| Message ID | 20200610191823.21143-1-michael@amarulasolutions.com |
|---|---|
| State | New |
| Headers | show |
| Series | [V2] nvme: Invalidate dcache before submitting admin cmd | expand |
On Thu, Jun 11, 2020 at 3:18 AM Michael Trimarchi <michael at amarulasolutions.com> wrote: > > From: Jagan Teki <jagan at amarulasolutions.com> > > This patch try to avoids eviction of dirty lines during DMA > transfer. The code right now execute the following step: > > - allocate the buffer > - start a dma operation using the non-coherent dma buffer > - invalidate cache lines associated with the buffer > - read the buffer > > This can lead to reading back not valid information, because the cache > controller could evict dirty cache lines belonging to the buffer *after* > the DMA operation has started to fill the DRAM. > In order to avoid this, a new invalidation is required *before* starting > the DMA operation. The patch just adds an invalidation before submitting > the DMA command. > > Example below shows the nvme disk scan result without the following > patch > > => nvme scan > nvme_get_info_from_identify: nn = 544502629, vwc = 100, > sn = dev_0T, mn = `?\?, fr = t_part, mdts = 105 > > So, invalidating the cache before submitting the admin command, > fix the cpu read. > > Cc: Andr? Przywara <andre.przywara at arm.com> > Reported-by: Suniel Mahesh <sunil at amarulasolutions.com> > Signed-off-by: Michael Trimarchi <michael at amarulasolutions.com> > Signed-off-by: Jagan Teki <jagan at amarulasolutions.com> > Tested-by: Suniel Mahesh <sunil at amarulasolutions.com> > --- > V1 -> V2: > Get feedback from Andr? Przywara and > Bin Meng. They help me not only to write a proper commit > message but even make in some English form. > --- > drivers/nvme/nvme.c | 3 +++ > 1 file changed, 3 insertions(+) > Reviewed-by: Bin Meng <bmeng.cn at gmail.com>
On Wed, Jun 10, 2020 at 09:18:23PM +0200, Michael Trimarchi wrote: > From: Jagan Teki <jagan at amarulasolutions.com> > > This patch try to avoids eviction of dirty lines during DMA > transfer. The code right now execute the following step: > > - allocate the buffer > - start a dma operation using the non-coherent dma buffer > - invalidate cache lines associated with the buffer > - read the buffer > > This can lead to reading back not valid information, because the cache > controller could evict dirty cache lines belonging to the buffer *after* > the DMA operation has started to fill the DRAM. > In order to avoid this, a new invalidation is required *before* starting > the DMA operation. The patch just adds an invalidation before submitting > the DMA command. > > Example below shows the nvme disk scan result without the following > patch > > => nvme scan > nvme_get_info_from_identify: nn = 544502629, vwc = 100, > sn = dev_0T, mn = `?\?, fr = t_part, mdts = 105 > > So, invalidating the cache before submitting the admin command, > fix the cpu read. > > Cc: Andr? Przywara <andre.przywara at arm.com> > Reported-by: Suniel Mahesh <sunil at amarulasolutions.com> > Signed-off-by: Michael Trimarchi <michael at amarulasolutions.com> > Signed-off-by: Jagan Teki <jagan at amarulasolutions.com> > Tested-by: Suniel Mahesh <sunil at amarulasolutions.com> > Reviewed-by: Bin Meng <bmeng.cn at gmail.com> Applied (back on June 23rd) to u-boot/master, thanks!
diff --git a/drivers/nvme/nvme.c b/drivers/nvme/nvme.c index 0357aba7f1..fc64d93ab8 100644 --- a/drivers/nvme/nvme.c +++ b/drivers/nvme/nvme.c @@ -466,6 +466,9 @@ int nvme_identify(struct nvme_dev *dev, unsigned nsid, c.identify.cns = cpu_to_le32(cns); + invalidate_dcache_range(dma_addr, + dma_addr + sizeof(struct nvme_id_ctrl)); + ret = nvme_submit_admin_cmd(dev, &c, NULL); if (!ret) invalidate_dcache_range(dma_addr,