phosphor-mboxd/test, branch master

mbox_msg: Move handler table to struct mbox_context

2018-04-04T07:57:31+00:00

This allows us to provide alternative implementations for the handlers as necessary. The vpnor feature, which enforces the read-only property of FFS partitions, requires this for handling CREATE_WRITE_WINDOW. Change-Id: Ia969a6f085244b194c500e66b62adca5e10bacba Signed-off-by: Andrew Jeffery

test: Move vpnor tests to vpnor directory

2018-04-04T07:57:31+00:00

In the spirit of things that are together should be kept together. The repository layout now better corresponds to upstream with the exception of the vpnor directory and some modifications to Makefile.am Change-Id: I16d59a3c9ee846065f6a8c83eb4459715d525f3f Signed-off-by: Andrew Jeffery

vpnor: Isolate relevant code in vpnor directory

2018-04-04T07:57:31+00:00

This is prepatory work for introducing more vpnor-specific behaviours to window handling. We will be introducing more objects to link, in order to hook some of the window command handlers. This change takes the opportunity to revert back to the upstream names for some of the original C files. Change-Id: I6b67ae466a2695054035e65ba752881be9c32d1a Signed-off-by: Andrew Jeffery

test: vpnor: Add create_write_window_unmapped

2018-04-04T07:56:23+00:00

Deny attempts to open write windows to flash space that is unmapped in the ToC. This gives explicit feedback that any data written would not be persisted if it were possible create the write window in the first place. Change-Id: I0e7967247b122aa8d0c1de38af43162ba0ccc8fa Signed-off-by: Andrew Jeffery

test: vpnor: Add create_write_window_rw_partition

2018-04-04T07:56:23+00:00

This test case should always pass. Ensure it does in the face of modifications to how the request is processed. Change-Id: I090aa6518750615c6b931404f5ad54b13cf95e28 Signed-off-by: Andrew Jeffery

test: vpnor: Add create_write_window_ro_partition

2018-04-04T07:56:23+00:00

The virtual PNOR implementation enforces the read-only attribute of partitions out of the box. This causes trouble when the host requests a write window over a read-only partition, as the flush command will fail. Further, by design, we have open-implies-close-implies-flush semantics, which means once a flush fails, any subsequent request to open a window also fails. We want the daemon to deny attempts to open write windows over a read-only partition during the CREATE_WRITE_WINDOW request, to avoid the cascading failures later on. Change-Id: Ib6bec3d34a8a47e517088dd504f7a74641882f5d Signed-off-by: Andrew Jeffery

mboxd_windows: Shrink windows accessing the end of flash

2018-04-04T07:56:23+00:00

The host may request a window over the end of the flash where the window size combined with the requested offset exceeds the limit of the flash. This issue was introduced with the virtual PNOR, as copy_flash() now may return a size less than requested. This leads to offset requests that are still block aligned, but the windows may no longer be aligned with respect to the flash size. This issue triggers the read error reported from the Petitboot environment in an earlier commit message: / # cat /dev/mtd0 > /dev/null [ 501.061616288,3] MBOX-FLASH: Bad response code from BMC 2 [ 501.150405995,3] MBOX-FLASH: Error waiting for BMC cat: read error: Input/output error / # echo $? 1 / # With the corresponding mboxd trace on the BMC: [ 1519966031.652036815] Received MBOX command: 4 [ 1519966031.652272613] Host requested flash @ 0x03f1a000 [ 1519966031.652411603] Tried to open read window past flash limit [ 1519966031.652500088] Couldn't create window mapping for offset 0x03f1a000 [ 1519966031.652607966] Error handling mbox cmd: 4 [ 1519966031.652661421] Writing MBOX response: 2 [ 1519966031.652762229] Error handling MBOX event Instead, shrink the request such that the resulting window exactly maps the flash limit, and no further. Change-Id: Id33ae3b14252eb40240ef1925311f22aceb103b4 Signed-off-by: Andrew Jeffery

mboxd_windows: Reset evicted windows

2018-04-04T07:56:23+00:00

After copying a portion of the backing store to a window, create_map_window() "resizes" the window to the aligned-up size reported by copy_flash(). This allows use of the window size as the content size elsewhere in the codebase. However, if we needed to evict a window to satisfy a request, the window properties were not reset. This lead to inefficient use of the reserved memory by limiting the effective window size to the minimum size of all requests that were previously allocated the window in question. Inefficient use of reserved memory isn't the only side effect; the host takes an eye-watering hit to throughput that gets exponentionally worse over time: From the petitboot shell without the patch applied: / # time cat /dev/mtd0 > /dev/null real 0m 49.77s user 0m 0.00s sys 0m 49.76s / # time cat /dev/mtd0 > /dev/null real 1m 33.57s user 0m 0.00s sys 1m 33.55s / # time cat /dev/mtd0 > /dev/null real 4m 45.37s user 0m 0.00s sys 4m 45.35s / # time cat /dev/mtd0 > /dev/null real 9m 17.77s user 0m 0.00s sys 9m 17.76s / # And with the patch applied: / # time cat /dev/mtd0 > /dev/null real 0m 43.00s user 0m 0.00s sys 0m 42.99s / # time cat /dev/mtd0 > /dev/null real 0m 42.40s user 0m 0.00s sys 0m 42.39s / # time cat /dev/mtd0 > /dev/null real 0m 42.41s user 0m 0.00s sys 0m 42.39s / # Reset the properties to allow use of the entire reserved memory region allocated to the window, improving memory efficiency, throughput, and minimising throughput variance. Change-Id: I7be78ec5e0a9ee0caf31133b0861e333844b8975 Signed-off-by: Andrew Jeffery

test: vpnor: Add create_read_window_remap

2018-04-04T07:56:23+00:00

Sanity check that requesting arbitrary offsets inside a partition will map to an existing window containing that partitions data. This ensures we don't have multiple windows mapping the same content and shooting ourselves in the foot with coherency issues. Change-Id: Ie13cc36a9f092381660d5c45ed6d2477c3a4d6ce Signed-off-by: Andrew Jeffery

test: vpnor: Add create_read_window_size

2018-04-04T07:56:23+00:00

Tests to make sure that the window size returned has not been shrunk inappropriately by previous requests. Change-Id: Ib86d0744c774b5cf57235833a402bc79ef9979b9 Signed-off-by: Andrew Jeffery