/pci/msi.c | 6 +++--- >> 1 files changed, 3 insertions(+), 3 deletions(-) >> >> diff --git a/drivers/pci/msi.c b/drivers/pci/msi.c >> index d5f90d6..604265c 100644 >> --- a/drivers/pci/msi.c >> +++ b/drivers/pci/msi.c >> @@ -831,7 +831,7 @@ int pci_enable_msi_block(struct pci_dev *dev, unsigned int nvec) >> int status, maxvec; >> u16 msgctl; >> >> - if (!dev->msi_cap) >> + if (!dev->msi_cap || dev->current_state != PCI_D0) >> return -EINVAL; >> >> pci_read_config_word(dev, dev->msi_cap + PCI_MSI_FLAGS, &msgctl); >> @@ -862,7 +862,7 @@ int pci_enable_msi_block_auto(struct pci_dev *dev, unsigned int *maxvec) >> int ret, nvec; >> u16 msgctl; >> >> - if (!dev->msi_cap) >> + if (!dev->msi_cap || dev->current_state != PCI_D0) >> return -EINVAL; >> >> pci_read_config_word(dev, dev->msi_cap + PCI_MSI_FLAGS, &msgctl); >> @@ -955,7 +955,7 @@ int pci_enable_msix(struct pci_dev *dev, struct msix_entry *entries, int nvec) >> int status, nr_entries; >> int i, j; >> >> - if (!entries || !dev->msix_cap) >> + if (!entries || !dev->msix_cap || dev->current_state != PCI_D0) >> return -EINVAL; >> >> status = pci_msi_check_device(dev, nvec, PCI_CAP_ID_MSIX); >
diff --git a/drivers/pci/msi.c b/drivers/pci/msi.c index d5f90d6..604265c 100644 --- a/drivers/pci/msi.c +++ b/drivers/pci/msi.c @@ -831,7 +831,7 @@ int pci_enable_msi_block(struct pci_dev *dev, unsigned int nvec) int status, maxvec; u16 msgctl; - if (!dev->msi_cap) + if (!dev->msi_cap || dev->current_state != PCI_D0) return -EINVAL; pci_read_config_word(dev, dev->msi_cap + PCI_MSI_FLAGS, &msgctl); @@ -862,7 +862,7 @@ int pci_enable_msi_block_auto(struct pci_dev *dev, unsigned int *maxvec) int ret, nvec; u16 msgctl; - if (!dev->msi_cap) + if (!dev->msi_cap || dev->current_state != PCI_D0) return -EINVAL; pci_read_config_word(dev, dev->msi_cap + PCI_MSI_FLAGS, &msgctl); @@ -955,7 +955,7 @@ int pci_enable_msix(struct pci_dev *dev, struct msix_entry *entries, int nvec) int status, nr_entries; int i, j; - if (!entries || !dev->msix_cap) + if (!entries || !dev->msix_cap || dev->current_state != PCI_D0) return -EINVAL; status = pci_msi_check_device(dev, nvec, PCI_CAP_ID_MSIX);
Currently, if device power state != PCI_D0, we still initialize device MSI/MSIX, but we won't write the MSI message to device MSI/MSIX registers. It's weird, we don't configure MSI/MSIX registers properly, but pci_enable_msi() or pci_enable_msix() return success, and even these registers will never be updated later. So I think it should return error if device power state != PCI_D0. --- v1->v2: keep __write_msi_msg() function device power state statement, because Ben Hutchings point out this function may be called during device in D3 state to set irq affinity. We should not touch that. Signed-off-by: Yijing Wang <wangyijing@huawei.com> --- drivers/pci/msi.c | 6 +++--- 1 files changed, 3 insertions(+), 3 deletions(-)